name of the person

timer Asked: Nov 19th, 2016

Question description

I need to a report based on these 6 papers below

It should be of 25 pages.

Computers and Electrical Engineering 44 (2015) 137–152 Contents lists available at ScienceDirect Computers and Electrical Engineering journal homepage: Bluetooth for Internet of Things: A fuzzy approach to improve power management in smart homes q M. Collotta ⇑, G. Pau Kore University of Enna, Faculty of Engineering and Architecture, Cittadella Universitaria, 94100 Enna, Italy a r t i c l e i n f o Article history: Received 14 June 2014 Received in revised form 7 January 2015 Accepted 9 January 2015 Available online 30 January 2015 Keywords: Power consumption Fuzzy logic controller Bluetooth Low Energy Internet of Things a b s t r a c t Thanks to the introduction of the Internet of Things (IoT), the research and the implementation of home automation are getting more popular because the IoT holds promise for making homes smarter through wireless technologies. There is a main requirement that make a wireless protocol ideal for use in the IoT, that is the energy efficiency. Bluetooth Low Energy (BLE) has a high potential in becoming an important technology for the IoT in low power, low cost, small devices. However, specific techniques can be used in such a way as to further reduce the energy consumption of BLE. To this end, this paper proposes a fuzzy logic based mechanism that determine the sleeping time of field devices in a home automation environment based on BLE. The proposed FLC determines the sleeping time of field devices according to the battery level and to the ratio of Throughput to Workload (Th/Wl). Simulation results reveal that using the proposed approach the device lifetime is increased by 30% with respect to the use of fixed sleeping time. Ó 2015 Elsevier Ltd. All rights reserved. 1. Introduction With the increasingly rapid development of various communication technologies, more and more devices are able to access the internet and to interact with it. When considering a global network of smart objects of all kinds, such as computers, appliances, clothes, sensors, interacting with each other through Internet protocols, the reference scenario is called Internet of Things (IoT). The devices that are part of the network of objects are called ‘‘smart objects’’ or ‘‘smart things’’, that unlike normal devices are able to interact within the communication system in which they are inserted since they have an active role. The devices can be identified by the following characteristics:  They are the real objects characterized by cost, shape, weight, etc.  They have limited resources in terms of processing capacity, memory, energy supply and routing [1].  They may be influenced and influence the surrounding environment by acting as actuators [2]. Thanks to the developments of wireless technologies and to studies about IoT, the ‘‘anywhere, any-time by anything’’ communication is no longer considered a true utopia. In fact, more and more devices, at any-time, even without receiving a physical input, can access the network and interact with the other connected devices [3]. The practical significance of q Reviews processed and recommended for publication to the Editor-in-Chief by Associate Editor Dr. Maurizio Palesi. ⇑ Corresponding author. E-mail addresses: (M. Collotta), (G. Pau). 0045-7906/Ó 2015 Elsevier Ltd. All rights reserved. 138 M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 the IoT is made possible through enabling technologies such as Wireless Sensor Networks (WSNs), mainly used for sensing operations. The nodes of a WSN are sensors arranged within an environment, with the aim to detect certain data that are send to one central node in order to process them. With the introduction of the IoT, the research and the implementation of home automation are getting more popular [4]. In fact, built on connections between digital devices and nearly anything that can be monitored or controlled electronically, the IoT holds promise for making homes smarter. Home Automation (HA) refers to the mechanization and automatic control of various residential activities. Typically, HA provides a centralized control of electrical appliances, such as air conditioners, lighting and security systems and even the home theater. Adding intelligence to home environment it would be possible to obtain excellent levels of comfort, and another feature taken into account is the energy savings. Moreover, the integration of several electrical devices in the household is an open challenge because of the absence of a cheap and standardized communication protocol between them [5]. Anyhow, at present, a wireless network exists in almost all homes. Smart-phones and tablets are natural devices to enable the control of electrical ones. In such a situation, the wireless protocols become an easy avenue for self-installation of HA systems. Moreover, a smart home has to meet several requirements such as:  Safety, i.e. the protection from possible malfunctions.  Security, i.e. authentication, authorization and data protection.  Energy saving, i.e. a smart mechanism to reduce power consumption. Several wireless technologies, such as Bluetooth Low Energy (BLE) [6], IEEE 802.15.4/ZigBee [7] and IEEE 802.11/Wi-Fi [8], that can support the remote data transfer, the sensing and the control, have been proposed in order to embed various levels of intelligence for smart home. BLE is being adopted by the health care industry for portable medical and lifestyle devices. On the other hand, the battle between ZigBee and low-power Wi-Fi technologies for home control and automation has just begun. The manufacturers of wireless devices are urgently looking for new revenue streams, and machine-to-machine communication and location-based services seem to be good places to make a bet. Both can use existing infrastructure and are very much a part of the emerging IoT market. Anyhow, there is a main requirement that make a wireless protocol ideal for use in the IoT, that is the energy efficiency. In many cases, the sensing nodes are battery-powered, so a low-power feature is a basic requirement. In the design of devices that implement a wireless protocol several mechanism can be adopted in order to reduce the power consumption, including low-leakage process technologies, best-in-class low-power non-volatile memory/flash memory technologies, architectural innovations and various clocking schemes. For battery-powered nodes, all of those techniques are needed in order to achieve the lowest possible power consumption. To cope with this problem, this paper proposes a fuzzy logic based mechanism in a home automation environment. The core of the proposed architecture is represented by a wireless network, organized in Wireless Automation Cells (WACs), based on Bluetooth Low Energy Protocol and managed by a Master node. The Bluetooth Low Energy protocol has been chosen from the results of an analysis carried out in the Section 2.2 of this paper. In each WAC there is a fuzzy module that aims at the energy saving of the network. In fact, the goal is to improve the low energy consumption of BLE through a fuzzy logic controller. The paper is organized as follows. Section 2 deals with the BLE support for IoT in home automation environments, while Section 3 shows main related works in order to deduce the innovations introduced with this work. Section 4 describes the system architecture and the proposed approach, showing the fuzzy logic controller module. Section 5 shows the performance obtained by the proposed approach and finally Section 6 concludes the paper and outlines some hints for future work. 2. Bluetooth Low Energy support for IoT in home automation In this section an analysis of wireless communication suitable for energy management in home automation applications is presented. In recent years, wireless communication technology has seen sudden growth and several approaches are also capable to support the energy management through widespread environmental sensing. This has been possible through the advancements of low-power and low-cost radio frequency wireless communication technologies, with smaller form factor, greater sensing density and longer functionalities lifetime. In the IEEE 802.11/Wi-Fi family, the nodes compete for the medium access according to the Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) protocol. However, contention-based approaches are not able to guarantee an upper bound on the medium access delay, so they are not adequate for time-constrained traffic and for applications that require low power consumption. In fact, Wi-Fi does not possess intrinsically ubiquitous characteristics, but the explosion of laptops and mobile devices has driven widespread adoption. However, in the field of energy management this adoption seems unlikely. In fact, the native design of Wi-Fi does not have the energy management in mind and for many applications, particularly environmental sensing, the IEEE 802.11 is considered too power hungry and in some cases the components are still too large. For this reason, other approaches have been developed. ZigBee, a pure wireless technology, is based upon the IEEE 802.15.4 standard for WPANs (Wireless Personal Area Networks). It is intended to be a low-cost, low-power wireless mesh network standard allowing for secure communication with a data rate of up to 250 kbps. ZigBee was originally conceived as an alternative to Wi-Fi/Bluetooth Classic around 1998 as a M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 139 solution for self-organizing ad hoc digital networks. ZigBee-based HA solutions operate in the 2.4 GHz range and have a range of 70 m indoors and up to 400 m outdoors. The networks are self organizing, and any member can be the controlling node. There are plenty of ZigBee home automation solutions already available in the market and also in the literature applications in industrial contexts have been proposed [9]. Technologies such as Bluetooth Low Energy (BLE) came up in part due to the unsuitability of Wi-Fi, ZigBee and other such technologies in order to provide networking with low power consumption, low cost, simplicity [10] and the ability to remain in a suspended state for extended periods of time [11]. Most of the HA technologies have been developed when Bluetooth Classic was unable to satisfy all of these criteria. However, over the last few years, thanks to the new specifications of Bluetooth 4.0, the BLE has slowly, but surely, gained a significant foothold in the home automation industry. The rise of BLE’s role in home automation has primarily come about due to two factors.  Networked nature of today’s electronics: commonly used electronic devices such as televisions and AV receivers have started becoming part of the home IP network. Even though the networking aspects were added to bring in support for on-line streaming services or network playback, it became an easy value-add for the manufacturers to expose control of the device functionality over the network.  Rise of mobile computing devices: the increasing rate of adoption of smart-phones and tablets was an added boon. These devices ensure that consumers have access to a portable ‘‘controller’’ for the electronics connected to the network. 2.1. Bluetooth Low Energy protocol Classic Bluetooth is used for short range wireless communication among devices in networks where nodes can easily come and go. It uses 79 channels with a bandwidth of 1 MHz on the 2.4 GHz ISM band with a pseudo-random frequency hopping sequence [12]. In a Piconet each Master device establishes the frequency hopping sequence and can have up to 7 Slave connections. A device can be in more than a single Piconet and overlapping Piconets are Scatternets. However, in the literature, several optimization approaches have been proposed in order to improve Classic Bluetooth, especially to make it usable even in industrial contexts [13,14]. Bluetooth 4.0, or BLE, implements an entirely new protocol stack along with new profiles and applications. Its core objective is to run for a very long time on a coin-cell battery. It also enables devices to connect to the internet, where traditionally they have not been able to, in an efficient way through its client/server architecture. BLE is designed to be easy to develop for at a cheap price. Bluetooth Low Energy operates in the 2.4 GHz ISM band with only 40 channels spaced 2 MHz apart (Fig. 1). It is capable of transmitting at a rate of 1Mbit/s using GFSK modulation. Like Bluetooth Classic, it uses frequency hopping, but it uses adaptive frequency hopping and at a slower rate. BLE uses 3 of the 40 channels to advertise which allow for device discovery. After a device is discovered and connected the remaining 37 channels are used to transmit data. There are 4 basic modes through which a BLE device can operate, that are master device mode, slave device mode, advertising mode, and scanning mode. The advertising mode is used by the device to periodically advertise information that can be used to establish a link. It can also use this mode in order to respond to additional queries that another device might make. The scanning mode is used to capture advertise packets. Slave and Master Modes are used once a link has been established between 2 devices, and their primary functions are to allow the devices to read, to write and to query each other. The device that starts out in advertise mode will assume the slave device mode and conversely the device that is initially scanning mode will assume the master device mode. Regarding to the packet format, there are 2 types of packets, Data and Advertise, each with variable lengths (Fig. 2). BLE Data packets consist of an 8 bit preamble, 32 bit access codes that are defined by the radio frequency channel used, a variable protocol data unit (PDU) ranging from 2 to 39 bytes and a 24 bit CRC. This means that the shortest packet can be as small as 80 bits or as long as 376 bits. It also means that a transmission time can range from 80 ls to 0.3 ms. Advertise packets on the other hand, have protocol data unit containing a 16 bit header and up to 31 bytes of data. There are three ways through which two devices can associate, Just Works, Out of Band, and Passkey Entry. An advertising device transmits packets on the advertise channels with a PDU containing the device address and up to 31 bytes of additional information. A scanning device is able to see the address and depending on the advertiser, additional information may be Fig. 1. Bluetooth Low Energy radio links: channels in green are used for advertising. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) 140 M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 Fig. 2. Bluetooth Low Energy data packet format. sent upon request. This means that a good amount of information can be obtained about the device without even establishing a connection. Advertising is done sequentially on all available channels at a rate from 20 ms to every 10 s depending on configuration. On the other hand a scanner device is configured with a scan window and a scan interval. Once a connection is made, the scanner will supply the advertiser with 2 critical pieces of information, that are the connection interval and the slave latency. The connection interval is used to determine the start time of connection events. A connection event is the exchange sequence of data packets. The other parameter, the slave latency, is the amount of connection intervals that a slave can ignore without losing the connection. This is done in order to optimize the power consumption. After a link is established the communication is carried out over the 37 channels. The PDU’s have up to 37 bytes of payload, along with a packet header, and a Message Integrity Check of 4 bytes. A communication event is initiated by a Master device, alternating between master/slave until one of them stops the transmission. The BLE protocol stack (Fig. 3) is partitioned into a Controller and a Host. The Controller handles the lower layers of the stack responsible for capturing physical packets and the radio frequency used by the radio. The Host handles the upper layers of the stack that include the application, the attribute protocol, and the L2CAP. The Host and Controller can be either collocated or the Host can run in the application processor with the application. In the second option, a host Controller Interface (HCI) is used by the Controller and the Host to communicate. The BLE protocol stack consists of:  Controller: the link layer controller captures the physical packets in the air band; it also manages the timing and the queue of incoming and outgoing packets. In short, it is responsible for the physical level data flow. This component can also be used as a firewall to the device by filtering packets from specific devices. Fig. 3. Bluetooth Low Energy protocol stack. M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 141  L2CAP: the controller communicates with the Logical Link Control and Adaptation Layer (L2CAP) protocol via the HCI or directly if collocated. The main function of this component is to provide data services to the upper level layers and for multiplexing and segmenting packets into fragments for the controller. Conversely, it reassembles packets from the controller before they are routed to the upper level layers.  GAP: the Generic Access Profile (GAP) is responsible for defining generic procedures that are used in the pairing and linking of the device. It is the interface for the application layer that implement different Bluetooth modes (Advertising, Scanning, etc.).  SM: the Security Manager (SM) is used for the authentication and the encryption. It uses AES-128 bit encryption engine to do so and is also responsible for pairing and key distribution. This component is used by the Master device to ease the computing demands of security on the Slave device.  ATT: the Attribute protocol is a communication method designed in order to optimize the transmission of small packets. ATT are pairs of attributes and values that can be used to read, to write, or to discover other devices.  GATT: the Generic Attribute Profile (GATT) is responsible for describing the different service frameworks and is an extension to ATT that is specific to Bluetooth LE 4.0. It interfaces with the application layer through the application profiles. Each application profile defines data formatting and how it should be interpreted by the application. The profiles improve on power efficiency by reducing the amount of data being exchanged. They are designed for specific functionality, for example there is a Heart Rate profile, a Glucose profile, and an Alert Notification profile among dozens and dozens of others. This makes it easy for developers to create applications aimed at specific functionality using the predefined attributes/value pairs found in each profile. Regarding to the power consumption, it is necessary to note that Bluetooth 4.0 achieves an increased power efficiency. First, it uses a lower duty cycle, this means it goes to sleep for longer periods of time and wakes up less frequently to send or receive packets. Second, using the GATT profiles BLE is able to send smaller data packets in short bursts in order to save power. The data transmission can be triggered by a local event and is available for a client to access at any time. Lastly, BLE does not maintain links with devices whenever are not communicating. Whereby, the device goes to sleep and ends the link once the exchange is complete. A link is re-established rapidly upon the next communication exchange. The transmit time of Bluetooth Classic is 100 ms whereas that of BLE is 3 ms. Moreover, the current consumption peak of BLE is 10 mA less than Bluetooth Classic and in best case scenarios has a power consumption that is 1/100th than that of Bluetooth Classic. BLE 4.0 is not designed to stream large amounts of data; it is designed to periodically send short bursts of data. There are 2 types of BLE 4.0 devices, dual mode which is backwards compatible with previous BT versions, and single mode which only supports BLE 4.0. Dual mode devices that perform high data rate streaming do not benefit from the low power consumption of BLE 4.0, which is only accomplished when BLE low data rate mode is used. 2.2. Wireless protocols comparison For IoT, a required feature of the chosen short-range technology is support for mobile use cases where a smart-phone or other mobile device can be used as a temporary gateway. Some of the important features when selecting the appropriate short-range wireless technology for IoT use cases are the following:  Cost of the radio technology: as many of the devices (sensors) are small low-cost devices, the radio must not add too much additional cost to the bill-of-material. This also implies that the radio and device application in many cases need to share the same computing engine (micro-controller).  Power consumption: many use cases require battery or some kind of energy harvesting technology as a power source.  Ease-of-use: it must be easy to associate a device to the network and to the Internet services.  Security: the authentication and the encryption must be adequately supported by the wireless technology and sometimes end-to-security (all the way from sensors to the Web services) is required.  Available ecosystem: possibility to connect to smart-phones, tablets, PCs, home gateways, etc. is important. This requirement also drives volumes and has an important impact on the cost (a good example is Bluetooth Classic where the large volumes of phones and phone accessories have lowered the costs).  Range: it is necessary the capability to cover an enough range or have some capabilities to extend the coverage (repeaters, routers, etc.) without having too big impacts on the system cost. Table 1 shows how different wireless technologies fit specific verticals. It is necessary to note that the green check-mark indicates that the protocol behaves well in the reference application, unlike the orange color that indicates a just satisfactory behavior. On the contrary, the red cross indicates that the protocol is not suitable for that application. NFC can be ruled out except for very specific use cases/verticals. IEEE 802.15.4 based technologies will become a niche technology especially in those areas in which it is already used such as home and building automation and smart energy. According to Table 2 (currently most commonly wireless solutions used in home automation contexts) the following conclusions can be deduced: 142 M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 Table 1 Technologies and verticals. BLE IEEE 802.15.4 IEEE 802.11 NFC Remote control Security Health and fitness Home and building Industrial Positioning Payment Automotive Table 2 Comparison of wireless technologies and their usefulness in IoT. BLE IEEE 802.15.4 IEEE 802.11 Cost Security Power consumption Ecosystem Reliability Ease of use Range  All three technologies have built-in link layer authentication and encryption which sometimes needs to be completed with end-to-end security from the sensor to the Web application. Some IoT use cases may be fully behind an enterprise firewall (e.g. a use case inside a factory where the IoT Internet Service runs on a local server). There are also IoT systems operating on a wide-area network but acting as local networks by the use VPN tunnels or similar security mechanisms.  Correctly used, Bluetooth Low Energy has the potential for less power consumption than IEEE 802.15.4 (less overhead).  The lack of native support for IEEE 802.15.4 in mobile devices (smart-phones, tablets, laptops, etc.) is a problem especially for mobile or temporarily mobile use cases.  The ecosystem with phones, tablets, laptops and phone accessories will drive down the cost for BLE.  IEEE 802.15.4 has a main advantage in its range since many IEEE 802.15.4 based technologies (e.g. ZigBee) support mesh whereby coverage can be extended by using routers.  Bluetooth Low Energy is very reliable with its support for Adaptive Frequency Hopping (AFH) and other features inherited from Bluetooth Classic.  IEEE 802.11/Wi-Fi can be used in devices with less demands on low power consumption and as a wireless backbone in combination with other technologies. Bluetooth Low Energy has a high potential in becoming an important technology for the Internet of Things in low power, low cost, small devices. However, there are still use cases where IEEE 802.15.4 based technologies are used especially in areas where it is already established. In spite of its installed base in smart energy, home and building automation applications, IEEE 802.15.4 faces competition in BLE in these applications as well. Anyhow, Bluetooth Low Energy has a lower energy consumption than IEEE 802.15.4 and for this reason may be the best choice in applications of home automation. IEEE 802.11 is used in devices where cost, low power is less important and as a wireless backbone combined with the other wireless technologies. The analysis carried out in this Section clearly explains why in this work BLE has been chosen for a home automation application. As mentioned above, one of the biggest advantages of BLE is precisely its low power consumption. In any case, specific techniques can be used in such a way as to further reduce the energy consumption of BLE. 3. Energy saving approaches The energy saving is one of the main reasons for the emergence of smart home automation concept. Most wireless autonomous devices are usually battery-powered. Therefore it is essential to manage the smart devices to best utilize the scarce power resources over long time. In the literature adequate research works that describes the application of a fuzzy logic controller in order to reduce the power consumption over Bluetooth Low Energy wireless networks in the new smart homes introduced by the Internet of Things are missing. For this reason, in the following subsections the research works that analyse the energy savings in BLE applications and fuzzy approaches in order to reduce the power consumption will be analyzed separately. M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 143 3.1. Power consumption in Bluetooth Low Energy The authors of [15] measure and model the energy consumption through, real device, of Bluetooth Low Energy protocol and compare it to the one of IEEE 802.15.4/ZigBee protocol. The results show that when compared to ZigBee the BLE is very energy efficient in terms of number of bytes transferred per Joule spent. Moreover, the authors analyse the energy overhead of IPv6-based communication over BLE, which is relevant for IoT scenarios, and the results show that it remains reasonable. All these results show that the energy efficiency of BLE could be further improved by allowing more packets to be sent within a connection event and by implementing a frequency hopping in order to prevent interference. In [16] the power consumption of ANT, ZigBee, and Bluetooth Low Energy protocols is analyzed. It is highlighted that it is not straightforward to predict the exact power consumption in a cyclic sleep scenario from the data sheet alone. For this reason the actual power consumption is determined by combination of interacting factors, not just the average receive, transmit, and sleep currents and data rate typically given in the data sheet. In fact, the parameters that affect the power consumption are not the active or sleep currents but rather the time required to reconnect after a sleep cycle. The results of the analysis show that Bluetooth Low Energy achieve the lowest power consumption, followed by ZigBee and ANT. Anyhow, the power consumption of tested protocols might change depending on other factors such as packet size variations, transmitter and receiver distance, and hub parameters. A quantitative analysis for assessing the energy performance of BLE advertiser device is presented in [17]. In fact, although there are some prior arts focusing on BLE energy performance, it still lacks a thorough study on the important aspect of device discovery. Such energy cost, introduced by intermittent scanning or connection set-up, could seriously affect the battery endurance ability of the devices. The authors propose an accurate mathematical model for the device discovery dynamics, and derive the performance for the advertiser under the condition of various parameter settings. This mathematical model allows to study in details the system behavior in different parameters and to investigate the potential performance trade-off between achievable energy and accessing latency. In [18] a design of an ultra-low power and highly-integrated portable health monitoring system, based on Bluetooth Low Energy, capable of measuring a subject’s ECG, respiration, and body temperature is presented. Compared to a former design using MSP430 MCU and Bluetooth 2.1 proposed by the same authors, the novel design can save as much as 75% power consumption. For this reason, the paper analyses the power consumption in detail and tests the performance of the proposed approach in a real device. The results show that the proposed design implemented in a device based on BLE make it suitable to monitor the ECG, the respiration and the body temperature. The authors of [19] provide a performance evaluation BLE technology and explore its potential applications. The analysis notes that In BLE there is a trade-off between energy consumption, latency, piconet size, and throughput that mainly depends on the connInterval and connSlaveLatency parameters. Evaluation results show how these parameters can be tuned wisely in order to meet the application requirements, such as power consumption. In conclusion, it is clear that Bluetooth Low Energy emerges as a strong low-power wireless technology for single-hop communication use cases. In fact, BLE allows the connection of a large amount of new devices to the applications of Internet of Things, such as smart homes. 3.2. Fuzzy approaches for energy saving The concept of smart home energy management involves the integration of various appliances with a smart control unit capable of bidirectional wire-line or wireless communication between the controller and the utility. One problem that arises with these features is various compatibility issues between the different appliances, various smart controllers, and communication protocols. Fuzzy logic controllers are becoming increasingly popular in everyday life. Home automation, health, industry and Intelligent Transportation System are some examples of application fields. Many studies, conducted in different scenarios in which fuzzy logic controllers are applied, focused on power consumption. An automated energy management system is presented in [20]. The proposed system is composed of a fuzzy controller and by an intelligent lookup table. Through its membership functions, the fuzzy controller evaluates the appropriate outputs for the intelligent lookup table (a neural network) that maps inputs to desired outputs. Simulation results show that the proposed automated energy management system is able to find the best energy efficiency scenario in different situations. The author of [21] propose a smart energy control for house energy consumption with maximizing the use of solar energy and reducing the impact on the power grid while satisfying the energy demand of house appliances. A fuzzy-based energy management control is proposed in order to reduce the consumed energy of the building while respecting a fixed comfort. The results demonstrate that the proposed energy management system controller provides a better strategy compared to the conventional method for cost saving (almost 20%). A fuzzy logic system is presented in [22] in order to save battery life of wireless sensor network nodes and to have an efficient, robust and cost effective sensing network that can monitor events of interest, for example those of a home automation environment. The proposed fuzzy logic system helps efficiently to decide the situation of ON/OFF state for active/ sleep mode of the processing and communication parts of the sensor node. Simulation results show that the proposed system is energy efficient and have property of liveness, soundness, without any deadlock state during execution. 144 M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 In [23] fuzzy logic is used to reduce the energy consumption of Wireless Sensor Network (WSN) nodes. The lifetime of the wireless sensor networks has always been a crucial challenge. Most of the energy is consumed in data transmission to the sink. Authors propose an approach in which the movement in a circular path of sink node is governed by fuzzy logic based mechanism. Simulation results show that wireless sensor network having base station moving on circular path gives more lifetime than that of stationary base station case. A fuzzy logic controller which adapts the MAC protocol parameters by employing local node inputs such as battery power and average packet traffic is presented in [24]. MAC parameters are selected individually by each node according to local information and based on two fuzzy descriptors: energy and traffic. In order to evaluate the proposed approach the authors perform analytical computation of power consumption and compared it with a constant sleeping period approach, showing that the fuzzy selector improves the life of the network by reducing the power consumption of nodes that are close to the base station and nodes that have low battery charge. The analysis of this research works shows that fuzzy logic can be applied in order to reduce the power consumption in several contexts. Therefore, although BLE has a low power consumption, it is clear that a fuzzy mechanism could be introduced in order to further improve the power consumption in home automation environments. 4. System architecture and requirements Home automation environments are characterized by applications in which small embedded devices, like sensors and actuators, spend most of their time in a sleep state and wake up with a given periodicity or when a critical event occurs. Several methodologies and techniques, such as networking intelligent computational devices, data/resources/services management modules, and so on, have to be merged together in order to design such environments. 4.1. System requirements Several requirements must be met in order to use wireless networks in home automation environments. System requirements that drive the design of the proposed architecture can be summarized as follow:  The system must be capable to use wireless protocols in home automation environments that are increasingly characterized by noise and multiple propagation behavior.  Reduce non-determinism in wireless communication: the system should reduce the sources of non-determinism as much as possible on the wireless channel and on the wired medium.  The system must support multiple wireless cells. In fact, when it is necessary to cover a large area, multiple wireless connection points are required in order to provide coverage.  Interoperability and interchangeability: the devices developed by different manufacturers shall interoperate with each other within the architecture. Moreover, It must be possible to replace a component from one manufacturer by an equivalent one from other manufacturers.  Fault tolerance: the architecture must prevent performance degradation in case of fault to any part of the system. It is necessary to conduct a fault analysis of the system performance in order to provide fault tolerance mechanisms. Moreover, the system must be self-healing, thus it must be able to detects communication errors and fix them by its own means.  Communication reliability: the system must provide reliability in terms of communication services. Whereby, the message error rate must be kept acceptable for the automation application.  Resources allocation: this mechanism should be provided for communications between coordinators and end nodes in order to grant several system resources like bandwidth, reaction time etc.  Quality of Service (QoS) provisioning: the system must implement advanced QoS mechanisms and a clear policy to ensure guaranteed performance for predefined processes. The system must provide high QoS degree for all kind of operations involved in the system.  Differentiated QoS classes: as end-users require multiple types of traffic with different constraints within the same network, the system should support different QoS class for the different classes. 4.2. The proposed architecture In order to satisfy the above mentioned requirements and to realize the functionalities here described, the proposed architecture is composed by several independent Wireless Automation Cells (WACs) (Fig. 4), managed by a BLE Master device. The wireless network for home automation, based on BLE, is composed by all the WACs within which there are the Field Devices (FDs), that are Bluetooth Low Energy devices dealing with a specific task. In fact, the proposed wireless home automation network enables a variety of use cases, for this reason a non-exhaustive list of examples is provided below: M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 145 Fig. 4. Home automation BLE architecture.  Light control: lights can be controlled from any switch in order to reduce the need of new wired connections. Lights can also be activated in response to a command from a remote control. Moreover, they can be turned on automatically when presence and luminance sensors detect that people are in a poorly illuminated room.  Smart energy: window shades, HVAC, central heating, and so on may be controlled depending on the information collected by several types of sensors that monitor parameters such as temperature, humidity, light, and presence. Whereby, unnecessary waste of energy can thus be avoided. Furthermore, smart utility meters can be used to detect usage peaks and alert the household devices that may be causing them.  Security and safety: advanced security systems can be based on several sensors (e.g., smoke detectors, glass-break sensors, and motion sensors) in order to detect possible risk situations that trigger appropriate actions in response. For example, smoke detectors may activate fire alarms.  Remote control: over the years, infrared technology has been used for wireless communication between a remote control and devices such as TVs, Hi-Fi equipment, and heating, ventilating, and air conditioning systems. However, infrared requires line-of-sight and short-distance communication. Radio frequency technology overcomes these limitations. Some of the WACs of the proposed architecture behave as WSNs and then they can continuously monitor environments with less human effort and are low cost and low power. Through wireless communication, the Master node receives and processes data detected by FDs placed inside the WAC. It also allows sending user commands or system commands to FD nodes. Moreover, wireless links allow the communication among mobile devices (such as smart-phones or tablets) and the WACs. Whereby, people can authenticate themselves inside the home automation system and subsequently monitor data detected by sensor nodes and, in case of need, send commands. The range of BLE radio may be optimized according to application. The majority of Bluetooth devices on the market today include the basic 30 foot, or 10 m, range of the Bluetooth Classic radio, but there is no limit imposed by the specification. With BLE, manufacturers may choose to optimize range to 200 feet (about 67 m) and beyond, particularly in home sensor applications where longer range is a necessity. Anyhow, it is necessary to take into account the energy consumption of the devices. For this reason, the Energy Saving Fuzzy Controller is necessary in order to ensure the power consumption management. This module dynamically manages sampling times in order to prolong sleeping periods of Field Devices. In this way, it is possible to improve energy savings and, at the same time, prolong batteries and the network life-cycle. 4.3. Energy Saving Fuzzy Controller In home automation applications the end nodes continuously work inside the monitored environment. For this reason it is necessary to develop an energy management paradigm to ensure network flexibility, adaptability and scalability. Moreover, this mechanism must be able to optimize power resources and, at the same time, increase the life-cycle of devices. The behavior of the proposed wireless network is similar to a WSN scenario, where it is not possible to determine ‘‘a priori’’ nodes behavior since they are often used to monitor sporadic events. However, traffic flows generated by WSNs can be considered as periodic [25]. A centralized mechanism is proposed in order to regulate the sleeping time of the field devices in a Bluetooth Low Energy network, with the aim of reducing their power consumption. Each device sends information about its operating conditions to 146 M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 Fig. 5. Architecture of the proposed FLC. Fig. 6. Periodic packets count. the Master node when its sleeping time is expired. The Master node is a special device properly equipped for executing computational tasks. In the proposed approach the Master node uses a fuzzy logic controller (FLC) in order to calculate the new values of the sleeping time of each FD, as shown in Fig. 5. The FLC determines the sleeping time of the FD according to the battery level and to the ratio of Throughput to Workload (Th/Wl). The throughput is the sum of both periodic and aperiodic packets sent by the device. The workload is the total number of packets that the device has to send. Since the number of aperiodic packets can not be known a priori and, as a consequence, can be considered as a random variable, the calculation of the number of periodic packets (P periodic ) can be done by increasing a counter when the following equation is satisfied: $ Pperiodic % T end 1 X ¼ i Tsi i¼T ð1Þ start where T start and T end represent the initial and final instants of the sleep phase, respectively. Tsi is the sampling time of the i-th node that coincides with the packet emission time. Let’s consider an example, reported in Fig. 6, in which T start ¼ 5; T end ¼ 8 and Tsi ¼ 2. Eq. (1) is satisfied two times, i.e. for T ¼ 6 and T ¼ 8. According to this example, the Th/Wl parameter is calculated as: Th=Wl ¼ 1þk 2þk ð2Þ where k is a random value that refers to the number of aperiodic packets to send, 1 refers to the last periodic packet that the node has to transmit while 2 are the number of periodic packets that fall within the sleep time window. The proposed FLC considers three triangular membership functions (Low, Medium, High) for each input variable. These functions fuzzify the crisp inputs, while the ranges of which are:  Th/Wl: [0, 100];  battery_level: [0, 1024], where 0 represents the lowest level of battery while 1024 represents the highest level of battery and it is the maximum value at the output of a 10-bit AD converter (with an appropriate electronic signal conditioning circuit). In the same way, three triangular membership functions (Low, Medium, High) are defined for the sleeping time. In this case, the range of the crisp values of this output variable is:  sleeping time : ½0; 10  sampling time (sec). where the sampling_time value is a constant value defined at design time for each field device. Fuzzy membership functions of the Th/Wl, the battery_level and the sleeping_time are depicted in Fig. 7, where the degree of membership is represented by normalized values [0–1]. As shown in Table 3, the output value is determined through 9 fuzzy rules based on the IF-THEN statement of classic programming languages. For example, considering rule 1, if Th/Wl is Low and battery level is Low, sleeping time will be Medium. Finally, the output value is defuzzified using the centroid mechanism: M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 147 Fig. 7. Membership functions of input and output parameters for the FLC. Table 3 FLC inference rules - A = Antecedent, C = Consequent. Pn sleeping time ¼ Rule A (Th/Wl) A (battery_level) C (sleeping_time) 1 2 3 4 5 6 7 8 9 Low Low Low Medium Medium Medium High High High Low Medium High Low High Medium Low High Medium Medium Low Low Medium Low Medium High High High Outi  C i Pn i¼1 C i i¼1 ð3Þ where Out i is the output of rule base i, and C i is the center of the output membership function. 5. Performance evaluation 5.1. Simulation model description The approach proposed in Section 4.3 can be applied regardless of the network topology. As depicted in Fig. 8, a star topology of a generic WAC has been chosen for the performance evaluation. Both the Master node and the field devices are equipped with a micro-controller and a wireless module BLE compliant. More in detail, the numerical values used in the simulations are extracted from the following devices:  16 bit MCU – Microchip PIC24F family (PIC24FJ256GB108) [26];  BLE121LR Radio Frequency Transceiver – Bluetooth Low Energy [27]. The simulations have been conducted with a model which has been built in Simulink/Matlab (Fig. 9). The main aim of this model is to simulate the behavior of the Master node and of a field device. The Field Device block manages the battery consumption of the field device. The sleeping time and the transmission power are acquired as input parameters of this block through a feedback loop system. This block produces two output variables that are used as input variables by the fuzzy logic controllers: Th/Wl and the battery_level. The activities of the micro-controller and of a connected sensor have a low impact on the battery consumption. In fact, their energy requirement is estimated as 50 mA (MCS). This value has been measured in an electronic board characterized by the PIC24FJ256GB108 micro-controller and a DS18B20 [28] temperature sensor. Whereby, the battery consumption mainly depends on the working state of the device. When the device is sleeping, the battery consumption is 50 mA þ 0:5 lAð0:5 lA 148 M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 Fig. 8. An example of WAC star topology. Fig. 9. Model scheme. Fig. 10. Battery consumption flow chart. is the power consumption of the BLE module in sleeping mode [27]). On the contrary, when the device is transmitting, the transmission power heavily affects the energy consumption. The battery consumption trend is evaluated in relation to the sleeping time by means of the Simulink/Stateflow environment. This tool uses flow charts and finite state machines in order to represent the evolution of a system. The Chart section, created in Simulink/StateFlow, related to the behavior of the battery consumption is represented in Fig. 10. Considering a 10:8 V lithium-ion battery, the maximum level of the battery when it is fully charged is 3100 mA, while the corresponding digital value, acquired through a 10 bit AD Converter, is 1024 (MaxBit). It is necessary to note that when the device is in sleeping mode, the power consumption is mainly due to the micro-controller and to the sensor, since the RF transceiver consumption is negligible. In this case, the consumption is 0.0046 bit/s. This value is determined by the following relation: Consumption Sleep Mode ¼ MCS  MaxBit FullBattery  3600 ð4Þ M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 149 where 3600 is the number of seconds in an hour. By applying this formula the power consumption of each device in sleeping mode is: Consumption Sleep Mode ¼ 50  1024 ¼ 0:0046 3100  3600 ð5Þ Whereby, when the device is transmitting, the battery consumption is 0.0046 bit/s, while in case of maximum transmission power (0 dB) the RF transceiver consumption is 36 mA [27]. As depicted in Fig. 9 the TXPower value is obtained as follow: Consumption Transmission Mode ¼ 36  1024 ¼ 0:0033 3100  3600 ð6Þ and it is necessary to underline that this value is fixed since the considered transceiver does not allow to change the transmission power [27]. The proposed FLC has the Th/Wl and the battery_level as input variables and dynamically produces the sleeping time considered as multiple value (from 1 to 10 times) of the sampling time (1 s). 5.2. Simulations results In order to validate the proposed approach several simulations have been carried out. The battery level during a simulation period of 48 h is depicted in Fig. 11. In particular, the proposed approach has been compared with the results obtained Fig. 11. Battery consumption comparison. Fig. 12. Th/Wl behavior. 150 M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 without FLC, i.e., assuming that the transmission power (0 dB) and the sleeping time (equal to 1, coinciding with the sampling time) are both fixed. As it is possible to see, the power consumption reduction, obtained with the proposed approach, prolongs the device lifetime. In fact, due to the proposed FLC the battery will be fully discharged after 170,600 s, i.e. after 47 h and 38 min. On the contrary, in the standard case, the battery will be fully discharged much earlier, after only 120,100 s, i.e. after about 33 h and 36 min. Using the proposed approach, an improvement of about about 30% has been obtained. This is a good result that emphasizes the goodness of the approach proposed in this paper. It, therefore, lends itself well in IoT applications where it is necessary to have a very low energy consumption. In Fig. 12 the Th/Wl behavior using the proposed approach is depicted. As it is possible to see, the Th/Wl fluctuates between 75% and 35% and these values are acceptable especially in those application fields with a moderate variation of data, e.g. temperature detection, where the most important thing is to prolong, as much as possible, the battery lifetime rather than to ensure high Th/Wl performance. In order to further validate the proposed approach another simulation has been carried out to make it suitable also in contexts characterized by real time constraints. For this reason, the proposed FLC has been modified so that the sleeping time does not increase too much since the main aim is to obtain better performance in term of Th/Wl. In fact, as depicted in Fig. 13, a greater Th/Wl has been obtained (between 65% and 85%). However, also an increased battery consumption has been measured as a consequence. The battery consumption measured in this simulation scenario is depicted in Fig. 14. The Fig. 13. Th/Wl behavior in time-constrained environments. Fig. 14. Battery consumption comparison in time-constrained environments. M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 151 performance of the proposed approach (about 42 h) are lower compared to the previous simulation but, even in this case, are still better than the standard case. Considering the results obtained from both simulations, it is clear that the use of a FLC provides considerable benefits in terms of power consumption reduction. In a non-real-time context, it is possible to set the FLC in such a way as to increase the sleeping time of network nodes in order to maximize the the batteries lifetime, while in the presence of real-time constraints then the sleeping time should be reduced in order to obtain the best network performance in terms of Th/Wl for example. However, it is necessary to highlight that the results obtained from the simulations are closely related to the type of membership function used by the FLC, which are triangular in this paper. The use of different membership functions can greatly change the results obtained with the proposed approach. For example, as shown in [29], the use of gaussian membership functions [30] can increase the accuracy greatly, without degrading the computational performance. Then the use of these membership functions could improve even more the performance achieved by the approach proposed in this work. 6. Conclusions In this paper a fuzzy logic based mechanism is presented in order to improve the lifetime of devices in a home automation wireless network. A thorough analysis has been done in order to determine the wireless communication protocol and the results showed that Bluetooth Low Energy has a high potential in becoming an important technology for the Internet of Things because one of the its biggest advantages is precisely its low power consumption. Anyhow, specific techniques can be used in such a way as to further reduce the energy consumption of BLE. In fact, the main aim of this work is to dynamically changes the sleeping time in order to increase the battery duration of the field devices. Simulations results are very promising and demonstrate that using the proposed FLC a substantial reduction of the energy consumption is obtained compared to simulations carried out with fixed sleeping time. Reference values for the execution of the simulations are obtained from the data-sheets of two off-the-shelf devices, i.e. Microchip micro-controller (16 bit MCU PIC24F family – PIC24FJ256GB108) and Bluegiga radio frequency transceiver (BLE121LR Bluetooth Low Energy). Future work will deal with a real implementation of the proposed approach with these devices, in order to confirm the results obtained by the simulations. Moreover, the membership functions can be represented by Gaussian functional shapes in order to increase the accuracy greatly, without degrading the computational performance. References [1] Atzori L, Iera A, Morabito G. The internet of things: a survey. Comput Netw 2010;54(15):2787–805. [2] Miorandi D, Sicari S, Pellegrini FD, Chlamtac I. Internet of things: vision, applications and research challenges. Ad Hoc Netw 2012;10(7):1497–516. [3] Gubbi J, Buyya R, Marusic S, Palaniswami M. Internet of things (iot): a vision, architectural elements, and future directions. Future Gener Comput Syst 2013;29(7):1645–60. [4] Liu Y. Study on smart home system based on internet of things technology. Inform Manage Sci 2013;207(4):73–81. [5] Silva I, Leandro R, Macedo D, Guedes L. A dependability evaluation tool for the internet of things. Comput Electr Eng 2013;39:2005–18. [6] Bluetooth Specification Version 4.0, Bluetooh SIG; 2010. [7] 802.15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (LR- WPANs), IEEE standard for information technology. Part 15.4; 2006. [8] IEEE Std 802.11-2007 for Information technology – Telecommunications and information exchange between systems – Local and metropolitan area networks – Specific requirements – Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, C1-1184; 2007. [9] Collotta M, Gentile L, Pau G, Scata G. Flexible ieee 802.15.4 deadline-aware scheduling for dpcss using priority-based csma-ca. Comput Ind 2014;65(8):1181–92. [10] Collotta M, Messineo A, Nicolosi G, Pau G. A self-powered bluetooth network for intelligent traffic light junction management. WSEAS Trans Inform Sci Appl 2014;11:12–23. [11] Shrestha G, Imtiaz J, Jasperneite J. An optimized opc ua transport profile to bringing bluetooth low energy device into ip networks. In: 2013 IEEE 18th Conference on emerging technologies factory automation (ETFA); 2013. p. 1–5. [12] Collotta M, Lo Bello L, Mirabella O. An innovative frequency hopping management mechanism for bluetooth-based industrial networks; 2010. p. 45– 50. [13] Collotta M, Lo Bello L, Mirabella O. Deadline-aware scheduling policies for bluetooth networks in industrial communications; 2007. p. 156–63. [14] Collotta M, Pau G, Scatá G. Deadline-aware scheduling perspectives in industrial wireless networks: a comparison between ieee 802.15.4 and bluetooth. Int J Distrib Sensor Netw 2013:11. Article ID 602923. [15] Siekkinen M, Hiienkari M, Nurminen J, Nieminen J. How low energy is bluetooth low energy? comparative measurements with zigbee/802.15.4. In: IEEE wireless communications and networking conference workshops (WCNCW) 2012. p. 232–7. [16] Dementyev A, Hodges S, Taylor S, Smith J. Power consumption analysis of bluetooth low energy, zigbee and ant sensor nodes in a cyclic sleep scenario. In: IEEE international wireless symposium (IWS); 2013. p. 1–4. [17] Liu J, Chen C, Ma Y, Xu Y. Energy analysis of device discovery for bluetooth low energy. In: IEEE 78th vehicular technology conference (VTC Fall); 2013, p. 1–5. [18] Zhou B, Chen X, Hu X, Ren R, Tan X, Fang Z, et al. A bluetooth low energy approach for monitoring electrocardiography and respiration. In: IEEE 15th international conference on e-health networking, applications & services (Healthcom); 2013. p. 130–4. [19] Gomez C, Oller J, Paradells J. Overview and evaluation of bluetooth low energy: an emerging low-power wireless technology. Sensors 2012;12:11734–53. [20] Shahgoshtasbi D, Jamshidi M. Energy efficiency in a smart house with an intelligent neuro-fuzzy lookup table. In: 6th international conference on system of systems engineering; 2011. p. 288–92. [21] Chehri A, Mouftah H. Feman: fuzzy-based energy management system for green houses using hybrid grid solar power. J Renew Energy 2013;2013:6 pages (Article ID 785636). [22] Tripathy M, Gaur K, Sharma S, Virdi G. Energy efficient fuzzy logic based intelligent wireless sensor network. In: Progress in electromagnetics research symposium; 2010. p. 91–5. [23] Singh A, Alkesh A, Purohit N. Minimization of energy consumption of wireless sensor networks using fuzzy logic. In: International conference on computational intelligence and communication networks (CICN); 2011. p. 519–21. 152 M. Collotta, G. Pau / Computers and Electrical Engineering 44 (2015) 137–152 [24] Sanchez E, Montrucchio B, Murillo L, Rebaudengo M. Adaptive fuzzy-mac for power reduction in wireless sensor networks. In: 4th IFIP international conference on new technologies, mobility and security (NTMS); 2011. p. 1–5. [25] Khader O, Willig A, Wolisz A. Distributed wakeup scheduling scheme for supporting periodic traffic in wsns. In: European wireless conference (EW); 2009. p. 287–92. [26] PIC24FJ256GB108 Family Data Sheet, Microchip Technology; 2014. . [27] BLE121LR Data Sheet, Bluegiga; 2014. . [28] DS18B20 Data Sheet, Maxim Integrated; 2014. . [29] Olunloyo V, Ajofoyinbo A, Ibidapo-Obe O. On development of fuzzy controller: the case of gaussian and triangular membership functions. J Signal Inform Process 2011;2(4):257–65. [30] Preuss H, Tresp V. Neuro-Fuzzy. Automatisierungstechnische Praxis; 1994 [in German]. Mario Collotta is Assistant Professor with tenure in the Faculty of Engineering and Architecture at the Kore University of Enna, Italy, and since 2011 he is scientific responsible and director of Telematics Engineering Laboratory. His research activity is mainly focused on the study of real-time networks and systems. He is a member of IEEE and has published 2 book chapters, and over 40 refereed international journals and conference papers. Giovanni Pau is currently a PhD student at the Kore University of Enna, Italy. His research interest includes wireless sensor networks, soft computing techniques and real-time systems. In each of these research fields, he has produced several publications in international conferences and journals.
Computer Communications 28 (2005) 37–50 A self-adaptive zone routing protocol for Bluetooth scatternets Chenn-Jung Huanga,*, Wei-Kuang Laib, Sheng-Yu Hsiaob, Hao-Yu Liub b a Institute of Learning Technology, National Hualien Teachers College, 123 Huashi, Hualien 97043, Taiwan, ROC Department of Computer Science and Engineering, National Sun Yat-Sen University, Kaohsiung 804, Taiwan, ROC Received 12 August 2003; revised 10 July 2004; accepted 28 July 2004 Available online 26 August 2004 Abstract This work presents a routing protocol that utilizes the characteristics of Bluetooth technology for Bluetooth-based mobile ad hoc networks. The routing tables are maintained in the master devices, and the routing zone radius for each table is adjusted dynamically via a fuzzy inference system. Given that some useless routing packets exist which increase the network loads in the existing ad hoc routing protocols, this work selectively employs multiple unicasts or a single broadcast when the destination device moves beyond the routing zone radius coverage of the routing table. The simulation results demonstrate that the dynamic adjustment of the routing table size in each master device results in considerably faster routing request reply time, as well as fewer request packets and useless packets compared with two representative protocols, Zone Routing Protocol (ZRP) and Dynamic Source Routing. q 2004 Elsevier B.V. All rights reserved. Keywords: Fuzzy logic; Bluetooth scatternet; Zone routing protocol; Reactive routing; Proactive routing 1. Introduction A Mobile Ad Hoc Network (MANET) lacks a fixed infrastructure. All devices in a MANET must participate in routing and forwarding since a MANET contains no Access Point (AP), base station, or router. When a source device wishes to communicate with a destination device, it must establish a routing path between the source and destination. Node mobility, available bandwidth and transmission power influence the design of the ad hoc network routing protocol. Bluetooth is primarily perceived as an affordable technology enabling peer-to-peer communication between a central terminal and peripheral devices. The characteristics of low-power consumption and high security make Bluetooth a good choice for MANET deployment. However, Bluetooth-based MANETs do differ from traditional ad hoc networks in some important ways. First, the connection range is smaller in a Bluetooth * Corresponding author. Tel.: C886-3833-5657; fax: C886-3833-9736. E-mail address: (C.-J. Huang). 0140-3664/$ - see front matter q 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.comcom.2004.07.009 MANET owing to the low power of Bluetooth devices. Second, the number of neighboring nodes for a Bluetooth device is limited since the piconet scenario in a Bluetooth-based ad hoc network comprises one master device and up to seven slave devices. Third, a large routing table is inappropriate in most Bluetooth devices due to limited storage space. Fourth, it is common for a moving Bluetooth device to be out of connection with the joined piconet owing to the short communication range in a Bluetooth MANET. To address these challenges, this work presents a selfadaptive zone routing protocol (SAZRP) for Bluetooth scatternets. The proposed algorithm establishes a limited routing table in every master device, while keeping the size of the routing table adjustable depending on the computational result of a fuzzy inference system. The simulation results demonstrate that the SAZRP requires less routing request reply time, and generates fewer request packets and useless packets than other representative routing protocols used in ad hoc networks. The remainder of the paper is organized as follows. Section 2 gives a brief description of the Bluetooth technology and the related routing protocols, such as Dynamic Source 38 C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 Routing Protocol (DSR) and Zone Routing Protocol (ZRP). Section 3 shows the details of SAZRP. Section 4 reviews the simulation results and comparisons. Conclusions are made in Section 5. 2. Background and related work Bluetooth is a radio interface for short-range and lowpower connections between electronic devices [1–4]. Bluetooth uses a frequency-hopping scheme in the unlicensed, scientific, and medical band at 2.4 GHz. The normal range is 10 m, but can be increased to 100 m. A frequency-hopping transceiver is used for reacting to interference and fading. Frequency-hopping spread spectrum (FHSS) possesses various properties that make it a good choice for an ad hoc radio system. There are several equally spaced channels exist, each with 1 MHz. A data rate of 1 Mb/s is achieved with FSK modulation. Seventy-nine hop channels have been defined in the 2.45 GHz ISM band. During a connection, transceivers hop among channels pseudorandomly, and receivers must simultaneously hop among channels using the same pattern. Bluetooth channels use a Frequency Hopping/Time Division Duplex (FH/TDD) scheme. The channel is divided into consecutive slots of 625!10K6 s each. Consecutive slots utilize different hop channels. The master determines the hopping sequence and uses even number slots, while the salves use odd number slots. When two Bluetooth devices establish a Bluetooth link, one acts as a master and the other acts as a slave. The Bluetooth specification permits any Bluetooth device to assume either role, and a device can act as a master for one communication link and as a slave for another. The role of a master does not involve special privileges or authority, but merely involves governing the synchronization of the FHSS communications between devices. Two kinds of links exist between masters and slaves in Bluetooth environments. One kind of link is real-time, constant rate, synchronous connection-oriented links (SCO), while the other kind is variable rate, asynchronous connectionless links (ACL). SCO links are established by reserving two consecutive slots for master-to-slave transmission and slave-to-master transmission, respectively. ACL links support point-to-multipoint and packet-switched connections used for burst data transmissions. Master devices use a polling scheme to schedule the transmission order of ACL links and avoid channel collisions when multiple slaves have data to transmit. Most data transmissions establish ACL links except in situations involving real-time requirements. A slave device in a Bluetooth environment possesses four different modes for achieving energy conservation. In the active mode, a slave always listens for transmissions from the master. An active slave must listen to all packets originating from the master. The active mode minimizes response time at the cost of increased power consumption. In the sniff mode, a slave becomes active periodically. The master agrees to transmit packets destined for a particular slave only at certain regular intervals. The slave therefore only needs to listen to the packets from the master during these intervals. Meanwhile, the hold mode resembles the sniff mode but has lower energy consumption. In the hold mode, a slave may stop listening to the packets completely for a specific time interval. The slave in the hold mode may be less responsive than that in the sniff mode, and can obtain greater power savings. Finally, in the parked mode, a slave only synchronizes with the master. The parked mode enables the master to orchestrate communications with more than seven devices in a piconet by exchanging the slaves in the active and parked modes and maintaining up to seven active connections while keep the remaining slaves in the piconet parked. A wireless MANET is a collection of self-configuring wireless mobile hosts forming a temporary network without any centralized administration and fixed infrastructure. A routing protocol is required because there is no AP and two hosts that wish to exchange packets may not be able to communicate directly. A good routing protocol can not only determine the shortest routing path, but also is suitable for the mobility characteristics of ad hoc networks. The routing protocols in MANET may be primarily classified as proactive and reactive. Proactive routing protocols require that all mobile devices have complete network knowledge. Unfortunately, most mobile devices have limited space for storing the routing information. Moreover, devices do not maintain routing tables in reactive routing protocols. Reactive routing protocols involve two main functions, route discovery and route maintenance. A source uses the route discovery function, generally implemented via some form of flooding, to establish a routing path to the destination. Route maintenance is responsible for avoiding routing along an unavailable path in situations involving topological changes. Numerous ad hoc routing protocols have been presented in the literature [5]. The following briefly introduces two representative protocols, Dynamic Source Routing Protocol (DSR) [6] and ZRP [7–8]. DSR is a reactive routing protocol based on source routing, and each packet determines a routing path to the destination itself. In Route Discovery, the source device broadcasts a ROUTE REQUEST packet that is flooded through the network in a controlled manner and answered by a ROUTE REPLY packet from the destination device. Additionally, the routing fields of the ROUTE REQUEST record the traversed devices from the source to the destination. Route Maintenance is performed when a packet cannot be successfully forwarded to the next-hop device. In this situation the next link of the source route is declared broken. The source device then is informed of this broken link. ZRP is a hybrid reactive/proactive routing protocol. On the one hand, ZRP limits the scope of the proactive C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 procedure to the local neighbors of the node. On the other hand, network searching is performed when a device cannot find the destination through proactive routing. The ZRP comprises two procedures, the IntrAzone Routing Protocol (IARP) and the IntErzone Routing Protocol (IERP). The IARP is used within the routing zone, while IERP is used when the distance between the source and destination exceeds the radius of the routing zone. Each device must maintain the routing information of all devices in its routing zone, and updates the information in the case of topological change. When the distance to the destination is less than the zone radius, the destination can be located by IARP based on the routing information in each device. However, if the destination is located outside its zone, the IERP will broadcast a route request to identify the destination. Each device that receives the route request will repeat the above procedure until the destination is found. Using a mixture of reactive and proactive routing, ZRP can control routing information storage space and number of broadcasts. Thongpook and Thumthawatworn [9] further developed an adaptive ZRP by using fuzzy rule-base to permit dynamic adjustment of the zone radius of the routing table for the IARP to react appropriately to network configuration change. Although numerous ad hoc routing protocols were proposed or reviewed in Refs. [5,9], they are not well suited for Bluetooth scatternets before being adapted to the specifics of Bluetooth. Recently, Prabhu and Chockalingam [10] presented a routing protocol for increasing gain in network life time, but this protocol still did not address the issues of reducing routing request reply time and request packets and reducing useless packet path length. Kapoor and Gerla [11] also established a routing scheme for Bluetooth scatternets based on the ZRP, but failed to resolve the issue of allowing individual nodes to identify and react to changes in network behavior by adjusting the routing zone radius. To address the above challenges, this work proposes a novel routing protocol for a Bluetooth MANET that can adapt ZRP to the characteristics of Bluetooth technology. Notably, the proposed routing scheme uses a fuzzy inference system to tune the radius of the routing zone such that the collaboration of the proactive and reactive protocols can promptly accommodate the topological change in Bluetooth scatternets. 39 3. Self-adaptive zone routing protocol (SAZRP) After observing the Bluetooth-based ad hoc networks, we find several characteristics which are different from traditional ad hoc networks. 1. The number of neighboring devices is limited and small. For other ad hoc networks, the neighboring devices may be large. However, in Bluetooth-based ad hoc networks, a master device connects up to seven slave devices, and a slave also connects to limited master devices. 2. For a master device A in a Bluetooth MANET, if there are other master devices within the same network, there exists at least one master device whose distance to device A is no more than two hops. Fig. 1 shows two possible conditions for the distance between two masters. It is two hops if a slave, said B, is connected to the masters of two piconets (Fig. 1(a)) and it is one hop if the device B is a slave in one piconet I and the master in another (Fig. 1(b)). Based on the above observations, we draw some conclusions as follows. 1. If routing tables are built in all master devices, all devices of ad hoc networks can be covered. It is not necessary to have routing tables in slave devices. 2. When the routing table in a master device covers devices within two hops, the master can use this routing table to find other nearby master devices. 3. The size of a routing table in the master is smaller than traditional ad hoc networks in general because there are at most seven active slaves within a piconet. The masters have more room to adjust the routing zone radius following the change of network and node behavior. 4. If we can reduce the number of broadcasts, we can also diminish the number of nodes involved in unnecessary transmissions that may considerably interfere with the reply of establishing a connection. Meanwhile, we can reduce the time in finding a path to the destination which in turn alleviates the effects of topology changes due to node mobility. Fig. 1. The distance between two masters of a scatternet is no more than two hops. 40 C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 Fig. 2. An example of Bluetooth MANET. Next we shall introduce the operations of the master device and the slave device of SAZRP in details. Then a fuzzy inference system is presented to adapt the routing zone radius of the routing table to a shift of the network operating conditions. 3.1. Master device Fig. 2 illustrates an example of a Bluetooth network. The example displayed includes three piconets, whose masters are B, E and F, respectively. Tables 1(a) and (b) display the routing tables of master B and master F, respectively. The routing table of master E can be built similarly. Table 1 shows that the routing tables are built using link lists comprising lists of ID–Type pairs. The first node in each ID–Type pair records those devices which are separated from the master for one hop. The second node of the ID– Type pair identifies devices recorded in the first node as either masters or slaves. For example, node C is stored in the first node of second list in Table 1(a). To its right, node C is identified as a slave node in the second node. Notably, if node C is a master of another piconet, it is identified as a master node rather than as a slave node of master B. Then in the third and the fourth nodes, devices with a distance to the master of two hops are recorded, and are identified Table 1 The routing table of master is built by link lists as masters or slaves, respectively. Furthermore, if node C is connected to multiple masters, the fifth and sixth nodes are used for the second master, the seventh and the eighth nodes for the third master, and so on. The source routing approach is used in SAZRP. The ROUTE REQUEST and ROUTE REPLY packets both have a type field and several routing fields which record the routing path from the source to the current nodes, as illustrated in Fig. 3(a) and (b). When a master device receives a ROUTE REQUEST packet, it first checks whether the destination is itself or instead is a device contained in its routing table. If the destination is itself, it sends the ROUTE REPLY packet to the source. The destination device reverses the routing path in the routing fields of the ROUTE REQUEST to switch the roles of the destination and the source before including them in the routing fields of ROUTE REPLY. The destination forwards the ROUTE REPLY to the neighboring device depending on routing fields. If the destination is in its routing table, the device will add its ID following the last routing field in the ROUTE REQUEST, and then send the ROUTE REQUEST to the destination or a device neighboring the destination. If the destination is not contained in the routing table, the master device will append its unique ID to the last routing field C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 Fig. 3. Formats of ROUTE REQUEST and ROUTE REPLY packets. 41 Each device except the source receiving the ROUTE REPLY in the network must seek routing fields and then send the ROUTE REPLY to the next specific device in ROUTE RECORD. The routing operation is complete when the source device, located at the final position in the ROUTE RECORD of ROUTE REPLY receives the ROUTE REPLY. 3.2. Slave device and forward the ROUTE REQUEST to its neighboring devices via multiple unicasts or a broadcast, depending on the number of neighboring devices. If the number of neighboring devices, which are either masters themselves or connected to master devices, exceeds a certain threshold, broadcast is used to forward the ROUTE REQUEST, otherwise the ROUTE REQUESTs is forwarded via multiple unicasts to avoid sending redundant ROUTE REQUESTs to numerous nodes. However, neighboring devices receive the ROUTE REQUEST only if they satisfy either of the following two conditions: (a) the neighboring device acts as master in another piconet or (b) the neighboring device of the master has one connection to a remote master not contained in the routing fields, and the master has not previously sent or forwarded this ROUTE REQUEST. Fig. 4 illustrates in detail the operations when a master device receives a ROUTE REQUEST. The slave devices do not build the routing table and simply broadcast the ROUTE REQUEST. On receiving a ROUTE REQUEST, a slave device will first check whether the destination is itself. If this is the case, the slave device will send the ROUTE REPLY to the source device. Meanwhile, if the destination is a neighboring device, the slave device will add its own unique ID after the last routing field of the ROUTE REQUEST, and send the ROUTE REQUEST to the destination via unicast. Moreover, if the destination is neither itself nor a neighbor, the slave device will add its own unique ID following the last routing field of the ROUTE REQUEST, and then unicast the ROUTE REQUEST to all its neighboring devices individually. The slave device which receives a ROUTE REPLY also must forward it to the next specific device in the ROUTE RECORD (Fig. 5). Fig. 4. The pseudo code for the operations of a master device that receivs a ROUTE REQUEST. 42 C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 Fig. 5. The algorithm in slave device that receives a ROUTE REQUEST. 3.3. A fuzzy routing zone radius estimation scheme The fuzzy logic has been used to solve several routing protocols and handover problems efficiently in wireless networks in the literature [12–14]. There are lots of solutions on VLSI chips which allow fuzzy inferences to be hardware-computed, and high-speed low cost fuzzy chips have been introduced recently, the implementation of fuzzy logic by hardware thus becomes feasible nowadays [15–17]. In our scheme, a fuzzy logic approach is attempted to offer the self-tuning capability in the routing zone radius estimation mechanism. The proposed fuzzy routing zone radius estimator is encompassed in the dotted frame as shown in Fig. 6. The basic functions of the components employed in the scheme are described as follows. †Fuzzifier. The fuzzifier performs the fuzzification function that converts three types of input data from the fuzzy routing zone radius scheme into suitable linguistic values which are needed in the inference engine. Notably, the input to the fuzzifier v represents node velocity, which is a measure of network reconfiguration rate. The input n denotes the node density, which is the number of neighboring nodes of the master, and the input r stands for the route query rate observed by the master node. †Fuzzy rule base. The fuzzy rule base is composed of a set of linguistic control rules and the attendant control goals. †Inference engine. The inference engine simulates human decision-making based on the fuzzy control rules and the related input linguistic parameters. The max–min inference method is used to associate the outputs of the inferential rules [18,19], as described later in this subsection. †Defuzzifier. The defuzzifier acquires the aggregated linguistic values from the inferred fuzzy control action and generates a non-fuzzy control output, which represents the estimated routing zone radius adapted to the new network and node conditions. The Tsukamoto defuzzification method is employed to compute weighted average of the aggregated output of the inferential rules due to its simplicity in computation [18,19]. Figs. 7–9 illustrate the mapping of inputs of the fuzzifier into some appropriate linguistic or membership values, which are expressed by the values within the range of 0 and 1. The set of membership functions for the node velocity, the node density, and the route query rate, are presented in Figs. 7–9, respectively. All the inputs v, n and r are mapped into three linguistic term sets, ‘low’, ‘medium’ and ‘high’. The output parameter of the inference engine, Rz, is defined as the routing zone radius control action of our scheme. The fuzzy linguistic variables for the output are ‘low’, ‘medium’ and ‘high’, which are represented by the membership functions as shown in Fig. 10. The input and output fuzzy sets are correlated to establish the inferential rules of the fuzzy routing zone radius Fig. 6. The fuzzy routing zone radius estimator. C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 Fig. 7. Membership function for the node velocity. estimator as listed in Table 2, which are correspondent with the observation made by Pearlman and Haas [20]. By way of illustration, rule 1 can be interpreted as: IF network reconfiguration rate is ‘low’, AND the node density is ‘low’, AND the route query rate is ‘low’, THEN the weighting factor of the routing zone radius for the routing table is ‘low’. Fig. 11 illustrates the reasoning procedure for rule 20 in Table 2. The non-fuzzy output of the defuzzifier can then be expressed as the weighted average of each rule’s output after the Tsukamoto defuzzification method is applied: P27 iZ1 Rz;i wi zZ P (1) 27 iZ1 wi where Rz,i denotes the output of each rule induced by the firing strength wi. Notably, wi represents the degree to Fig. 8. Membership function for the node density. 43 Fig. 9. Membership function for the route query rate. which the antecedent part of each fuzzy rule constructed by the connective AND as shown in the above example is satisfied. SAZRP is found to have three advantages compared with the traditional ad hoc routing protocol. 1. Less number of broadcasting. In most ad hoc routing protocols, devices broadcast route requests if they do not know the locations of destinations. Broadcast messages then are continually delivered until the final destination is reached. Meanwhile, a master device in SAZRP uses multiple unicasts or a broadcast if the destination lies outside of the routing zone radius and otherwise uses the selected unicast. This approach can significantly reduce network load since it reduces the total number of broadcasts. For example, this work assumes that the current routing zone radius of device B is two hops, that Fig. 10. Membership function for weighting factor of estimated routing zone radius. 44 C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 Table 2 Fuzzy rule base for routing zone radius estimator Output (Rz,i) Rule Input v n r 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 Low Medium High Low Medium High Low Medium High Low Medium High Low Medium High Low Medium High Low Medium High Low Medium High Low Medium High Low Low Low Medium Medium Medium High High High Low Low Low Medium Medium Medium High High High Low Low Low Medium Medium Medium High High High Low Low Low Low Low Low Low Low Low Medium Medium Medium Medium Medium Medium Medium Medium Medium High High High High High High High High High device A is the source, and that device E is the destination, as shown in Fig. 2. In SAZRP, slave A unicasts a ROUTE REQUEST to master B. After receiving the ROUTE REQUEST, master B checks to see whether E is in its routing table. Because the distance between B and E is two hops, the position of E is recorded in the routing table of device B. Therefore, device B unicasts ROUTE REQUEST to C, and C forwards it to destination E. On the other hand, in most reactive ad hoc routing protocols, for example, DSR, device B is unaware of the path to destination E. Thus device B broadcasts the ROUTE REQUEST, and both devices C and D receive it. Unfortunately, device Low Low Low Medium Low Low Medium Low Low Medium Medium Low Medium Medium Medium High Medium Medium High High Medium High High Medium High High High D does not know the position of destination E, and thus also broadcasts the ROUTE REQUEST. Finally, the ROUTE REQUEST is passed to device F, G, which adds more traffic in the network and is clearly useless. 2. Lower storage spaces. Most proactive routing protocols require each network device to build a routing table. This requirement is extremely costly for all devices in MANETs. However, ZRP requires less storage space than traditional proactive routing protocols. ZRP controls routing table size via the routing zone radius. However, each device still must establish a routing table in the ZRP. In the SAZRP, only master devices need to build routing tables, and each master connects up to Fig. 11. The reasoning procedure for Tsukamoto defuzzification method. C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 Fig. 12. Total reply time of routing request under different node population. Fig. 13. Broadcast to unicasts ratio for FZRP, ZRP and DSR. Fig. 14. Broadcast to unicast ratio for SAZRP and FZRP. 45 46 C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 Fig. 15. Broadcast packer number under different node population. seven slaves. The master devices thus can lengthen the routing zone radius for the routing table if necessary. 3. Shorter time for reply to route requests. In an ad hoc mobile network, the longer a source takes to receive a ROUTE REPLY, the more likely the transmitted path is likely to be changed. The SAZRP has shorter reply time than the ZRP since the ZRP broadcasts more ROUTE REQUEST packets, which might interfere with the ROUTE REPLY and delay the arrival of the ROUTE REPLY at the source. 4. Simulation results This work randomly generates 50–150 Bluetooth devices in a 5625 m2 area. Device positions and speeds are also produced randomly. The speed of each Bluetooth device ranges from 3 to 30 m/s, and the connection range of each device is 10 m. A master device can connect up to seven slave devices, and a slave device can join up to 10 piconets. ACL links are established. Following network construction, the network devices are randomly selected as the connection source and destination points. The maximum number of connections is limited to one third of the node counts in the network and each connection begins at a random time. The source device must send the ROUTE REQUEST to a destination and receive the ROUTE REPLY from a destination for building a routing path. For comparison, a series of simulations were run for the ZRP, DSR, SAZRP, and SAZRP with a fixed routing zone radius (FZRP). The routing radius was set to two hops for the ZRP and FZRP schemes. The simulation was repeated 500 times and the average was taken as the final result. Fig. 12 compares the routing request reply time for the four schemes under different node populations. The reply Fig. 16. Unicast packet number under different node population. C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 47 Fig. 17. Total request packets comparison under different node population. time represents the time interval between the source sending a ROUTE REQUEST and receiving a ROUTE REPLY. Clearly, the reply time in the SAZRP scheme is significantly less than in the other three schemes. We believe that this is primarily because the DSR and ZRP both broadcast ROUTE REQUEST when devices are unaware of the destination positions. The packets will clearly be delayed when the network is congested with numerous ROUTE REQUEST broadcast messages since the number of connection sessions is up to one third of the number of network nodes. Although the SAZRP also broadcast when the destination is not within its zone radius coverage, the capability of self-adaptation on the routing zone radius results in the spread of markedly fewer broadcasts. The ratio of broadcast to unicast for the four schemes shown in Figs. 13 and 14 supports the speculation of this study. Notably, this study accounts the control packets that include the IARP and the IERP packets for broadcast and unicast traffic since this work focuses on the routing scheme. The comparison of the broadcast to unicast ratio is separated into two figures since the DSR and ZRP schemes demonstrate a significant lag. The SAZRP and FZRP schemes are difficult to differentiate if all the comparisons are shown in a single figure. Figs. 15 and 16 reveal why the difference in broadcast/unicast ratio between FZRP and SAZRP is very small compared with the other two schemes. The number of packets broadcast for the FZRP and SAZRP schemes appear significantly lower than for the other two schemes, while the number of unicast packets for the FZRP and SAZRP schemes is slightly higher than for the other two schemes. This explains why the broadcast/unicast ratio for the FZRP is slightly larger than one and the ratio for the SAZRP is even smaller than one, Fig. 18. Comparison of the useless ROUTE REQUESTs. 48 C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 Fig. 19. Total reply packet comparison under different node population. as illustrated in Fig. 14. Notably, the total number of control packets for the FZRP and SAZRP schemes is still much lower than the other two schemes, as illustrated in Figs. 15 and 16, although the number of unicast packets is somewhat higher than the other two schemes. Fig. 17 illustrates the comparison of total ROUTE REQUESTs received by each node. Every time a device receives a ROUTE REQUEST, the value of total received ROUTE REQUESTs increases by 1. Clearly, the SAZRP receives fewer ROUTE REQUEST packets than other schemes, particularly when the node population is large, because the SAZRP selectively uses either multiple unicasts or one broadcast depending on the situation of neighboring devices when the destination is out of the routing zone radius of the master. The figure further explains that the SAZRP has the shortest reply time because the network nodes receive fewer messages and can reply ROUTE REPLY to the source faster than other protocols. The routing path is identified when a source device receives a ROUTE REPLY from the destination device. However, some ROUTE REQUESTs may still be being sent via the network at this time. These ROUTE REQUEST packets do not give any help in building the routing path. The reason these packets remain alive is that some devices do not know that the routing path has been found, and consequently still forward the ROUTE REQUESTs to neighboring devices. Fig. 18 illustrates that SAZRP has considerably fewer useless ROUTE REQUESTs than other Fig. 20. Success ratio of routing request under different node population. C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 49 Fig. 21. Success ratio of routing request under different node speed. Fig. 22. Comparison of routing path lengths. schemes. Fig. 19 also displays the comparison of total ROUTE REPLY packets among the four schemes. As expected, the SAZRP scheme maintains its superiority over the other three schemes owing to the use of an adjustable routing zone radius in the SAZRP. The allowance of dynamic adjustment of routing zone radius also contributes to the shorter reply time of the SAZRP scheme compared to the other schemes as illustrated in Fig. 12. Figs. 20 and 21 illustrate the success ratio of routing requests under different node populations and speeds, respectively. The success ratio of routing requests is slightly better in the SAZRP than in other schemes for a small node population, and is significantly better for large node populations. Additionally, the success ratio or routing request for each node with different mobility is also better for the SAZRP scheme than for other schemes, as illustrated in Fig. 21. In DSR, the ROUTE REQUEST is continually broadcast until it reaches the destination. The ROUTE REQUEST packets reach almost every device in the network, but the routing path is not shorter than for other protocols, as illustrated in Fig. 22. Clearly, the flooding of ROUTE REQUEST packets cannot achieve the shortest routing path because the first flooding of the ROUTE REQUEST to the destination does not necessarily achieve the shortest distance to the destination. 5. Conclusion This work takes use of some characteristics of Bluetooth technology to design an efficient protocol called the SAZRP for Bluetooth-based MANETs. In SAZRP, routing table is built in each master device to reduce the space cost. In order 50 C.-J. Huang et al. / Computer Communications 28 (2005) 37–50 to reduce the flooding of broadcast, the SAZRP uses the unicast in master devices to replace the broadcast. SAZRP also checks if the neighboring device needs to receive the ROUTE REQUEST packet. A fuzzy inference system is used to decide the routing zone radius for the routing table based on three parameters observed by the masters. Simulation results demonstrate that the SAZRP has less reply time of routing request, smaller broadcast to unicasts ratio, fewer request and reply packets, and lower useless packet ratio, compared to the DSR and ZRP. Notably, the vector of Routing Vector Method (RVM) [21] can be incorporated into our scheme to replace the Bluetooth 48-bits address in order to further reduce overheads in networks since The RVM can reduce the size of ROUTE REQUEST and ROUTE REPLY packets. Acknowledgements The authors would like to thank the National Science Council of the Republic of China, Taiwan for financially supporting this research under Contract No. NSC 90-2213E-143-001. References [1] A. Das, A. Ghose, A. Razdan, H. Saran, R. Shorey, Enhancing performance of asynchronous data traffic over the Bluetooth wireless ad-hoc network, INFOCOM 2001 1 (2001) 591–600. [2] P. Johansson, M. Kazantzidis, R. Kapoor, M. Gerla, Bluetooth: an enabler for personal area networking, IEEE Network 15 (5) (2001) 28–37. [3] D. Groten, J.R. Schmidt, Bluetooth-based mobile ad hoc networks: opportunities and challenges for a telecommunications operator, IEEE Vehicular Technology Conference, 2001 2 (2001) 1134–1138. [4] B.A. Miller, C. Bisdikian, Bluetooth Revealed, Prentice-Hall, Englewood Cliffs, NJ, 2001. [5] E. Royer, C.-K. Toh, A review of current routing protocols for ad hoc mobile wireless networks, IEEE Personal Communications 1999; 46–55. [6] D.B. Johnson, D.A. Maltz, Dynamic source routing in ad-hoc wireless networks. in: T. Imielinski, H. Korth (Eds.), Mobile Computing, Kluwer, Dordecht, 1996, pp. 153–181. [7] Z.J. Haas, M.R. Pearlman, The performance of query control schemes for the Zone Routing Protocol, IEEE Transactions of the Networking 9 (4) (2001) 427–438. [8] M.R. Pearlman, Z.J. Haas, S.I. Mir, Using routing zones to support route maintenance in ad hoc networks, IEEE Wireless Communications and Networking Conference 3 (2000) 1280–1285. [9] T. Thongpook, T. Thumthawatworn, Adaptive zone routing technique for wireless ad hoc network, Proceedings of the ITC-CSCC 2002; 1839–1842. [10] B.J. Prabhu, A. Chockalingam, A routing protocol and energy efficient techniques in Bluetooth scatternets, IEEE International Conference on Communications 5 (2002) 3336–3340. [11] R. Kapoor, M. Gerla, A zone routing protocol for Bluetooth scatternets, 2003 IEEE Wireless Communications and Networking Conference 3 (2003) 1459–1464. [12] Y.F. Wong, W.C. Wong, A fuzzy-decision-based routing protocol for mobile ad hoc networks, 10th International Conference on Network 2002; 317–322. [13] G.V.S. Raju, G. Hernandez, Q. Zou, Quality of service routing in ad hoc networks, 2000 IEEE Wireless Communications and Networking Conference 1 (2000) 263–265. [14] Q. Liang, Soft handover for nonuniformly-loaded mobile multimedia cellular networks, 55th IEEE Vehicular Technology Conference 3 (2002) 1096–1100. [15] A. Rodriguez-Vazquez, R. Navas, M. Delgado-Restituto, F. VidalVerdu, A modular programmable CMOS analog fuzzy controller chip, IEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing 46 (3) (1999) 251–265. [16] K. Daijin, An implementation of fuzzy logic controller on the reconfigurable FPGA system, IEEE Transactions on Industrial Electronics 47 (3) (2000) 703–715. [17] M.N. Uddin, T.S. Radwan, M.A. Rahman, Performances of fuzzylogic-based indirect vector control for induction motor drive, IEEE Transactions on Industry Applications 38 (5) (2002) 1219– 1225. [18] W. Pedrycz, F. Gomide, An introduction to fuzzy sets: analysis and design (complex adaptive systems), MIT Press, Cambridge, MA, 1998. [19] J.J. Buckley, E. Eslami, E. Esfandiar, An introduction to fuzzy logic and fuzzy sets (advances in soft computing), Physica-Verlag, Wurzburg, 2002. [20] M.R. Pearlman, Z.J. Haas, Determining the optimal configuration for the Zone Routing Protocol, IEEE Journal of the Selected Areas in Communications 17 (8) (1999) 1395–1414. [21] P. Bhagwat, A. Segal, A Routing Vector Method (RVM) for routing in Bluetooth scatternets, 1999 Mobile Multimedia Communications 1999; 375–379.
Journal of King Saud University – Computer and Information Sciences (2016) xxx, xxx–xxx King Saud University Journal of King Saud University – Computer and Information Sciences An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks Sabeen Tahir a,b,*, Sheikh Tahir Bakhsh a, Abdulrahman H. Altalhi a a Faculty of Computing and Information Technology, King Abdulaziz University, Jeddah 21589, Makkah, Saudi Arabia Department of Computer and Information Sciences, Universiti Teknologi PETRONAS, Bandar Seri Iskandar, 31750 Tronoh, Perak, Malaysia b Received 13 April 2016; revised 28 July 2016; accepted 28 July 2016 KEYWORDS Bluetooth network; Route maintenance; Weak links; Prediction; Backup device Abstract Bluetooth is a widespread technology for small wireless networks that permits Bluetooth devices to construct a multi-hop network called scatternet. Routing in multi-hop dynamic Bluetooth network, where a number of masters and bridges exist creates technical hitches. It is observed that frequent link disconnections and a new route construction consume extra system resources that degrade the whole network performance. Therefore, in this paper an Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks (ERMP) is proposed that repairs the weak routing paths based on the prediction of weak links and weak devices. The ERMP predicts the weak links through the signal strength and weak devices through low energy levels. During the main route construction, routing masters and bridges keep the information of the Fall Back Devices (FBDs) for route maintenance. On the prediction of a weak link, the ERMP activates an alternate link, on the other hand, for a weak device it activates the FBD. The proposed ERMP is compared with some existing closely related protocols, and the simulation results show that the proposed ERMP successfully recovers the weak paths and improves the system performance. Ó 2016 King Saud University. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ( 1. Introduction * Corresponding author at: Faculty of Computing and Information Technology, King Abdulaziz University, Jeddah 21589, Makkah, Saudi Arabia. E-mail addresses: (S. Tahir), stbakhsh@kau. (S.T. Bakhsh), (A.H. Altalhi). Peer review under responsibility of King Saud University. Production and hosting by Elsevier Bluetooth specification is designed for the low power two-way radio communication to connect the network devices over short distances (Chih-Min and Yin-Bin, 2014). The Bluetooth physical layer operates on the 2.4 GHz Industry Scientific and Medical (ISM) frequency band. The key advantage of the ISM frequency band is that it makes the Bluetooth technology accepted worldwide (Laharotte et al., 2015). A Bluetooth device can play the role of a master, slave or bridge. A Bluetooth basic networking unit is called a piconet which behaves like a cluster in an ad-hoc network topology that contains the maximum of eight Bluetooth active devices out of 256 devices. 1319-1578 Ó 2016 King Saud University. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ( Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), 2 In a piconet, one Bluetooth device becomes a cluster head known as a master device and other devices become slaves. Within a piconet, direct slave to slave links are not allowed and all communicating links go through the master device. A master device treats the slave devices in a Round Robin (RR) fashion that basically overcomes the collision between devices (Singh and Agrawal, 2011). To decrease the interference between devices, Bluetooth technology uses the Frequency-Hopping Spread Spectrum (FHSS) Jin-Ho et al., 2014; Subhan et al., 2012 technique. The use of FHSS makes the Bluetooth technology less susceptible to signal congestion and eavesdropping than other wireless technologies. Each Bluetooth device uses 79 channels using a pseudo-random hopping sequence. Each Bluetooth device uses a running clock, and all slave devices stay synchronized to the piconet from time to time adding a timing offset to their clocks from the clock of the master (Etxaniz and Aranguren, 2015). In a Bluetooth network, each active slave device is also allocated a 3-bit Active Member Address (AM_Addr) by the master device. Bluetooth devices discover each other and make the connections. Initially, all the devices are in standby mode only the native clock (CLKN) runs. The Bluetooth clock monitors the timing and hopping sequence of the transceiver, it is typically executed as a 28-bit wrap-around counter (Yu and Lin, 2012; Sharafeddine et al., 2012). The master sends requests to the other Bluetooth devices to make a piconet. Devices which agree to make a piconet reply to the master’s request. The master device sends the ID packets on different frequencies and waits for the reply packets called Frequency Hopping Sequence (FHS) packets from the slave devices. The device sends not only a FHS packet but also its ID and clock value. A Bluetooth device in the inquiry scan mode decreases its hopping frequency; this mode gives consent to the inquirer to catch up with the transmit frequency of the Bluetooth device which is in the inquiry scan mode (Ramana Reddy et al., 2010; Yu, 2010). As the frequencies match, the scanned devices behave like slave devices and transmit their ID and clock information to the master device. The time interval between the starting of two consecutive page scan operations is considered as the scan interval time, Tinquiry_scan, that should be less than or equivalent to 2.56 s (Cui et al., 2010). When Bluetooth devices exceed more than eight devices, they make a scatternet (Subhan et al., 2011). A scatternet is a combination of interconnected piconets where a device of one piconet (master/slave) elects to contribute as a slave in other piconets. This intermediate device is called a bridge/relay device that relays the data between piconets. By using this approach, more interconnected piconets can make a large network (Xin and YuPing, 2009) as shown in Fig. 1. There are many Bluetooth multi-hop routing protocols which have been proposed for inter-piconet communication, but it has been analyzed that none of them consider the route breakage issue (Bo et al., 2012; Yu, 2012; Aldabbagh et al., 2015a,b; Tahir and et al., 2013). Most of the researchers proposed ideas for route optimization without considering route breaking issues. It is true that if the main routing link breaks, it creates some serious problems like delay, unnecessary energy consumption, packet loss, etc. Therefore, it has provided an opportunity to propose an Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. The proposed protocol overcomes the problem of frequent link breakage and provides a solution in terms of predicting weak links. S. Tahir et al. Figure 1 Bluetooth scatternet where one bridge device links three piconets. The remaining portion of this paper is organized as follows. Section 2 discusses the related work of the proposed protocol. To overcome the frequent link breakage problem, the proposed protocol is presented in Section 3. The performance analysis of the proposed protocol and its comparison with a few similar protocols is shown via simulation in Section 4 using the NS-2 (The NS-2 Simulator, 2014) and UCBT (Agrawal and Wang, 2007). Finally, the conclusion and possible future work is presented in Section 5. 2. Background and related work In a piconet, direct slave to slave communication is not possible so all communicating links (outgoing and incoming) data traffic go through the master device; therefore, the master is the most important device. If a master device goes down or moves, it may disconnect all linked devices. Similarly, within a scatternet, direct master to master communication is not possible as they always need an intermediate bridge device, therefore, the bridge device is considered as the most significant device because it connects multiple piconets (Ching-Fang and Shu-Ming, 2008). If it fails or moves, then it can disturb the whole network. When data are routed between two Bluetooth networks for slave to slave communication, it follows the rule that the data will go through ‘‘slave-to-master – bridge-master-slave”. Although, many researchers have proposed different ideas for routing and scatternet formation for a Bluetooth network, but in this research, we considered only the most relevant protocols. The authors proposed a dynamic energy-aware network maintenance (DENM) protocol Bakhsh and et al., 2011, where a master device maintains a table storing all the connected slave devices’ information. When the energy level of the master device reaches L1, it activates an auxiliary master from its slave devices. Similarly, if a bridge device energy level reaches L1, it sends a request message to the connected master devices for backup relay activation. This technique provides flexibility against frequent link discussion based on the energy level. The main drawback with this technique is that it does not consider device mobility which can cause frequent link disconnection. The DENM is explained through Fig. 2(a), where source device A transmits data to the destination device D passing Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), Route Maintenance Protocol for Dynamic Bluetooth Networks Figure 2 Route maintenance using DENM protocol. through A–M1–B–M2–D. Once the energy level of M1 reaches its limit, it activates C as a new master and starts transmission through the A–C–B–M2–D route as shown in Fig. 2(b). The second relevant protocol is the Novel Route Maintenance (ROMA) Sahoo et al., 2008 for Bluetooth ad-hoc networks. ROMA maintains a routing path for a Bluetooth scatternet. If devices are added or removed from the network, ROMA decreases the number of hops by adding new devices and reconstructs the routing path. In ROMA, if a new device joins or leaves the scatternet it maintains its routing link by considering the number of hops. The respective master device checks the new device to find out whether it can reduce the hop counts and make the routing path shorter; then, it changes the role of the new device and creates a new piconet(s) within the routing path. Although ROMA works well, it does not consider the change to the device’s role once the communication has ended. Moreover, ROMA does not provide any solution in the routing path for the devices that want to communicate with their original master device or slave device in the network. According to ROMA, if a Bluetooth device moves away from the piconet, it informs its master device for the device leaving procedure which takes extra time and consumes more resources. As an example shown in Fig. 3(a), source device A is transmitting data to the destination device D through routing path (A–M1–B–C–D) and data are also routed from the source device E to H through the routing path (E–J–H). Fig. 3(b) shows that during data transmission, a new device I enters into the piconet M1. M1 is a routing master, it checks that this device can provide the shortest path. As device I and D are direct within the radio range of each other, the routing master breaks the existing link and makes the new link A–I–D. Meanwhile, device F starts moving and goes out from the range of Figure 3 3 the routing masters, M2 and M3. When it starts moving, it informs the routing masters so they can find another device J, that can re-establish the route E–J–H as device E is now transmitting data through a new route; however, if suddenly, device J fails due to energy level, it would not be able to inform its routing master. In the above scenario, ROMA fails to provide any solution to continue transmission. Therefore, this has provided an opportunity to propose a new protocol for route maintenance to overcome the inefficiencies of existing protocols. BlueStar is a scatternet development technique, where each node, after the discovery stage, calculates its weight on the basis of its degree and other parameters, and compares them with its neighbors for determining its role as (master/ slave). The BlueStar protocol sets up its own separate piconets; a node that has a higher weight compared to its neighbors will become the master and the remaining nodes will become slaves. The master node selects a bridge node to connect with neighboring piconets. In the resultant BlueStar scatternet, each piconet can contain more than seven slave nodes. However, it is observed that having more than seven slave nodes in a piconet could degrade the performance. Another scatternet protocol called BlueMesh (Emre and Oznur, 2006) has been proposed for the improvement of the BlueStar protocol. BlueMesh activates in two stages; in the first stage, it determines one and two hop neighboring nodes and in the second stage, the Bluetooth nodes select their roles. BlueMesh defines a procedure for handling the topology variations based on the insertion and removal of Bluetooth nodes. When a new node wants to join the scatternet, it first broadcasts identity packets and then the receiving nodes reply with a FHS packet. New nodes can receive FHS packets from more than one neighbor; in this case, a decision is required for connection establishment. It has been analyzed that BlueMesh’s large control overhead is due to device role selection. 3. The proposed ERMP protocol The proposed Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks (ERMP) is discussed is this section. Bluetooth technology does not allow direct slave to slave or master to master communication in different piconets. Therefore, the Bluetooth devices need to follow the communication rules (slave-master) and (master-bridge-master). In the proposed ERMP, when a main link is established, each routing Routing problem in the ROMA scatternet. Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), 4 S. Tahir et al. master and routing bridge update their tables and save the fallback devices’ information. During transmission, if a weak link or weak device is notified, the proposed protocol creates a new link to overcome the frequent link disconnection using fallback devices. When a fallback device is unavailable, the proposed protocol performs a role switch operation for route maintenance. 3.1. System model for a connected scatternet Suppose N is the total number of Bluetooth devices (slaves, masters, and bridges) in a scatternet as follows. N¼S[M[B ð1Þ  Pi ðSij 2 Mi Þ ¼ 1 Sij connects Mi 0 Sij does not connect Mi ð2Þ Subject to Sij 6 7; 8i ð3Þ Distance EDðSij ^ Mi Þ < 10 m ð4Þ where S, M, and B denote the number of slaves, masters, and bridges, respectively. Where Pi is the ith piconet and Mi corresponds to the piconet master, Sij is the jth slave device in the ith piconet. Sij is set to 1 if there is a master–slave relationship between device i and device j, otherwise it is set to 0. Constraint (3) determines if each piconet (P) has the maximum seven slave devices and that the maximum distance between the master and slave devices is 10 m in constraint (4). A connected scatternet has intermediate devices (bridge/ relay) to provide communication in multiple piconets. Eq. (5) represents an intermediate device relay, where Pi and Pj are any two piconets and nb is an intermediate device between these piconets. Pi \ Pj ¼ nb where i – j ð5Þ In each piconet, the routing master maintains a Routing Master Information Table (RMIT) that contains the list of slave devices, clock offset, device ID, device status (active/inactive), energy level, signal strength, FallBack Masters, and FallBack Bridges. On the other hand, in a scatternet, each routing bridge maintains a Routing Bridge Information Table (RBIT) that contains the list of connected master devices, device status (active/inactive), signal strength, energy level, FallBack Bridges, FallBack Masters and master’s degree. (  k  1 for Rk1 connected mj Degree R1 ; Mj ¼ ð6Þ null Rk1 no connection with mj where R is a relay and k is the degree of relay, each relay with a connected master (M) has a value of 1 and null otherwise. The flow diagram of the proposed ERMP is shown in Fig. 4, where three colors are used, the gray color shows that both master and slave execute these steps, while, dark colored steps represent a master device and lighter colored steps represent a slave device. 3.2. Route maintenance protocol The proposed ERMP protocol is implemented in the dynamic Bluetooth network that forecasts the mobility of devices to improve the network stability, where devices can move any time with different speed and direction. Therefore, the Random Walk Mobility Model (RWMM) is used for the proposed protocol because it considers the speed and direction of the devices. RWMM is a simple mobility model based on the random directions and speeds from a predefined range (Kuo Hsing Chiang, 2004). The mobile device can move from its present position to a new position by randomly selecting the directions and speed. From Eq. (7), the time is divided into the time slots t = 0, 1, 2. . ., where at t = 0, the device can choose one of four adjacent units {(x + 1, y), (x  1, y), (x, y + 1), (x, y  1)} with the same probability of 1/4. This procedure is repeated in every subsequent time period. The position of a device at timeslot t = 0, 1, 2. . . is denoted by C(n) (t) that basically shows the unit where the device exists. ðx; yÞ; x; y 2 f0; 1; 2 . . . n  1g ð7Þ 3.2.1. Mobility-base link replacement Before starting communication, a routing link is established between intermediate devices, the device participating in the route are called routing devices. A routing link within a scatternet passes through multiple piconets; therefore, the routing link consists of several routing master devices. A piconet that holds the routing master device is known as the Bluetooth routing inter-piconet. Whenever a new device enters into the routing piconet it sends its complete information to the routing master. Each routing master and routing bridge device maintains their tables. The masters monitor routing devices Signal-to-Noise Ratio (SNR). The received power signal strength can be utilized to estimate the distance, as all the electromagnetic waves show the inverse square relationship between the distance and received power. The relationship between the distance and received power is as follows. Pr 1 1=d2 ð8Þ The received power between the transmitter and receiver is indicated by Pr and d is the distance. During transmission, a weak link is notified if any intermediate device between any pair of source and destination starts moving. When an intermediate device starts moving away from a piconet the routing master observes its weak signal strength because the device is going out of radio range. Eqs. (9) and (10) are used for calculating signal strength. PRe ¼ PTr  GRe  GTr  ð1=ð4  s  dÞÞ2 ð9Þ where PTr is transmit power, GRe is receiver antenna gain and GTr is transmitter antenna gain. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi d ¼ ðx2  x1Þ2 þ ðy2  y1Þ2 ð10Þ where d is the distance between the master and moving device and x and y are the current position of the devices. In the proposed protocol, each routing master device maintains its table in which it keeps the information of the signal strength (from the master device to the connected slave device), device status, energy level, clock off-set and the role of the device (FallBack master, FallBack Bridge). The routing bridge device keeps the information of the device ID, device status, signal strength (from the bridge device to connected master devices), energy level and the role of the device (FallBack master, FallBack Bridge) in its table. A threshold q = 45 db is defined for the signal strength, which is fixed for all routing Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), Route Maintenance Protocol for Dynamic Bluetooth Networks 5 Standby Inqury Role? Master Page Slave Inquiry Scan Page Scan Connect Master Slave Add (ID, Role) Dst? RRP RMIT Role? Master Bridge BRIT ρ<-45 db OR Ɵ=L1 FBD FBD? No Role Switch Yes Update RMIT Connect Transmission End Figure 4 System flow diagram. devices. If any intermediate device in a routing link starts moving, the signal strength between the routing master and the moving device gradually decreases. The routing master can easily predict a weak signal strength as the device is moving and it can choose an appropriate routing device to make a new link before the main route breaks. When a routing bridge device starts moving, the routing master device checks for a FallBack Device (FBD); the routing master device activates the FBD and maintains the routing link. If the FBD is not available, the routing master device chooses another device (that may be a slave device) and changes the role of the device as required. The master insures that the new device must be within the radio range of both the devices so it can maintain the routing link efficiently. A routing link between two directly connected devices is considered weak if the devices have a weak signal strength. Fig. 5 shows different links between two devices. Any routing device can be weak, slave, master, or bridge. If a device becomes weak multiple devices are available as FBDs, the highest priority is given to a bridge because it can connect multiple piconets. The second priority is given to a master as it can connect multiple slave devices within a piconet. A slave has the lowest priority due to its limited role in the network. Figure 5 Different links between source and destination. The proposed ERMP is explained through Fig. 6(a). The routing master M3 stores the active devices’ information in RMIT as listed in Table 1. On the other hand, B2 is a routing bridge and stores connected master inform in RBIT as listed in Table 2. Source A and destination C are communicating through the route A–M1–B2–M2–B4–M5–C and the second link is between source D and destination E which are communicating through the route D–M4–B3–M2–B6–M3–E. During transmission between A and C, the bridge device B2 starts moving upward, in the meanwhile the routing masters M1 and M2 predict that the bridge device B2 is going out of range. Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), 6 S. Tahir et al. Figure 6 Table 1 Device ID E F G H B5 B6 Table 2 Device ID M1 M2 Before and after link replacement. Routing Master Information Table (RMIT) for M3. Clock off-set c-offset(E) c-offset(F) c-offset(G) c-offset(H) c-offset(B5) c-offset(B6) Device status (active/inactive) Signal strength (dBm) Energy level Device role Active Inactive Inactive Inactive Inactive Active StM3-E = 70 StM3-F = 65 StM3-G = 76 StM3-H = 66 StM3-B5 = 60 StM3-B6 = 60 L3 L2 L2 L3 L1 L3 Slave Slave Slave Slave Bridge FBM (M/S bridge) Routing Bridge Information Table (RBIT) for B2. Device status (active/in active) Signal strength (dBm) Active Active StM1-B2 StM2-B2 = 45 = 50 Table 3 Figure 7 (a) Prediction of weak link and weak device. (b) Maintenance role switch operation. Energy level Device role L2 L2 Master Master Simulation parameters. Parameters Values Simulation area Number of devices Number of pairs Data packet type Traffic model Communication range Energy consumption Power class Device deployment Mobility model Mobility speed Polling algorithm Bridge scheduling algorithm Packet size Inquiry time Page time Packet interval Queue length Simulation time 80 m  80 m 15–90 84 DH3, DH5 CBR 10 m 0.0763  106 J/bit B Random deployment Random Walk Mobility Model 0.5–3.0 m/s Round Robin MDRP 100–500 bytes 10.24 s 128–256 s 0.1–0.5 s 50 packets 1000 s Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), Route Maintenance Protocol for Dynamic Bluetooth Networks The masters execute the device leaving procedure and find out which device can replace the weak link. As B7 is a potential FBD, that can replace B2, therefore, both masters execute the replacement procedure and transmit a request to B7 to make a new link between M1 and M2. The master M1 transmits an ID packet along with a Device Access Code (DAC) repeatedly until it receives a response. The master M1 does not know on which hop frequency B7 will wake up so that is why it broadcasts a train of the same DACs on different hop frequencies and listens for a reply. The master M1 utilizes the BD_Addr and the clock of the B7 device to obtain the page hopping sequence. B7 performs the page scan that is similar to the inquiry scan in which a Bluetooth device listens and replies. There exist 32 paging frequencies including a page hopping sequence which is obtained by the BD_Addr of the master. The B7 changes listening frequencies after 1.28 s. The master M1 freezes its predictable B7 clock to the value that triggered a reply from the paged device. It is equal to using the clock values estimation when receiving the B7 response. The frozen clock value is used at the content where the recipient’s access code is identified. Let N be a counter that starts from zero and increases by one for each time when CLKN1 is set to zero which matches the start of a master TX slot. Finally, B7 reconstructs the sub route by replacing the weak link. As shown in Fig. 6(b), the new link between M1 and M2 is established and the data will go through the new route. ROMA 3.2.2. Energy-base link replacement University of Cincinnati Bluetooth (UCBT) presents four levels of energy, i.e., L0, L1, L2 and L3. The proposed ERMP defines a threshold (h) value for the energy level. A device is called a weak device, if the energy level of any device gets to L1. The master device selects a FBD to create a new link. During the route request each device includes its available battery level to the route request packet and forwards it to the next hop. This procedure carries on until the route request packet arrives at the final destination. Each master device calculates the master’s average power MPavg using Eqs. (11) or (12) as below: MPavg ¼ iTx  1=2Tsniff þ iRx  1=2Tsniff V Tsniff where Tx and Rx are the transmission and receive slots. The voltage of the Bluetooth’s specific chip is indicated by V and Tsniff is the sniff mode interval. The slave device’s Rx and Tx average power is calculated as below: SPRx ¼ SPTx ¼ ðNsniff  tslot Þ  iRx V Tsniff attempt iTx  tslot þ ððNsniff 50 40 30 20 10 0 18 24 30 36 attempt ð13Þ  1Þ  tslot Þ  iTx Tsniff idle V ð14Þ DENM 12 ð12Þ ERMP 60 6 ð11Þ MPavg ¼ 1=2ðiTx þ iRx Þ  V 42 48 54 60 Successfully repeaired links (n) Successfully repaired links (n) DENM 7 ROMA ERMP 60 50 40 30 20 10 0 100 200 300 400 500 Packet size (bytes) Number of failed/weak links (n) (a) (b) Sucessfully repaired links (n) DENM ROMA ERMP 50 40 30 20 10 0 0.1 0.2 0.3 0.4 0.5 Pakcet transmission interval (s) (c) Figure 8 interval. (a) Repaired links vs. number of failed/weak links. (b) Repaired links vs. packet size. (c) Repaired links vs. packet transmission Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), 8 S. Tahir et al. where iTx is a task slave device when it transmits and iTx_idle is a task of a slave device when in idle mode or not transmitting. The routing links are defined for the pairs of source and destination devices, where the source device initiates the route request for the destination device. The source device forwards a RSP to the master device and the master device forwards the received packet to the connected bridges. Once the destination device receives the RSP, it sends the unicast RRP to the source device. As the energy level of the intermediate routing device is gradually decreasing, when it reaches the specified limit L1, the routing master updates its status as a weak device and chooses another appropriate FBD as given in Eq. (15). If data are forwarded from the bridge device to the weak master device and the master device has to forward the data to the slave or bridge device, then a weak master device will check if the next device is within the radio range of the bridge device or not. If the next device is within the radio range of the bridge device, the weak master device sends a request message to the bridge device to switch the role from bridge to master/slave bridge and make a direct link to the next device before the weak master device dies. Finally, the routing master activates a new link and forwards the data through the new link. FBD ¼ FBDl # Rk routing link between M1 and M2 becomes weak. In this case, when a master device itself becomes weak, it first of all finds a device that will become the master device; so, from its table, it selects device C and sends a request for the role switch operation from slave to master. Now, device C becomes a new master device so it updates its table. The master device C fines device D within its radio range and also within the range of M2; so, device D works as a bridge between these piconets. For route maintenance, devices C and D make new links. In Fig. 7(b), the new routing link is A-C-D-M2-E and data go through the new route. 4. Simulation results and discussion This section evaluates the simulation results by comparing numerous performance metrics: healing delay, control packet overhead, blocking connections, route recovery time and slot utilization. The proposed and base protocols implemented UCBT (Agrawal and Wang, 2007) based on The NS-2 Simulator, (2014). Table 3 shows the list of parameters (Sahoo et al., 2008) that were used in the simulation. The area where l and k represent the bridge degree ðnumber of connectionsÞ As shown in Fig. 7(a), devices M1 and M2 predict that device B starts moving from its position and it is also predicted that the energy level of M1 is gradually decreasing so the Throughput (kbps) DENM 1000 900 800 700 600 500 400 300 200 100 0 0 of 80 m  80 m was taken in which the Bluetooth devices were scattered randomly. The Bluetooth devices were varied between 15 and 90, and 84 pairs of source and destination ROMA 500 ð15Þ 1000 ERMP 1500 2000 Time (s) (a) ROMA ERMP 1000 900 800 700 600 500 400 300 200 100 0 100 Figure 9 DENM Throughput (kbps) Throughput (kbps) DENM 200 300 400 500 ROMA ERMP 1000 900 800 700 600 500 400 300 200 100 0 0.1 0.2 0.3 0.4 Packet size (bytes) Packet transmission interval (s) (b) (c) 0.5 (a) Throughput vs. time. (b) Throughput vs. packet size. (c) Throughput vs. packet transmission interval. Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), Route Maintenance Protocol for Dynamic Bluetooth Networks devices were selected. The numbers of nodes and links are considered as a realistic use of Bluetooth in our daily life for typical applications Emre and Oznur, 2006). The RWMM was used as a mobility model. The Round Robin (RR) algorithm was used for scheduling. A CBR traffic model was taken to generate the data traffic for each routing link. The whole simulation run time was set to 1000 s. The simulation was run ten times and the results were obtained by averaging those tentime simulations. Each parameter was evaluated against the simulation time, packet size and packet interval. Packet interval is a small pause between packet transmissions; it is analyzed that the packet interval seriously affects simulation results. A long packet interval may reduce throughput and slot utilization. On the other hand, a short packet interval may increase healing delay and increase control packet overhead. Therefore, the packet interval should be wisely selected. After getting the comparison results, it was analyzed that the proposed ERMP protocol outperformed the existing protocols. The successfully repaired links demonstrated how many failed links had been repaired. The proposed ERMP was compared with the previous DENM and ROMA protocols and it is observed from Fig. 8, that the proposed ERMP successfully repaired more links as compared to the ROMA and DENM protocols. The proposed ERMP predicted the weak links when the distance increased between the connected routing devices. Thus to overcome the link disconnection problem, the ERMP dynamically activated a new link before the main link disconnected. An example, shown in Fig. 3(b), shows how the ERMP repaired the weak routing links before the main route breakage. Once the main route broke, the devices started to search Contorl overhead (packets) DENM 9 for a new route through inquiry and the inquiry procedure; thus, unnecessary resources were consumed by the system. ROMA did not provide a solution for the route recovery on run time when an active routing link disconnected. It started a new link process from the inquiry scan which consumed more time. On the other hand, the DENM protocol just predicted the energy level of the devices. When it predicted a lower energy level, it activated a backup device but it did not provide any solution when a link broke because of mobility. Therefore, DENM has repaired fewer numbers of routes compared to ERMP. Throughput can be defined as the successful delivery of data packets per unit time across the network. It is observed from Fig. 9, that the overall throughput of the proposed ERMP was higher than the ROMA and DENM protocols. During the simulation, the same numbers of data packets were routed across the network using the ERMP, ROMA and DENM protocols. It was noted that the total numbers of received packets through the ERMP were more than the other protocols because it prevented the main routing link being broken and it always maintained the network transmission by predicting the status of the routing devices. Whereas, the throughput of the ROMA protocol was less than the proposed protocol, because once the main links broke it created all new links for the existing devices. Thus, ROMA increased the number of piconets which degraded the network performance. It was also analyzed that the throughput of the DENM protocol was also less than the ERMP. Due to the mobility; if the main link broke, it did not perform any action so it could not receive all of the data packets. Therefore, the proposed ERMP is conROMA ERMP 1200 1000 800 600 400 200 0 0 500 1000 1500 2000 Time (s) (a) ROMA 1100 1000 900 800 700 600 500 400 300 200 100 0 100 200 300 400 Packet size (bytes) (b) Figure 10 DENM ERMP Control overhead (packets) Control overhead (Packets) DENM 500 ROMA ERMP 1100 1000 900 800 700 600 500 400 300 200 100 0 0.1 0.2 0.3 0.4 0.5 Packet transmission interval (s) (c) (a) Control overhead vs. time. (b) Control overhead vs. packet size. (c) Control overhead vs. packet transmission interval. Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), 10 S. Tahir et al. DENM ROMA ERMP 100 Healing delay (s) 90 80 70 60 50 40 30 20 10 0 0 500 1000 1500 2000 Time (s) (a) DENM ROMA DENM ERMP ROMA ERMP 100 Healing delay (s) Healing delay (s) 100 80 60 40 20 0 60 40 20 0 100 200 300 400 500 Packet size (bytes) (b) Figure 11 80 0.1 0.2 0.3 0.4 0.5 Packet transmission interval (s) (c) (a) Healing delay vs. time. (b) Healing delay vs. packet size. (c) Healing delay vs. packet transmission interval. sidered as a more efficient protocol than the previous ROMA and DENM protocols. Bluetooth devices use control packets for transmission, synchronization and connection. It has been analyzed that due to limited battery power and mobility, Bluetooth frequently performs network restructuring consuming extra control packets that ultimately degrades the network performance. Therefore, if the battery power and mobility of a device is predicted, it can reduce frequent link disconnection that can ultimately reduce control overhead. Control message overhead is calculated as, the sum of MAC, NULL, bb and POLL. From Fig. 10, it is observed that the proposed ERMP has less control packet overhead compared to the DENM and ROMA protocols because both protocols did not perform network restructuring for route maintenance. Whereas, the proposed protocol reserved the information of the FBDs and whenever a weak link or weak device was predicted, it activated the FBD and saved the extra resource utilization. A device was selected as a backup device if it was connected with the same piconets and could support the disconnected links. Although ROMA is a route maintenance protocol, it changes the overall network structure which creates unnecessary control overhead. It can be analyzed that the proposed ERMP control packet overhead is the least in all three protocols. Fig. 11 shows the healing delay of the protocols. After getting the simulation results, it was observed that the proposed ERMP consumed the minimum route recovery time as compared to the ROMA and DENM protocols. The reason behind this was that the proposed ERMP is based on the prediction of weak links and weak devices and it always keeps the information of the FBDs for the route maintenance on run time. As the existing protocols start a new link procedure form inquiry and the inquiry scan procedure, which take 10.28 s. While, the proposed ERMP repairs the weak links, it starts from the paging process that takes the maximum time of 2.56 s for each link. The DENM protocol takes more time as compared to the ERMP because it predicts the devices based on the energy level; if a device having a critical energy level is predicted, it activates a backup device. When it repairs the links by activating the backup device, it starts from the paging process that takes 2.56 s for each link. In case a link fails due to the mobility, it starts to re-establish the link and performs the inquiry, inquiry scan, page and page scan processes; that takes a longer time. The DENM route recovery for each link takes (2.56 s + 10.24 s), therefore, it takes more time compared to the ERMP. On the contrary, the ROMA requires a longer healing time for route maintenance because it maintains the routing link if any device joins or leaves the network, however, it does not perform any action if suddenly a link breaks or device fails. ROMA reconstructs the routing paths if the communication between the networks stops; it also starts from the inquiry processes. Therefore, the ROMA also increases healing delay compared to ERMP. New connections were blocked when the numbers of requests increased or active links were disturbed. From Fig. 12, it is observed that the ROMA and DENM protocols blocked more connections. As DENM only considers the low energy level of a device, it did not provide the solution if any routing device moved away and broke the link. When devices were disconnected due to mobility between the main Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), Route Maintenance Protocol for Dynamic Bluetooth Networks DENM 11 ROMA ERMP Blocking connections (%) 20 18 16 14 12 10 8 6 4 2 0 0 500 1000 1500 2000 Time (s) (a) ROMA ERMP DENM ERMP 8 6 4 2 0 100 200 300 400 500 0.1 Packet size (bytes) 0.2 0.3 0.4 0.5 Packet transmission interval (s) (b) Figure 12 interval. ROMA 10 Blocking connections (%) Blocking connections (%) DENM 20 18 16 14 12 10 8 6 4 2 0 (c) (a) Blocking connections vs. time. (b) Blocking connections vs. packet size. (c) Blocking connections vs. packet transmission routing links, the DENM blocked more connections. On the contrary, ROMA did not provide the best solution for broken links. Mostly, source and intermediate devices were unable to forward the data to the destination device due to a link failure. It has been analyzed that the proposed ERMP reduced the blocking percentage compared to the previous DENM and ROMA protocols. It has also been observed that as the number of devices increased it also increased the network size, which increased the path length. For the longer path, more numbers of intermediate devices were used, which increased the chances of link breakage. It is analyzed from Fig. 13 that the overall slot utilization of the proposed ERMP was higher than the ROMA and DENM protocols. During the simulation, the same numbers of data packets were routed across the network using the ERMP, ROMA and DENM protocols. It was noted that the total number of received packets through the ERMP was more than the other protocols because it prevented the main routing link from being broken and it always maintained the network transmission by predicting the status of the routing devices. Whereas, the slot utilization of the ROMA protocol was less than the proposed protocol, because once the main links broke, it created all new links for the existing devices. Thus, ROMA increased the number of piconets which degraded the network performance. It was also analyzed that the slot utilization of the DENM protocol was also less than the ERMP. Due to the mobility, if the main link broke, it did not perform any action so it could not utilize available slots. The proposed ERMP is considered as more efficient in terms of slot utilization due to route maintenance functionality compared to ROMA and DENM. 5. Conclusion In this paper, an Efficient Route Maintenance (ERMP) Protocol is proposed for dynamic Bluetooth networks. During the main route construction each routing master and routing bridge device keeps the required information of all the connected devices. The proposed protocol predicts the weak links and weak devices through the signal strength and energy level of the devices and repairs the weak routing path by activating the FBDs. Our analysis showed that ERMP is more efficient compared to ROMA and DENM because both the protocols provide a solution for route breakage only when the main route is broken. On the contrary, the proposed protocol predicts the weak routing links and the weak devices, if a weak routing link is notified the proposed protocol introduces a new link before the main route is damaged. The simulation results reveal that the ERMP outperforms the existing protocols in terms of repairing connections, healing delay, control overhead, blocking connections and slot utilization. Every research work has a future direction, which provides the opportunity for future research continuation in the field. The proposed ERMP can be extended in future to select a Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), 12 S. Tahir et al. DENM ROMA ERMP 100 Slot utilization (%) 90 80 70 60 50 40 30 20 10 0 0 500 1000 1500 2000 Time (s) (a) DENM ROMA DENM ERMP Slot utilization (%) Slot utilization (%) 100 80 60 40 20 0 ERMP 80 60 40 20 0 100 200 300 400 500 Packet size (bytes) (b) Figure 13 ROMA 100 0.1 0.2 0.3 0.4 0.5 Packet transmission interval (s) (c) (a) Slot utilization vs. time. (b) Slot utilization vs. packet size. (c) Slot utilization vs. packet transmission interval. relay based on relay mobility. The plan is to enhance the ERMP in such a way that a backup device can be adjusted in a piconet where it is frequently communicating. This is necessary to support the problem of link disconnection due to user mobility. References Agrawal, D., Wang, Q., 2007. University of Cinicinnati Bluetooth Simulator (UCBT), Online: . Aldabbagh, G. et al, 2015a. QoS-Aware Tethering in a heterogeneous wireless network using LTE and TV white spaces. Comput. Netw. 81, 136–146. Aldabbagh, G. et al, 2015b. Distributed dynamic load balancing in a heterogeneous network using LTE and TV white spaces. Wirel. Netw., 1–12 Bakhsh, S.T. et al, 2011. Dynamic energy-aware network maintenance for Bluetooth. In: International Conference of Information Science and Applications. Bo, H. et al, 2012. Mobile data offloading through opportunistic communications and social participation. IEEE Trans. Mob. Comput. 11 (5), 821–834. Chih-Min, Y., Yin-Bin, Y., 2014. Reconfigurable algorithm for bluetooth sensor networks. Sens. J. IEEE 14 (10), 3506–3507. Ching-Fang, H., Shu-Ming, H., 2008. An adaptive interpiconet scheduling algorithm based on HOLD mode in bluetooth scatternets. IEEE Trans. Veh. Technol. 57 (1), 475–489. Cui, Y.-D. et al, 2010. Bluetooth energy-saving optimization: a queuing theory analysis. J. Chin. Univ. Posts Telecommun. 17 (Suppl. 1), 54–57. Emre, A., Oznur, O., 2006. A classification and performance comparison of mobility models for Ad Hoc networks. In: ADHOC-NOW. Springer, Istanbul, Turkey, pp. 444–457. Etxaniz, J., Aranguren, G., 2015. Modeling of the data transportation network of a multi-hop data-content-sharing home network. IEEE Trans. Consum. Electron. 61 (1), 31–38. Jin-Ho, C., Guang, G., Kyeongcheol, Y., 2014. New families of optimal frequency-hopping sequences of composite lengths. IEEE Trans. Inf. Theory 60 (6), 3688–3697. Kuo Hsing Chiang, 2004. A 2D Random Walk Mobility Model for Location Management Studies in Wireless Networks. Laharotte, P.A. et al, 2015. Spatiotemporal analysis of bluetooth data: application to a large urban network. IEEE Trans. Intell. Transp. Syst. 16 (3), 1439–1448. Ramana Reddy, G. et al, 2010. An efficient algorithm for scheduling in bluetooth piconets and scatternets. Wirel. Netw. 16 (7), 1799–1816. Sahoo, P.K., Chang, C.-Y., Chang, S.-W., 2008. Novel route maintenance protocols for the Bluetooth ad hoc network with mobility. J. Netw. Comput. Appl. 31 (4), 535–558. Sharafeddine, S., Al-Kassem, I., Dawy, Z., 2012. A scatternet formation algorithm for bluetooth networks with a non-uniform distribution of devices. J. Netw. Comput. Appl. 35 (2), 644–656. Singh, P., Agrawal, S., 2011. A servey on bluetooth scatternet formation. In: Wyld, D. et al. (Eds.), Advances in Computing and Information Technology. Springer, Berlin Heidelberg, pp. 260– 269. Subhan, F. et al, 2011. Handover in bluetooth networks using signal parameters. Inf. Technol. J., 965–973 Subhan, F. et al, 2012. Analysis of bluetooth signal parameters for indoor positioning systems. In: International Conference on Computer & Information Science (ICCIS), pp. 784–789. Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016), Route Maintenance Protocol for Dynamic Bluetooth Networks Tahir, S. et al, 2013. Hybrid congestion sharing and route repairing protocol for bluetooth networks. In: International Conference on Electronics, Signal Processing and Communication Systems, pp. 33–39. The Network Simulator (NS-2), 2014. Online: . Xin, Y., YuPing, W., 2009. Secure constructing bluetooth scatternet based on BTCP. In: Information Assurance and Security, 2009. IAS ‘09. Fifth International Conference on. 13 Yu, C.-M., 2010. A study on the topology control method for bluetooth scatternet formation. In: Pan, J.-S., Chen, S.-M., Nguyen, N. (Eds.), Computational Collective Intelligence. Technologies and Applications. Springer, Berlin Heidelberg, pp. 263–271. Yu, C.M., 2012. Global configured method for blueweb routing protocol. IET Commun. 6 (1), 69–75. Yu, C.M., Lin, J.H., 2012. Enhanced bluetree: a mesh topology approach forming bluetooth scatternet. IET Wirel. Sens. Syst. 2 (4), 409–415. Please cite this article in press as: Tahir, S. et al., An Efficient Route Maintenance Protocol for Dynamic Bluetooth Networks. Journal of King Saud University – Computer and Information Sciences (2016),
Ad Hoc Networks 2 (2004) 185–202 Ad hoc networking with Bluetooth: key metrics and distributed protocols for scatternet formation Tommaso Melodia *, Francesca Cuomo INFOCOM Department, University of Rome ‘‘La Sapienza’’, Via Eudossiana 18, 00184 Rome, Italy Received 13 June 2003; received in revised form 29 July 2003; accepted 31 July 2003 Abstract Bluetooth is a promising technology for personal/local area wireless communications. A Bluetooth scatternet is composed of simple overlapping piconets, each with a low number of devices sharing the same radio channel. A scatternet may have different topological configurations, depending on the number of composing piconets, the role of the devices involved and the configuration of the links. This paper discusses the scatternet formation issue by analyzing topological characteristics of the scatternet formed. A matrix-based representation of the network topology is used to define metrics that are applied to evaluate the key cost parameters and the scatternet performance. Numerical examples are presented and discussed, highlighting the impact of metric selection on scatternet performance. Then, a distributed algorithm for scatternet topology optimization is introduced, that supports the formation of a ‘‘locally optimal’’ scatternet based on a selected metric. Numerical results obtained by adopting this distributed approach to ‘‘optimize’’ the network topology are shown to be close to the global optimum.  2003 Elsevier B.V. All rights reserved. Keywords: Ad hoc networks; Bluetooth; Scatternet formation; Topology optimization; Distributed protocols 1. Introduction Bluetooth (BT) is a promising technology for ad hoc networking that could impact several wireless communication fields providing WPAN (Wireless Personal Area Networks) extensions of public radio networks (e.g., GPRS, UMTS, Internet) or of local area ones (e.g. 802.11 WLANs, Home RF) [1,2]. The BT system is described in the Bluetooth Specifications 1.1 [3] and supports a * Corresponding author. E-mail addresses: (T. Melodia), (F. Cuomo). 1 Mbit/s gross rate in a so-called piconet, where up to 8 devices can simultaneously be interconnected. The radius of a piconet (transmission range––TR) is about 10 m for Class 3 devices. A BT based standard has been released by the IEEE 802.15, which also addresses coexistence with the 802.11 wireless LAN technology, in the un-licensed 2.4 GHz ISM (Industrial, Scientific and Medical) band [4]. One of the key issues associated with the BT technology is the possibility of dynamically settingup and tearing down piconets. Devices (named also nodes in the following) can join and leave piconets. Different piconets can coexist by sharing the spectrum with different frequency hopping sequences, and interconnect in a scatternet. When all nodes 1570-8705/$ - see front matter  2003 Elsevier B.V. All rights reserved. doi:10.1016/S1570-8705(03)00054-4 186 T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 are in radio visibility, scenario which we will refer to as single hop, the formation of overlapping piconets allows more than 8 nodes to contemporary communicate and may enhance system capacity. In a multi-hop scenario, where nodes are not all in radio vicinity, a scatternet is mandatory to develop a connected platform for ad-hoc networking. This paper addresses the scatternet formation issue by considering topological properties that affect the performance of the system. Most works in literature aim at forming a connected scatternet while performance related topological issues typically remain un-addressed. To this aim we introduce a matrix based scatternet representation that is used to define metrics and to evaluate the relevant performance. We then propose a distributed algorithm that performs topology optimization by relying on the previously introduced metrics. We conclude by describing a two-phases scatternet formation algorithm based on the optimization algorithm. To the best of our knowledge, this is the first scatternet formation algorithm explicitly aimed at optimizing network topology. The paper is organized as follows. Section 3 briefly summarizes the state of the art in scatternet formation, while in Section 4 we present a framework for scatternet analysis, based on a matrix representation of the scatternet. Section 5 presents some metrics that can be used to evaluate a scatternet; related numerical results are shown in Section 6. Section 7 presents the Distributed Scatternet Optimization Algorithm (DSOA) while Section 8 describes a two-phase scatternet formation algorithm based on DSOA. Section 9 concludes the paper. slave exchange data by hopping at a frequency of 1600 hops/second on the 79 available channels. Different hopping sequences are associated to different masters. A master can connect with up to 7 slaves within a piconet. Devices belonging to the same piconet share a 1 Mbit/s radio channel and use the same frequency hopping sequence. Only communications between master and slaves are permitted. Time is slotted and the master, by means of a polling mechanism, centrally regulates the access to the medium. Thanks to the FHSS, which is robust against interference, multiple piconets can co-exist in the same area. Considerable performance degradation only occurs for a high number of co-located piconets (in the order of 50) [5]. To overcome the limits imposed by the low number of devices that can simultaneously communicate (up to 8) and by the channel capacity (less than 1 Mbit/s) in a piconet, the BT specifications introduced the concept of ‘‘scatternet’’, defined as an interconnection of overlapping piconets. Each device can join more than one piconet, and participates to communications in different piconets on a time-division basis. Devices that belong to more that one piconet are called gateways or BridGing units (BG). In Fig. 1 we show an example of scatternet, composed of 8 devices organized in 3 piconets. Devices number 1, 5 and 6 are masters; devices 4 and 7 are slaves in two different piconets. Thus, they act as BGs between them, i.e., they can forward traffic from and to devices belonging to the two different piconets. 2. From piconets to scatternets BT exploits an 83.5 MHz band, divided into 79 equally spaced 1 MHz channels [1,3]. The multiple access technique is the FHSS–TDD (frequency hopping spread spectrum–time division duplexing). Two BT units exchange information by means of a master-slave relationship. Master and slave roles are dynamic: the device that starts the communication acts as master, the other one as slave. After connection establishment, master and Fig. 1. An example of scatternet made up of 3 piconets. T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 It is easy to see that there are many topological alternatives to form a scatternet out of the same group of devices. The way a scatternet is formed considerably affects its performance. 3. Related works Scatternet formation in BT has recently received a significant attention in the scientific literature. Existing works can be classified as single-hop [6–9] and multi-hop solutions [10–15]. Paper [6] addresses the BT scatternet formation with a distributed logic that selects a leader node which subsequently assigns roles to the other nodes in the system. The scheme works for a number of nodes 6 36. In [7] a distributed formation protocol is defined, with the goal of reducing formation time and message complexity. In both [7,8], the resulting scatternet has a number of piconets close to the theoretical minimum. The works in [9–11] form tree shaped scatternets. The tree structure is shown to be simple to realize and efficient for packet scheduling and routing. The work [9], by Tan et al., presents the TSF (Tree Scatternet Formation) protocol. The topology of the scatternet is a collection of one or more rooted spanning trees, each autonomously attempting to merge and converge to a topology with a smaller number of trees. TSF assures connectivity only in single-hop scenarios since trees can merge only if their root nodes are in transmission range of each other. Zaruba et al. propose a protocol which operates also in a multi-hop environment [10]. This latter protocol is based on a process that is initiated by a unique node (named blueroot) and repeated recursively till the ‘‘leaves’’ of the tree are reached. In order to operate in a distributed way and to avoid deadlocks the algorithm is based on time-outs that could affect the formation time. SHAPER [11] also forms tree-shaped scatternets, but is fully distributed, works in a multi-hop setting, has very limited formation time and assures self-healing properties of the network, i.e. nodes can enter and leave the network at any time without causing loss of connectivity. The Bluenet protocol in [12] forms a scatternet which has reasonably good connectivity. 187 A second class of multi-hop proposals is concerned with algorithms based on clustering schemes. These algorithms principally aim at forming connected scatternets. In [13,14] the BlueStars and BlueMesh protocols are described respectively. The BlueStars protocol has three phases: device discovery, partitioning of the network into Bluetooth piconets and interconnection of the piconets into a connected scatternet. It is executed at each node with no prior knowledge of the network topology, thus being fully distributed. The selection of the Bluetooth masters is driven by the suitability of a node to be the ‘‘best fit’’ for serving as a master. Finally, the generated scatternet is a connected mesh with multiple paths between any pair of nodes, which guarantees robustness. Simulation results are provided which evaluate the impact of the Bluetooth device discovery phase on the performance of the protocol. Also the protocol in BlueMesh forms scatternets without requiring the BT devices to be all in each otherÕs transmission range. BlueMesh scatternet topologies are meshes with multiple paths between any pair of nodes. BlueMesh piconets are made up of no more than 7 slaves. Simulation results in networks with over 200 nodes show that BlueMesh is effective in quickly generating a connected scatternet in which each node, on average, does not assume more than 2.4 roles. Moreover, the route length between any two nodes in the network is comparable to that of the shortest paths between the nodes. Also [15] defines a protocol that limits the number of slaves per master to 7 by applying the Yao degree reduction technique, assuming that each node knows its geographical position and that of each neighbor. Recently, the work in [16] proposed a new ondemand route discovery and construction approach which, however, requires substantial modifications to the Bluetooth standard to guarantee acceptable route-setup delay. Some other works discuss the optimization of the scatternet topology. This issue is faced in [17,20] by means of centralized approaches. In [17] the aim is minimizing the load of the most congested node in the network while [20] discusses the impact of different metrics on the scatternet topology. A distributed approach based on simple 188 T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 heuristics is presented in [21]. In [18], an analytical model of a scatternet based on queuing theory is introduced, aimed at determining the number of non-gateway and gateway slaves to guarantee acceptable delay characteristics. In this framework the objectives of our contribution are • provide a framework for scatternet topology analysis based on matrices which turns out to be a very simple and effective design tool; • identify metrics that can be used to form and to evaluate scatternets; we emphasize differences between traffic dependent metrics and traffic independent ones and we show selected numerical results; • to present the building blocks for the implementation of a distributed algorithm that optimizes the scatternet topology. 4. The scatternet formation issue Before addressing the issue of scatternet formation, we introduce a suitable scatternet representation. 4.1. Scatternet representation Let us consider a scenario with N devices. The scenario can be modeled as an undirected graph GðV ; EÞ, where V is the set of nodes and an edge eij , between any two nodes vi and vj , belongs to the set E iff distðvi ; vj Þ < TR, i.e., if vi and vj are within each otherÕs transmission range. GðV ; EÞ can be represented by a N  N adjacency matrix A ¼ ½aij , whose element aij equals 1 iff device j is in the TR of device i (i.e., j can directly receive the transmission of i). Besides the adjacency graph GðV ; EÞ, in accordance to, we use a bipartite graph GB ðVM ; VS ; LÞ, to model the scatternet, where jVM j ¼ M is the number of masters, jVS j ¼ S is the number of slaves, and L is the set of links (with N ¼ M þ S, VM \ VS ¼ f0g, VM \ VS ¼ V ). A link may exist between two nodes only if they belong to the two different sets VM and VS . Obviously, for any feasible scatternet, we have L E. This model is valid under the hypothesis that a master in a piconet does not assume the role of slave in another piconet; in other words, by adopting this model, the BGs are slaves in all the piconets they belong to. We rely on this hypothesis to slightly simplify the scatternet representation, the complexity in the description of the metrics and to reduce the space of possible topologies. Moreover, intuitively, the use of master/slave BGs can lead to losses in system efficiency. If the BG is also a master, no communications can occur in the piconet where it plays the role of master when it communicates as slave. However, to the best of our knowledge, this claim has never been proved to be true. Future work will thus extend the results presented in this paper to non-bipartite graphs. The bipartite graph GB can be represented by a rectangular M  S binary matrix B [19]. In B, each row is associated with one master and each column with one slave. Element bij in the matrix equals 1 iff slave j belongs to master iÕs piconet. The scatternet of Fig. 1 may be represented by the following matrix B (Eq. 1a). In addition, a path between a pair of nodes (h, k) can be represented by another M  S matrix Ph;k ðBÞ, whose element pijh;k equals 1 iff the link between master i and slave j is part of the path between node h and node k ð1 6 i; j; h; k 6 N Þ. As an example, and referring again to Fig. 1, the path between nodes 2 and 8 can be represented by the matrix P2;8 ðBÞ of Eq. (1b) 1 2 1 3 1 4 1 7 0 8 0 B= 5 0 0 1 1 0 6 0 0 0 1 1 1 (a) P 2,8 (B)= 5 6 2 1 3 0 4 1 7 0 8 0 0 0 1 1 0 0 0 0 1 1 : (b) ð1Þ Given a set V of N nodes, and an adjacency matrix A, we can build the M  S matrix B ¼ ½bij  by associating its rows to a VM non-empty subset of M nodes in V , and the columns to a VS non-empty subset of S nodes in V (with N ¼ M þ S, VM \ VS ¼ f0g and VM [ VS ¼ V ), and by selecting a subset of links in E. The resulting matrix B represents a ‘‘BT-compliant’’ scatternet with M masters and S slaves if the following properties apply: T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 1. Each PS master is connected at least to a slave, i ¼ 1; . . . ; M. j¼1 bij P 1; 2. No more than seven slaves belong to a piconet, PS b 6 7; i ¼ 1; . . . ; M. j¼1 ij 3. P Each slave is connected at least to a master, M j ¼ 1; . . . ; S. i¼1 bij P 1; 4. The resulting network is connected, the matrix B does not have a block structure, row permutations notwithstanding. 4.2. A framework for scatternet topology analysis By exploiting the above matrix representations, we are interested in (a) characterize the space of solutions; i.e., all the B matrices compliant with rules 1–4; (b) define metrics to evaluate the scatternet performance; (c) single out the optimal scatternet (B ) with respect to a selected metric; (d) analyze the topological properties of the extracted solutions. We stress that our objective is to enucleate scatternet characteristics related to specific metrics that could be adopted in the formation process. We do not aim at proposing sophisticated algorithms that elaborate the matrix A to derive the optimal scatternet, also because, due to the complexity of the problem, they probably shall operate in a centralized manner while scatternet formation should be solved by adopting distributed operations performed by all network nodes. To work out points (a) and (c) we rely on space state enumeration that pays the complexity of examining and listing all potential scatternets but on the other hand allows us to completely characterize the space of possible solutions. In the following we briefly describe our approach to go along steps a–d, while further details can be found in [20]. We randomly generate communication scenarios taking as inputs the number of devices, N , and the dimensions of the area where the nodes are located; the scenario is represented by the A matrix. We then identify and enumerate all the ‘‘BTcompliant’’ scatternets that may be obtained from the scenario; if we let M be the number of masters 189 in the scatternet, with Mmin 6 M 6 Mmax the number of possible choices for the network    PMmax of Nroles N nodes is equal to M¼M , since there are M M min possible ways of selecting M masters among N nodes. Each choice implies a set L0 of possible links (with L0 E) and we consider every subset L L0 which gives rise to a scatternet that respects properties 1–4. All these scatternets constitute our B matrices. We consider Mmin ¼ dN =ð7 þ 1Þe and Mmax ¼ bN =2c. A number of masters greater than half the nodes introduces inefficiencies (e.g., interference) without bringing benefits to the scatternet. The B matrices are evaluated by applying suitable metrics described in Section 5. For a given metric, the output of the overall process is the identification of the optimal B (indicated with B ), which represents the scatternet with the optimal topology, and a distribution of the metric values. 5. Metrics for scatternet evaluation Metrics for scatternet evaluation can either be dependent on or independent of the traffic loading the scatternet. In the traffic independent (TI) case, the scatternet is formed without a priori knowledge of traffic patterns among involved devices. The scenario is described only by means of the adjacency matrix A, without associating to possible pairs of devices a description of the related exchanged traffic. On the other hand, it may be useful to form a scatternet by taking into account traffic patterns, if such information is available. In that case, the traffic patterns can conveniently be described by a traffic matrix, T. In the following we refer to this case as traffic dependent (TD). In the following, we introduce several metrics; each of them has pros and cons. 5.1. TI metrics: scatternet with maximum capacity A first traffic independent metric is the overall capacity of the scatternet. Evaluating such a capacity is not an easy task, since it is related to the capacity of the composing piconets which in turn depends on the intra-piconet and interpiconet scheduling policies. To the best of our knowledge, 190 T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 no such evaluation is available in literature. In the following, we introduce a simple model to estimate the capacity of a scatternet and we exploit this evaluation for scatternet formation. In the model we assume that As an example, let us consider the scatternet of Fig. 1. The matrices OTI ðBÞ and RTI ðBÞ are • a master may offer the same amount of capacity to each of its slaves by equally partitioning the piconet capacity; • a BG slave spends the same time in any piconet it belongs to. ð5Þ These assumptions are tied to intra and inter piconet scheduling; here, for the sake of simplicity, we assume policies that equally divide resources; however the model can be straightforwardly extended to whatever scheduling policy. The scatternet capacity will be evaluated by normalizing its value to the overall capacity of a piconet (i.e., 1 Mbit/s). Let us define two M  S matrices, OTI ðBÞ ¼ ½oij , and RTI ðBÞ ¼ ½rij  with oij ¼ bij =si and rij ¼ bij =mj , where si denotes the number of slaves connected to master i and mj denotes the number of masters connected to slave j (for j ¼ 1; . . . ; S and i ¼ 1; . . . ; M): M X mj ¼ bij for j ¼ 1; . . . ; S; and the related matrix CTI ðBÞ is ð6Þ The resulting normalized capacity is cTI ðBÞ ¼ 3 ( ¼ 3 Mbit/s). Eq. (4) is valid in the ideal case where both interference from co-located piconets and switching overhead, caused by BGs that change piconet, are neglected. In order to include these two effects, Eq. (4) can be rewritten as cTI ðBÞ ¼ ðcTI ðBÞ  DðBÞÞ  IðBÞ; ð7Þ i¼1 si ¼ S X bij for i ¼ 1; . . . ; M: ð2Þ j¼1 The matrix OTI ðBÞ represents the portions of capacity a master may offer to each of its slaves. The RTI ðBÞ matrix represents the portions of capacity a slave may ‘‘spend’’ in the piconet it is connected to. The overall capacity of the scatternet is given by the sum of the capacities of all links. The capacity cij of link (i; j) is the minimum between the capacity oij and the capacity rij . Let us define the matrix CTI ðBÞ, whose elements represent the normalized link capacity, as CTI ðBÞ ¼ ½cij  ¼ ½minðoij ; rij Þ: ð3Þ The associated metric is the normalized capacity cTI ðBÞ of a scatternet defined as cTI ðBÞ ¼ M X S X i¼1 j¼1 minðoij ; rij Þ: ð4Þ where DðBÞ represents a loss of capacity due to the switching overhead and IðBÞ is a decreasing factor that accounts for interference from co-located piconets. 5.2. TD metrics: scatternet with maximum residual capacity or minimum average load We consider two traffic dependent metrics: (i) the so-called residual capacity (i.e., the capacity that remains available in a scatternet, after that all pre-defined traffic patterns are accommodated); (ii) the nodes’ average load. The evaluation of the above metrics is TD, and as such, is dependent on the adopted routing strategy too. As an example, given a traffic pattern, for instance a data flow between device h and device k (with 1 6 h, k 6 N ), the capacity that such flow requires from the overall scatternet depends on the number of hops that make up the path T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 between device h and device k. In our analysis, we assume without loss of generality, that a shortest path routing algorithm is adopted. To evaluate the metrics, we start by describing the traffic relationships with an N  N traffic matrix T ¼ ½thk , whose element thk represents the capacity, normalized with respect to the piconetÕs capacity, required by the (h; k) relationship, (1 6 h; k 6 N ). We assume that the thk (normalized traffic rates) are fixed for each source–destination couple. We also denote by R the number of traffic relationships expressed by this matrix. It is easy to see that the capacity required on each link by the traffic relationship between node h and node k is given by thk Ph;k ðBÞ. The matrix of the overall normalized capacity required for each link to transport the traffic patterns expressed by T is given by N N X X CTD ðBÞ ¼ thk  Ph;k ðBÞ h¼1 k¼1;k6¼h " ¼ dij ¼ N N X X # thk  pijh;k : ð8Þ h¼1 k¼1;k6¼h It is to be noted that the traffic relationships defined by the matrix T can effectively be supported by the scatternet represented by B if the following conditions, that assure the steady state, are verified: masters are not over-loaded ) S X dij 6 1 8i 2 M; ð9Þ j¼1 slaves are not over-loaded ) M X dij 6 1 8j 2 S: according to Eq. (3), the residual capacity matrix, is given by Cres;TD ðBÞ ¼ CTI ðBÞ  CTD ðBÞ ¼ bfij c: ð12Þ The information contained in this matrix can be summarized in a single parameter, the residual capacity, cres;TD ðBÞ, given by cres;TD ðBÞ ¼ M X S X i¼1 fij : ð13Þ j¼1 Also in this case considerations about decreasing factors due to interference and switching overhead can be applied. According to this metric, a scatternet is optimal, when the value of cres;TD ðBÞ is maximized. Alternatively, we can adopt as metric the nodes’ average normalized load. In accordancePto Eq. (8) the normalized load on M a slave j is lSj ¼ i¼1P dij , while the normalized load S on a master is lM j¼1 dij . i ¼ The normalized load averaged over the N nodes, which we denote as lðBÞ, is PS S PM M j¼1 lj þ i¼1 li lðBÞ ¼ N PM PS 2  i¼1 j¼1 dij 2  cTD ðBÞ : ð14Þ ¼ ¼ N N With this metric, the target is the minimization of lðBÞ; we point out that the minimization of the average load goes in the direction of a minimization of the average energy consumption of the scatternet. 5.3. Metrics associated to the path length ð10Þ i¼1 The total capacity required by the traffic relationships of matrix T is M X S X dij : ð11Þ cTD ðBÞ ¼ i¼1 191 j¼1 Based on the above definitions, we can finally introduce two TD metrics. The first measures the capacity that remains available in a scatternet, after all traffic patterns are accommodated. Recalling that the capacity of each link is assigned As will be shown in Section 6 path lengths have a considerable impact on scatternet performance. As a consequence, in this section, we define three metrics that do take into account path lengths. The first two are not dependent on the traffic loading the scatternet and are defined as follows. Let us denote, for a scatternet represented by a matrix B, the length of the path between device h and device k (expressed in number of hops) as qh;k ðBÞ ¼ M X S X i¼1 j¼1 pijh;k : ð15Þ 192 T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 The first metric that we introduce is the average path length, which is the path length averaged over all possible source–destination couples, and is given by qTI ðBÞ ¼ N N X X h¼1 k¼1;k6¼h qh;k ðBÞ ; N  ðN  1Þ ð16Þ the optimization target of this metric is the minimization of qTI ðBÞ. Given the capacity of a scatternet cTI ðBÞ and the relevant average path length qTI ðBÞ, on average the capacity available for the generic source–destination couple among the nodes in B is aTI ðBÞ ¼ cTI ðBÞ : qTI ðBÞ  N  ðN  1Þ ð17Þ We will call this last metric average path capacity. The last metric we introduce depends on traffic and thus it depends on the matrix T, which defines R traffic relationships. The metric is the path length averaged over the R traffic relationships, instead that over all possible N  ðN  1Þ relationships, as done in Eq. (15); it is given by qTD ðBÞ ¼ X ðh;kÞ2T ;h6¼k qh;k ðBÞ : R ð18Þ The associated target consists in minimizing this path length; this metric goes in the direction of minimizing the transfer delay related to the number of hops. Table 1 summarizes the metrics introduced and briefly reports the significance of the metric themselves. Our analysis does not aim at identifying all the possible metrics, however by following this methodology many other metrics can be identified, as metrics related to the behavior of specific nodes/ links in the network (i.e., minimize the energy consumption of the bottleneck node, maximize the minimal residual capacity among the links). The six metrics in Table 1 have been introduced in order to measure some significant performance parameters of the overall scatternet. In particular, while within TI metrics cTI ðBÞ simply measures the potential of the scatternet in terms of capacity, the other two are introduced to evaluate the scatternet by considering a generic source–destination couple obtained by averaging over all possible ones and by assuming uniform traffic patterns. As for the TD metrics, the residual capacity has been introduced to evaluate how the scatternet is able, in terms of capacity, to sustain adjunctive flows. The other two metrics, that are related to the path lengths, are in favor of scatternets that efficiently support the given traffic matrix T. If traffic patterns are balanced (i.e., all the traffic relationship require the same capacity) the optimization of the two metrics qTD ðBÞ and lTD ðBÞ gives rise to the same optimal scatternet B . 6. Numerical results In this Section, we present numerical results relevant to the metrics defined above obtained in accordance to Section 4.2. Each of the following figures represents the area containing the scatter- Table 1 A summary of the defined metrics Traffic independent Traffic dependent Normalized capacity cTI ðBÞ Average path length Average path capacity qTI ðBÞ aTI ðBÞ Residual capacity cres;TD ðBÞ Average path length qTD ðBÞ Average normalized load lTD ðBÞ Capacity of the overall scatternet normalized with respect to the capacity of a piconet Average number of hops of shortest paths between nodes Capacity available for the generic source–destination couple Residual capacity of the scatternet once traffic in T has been allocated Average number of hops of shortest paths between nodes for patterns in T Average load on nodes in the scatternet to support patterns in T T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 net; x and y axes are measured in meters. The TR is assumed to be equal to 10 m. Nodes are numbered according to the order they are generated in the communication scenario. Figures report nodes, their roles (master, slave or bridge) and radio links connecting them. Since number of possible scatternets quickly becomes overwhelming with respects to the number of nodes, in this section we analyze networks with a limited number of devices. 6.1. Traffic independent metrics This subsection shows examples of scatternets resulting from TI optimization. The scenario consists of 12 devices not all the nodes are in radio visibility. The scatternet in Fig. 2 is obtained by selecting the one with maximum normalized capacity. The interference effect in Eq. (7) is evaluated by relying on the model presented in [22]. The switching overhead, which depends on the adopted interpiconet scheduling policy [2], has been considered here with simple assumptions on the switching frequency. It can be immediately noticed that the scatternet presents a linear structure, i.e., every node is connected with two other nodes only. The value assumed by cTI ðBÞ, evaluated as in Eq. (4) is 5.5. When switching overhead is considered cTI ðBÞ decreases to 5.2727 and by considering also the 50 master slave bridge Node: 12 45 Node: 7 40 meters Node: 8 Node: 6 Node: 4 35 Node: 2 30 Node: 3 25 Node: 9 20 15 65 Node: 1 Node: 10 Node: 5 70 75 80 85 meters 90 Node: 11 95 100 Fig. 2. Scatternet with maximum normalized capacity. 193 interference effect it becomes 4.7151. Although this scatternet is the one with maximum normalized capacity, it has characteristics that make its topology undesirable: it presents large values of the average path length that could lead to high transfer delays, and, since for Class 3 devices no power control mechanisms have been defined, increasing the number of hops per traffic relationship does not bring any benefit in terms of power consumption and consumes capacity as function of the involved links. As an example, a single 500 kbit/s bi-directional flow between node 10 and node 12 in Fig. 2 would use all scatternet capacity: nodes along the path would spend half their time receiving traffic from one of the two directions and the remaining time relaying traffic in the opposite direction. The peculiar structure produced by this metric is due to the following reasons: • The metric tends to favor scatternets formed by a large number of piconets, since each new piconet increases the overall capacity with its contribution. • The interference effect is not significant since the number of co-located piconets is low (<50). • When the switching overhead effect is taken into account, a BG loses capacity as a function of the number of piconets it is connected to. Thus, high performance, in terms of capacity, is attained when a bridge node is connected to only two piconets. These considerations explain why path lengths have to be taken into account. However, minimizing the path lengths without considering capacity at the same time, could lead to undesirable scatternet topologies too since, if the nodes are distributed in a small area, the resulting scatternet presents a fully meshed topology where every slave is connected to every master. In this case, the resulting capacity is low because of the high number of BGs connected to a high number of piconets. Fig. 3 refers to the metric that minimizes the average path length in the same scenario of Fig. 2. In this case the nodes in the lower part of the figure (which are in radio visibility of each other) are all connected. 194 T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 50 900 master slave bridge 45 800 Numberof Scatternets Node: 12 Node: 7 40 Node: 6 meters Node: 8 Node: 4 35 Node: 2 30 Node: 1 Node: 3 25 Node: 9 Mean value (centralized approach) 700 600 500 400 300 200 100 20 15 65 Node: 10 Node: 5 Node: 11 70 75 80 85 meters 90 95 0 Cmin=0.0087 0.0132 0.0166 Cmax=0.0188 Average Path Capacity 100 Fig. 5. Distribution of the values of average path capacity. Fig. 3. Scatternet with minimum average path length. Let us now look at Fig. 4, which shows a scatternet obtained with a metric that maximizes the average path capacity. This metric seems to be the most suited to maximize network performance, since both capacity and path length are taken into account. The scatternet of Fig. 4 presents a capacity, cTI ðBÞ equal to 4.667 (4.5271 taking into account the switching overhead and 4.0483 taking into account also the interference effect). The overall capacity is smaller than the value obtained by maximizing the normalized capacity, but, while in that case the average path length was 4.33, and 50 master slave bridge Node: 12 45 Node: 7 40 meters Node: 8 Node: 6 Node: 4 35 Node: 2 30 Node: 1 Node: 3 25 20 15 65 Node: 9 Node: 10 Node: 5 Node: 11 70 75 80 85 meters 90 95 100 Fig. 4. Scatternet with maximum average path capacity. the resulting capacity available for a generic source–destination pair was 0.0082, in the case of Fig. 4 the average path length is 2.81 and the resulting capacity per source–destination pair is 0.0109. As a result, we select as suitable TI metric the average path capacity. In order to better analyze the space of feasible scatternets Fig. 5 shows the histogram of the average path capacity of all possible ‘‘BT-compliant’’ scatternets obtained in a scenario constituted by 10 nodes distributed in an area of 25 · 25 m but, as will be shown later, a similar distribution holds in general. As indicated in Section 4.2 the number of different feasible topology is huge. The values of aTI ðBÞ are distributed in a range starting form aTI;min ðBÞ ¼ 0:0087 (8 kbit/s for every possible node pair) to aTI;max ðBÞ ¼ 0:0188 (19 kbit/s per pair); the mean value of aTI ðBÞ is also shown (equal to 0.0132). Note that the mean value is quite distant from the maximum value, which corresponds to the one associated with the optimal scatternet. Moreover, a few scatternets have a high value of aTI ðBÞ and are thus contained in the right tail of the histogram. This is an interesting result because it indicates that topology optimization is going to be a fundamental issue for Bluetooth scatternets: in fact, this distribution of the metric values means that it is highly unlikely to obtain a high performance scatternet by randomly selecting a topology. We need to deploy protocols that not only T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 195 Table 2 Traffic relationships Traffic relationship Required normalized capacity Traffic relationship Required normalized capacity 1$2 1$5 3$4 3$9 4$9 0.08 0.1 0.1 0.1 0.1 5$6 6$2 7$8 10 $ 7 10 $ 8 0.125 0.125 0.05 0.1 0.05 search for a connected scatternet but also explicitly aim at maximizing its performance. 20 Node: 7 master slave bridge 18 16 6.2. Traffic dependent metrics Node: 6 In this Section, we show results derived by applying the TD metrics. The scenario is composed of 10 nodes; the relevant traffic relationships are shown in Table 2. In Fig. 6 we depict the scatternet with maximum residual capacity. It can be noticed that this metric suffers from the same drawbacks of the metric it is derived from (i.e., normalized capacity, see Fig. 2): the resulting scatternet is likely to present a linear structure. In Fig. 7 we show the scatternet presenting the minimum average path length per traffic relationship, see Eq. (18). In this case the scatternet presents a more connected structure, with respect to that of Fig. 6. Finally, in Fig. 8 we minimize the average normalized load. The resulting normalized loads are meters 14 12 Node: 5 10 8 Node: 9 Node: 2 6 Node: 8 Node: 3 4 2 Node: 1 Node: 10 Node: 4 0 0 5 10 meters 15 20 25 Fig. 7. Scatternet with minimum average path length per traffic relationship. 20 master master slave slave bridge bridge Node: 7 0.15 18 16 Node: 6 14 Node: 7 0.15 18 master slave slave bridge bridge 16 meters 2 Node: 9 0.5 Node: 2 0.66 8 Node: 3 0.7 6 Node: 8 0.7 4 Node: 10 0.15 2 0 0 0 Node: 1 0.18 5 Node: 8 0.7 Node: 4 0.7 10 15 meters Node: 10 0.15 20 25 Fig. 8. Scatternet with minimum average load. Node: 4 0.5 Node: 1 0.48 0 Node: 3 0.2 4 Node: 5 0.73 Node: 9 0.2 Node: 2 0.205 6 12 10 Node: 5 0.935 10 8 Node: 6 0.55 14 0.25 12 meters 20 5 10 meters 15 20 Fig. 6. Scatternet with maximum residual capacity. 25 reported in the figure, next to each node. The scatternet average load is 0.3058. It can be noticed that nodes that are located in specific positions 196 T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 have to take care of all the forwarding load (i.e., traffic handled on behalf of other nodes). Nevertheless, since reducing load means reducing energy consumption, this scatternet consumes 70% of the energy required by the scatternet of Fig. 6 (the average load for the latter scatternet is 0.4267). Since with DSOA the nodes sequentially select how to connect, each node must be in TR of at least another already entered node. The following proofs that it is always possible to obtain such an ordering of the nodes, i.e. that this procedure always ends. 7. A distributed algorithm for topology optimization Theorem 1. Given a connected graph GðV ; EÞ, the procedure ORDER_NODES always terminates, and jW j ¼ N . In this section we describe a Distributed Scatternet Optimization Algorithm (DSOA) that aims at optimizing the topology to obtain a performance (in terms of the chosen metric) as close as possible to the optimum. Note that the selection of the optimized topology is decoupled from the establishment of the links that compose it, as will become clearer in Section 8, where we will describe a two-phases distributed scatternet formation algorithm based on DSOA. 7.1. Distributed scatternet optimization algorithm (DSOA) We consider the adjacency graph GðV ; EÞ. First, We aim at obtaining an ordered set of the nodes in V . The first procedure orders the nodes in the graph according to a simple property: a node k must be in transmission range of at least one node in the set 1 . . . k  1. ORDER_NODES Input: GðV ; EÞ Output: ordered set of the nodes in V , W ¼ fwk g, k ¼ 1; 2; ::; N , N ¼ jV j begin w1 ¼ random selection of a node v from V W ¼ fw1 g for k ¼ 1 : N wk ¼ random selection of a node v from V such that: (1) m 62 W (2) 9 u 2 W such that distance ðu; vÞ 6 TR W ¼ W [ fwk g endfor end Proof. Suppose that at some step k of the procedure, k < N , we have W ¼ fw1 ; w2 ; . . . ; wk1 g and no couple (v; w) with v 2 V n W , w 2 W exists such that distðv; wÞ < TR. Therefore, since W V , there exist two disconnected components, namely W and V n W , of GðV ; EÞ. h At the end of this procedure, then, node k is in transmission range of at least one of the nodes 1; 2; . . . ; k  1. The second procedure is the core of the algorithm. Here we let eij be the link between the nodes wi and wj of a scatternet ði; j 2 1; . . . ; N Þ. This part of the algorithm is dependent on the selected metric M. At each step k, node wk ‘‘enters’’ in the scatternet in the best possible way, according to M. SCATTERNET_OPTIMIZATION_ALGORITHM (SOA) Input: W , GðV ; EÞ, M Output:locally optimal scatternet B begin VM ¼ Ø VS ¼ Ø VM ¼ VM [ w 1 VS ¼ VS [ w2 B2 ¼ ½1 for k ¼ 3 : N case 1) consider wk in VM * derive all BT-compliant matrices Bk with jVM j þ 1 rows and jVS j columns calculate values of MðBk Þ case 2) consider wi in VS * derive all BT-compliant matrices Bk with jVM j rows and T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 jVS j þ 1 columns calculate values of MðBk Þ select the Bk with optimal MðBk ) if optimum in case 1) then VM ¼ VM [ wk else if optimum in case 2) VS ¼ VS [ w k else RECONFIGURE(Bk1 ; VM ; VS Þ endif endif endfor end The RECONFIGURE procedure is executed in the (unlikely) case when wk is only in transmission range of master nodes that have already 7 slaves in their piconet. For the sake of simplicity, details of this procedure are only given in the following proof of correctness. In this case, one of the 7 slaves is forced to become master of one of the other slaves. This is shown to be always possible. The following proves the correctness of SOA, i.e. it is always possible for a node to enter the network respecting the Bluetooth properties. Proof of correctness. Node w2 is in transmission range of w1 , thus the two nodes can connect. Each node wk , with k > 2 can always establish a new piconet, thus connecting as a master, whenever a node v 2 fw1 ; w2 ; . . . ; wk1 g exists s.t. v 2 VS and distance ðwk ; vÞ 6 TR, i.e. one of the slave nodes already in the network is in transmission range of wk . If no slaves are in transmission range of wk , whenever a node v 2 VM exists, with distance ðwk ; vÞ 6 TR, and slaves ðvÞ <¼ 7, wk can be a slave of v. Otherwise, at least one node wi 2 VM must exist, with distance ðwk ; vÞ 6 TR, and slaves ðvÞ ¼ 7, with i 6 k. The RECONFIGURE procedure can always be executed in this way. If at step i node wi selected more than 1 slave, it can disconnect from the slave that causes the minimum decrease/increase in the metric value. The topology is still connected, and wk can select wi as its slave. If, otherwise, wi selected only one slave at step i, this cannot be disconnected, since this could cause loss of connectivity for the network. Thus, one of 197 the other 6 slaves has to be disconnected. However, it was proven in [10] that in a piconet with at least 5 slaves, at least 2 of them are in TR of each other. Thus, at least one of the slaves can become master and select another slave. The network can therefore be reconfigured by forcing the seventh slave that connected to wi to become master of another slave of wi , to minimize reconfigurations. If it is not in TR of any other slave of wi , we can try with the sixth, and so on. At least one of the six slaves must be able to become master and select one of the other 5 as its slave. The local optimization in SOA (steps with mark *) can be performed by means of state space enumeration, as in the simulations results we show, or, e.g., by means of randomized local search algorithms. The distributed version of the SOA (distributed SOA, DSOA) straightforwardly follows. At each step k, a new node wk receives information on the topology selected up to that step ðBk1 matrix) and selects the role (master or slave) it will assume and the links it will establish, with the aim of maximizing the global scatternet metric. If the node becomes a master it will select a subset of the slaves in its TR already in the scatternet; if it becomes a slave it will select a subset of the masters in its TR, already in the scatternet. ORDER_NODES is needed to guarantee that, when node k enters, it can connect to at least one of the previously entered nodes. DSOA can be classified as a greedy algorithm, since it tries to achieve the optimal solution by selecting at each step the locally optimal solution, i.e. the solution that maximizes the metric of the overall scatternet, given local knowledge and sequential decisions. Greedy algorithms do not always yield the global optimal solution. As will be shown in the next subsection, the results obtained with DSOA are close to the optimum. 7.2. Examples and numerical results In this section we show some results obtained with DSOA, by using average path capacity as a metric. As previously discussed, we believe that average path capacity is a good metric since it takes into account both capacity and average path length of the scatternet. 198 T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 Node: 2 Node: 3 900 master slave bridge 86 800 Number of Scatternets 88 84 82 Node: 9 Node: 7 meters 80 Node: 1 78 Node: 10 Node: 8 76 Mean value (centralized approach) 700 600 Mean value (distributed approach) 500 400 300 200 74 Node: 4 100 Node: 6 72 70 0 a TI,min (B)=0.0087 Node: 5 68 40 45 50 55 60 65 meters 0.0132 0.0166 Average Path Capacity a (B)=0.0188 TI,max Fig. 11. Comparison between results obtained with DSOA and results derived from scatternet space enumeration. Fig. 9. Scatternet formed with DSOA. 88 Node: 2 86 master bridge Node: 3 84 82 Node: 9 meters 80 Node: 7 Node: 1 78 Node: 10 Node: 8 76 Node: 4 74 Node: 6 72 70 Node: 5 68 40 45 50 55 60 65 meters Fig. 10. Optimal scatternet. An example is shown in Figs. 9 and 10, where 10 nodes are distributed in an area of 25 · 25 m (the same scenario of Fig. 5). The first figure reports the scatternet formed with the DSOA (with an aTI ðBÞ ¼ 0:0145). Fig. 10 depicts the optimal scatternet (that presents an aTI ðBÞ ¼ 0:0188Þ. Fig. 11 shows a comparison between the optimal aTI ðBÞ and the one obtained with the DSOA. The dotted curve shows the histogram of the average path capacity of all possible ‘‘BT-compliant’’ scatternets feasible in this scenario. The values of aTI ðBÞ are distributed in a range starting form aTI;min ðBÞ ¼ 0:0087 (8 kbit/s for every possible node pair) to aTI;max ðBÞ ¼ 0:0188 (19 kbit/s per pair); the mean value of aTI ðBÞ is also shown (equal to 0.0132). Note that the mean value is quite distant from the maximum value, which corresponds to the one associated with the optimal scatternet. Moreover, a few scatternets have a high value of aTI ðBÞ and thus are contained in the right tail of the histogram. This is an interesting results because it suggests that topology optimization is a fundamental issue for Bluetooth scatternets: in fact, this distribution for the metric values means that it is highly unlikely to obtain a high performance scatternet by randomly selecting a topology. We need to deploy protocols that explicitly aim at maximizing performance. As regards the DSOA, the vertical lines in Fig. 11 correspond to the values of aTI ðBÞ for 100 different scatternets formed by using 100 different randomly chosen sequential orders. The lines are concentrated in the right part of the figure (i.e., the scatternets formed have a value of aTI ðBÞ greater than the overall mean value of all possible scatternets). The mean value of aTI ðBÞ of these 100 DSOA scatternets is equal to 0.0166. The unnormalized values of the average capacity per path obtained with DSOA is about 17 kbit/s, while the maximum possible value is 19 kbit/s; this confirms the good behavior of the DSOA. Fig. 12 shows a similar distribution in a scenario with 15 nodes in a multi-hop context. In Fig. 13 a distribution mediated on 100 different scenarios, with varying number of nodes is shown, while Fig. 14 reports the distribution of the values obtained T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 Fig. 12. Distribution of average path capacity for 15 nodes. 199 Fig. 14. Distribution of average path capacity for DSOA scatternets. spondingly, the value of the metric obtained with DSOA is close (sometimes equal) to the one obtained with the centralized approach. The same behavior has been observed in numerous experiments, carried out with different metrics and number of nodes. 8. A two-phases scatternet formation algorithm The actual Distributed Scatternet Formation Protocol is divided in two phases: 1. Tree scatternet formation (SHAPER). 2. DSOA and new connections establishment. Fig. 13. Distribution of average path capacity on different scenarios. with DSOA in the same scenarios. The probability of obtaining a value of the metric between the optimal and 70% of the optimal by randomly selecting a topology is very low; by using DSOA this probability is close to 1. For a higher number of nodes, the state-space enumeration approach, which has been useful in obtaining the distribution of the metric values, becomes unfeasible. The conclusion we can draw from the above figures is that scatternets formed with DSOA have a structure quite similar to the optimal ones, obtained with the centralized approach. Corre- To implement DSOA we need a mechanism to distribute the ‘‘right’’ to enter in the network to every node k at step k, and to convey the topology selected by the previous k  1 nodes (Bk1 matrix). The distributed implementation in Bluetooth however is not simple since the system lacks a shared broadcast medium that would allow signaling among nodes. A good solution which guarantees: (i) the required ordering of the nodes; (ii) synchronization of the decisions; (iii) a shared communication medium, is to form a tree-shaped ‘‘provisional’’ scatternet. A tree-shaped scatternet can asynchronously be formed in a distributed fashion. In [11], we 200 T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 proposed a new protocol for tree scatternets (SHAPER), which works in an asynchronous and totally distributed fashion, thus allowing the selforganized formation of a tree shaped scatternet in a multi-hop context. We showed that a tree scatternet can be formed in a few seconds time, and that less time is required when nodes are denser. After the tree has formed, a simple recursive visit procedure can be executed on it, which allows implementing the DSOA topology optimization process. It is easy to see that a sequential visit of all nodes in the tree, from the root down to the leaves, guarantees the order provided by ORDER_NODES. We let parent(v) be the parent of v in the tree and children(v) be the set of children nodes for v. Step k of the distributed procedure is executed on a node when it receives an execute_enter (Bk1 ; k) message from its parent. Bk1 is the matrix representing the topology selected by the previously visited nodes. The root node resulting from SHAPER starts the distributed execution of such procedure at the expiration of a timeout. PROCEDURE ENTER (Bk1 ; k) Bk ¼ DSOAðBk1 Þ foreach v 2 children send (execute_enter (Bk ; k þ 1), v) wait_answer() ½Bkþc ; c ¼ answerðvÞ k ¼kþc endforeach send (branch_entered (Bk ; k), parent) When a given node v starts the ENTER procedure, it executes the DSOA, i.e. it decides how to enter in the network. Then, the node randomly picks up one of its children nodes, and sends it the execute_enter(Bk , k þ 1) message. This causes the execution of the ENTER procedure on the child. After sending the execute_enter command, v waits for an answer message (branch_entered) from the child. This contains information about the topology selected by the whole branch which goes down from v to the leaf nodes. After the answer from the child is received, v selects another child and does the same. When v receives the answer from its last child, it informs its parent of the topology selected by itself and by all of its descendents with the branch_entered message. When the root node receives the answer from its last son, all nodes 1,11,25 1 2,10 7 2 3,5,7,9 3 8 12,24 13,23 8 4 5 4 14,16,18,20,22 6 9 6 10 15 11 12 17 19 13 21 Fig. 15. Visit procedure on the tree. have taken their decision. Fig. 15 shows how a simple tree, composed of 13 nodes, can be visited in 25 steps to execute DSOA. The numbers inside the circles represent the order in DSOA, i.e., the order in which nodes enter the network. The numbers outside the circles represent the ‘‘path’’ followed by the B matrix in the visit procedure. The last step concerns the actual connection establishment. The root node broadcasts the matrix representing the final scatternet structure. The matrix is recursively broadcasted at every level of the tree. Once a node has broadcast the matrix down to its children in the tree, it enters the reconfiguration phase. During reconfiguration it can start establishing the connections that will compose the ‘‘optimized’’ scatternet. Every link that is not already part of the tree topology has to be established. Redundant links have to be torn down. Every node alternates between a communication state and a formation state. During the latter the node tries to establish the new links, while during the former user data is transmitted so as to guarantee the continuity of service during the reconfiguration phase. If a node has a master role in the optimized scatternet, it pages its first slave. When the connection is established, it continues with the other ones. If the node has a slave role, it will page scan for incoming connections. Priority is given to previously entered masters so as to avoid dead- T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 locks. Every node starts tearing down the old links only when the new ones have been established, so as to preserve connectivity. Since all nodes know the overall topology, the routing task is also simplified. Route discovery algorithms have to be implemented only when mobility has to be dealt with or in other particular situations. The most time consuming phase of the algorithm is the formation of the tree, which, as said before, becomes necessary because Bluetooth lacks a shared broadcast medium. However, we showed in [11] that the tree can be formed in a few seconds. During the tree formation phase data exchange among nodes can start, so users donÕt have to wait for the overall structure to be set up. Data exchange can continue on the provisional tree scatternet during the optimization process. Work is in progress to add self-healing functionalities to the algorithm (nodes can enter and exit the network which is re-optimized periodically) and to simulate the integration of SHAPER and DSOA with the Blueware [23] simulator. 9. Conclusions In this paper, we discussed the scatternet formation issue in Bluetooth, by setting a framework for scatternet analysis based on a representation in a matrix form, which allows developing and applying different metrics. We identified several metrics both in a traffic independent and in a traffic dependent context, and we showed the relevant numerical results. The analysis of these results allows selecting the most suitable metric for a given scenario. A distributed algorithm for scatternet topology optimization, DSOA, was then described. The performance of DSOA has been evaluated and is encouraging: the distributed approach gives results very similar to a centralized one. The integration with the SHAPER Scatternet Formation Algorithm and other implementation concerns have been discussed. Ongoing activities include the full design of a distributed scatternet formation algorithm which implements DSOA and deals with mobility and failures of nodes, as well as a simu- 201 lative evaluation of the time needed to set-up a scatternet and its performance in presence of different traffic patterns. References [1] J.C. Haartsen, The Bluetooth radio system, IEEE Personal Communications 7 (1) (2000) 28–36. [2] P. Johansson, M. Kazantzidis, R. Kapoor, M. Gerla, Bluetooth––an enabler for personal area networking, IEEE Network 15 (5) (2001) 28–37. [3] Bluetooth SIG Specification of the Bluetooth System, Version 1.1––Core, February 2001. [4] IEEE Std 802.15.1TM -2002, Available from . [5] S. Zurbes, Considerations on Link and System Throughput of Bluetooth Networks, 11th IEEE International Symposium on Personal Indoor and Mobile Radio Communications, vol. 2, 2000, pp. 1315–1319. [6] T. Salonidis, P. Bhagwat, L. Tassiulas, R. LaMaire. Distributed Topology Construction of Bluetooth Personal Area Networks, IEEE INFOCOMÕ01, 2001, pp. 1577–1586. [7] C. Law, A. Mehta, K. Siu, Performance of a new Bluetooth formation protocol, ACM Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc), October 2001. [8] H. Zhang, J.C. Hou, L. Sha, A Bluetooth loop scatternet formation algorithm, Proceedings of the IEEE ICC 2003, Anchorage, 2003, pp. 1174–1180. [9] G. Tan, A. Miu, J. Guttag, H. Balakrishnan, An efficient scatternet formation algorithm for dynamic environments, IASTED Communications and Computer Networks (CCN), Cambridge, MA, November 2002. [10] G.V. Zaruba, S. Basagni, I. Chlamtac, Bluetrees-scatternet formation to enable Bluetooth-based ad hoc networks, IEEE International Conference on Communications (ICCÕ01), vol. 1, 2001, pp. 273–277. [11] F. Cuomo, G. Di Bacco, T. Melodia, SHAPER: A selfhealing algorithm producing multi-hop Bluetooth scattERnets, Proceedings of the IEEE Globecom 2003, San Francisco, USA, in press. [12] Z. Wang, R. Thomas, Z. Haas, bluenet––a new scatternet formation scheme, Proceedings of the Hawaii International Conference on System Science (HICSS-35), 2002. [13] C. Petrioli, S. Basagni, I. Chlamtac, Configuring bluestars: multihop scatternet formation for Bluetooth networks, IEEE Transactions on Computers (Special Issue on Wireless Internet) 52 (6) (2003) 779–790. [14] C. Petrioli, S. Basagni, I. Chlamtac, BlueMesh: degreeconstrained multihop scatternet formation for Bluetooth networks, Mobile Networks and Applications (Special Issue on Advances in Research of Wireless Personal Area Networking and Bluetooth Enabled Networks), 9 (1) (2004) 33–47. [15] I. Stojmenovic, Dominating set based scatternet formation with localized maintenance, Proceedings of the Workshop on Advances in Parallel and Distributed Computational Models, April 2002. 202 T. Melodia, F. Cuomo / Ad Hoc Networks 2 (2004) 185–202 [16] Y. Liu, M.J. Lee, T.N. Saadawi, A Bluetooth scatternetroute structure for multihop ad hoc networks, IEEE Journal on Selected Areas in Communications 21 (2) (2003) 229–239. [17] M. Ajmone Marsan, C.F. Chiasserini, A. Nucci, G. Carrello, L. De Giovanni, Optimizing the topology of Bluetooth Wireless Personal Area Networks, IEEE INFOCOMÕ02. [18] R. Kapoor, M.Y.M. Sanadidi, M. Gerla, An analysis of Bluetooth scatternet topologies, Proceedings of the ICC 2003, 2003. [19] P. Bhagwat, S.P. Rao, On the characterization of Bluetooth scatternet topologies, Available from . [20] F. Cuomo, T. Melodia, A general methodology and key metrics for scatternet formation in Bluetooth, IEEE Globecom 2002, November 2002. [21] C.F. Chiasserini, M. Ajmone Marsan, E. Baralis, P. Garza, Towards feasible distributed topology formation algorithms for Bluetooth-based WPANs, Hawaii International Conference on System Science (HICSS-36), Big Island, Hawaii, 6 January 2003. [22] E.-H. Amre, Interference between Bluetooth networksupper bound on the packet error rate, IEEE Communications Letters 5 (2001) 245–247. [23] G. Tan, Blueware: Bluetooth simulator for ns, MIT Laboratory for Computer Science, Cambridge, October 2002. Tommaso Melodia received his ‘‘laurea’’ degree in Telecommunications Engineering from the University of Rome ‘‘La Sapienza’’ in 2001. He is currently a Ph.D. student in Information and Communication Engineering at the same University. From February to August 2003 he was a Visiting Researcher at the Broadband and Wireless Networking Laboratory at the Georgia Institute of Technology. His main research interests are related to Computer Networks, Wireless ad hoc Networks, Wireless Sensor Networks, Personal and Mobile Communications. Francesca Cuomo received her ‘‘Laurea’’ degree in Electrical Engineering in 1993, magna cum laude, from the University of Rome ‘‘La Sapienza’’, Italy. She earned the Ph.D. degree in Information and Communication Engineering in 1998, also from the University of Rome ‘‘La Sapienza’’. Since 1996 she is a ‘‘Researcher’’ (Assistant Professor) at the INFOCOM Department of this University. She teaches courses in Telecommunication Networks. Her main research interests focus on: Modeling and Control of broadband integrated networks, Signaling and Intelligent Networks, Architectures and protocol for fixed an mobile wireless networks, Mobile and Personal Communications, Quality of Service guarantees and real time service support in the Internet and in the radio access, Reconfigurable radio systems and Wireless ad hoc networks. She participated in: (I) the European ACTS INSIGNIA project dedicated to the definition of an Integrated IN and B-ISDN network (1995–1998); (II) RAMON project, funded by the Italian Public Education Ministry, focused on the definition of a reconfigurable access module for mobile computing applications (2000–2002); (III) National project ‘‘5% Multimedialita’’ CNR-MURST. She is now participating to the European IST WHYLESS.COM project focusing on adoption of the Ultra Wide Band radio technology for the definition of an Open Mobile Access Network (2000–2003). In this project she is leader of the WP4 (Network Resource Manager). As for current national projects: (I) she is involved in FIRB project VIRTUAL IMMERSIVE COMMUNICATIONS (VICOM) where she is responsible of the research activities on the BAN and VAN networks; (II) she is responsible of the research unit at the University of Rome ‘‘La Sapienza’’ in the EURO project funded by the Italian Public Education Ministry. In 1995 she joined Coritel, a research institute on telecommunications, and she has been responsible for two years of the SWAP project in the Radio Access area.
Available online at ScienceDirect Procedia Computer Science 45 (2015) 524 – 527 International Conference on Advanced Computing Technologies and Applications (ICACTA2015) Bluetooth Smart based Attendance Management System Riya Lodhaa, Suruchi Guptaa, Harshil Jaina, Harish Narulaa a D. J. Sanghvi College of Engineering, Mumbai-400014, India Abstract Bluetooth Smart is a wireless technology aimed at innovative applications in the healthcare, fitness, beacons, security, and home entertainment industries. The technology makes use of electronic tags to facilitate automatic wireless identification, with a Bluetooth Smart enabled device. We are attempting to solve the problem of attendance monitoring using a Bluetooth Smart based system in this paper. This application of Bluetooth Smart to student attendance improves the time taken during manual attendance and human errors and provides administrators the statistics of attendance scores for use in further managerial decisions. © 2015 2015 The The Authors. Authors. Published Published by © by Elsevier Elsevier B.V. B.V. This is an open access article under the CC BY-NC-ND license ( Peer-review under responsibility of scientific committee of International Conference on Advanced Computing Technologies and Peer-review under responsibility of scientific committee of International Conference on Advanced Computing Technologies and Applications (ICACTA-2015). Applications (ICACTA-2015). Keywords: Bluetooth Smart; Lecture; Attendance; Tag; Student; Database 1. Introduction Valuable lecture time is dedicated to attendance taking and sometimes it is inaccurate. It is time-consuming and laborious. Besides, the process being manual, it is very prone to personal errors. Making this problem automated through the use of Bluetooth Smart technology makes it more efficient and effective. Bluetooth Smart technology can be used since it is an automated identification and data collection technology. It is accurate and ensures timely entry of data. It come into existence only recently, but has tremendous future scope due to its low cost and compatibility with a large number of mobile phones, tablets and computers. Bluetooth Smart combines microchip technologies and radio frequency to create a secure system which can be used for identification, monitoring and for maintenance of object inventories. Bluetooth Smart systems use tiny chips * Corresponding author. Tel.:+91-9773777671. E-mail address: 1877-0509 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ( Peer-review under responsibility of scientific committee of International Conference on Advanced Computing Technologies and Applications (ICACTA-2015). doi:10.1016/j.procs.2015.03.094 Riya Lodha et al. / Procedia Computer Science 45 (2015) 524 – 527 called tags. The tags themselves contain and then transmit some piece of identifiable information to a Bluetooth Smart enabled device. Bluetooth Smart systems can deliver accurate and precise data about tagged items that improves efficiency and this ability will bring many other benefits to the business community and to consumers alike in the near future. In this paper, we present a smart Bluetooth Smart based lecture attendance management and control system tailored around the Mumbai University (MU) policy of ensuring a 75% course attendance by students for a course before likelihood of writing a semester examination for any course. The application of Bluetooth Smart Technology to student lectures attendance monitoring problem in our proposal will lead to the creation of a student database management system that is not manipulated by anyone and not prone to errors, it will eliminate/reduce the wastage of quality time during manual collection of attendance, and most importantly it helps in better management of the classroom statistics for allocation of attendance scores, in a particular course, in the final grading of student performance. 2. Review of Related Work An automated attendance management system using both a stationary RFID reader with four circulatory antennae 1 and a handheld RFID reader was implemented in mobile and electronic platform respectively . A system comprising of an antenna placed at the entrance of the classroom and a student database is depicted by the attendance management system in the electronic platform. As students enter their class, their names are shown on the screen in order to ascertain that their attendance has been marked in the professor’s database. However, one major drawback of this system is that as the distance between the RFID tags and electronic device decreases the RFID tag read rates tremendously. A different type of automatic attendance system was proposed that uses 2 fingerprint verification technique . The technique of extraction of abnormal point on the ridge of user’s fingerprint or minutiae made the fingerprint technique verification achievable. This verification is used to confirm the authenticity of a user who is authorized by comparing the captured fingerprint template with the stored templates in 3 the database. Another system is based on true or false value of previous verification of person’s authenticity . Authors also reviewed and proposed biometric system for attendance automation of employees in an organization 4 using fingerprint identification . In 5, a RFID based system detects and identifies tags and is used to mark students’ attendance. A computer has been used as the medium to perform this task. The RFID reader detects the presence of a tag and the system processes this information on the computer according to the programmed instructions. The tractability, availability and receptiveness of the technology highly affect the ease with which RFID system can be integrated into current operations. The system provides an effective solution to lecture attendance problem by organizing the design of software and hardware along with efficient exchange of data between the RFID tag and reader which is interfaced with the computer. 3. Proposed Model and Working For RFID based systems, in case of scanning the tag, the tag must be close or in contact with the RFID reader to send the information to an established database, which interprets the data stored on that tag. This process needs monitoring as to who is scanning the tag to avoid proxies. The scanning time is approximately same as the time it takes to manually count the students in the class. The development of industrial wireless sensors has led to important demands for the wireless technologies like low energy consumption and a resource saving simple protocol stack. Bluetooth Smart is a rather new wireless standard which will completely fulfil these fundamental requirements. The proposed Bluetooth Smart system offers many advantages because electronic tags can be embedded into the student identification cards (student ID card); has low power consumption; the electronic tag can be read during motion as well and no line of sight is required for wireless communication between the tag and the reader. Tags can be read even if submerged or covered with dirt, are almost indestructible, and have unalterable permanent serial 525 526 Riya Lodha et al. / Procedia Computer Science 45 (2015) 524 – 527 code that prevents tampering. Fig.1. Flowchart showing the mode of operation of the student attendance management system A Bluetooth Smart chip is programmed and configured such that it works in connection with the Android application via Bluetooth. Every student is given a specific tag, which can then be detected by the application via Bluetooth Low Energy. When he/she attends the lecture, a serial number (related to each student’s SAP number) of the tag is associated with the student database entry. Therefore, every time a student carries his/her card and is attending the lecture the entries will be entered into the database with the time stamp as the lecturer moves around the class and the application detects the tags. Also, the application is configured to detect tags only within a particular range in order to avoid detection of tags that are outside the class. Fig.2. Depiction of Bluetooth Smart system operation Riya Lodha et al. / Procedia Computer Science 45 (2015) 524 – 527 Bluetooth Smart technology operates in the same spectrum range (the 2.400 GHz-2.4835 GHz ISM band) as Classic Bluetooth technology. Bluetooth connects the tag with the mobile device. The application checks for the validity of the tag. If the tag is valid, it continues to the database program and then registers the student’s attendance for the particular course. If the tag is found to be invalid, the application provides a notification that the tag has not been registered to any student and the user is required to supply a valid tag. The professor can use queries provided by the application to obtain more information about the attendance of a particular student or the entire class. The professor can grade students based upon their attendance for a particular course by entering the specific parameters in the application as specified by the university and can also generate reports weekly, monthly or for an entire semester. Additionally, the system can be used to notify the parents of defaulting students. The administrator assigns tags to the students and can not only designate new tags but also assign an existing tag linked with a particular student to another student. 4. Conclusion In this paper, we have discussed an automated attendance recording system that utilizes the capabilities of Bluetooth Smart technology. The major advantages of a Bluetooth Smart based system are: • Low power consumption • High data transfer rate • Small size of chips and low cost • Simple implementation of Bluetooth Smart based wireless sensors As Bluetooth Smart technology evolves, sophisticated applications in a variety of fields like healthcare, inventory management and sports can make use of this technology to design simpler, cheaper and more efficient solutions. Acknowledgements We would like to thank Prof. Harish Narula for guiding us with the research. We are also grateful to the principal of D. J. Sanghvi College of Engineering, Dr. Hari Vasudevan, and Head of Department of Computer Engineering, Dr. Narendra Shekokar, for giving us the facilities and providing us with a propitious environment for working in college. We would also like to thank S.V.K.M. for encouraging us in such co-curricular activities. References 1. FIDSensNet Lab (2005), A white paper on Automatic Attendance System. Texas A & M University, Texas, USA. 2. Chitresh, S and Amit K (2010),”An efficient Automatic Attendance Using Fingerprint Verification Technique”, International Journal on Computer Science and Engineering (IJCSE),Vol. 2 No. 2,pp 264-269. 3. Henry. S, S. Arivazhagan and L. Ganesan (2003), “Fingerprint Verification Using Wavelet Transform”, International Conference on Computational Intelligence and Multimedia Applications, 2003. 4. Maltoni D, D. Maio, A. K. Jain, S. Prabhaker (2003), “Handbook of Fingerprint Recognition”, Springer, New York, Pp 13-20. 5. Arulogun O. T., Olatunbosun, A., Fakolujo O. A., and Olaniyi, O. M (2013), “RFID-Based Students Attendance Management System”, International Journal of Scientific & Engineering Research Volume 4, Issue 2, February-2013, ISSN 2229-5518. 527

Tutor Answer

(Top Tutor) Studypool Tutor
School: Purdue University
Studypool has helped 1,244,100 students
flag Report DMCA
Similar Questions
Hot Questions
Related Tags
Study Guides

Brown University

1271 Tutors

California Institute of Technology

2131 Tutors

Carnegie Mellon University

982 Tutors

Columbia University

1256 Tutors

Dartmouth University

2113 Tutors

Emory University

2279 Tutors

Harvard University

599 Tutors

Massachusetts Institute of Technology

2319 Tutors

New York University

1645 Tutors

Notre Dam University

1911 Tutors

Oklahoma University

2122 Tutors

Pennsylvania State University

932 Tutors

Princeton University

1211 Tutors

Stanford University

983 Tutors

University of California

1282 Tutors

Oxford University

123 Tutors

Yale University

2325 Tutors