Mobile Phone Based Acoustic Localization using Doppler shift for Wireless Sensor Networks

Transcription

1 Mobile Phone Based Acoustic Localization using Doppler shift for Wireless Sensor Networks Amarlingam M, Charania Navroz Firoz, P Rajalakshmi Department of Electrical Engineering Department of Computer Science and Engineering Indian Institute of Technology Hyderabad Hyderabad, India ee13p13, cs14mtech113, raji@iith.ac.in Abstract Numerous applications of Wireless Sensor Network (WSN) requires a location map of sensor nodes for smart interaction, to ease the manual servicing of nodes etc. To create a location map, the primary task involves finding the locations of sensor nodes with an assumed reference origin (localization). This study proposes a mobile phone (commercial off-the-shelf mobile hardware) based acoustic localization technique that can be used along with physical or logical topology map to ease the search of a specific sensor node. The proposed method makes use of Doppler effect in guiding deployer towards the node of interest. A realtime deployment of the proposed acoustic localization algorithm is performed in an area of 2 m 2 m using in-house developed IITH Motes and mobile phone. From the experimental analysis it is observed that a reduction of 62.5% and 66.6% in searching time by the user is achieved upon using the proposed algorithm when compared to labeling methodology with both the logical and physical topology scenarios respectively. I. INTRODUCTION Location map is a map which associates physical locations to the locations of the respective sensor nodes. Every location map is created with a reference point and a co-ordinate system. In real field, searching a specific sensor node with only location map (physical topology map) or logical topology map is an arduous and tedious task. A basic acoustic localization technique for this problem is coined in [1]. In this article we propose an algorithm that can be integrated along with already proposed solution to make localization system more interactive and full-featured. With the rapid growth in technological advances in the areas of Internet of Things (IOT) and WSN, the sensor nodes in a network require efficient management system for control and monitoring. This system should also be able to precisely locate the sensor nodes for sensor network applications like forest fire detection, habitat sensing, pollution monitoring etc. In the literature various localization techniques have been proposed for addressing the aforementioned challenges. Localization algorithms based on Received Signal Strength Indication (RSSI) are given in [2]-[8]. Range based 2D localization algorithms using RSSI and path-loss model are proposed in [2]-[3]. Authors of [4] developed probabilistic model with RSSI for estimating 2D location of sensor nodes. N. Patwari et al. [5] presented relative 2D location estimation algorithm using time-of-arrival (ToA) and RSSI measurements from beacon sensor nodes. Centroid based 2D localization algorithm by using connectivity matrix from reference nodes is proposed in [6]. Applications like smart buildings, industrial security system etc., requires 3D locations of sensor nodes. Authors of [7] and [8] described RSSI based 3D localization algorithms. Acoustic localization techniques are more desirable than RF based localization techniques, the reason being the relatively low speed of acoustic waves that causes the localization system to be less sensitive to errors in time-of-flight measurements. Authors of [9] and [1] proposed acoustic localization algorithms for sensor networks. Angle-of-arrival based localization algorithms are proposed in [11]-[12]. G. Mao et al. [13] have done a survey on localization techniques for wireless sensor networks. All of these localization algorithms are proposed to create a location map for sensor nodes which associates the physical locations to the locations of the respective sensor nodes. These algorithms consider a known location as the reference point together with the help of a coordinate system to create a localization map. In real field deployment of sensor networks, sensor nodes are labeled with unique ID s (ID s usually includes numbers 1, 2,.. or unique symbols A, B,... etc.) for identification which is also known as labeling method. In general, localization maps embed the information such as node ID along with their co-ordinates. Fig. 1: Deployment Adviser tool showing network topology along with redundant and hot-spot nodes In [14], we proposed and developed a mobile phone application called as Deployment Adviser tool for the purpose of field deployment of sensor nodes. The Adviser tool shows logical topology map, redundant relay nodes and hot-spot relay node in the network as shown in Fig. 1. The Adviser tool advises the deployer for placing nodes targeting improved

2 lifetime of the network by distributing the power consumption among all the nodes present in the network. To increase the network lifetime, extra relay nodes around the hot-spot relay node (node which consumes more energy compared to all other nodes in the network) have to be placed. In real field deployment removing redundant relay nodes and placing extra relay nodes around hot-spot relay node involves manual intervention and exact knowledge of their physical locations. One example where a redundant node is removed and an extra relay node is placed using general labeling method is illustrated in Fig. 2. For large-scale network, searching a particular labeled node in real field by using logical topology map or physical topology map alone is time consuming and difficult. There is a need for an additional localization system connected to network topology mapping system, which makes searching of the sensor nodes easier. All the above discussed constraints motivated us towards this study and we proposed a mobile phone based acoustic localization system for wireless sensor networks. Fig. 2: Removing redundant node and placing extra relay nodes next to hot-spot In prior research, we proposed and implemented a basic mobile phone based acoustic localization system [1] which finds physical location of the sensor node in a sensor network with the help of network topology map. This system shows whether deployer is moving towards or away from the sensor node based on Doppler effect caused due to deployer s mobility. Although the information whether the deployer is moving towards or away from sensor node is obtained, it is still onus on deployer to localize a sensor node in a large-scale network precisely. To facile deployer for quick and precise localization of the sensor node, we propose an algorithm based on Doppler effect. Proposed method improves the process of localization by showing exact direction in which the deployer has to move to reach the required sensor node. With the help of these algorithms, a complete system of mobile phone based acoustic localization along with network topology map (physical topology map or logical topology map) is designed and developed. One of the main applications of the proposed algorithm includes servicing (sensors replacement, trouble shooting of sensors circuitry etc.) of the sensor nodes in a sensor network. R. J. Kozick et al. [15] proposed a localization algorithm using acoustic Doppler shift with a mobile access point. Here, the mobile access point need to know its location. The nodes are equipped with acoustic receivers (i.e. microphones) and information of the mobile access point location is known prior. Sensor nodes localize themselves by finding Doppler shift of the acoustic wave which is emitted by mobile access point with the help of its location. Y. Nishimura et al. [16] proposed a method which finds direction between two mobile phones using Doppler effect. The authors used pulse-pair method to calculate Doppler shift where both devices transmit and receive multiple pulse type acoustic waves (pulse type chirp signal). Direction between two mobile phones is estimated by detecting change between transmission and reception acoustic pulse intervals. W. Huang et al. [17] showed direction finding and indoor localization techniques using acoustic signal Doppler effect. They used inertial sensors like accelerometer to estimate velocity of the mobile acoustic receiver. Phase Locked Loop (PLL) is implemented to track phase change of the received acoustic wave. In their application, to find velocity of the acoustic receiver and phase, a complex system with PLL, Band Pass Filter (BPF) and Automatic Gain Control (AGC) is implemented. However, PLL, BPF, AGC are more computational intensive applications which poses multiple constraints when implementing on low end (commercial offthe-shelf) COTS (mobile phone) devices. In case of WSN, nodes (motes) have low computation and poor processing capability compared to mobile phones. To generate and detect received acoustic waves, extra hardware overhead is required which can be interfaced with sensor nodes. For detecting acoustic wave, a specially designed complex hardware consisting of custom sensor board with microphone, ADC, FPGA and DSP interfaced with Mica2 mote [18] is used. High complexity in hardware increases power consumption, which leads to reduction in sensor node s lifetime. Our method of localizing sensor node in a sensor network is direction oriented. In the proposed acoustic localization technique, we developed a low complex acoustic module designed using simple RLC resonant circuit followed by an amplifier. In our system, Doppler effect is used to find the direction of sensor node with respect to mobile phone (carried by deployer). Here, sensor node emits acoustic wave and Doppler effect is caused due to deployer s mobility. To emit acoustic wave, sensor node is equipped with an in-house developed low complex acoustic module. Rest of the paper describes the technical aspects in brief and is arranged as follows. In Section II, we discussed the technical aspects of the proposed localization method and III describes implementation of the algorithm. Section IV explains experimental evaluation and Section V concludes the paper. II. PROPOSED LOCALIZATION METHOD USING DOPPLER EFFECT Let the speed of the transmitted sound wave from the source be v s and the average moving velocity of the listener be v l as shown in Fig. 3. θ indicates the angle between sound source and listener velocity (i.e. angle between v l and v s ). Let the relative velocity of sound wave observed at the listener be v ob. If listener s moving angle θ {θ 1 θ 1 < 9 },

3 then v l cos θ will be positive and hence resultant velocity v ob will be v s from equation (1). In another case, if listener moving angle θ {θ 1 9 < θ 1 18 }, then v l cos θ will be negative and hence resultant wave velocity v ob will be v s from equation (1). Received sound wave frequency f l at the listener will be calculated using equation (2) (a) (b) Fig. 4: (a) Listener swinging mobile at an angle of θ, (b) Doppler frequency change Vs. angle θ with v l =.6 m/s, v s = 34 m/s and f s = 2 khz. Fig. 3: Listener moving at an angle of θ with a velocity v l 1 1 v ob = v s + v l cos θ (1) f l = v ob v s f s (2) Then above equation becomes, where f l = (1 + v l cos θ v s ) f s (3) f l = f s + f d (4) f d = ( v l cos θ v s ) f s (5) Here, f s is the source frequency and f d is the Doppler frequency. When listener approaches the sound source θ {θ 1 θ 1 < 9 }, from equation (4) Doppler frequency f d will be positive. The observed sound wave frequency at the listener will be f l f s from equation (3). When listener is moving away from sound source θ {θ 1 9 < θ 1 18 }, Doppler frequency f d will be negative from equation (4) and the observed sound wave frequency at listener will be f l f s from equation (3). In application [1], the mobile phone detects deployer s direction of movement and advise the deployer by indicating whether his movement is towards the sensor node or away from it based on received frequency. Even with this information, it is still difficult to locate the sensor node in a large-scale sensor network. Hence, we propose an algorithm which estimates the direction angle and make use of the same by integrating with the existing system [1]. A. Direction angle estimation Direction of sensor node can be estimated in a simple way by generating Doppler effect. One method for generating Doppler effect is by swinging acoustic receiver horizontally. Swinging of mobile phone in 36 with respect to World Coordinate System (WCS) can detect acoustic source direction by using earth magnetic field. Consider a scenario where a sound source is exactly in front of the deployer and deployer swings mobile phone (acoustic receiver) horizontally in (a) θ {θ 1 θ 1 9 } (c) θ {θ 1 18 θ 1 27 } (b) θ {θ 1 9 θ 1 18 } (d) θ {θ 1 27 θ 1 36 } Fig. 5: Doppler frequency patterns of four quadrants with v l =.6 m/s, v s = 34 m/s, f s = 2 khz and random swing starting points θ = 1, 12, 19, 29. The angle θ between mobile phone swinging direction and the acoustic source direction can vary from to 36 as shown in Fig. 4a. From equation (4) Doppler frequency of acoustic signal f d with respect to change in θ (from to 36 ) follows sinusoidal pattern as shown in Fig. 4b. Here Doppler frequency crosses zero two times in one 36 swing as shown in Fig. 4b. Pattern will be same for both clock-wise and anti-clock-wise swings. The angle with respect to WCS at which the Doppler frequency pattern crosses its first zero can be concluded as the acoustic source direction. In other words, the point at which the angle between acoustic source to the mobile swing direction (θ) becomes 9 can be concluded as the direction of the acoustic source. Deployer can start swinging from any arbitrary direction with respect to acoustic source. Patterns of the Doppler frequency with the starting angle of the swing chosen randomly in four different quadrants is shown in Fig. 5, here swinging angle is wrapped around the range to 36. Also there are two zero crossing points in every swing. To decide which point is exact direction to the acoustic source is the question to be answered. In a window size of 18 around zero crossing, if the Doppler frequency pattern changes from positive to

4 negative, then the direction of the acoustic source with respect to mobile phone is coherent and opposite otherwise i.e. when the pattern changes from negative to positive. Pseudo code for the proposed algorithm is described in Algorithm 1. Algorithm 1 Estimation of direction angle X Require: Source transmitted frequency f s and set of received frequencies {f li } and respective sampled magnetic sensor values {M gi } where i (1, N) Take X, X 1, X 2 2: T emp f l1 f s for i=2:n do 4: f d f li f s if f d < and T emp > then 6: X 1 Mg i 1 + Mg i 2 else if f d > and T emp < then 8: X 2 Mg i 1 + Mg i else 1: T emp f d end if 12: end for if X 1 and X 2 then X1 + X2 14: X 2 else 16: if X 1 then X X 1 18: else X X 2 2: end if end if 1) Detection of LoS component: In proposed method, received acoustic wave sampling frequency is 44.1 khz. Continuous received signal is divided into windows of size milliseconds in time containing 2,25 samples. Each window contains combination of multiple received acoustic waves due to multipath. Usually Line of Sight (LoS) component will have higher strength than non-line-of-sight (NLoS) components. To pick the frequency changes corresponding to LoS component in a window, we are taking maximum peak power frequency and calculating an adaptive threshold which is equal to 85% of maximum received power of that window. Picking frequencies that are greater than the threshold will give almost all frequency changes of LoS component. Empirically, we observed that user swings mobile phone with an average tangential velocity of.6 meter per second. We considered deployer arm length r = 7 cm [16]. In a window time of ms, deployer swings approximately θ = 3 as shown in Fig. 6. The variation in θ = θ 2 θ 1 (θ θ 2 θ 1 2θ with consideration of d r) with in ms lies between 3 and 6. The maximum variation of Doppler frequency when θ changes by 6 is 3.62 Hz with v l =.6 m/s and v s = 346 m/s [2]. In other words we pick frequencies from a window such that the difference between maximum frequency and minimum frequency of that window is less than 3.62 Hz with respect to maximum peak power frequency. Average change in frequency over a window gives the mean value of Doppler shift at corresponding approximate angle with respect to WCS. This procedure eliminates most of the multipath components in a window. 2) Environmental effects mitigation: The proposed algorithm finds a point where the angle between mobile swinging Fig. 6: Change of angle in a window time direction and the acoustic source θ becomes 9 i.e when the received frequency of acoustic wave is same as the transmitted frequency. θ is independent from acoustic wave speed v s. The Doppler frequency becomes zero when θ equals 9 and this point is taken as acoustic source direction. Estimated direction point is independent from environmental factor changes like variation in humidity, temperature, pressure etc., as θ is independent of acoustic wave velocity. III. ACOUSTIC LOCALIZATION IMPLEMENTATION Proposed acoustic localization system is depicted in Fig. 7. Google Nexus 5 mobile phone with Android OS is used for mobile application development and testing. TinyOS is used for sensor node programming and JAVA made use for server programming. To create sensor network, IITH motes are used as sensor nodes [19]. Each sensor node in the network is equipped with simple acoustic module as shown in Fig. 8. Fig. 7: Acoustic localization implementation Sensor nodes usually have very low computation capability due to minimal on-board resources. Now a days mobile phones opened the feasibility for highly intensive computations by utilizing extensive on-board resources such as huge internal memory, multiple radios etc., Acoustic wave generation is simpler than detection and processing. Thus, sensor nodes are selected as sound wave sources and mobile phone as listener. Sensor node is programmed to generate square pulses from one GPIO (General Purpose Input/Output) pin. Interfaced circuitry converts 2 khz square pulses to 2 khz single tone amplified sinusoidal wave which is input to the speaker as shown in Fig. 9. From the speaker 2 khz sound wave is emitted. 2 khz frequency is chosen for the proposed system implementation because of low ambient noise levels.

5 Fig. 8: Sensor node interfacing with acoustic module The circuitry which converts square wave to sinusoidal is a simple RLC resonant circuit that consumes less power. For amplification, we used a simple acoustic wave amplifier IC LM386 based circuitry. Fig. 9: Acoustic wave generation circuitry A. Mobile application for finding direction Deployer swings mobile phone horizontally in 36 to find direction between the mobile phone and the sensor node that emits acoustic wave. While swinging, the mobile application receives sound signal from microphone of the mobile phone. Further, Fast Fourier Transform (FFT) is applied on the received acoustic signal and the frequency f l is calculated. From equation (3) Doppler frequency can be calculated from f l and known sensor node frequency f s. Original sound source frequency f s information will be stored in mobile phone. When swing starts, magnetic sensor of the mobile stores WCS values using earth magnetic field direction. Doppler frequency arrays will be processed to find the zero crossing point index and the same index of magnetic sensor array will be the solution if application considers first zero crossing index. In case if application considers second zero crossing index of Doppler frequency array, then resultant angle of the sound source direction is the sum of magnetic sensor estimated angle value at that index and angle of 18. Mobile phone displays GUI with an arrow that shows sensor node direction is shown in Fig. 1. Fig. 1: Android application showing direction to the acoustic source IV. EXPERIMENTAL RESULTS AND ANALYSIS For experimentation, a sensor network with 1 randomly deployed nodes in an area of 2 m 2 m at outdoor (a) Fig. 11: (a) physical topology map, (b) logical topology map environment is considered. To analyze the proposed technique, two cases are considered. In first case, deployer knows logical topology map of the network. In second case deployer knows physical topology map. To search a node in network, first deployer (a volunteer) has to enable swing option which is developed in GUI and then he has to swing the mobile phone. Application shows direction towards the sensor node with respect to mobile phone as shown in Fig. 1. If user enables walk option, application guides the deployer by indicating whether the deployer is approaching or leaving the sensor node [1]. Experiments are performed with the help of 1 volunteers. Each volunteer is asked to search a particular node in network by starting at random location with the help of given network topology map under both general labeling and proposed methods. A. Case 1: Searching a sensor node with logical topology map We deployed a sensor network as shown in Fig. 11a and each node is labeled with a unique ID. Initially volunteers are asked to find a particular node without using acoustic application (here after we call proposed application as sensor application), only with the help of logical topology map. Next we changed node numbers but maintained the same topology. Volunteers are again asked to find a the target node with the help of logical topology map along with sensor application (walk+swing). Average time consumption of all volunteers in both cases are calculated. To compare the full functioning system with prior proposed system [1], same experiment is repeated by considering a case where deployer uses only walk option (which is proposed in [1]) to search a node. Resultant average time consumption of the mentioned three cases is shown in Fig. 12. B. Case 2: Searching a sensor node with physical topology map Network is deployed as shown in Fig. 11b. Volunteers are asked to find a particular node without using sensor application with the help of physical topology map (2D location map). Next we changed node numbers but maintained the same topology. Volunteers are again asked to find a target node with the help of physical topology map and sensor application. Average time consumption of all volunteers in both cases are calculated. To compare the full functioning system with the prior proposed system [1], same experiment is repeated by using only walk option as discussed in subsection IV.A. (b)

6 Time in seconds Logical topology map Average time required labeling method walk proposed method (walk+swing) Physical topology map Fig. 12: Average time consumed for searching a sensor node in network Average time consumption of all considered cases are shown in Fig. 12. C. Results and analysis From Fig. 12, we can observe that searching a particular sensor node by using logical topology map or physical topology map without sensor application is a time consuming process, which clearly implies that it is onus to the deployer. In logical topology case, an average time consumed for searching a node using labeling method is 85 s, with walk option s where as the complete system it is taking only 3 s. In physical topology case for searching a particular node, the average time consumed with labeling method, walk option and complete system are 6 s, 4 s, 2 s respectively. From the above analysis one can infer the improvement in ease of localizing a specific sensor node being provided with the proposed method. D. Discussion To search a particular node in a sensor network, coarsely accurate measurements are done based on direction finding with tolerable mismatch are sufficient. Average angle estimation error using the proposed algorithm is observed to be which is sufficient to allow user to search sensor node in real field easily. Current consumption of designed low power acoustic module is observed to be 8.5 ma (which is practically measured) at a voltage of 5 V. Acoustic module will be ON when user wants to find physical location of the sensor node in real field. Acoustic module will be turned OFF for the remaining time which means that there will be a very minimal power consumed by the acoustic module. Time consumption for searching a sensor node depends on network structures and sizes. In this article, results are analyzed with network size of 1 nodes deployed in 2 m 2 m area for evaluating the performance of the proposed method. V. CONCLUSION In this paper, we proposed an acoustic localization method that finds the physical location of the sensor node in a sensor network with the help of network topology map and Doppler effect. The proposed method is implemented and tested using IITH motes and off the shelf mobile phone. Thus developed sensor application assists the deployer in reaching a sensor node of interest by guiding the deployer with appropriate direction. Experimental results shows that, for searching a specific node consumed time is only 37.5% and 33.3% of time consumed in case of labeling method used alone with the both logical and physical topology scenarios accordingly. It is clearly indicating that, proposed method reduces onus on deployer and facilitate of searching the sensor nodes in sensor networks. REFERENCES [1] M. Amarlingam, P. Rajalakshmi, M. Yoshida, and K. Yoshihara, Mobile phone based acoustic localization for wireless sensor networks, in Proc. WF-IoT 215. [2] G. Wang and K. Yang, A new approach to sensor node localization using RSS measurements in wireless sensor networks, IEEE Trans. Wireless Commun., vol. 1, no. 5, pp , May 211. [3] S. Gansemer, U. GroBmann, and S. Hakobyan, RSSI-based euclidean distance algorithm for indoor positioning adapted for the use in dynamically changing WLAN environments and multi-level buildings, in Proc. Int. Conf. Indoor positioning and Indoor Navigation(IPIN) 21. [4] C. Papamanthou and P. Franco, Algorithms for location estimation based on RSSI sampling, Algorithmic Aspects of Wireless Sensor Netw., Springer Berlin Heidelberg, 28, pp [5] N. Patwari, A. O. Hero, M. Perkins, N. S. Correal, and R. J. ODea, Relative location estimation in wireless sensor networks, in IEEE Trans. Signal Process., vol. 51, no. 8, pp , Aug. 23. [6] N. Bulusu, J. Heidemann, and D. Estrin, GPS-less low cost outdoor localization for very small devices, IEEE Pers. Commun., vol. 7, no. 5 pp.28-34, Oct. 2. [7] M. Amarlingam, P. Rajalakshmi, V. K. Netad, M. Yoshida, and K. Yoshihara, Centroid based 3D localization technique using RSSI with a mobile robot, in Proc. Int. Symp. Wireless Personal Multimedia Commun. (WPMC), pp , 7-1 Sep [8] M. Yoshida, K. Yoshihara, M. Amarlingam, V. K. Netad, and P. Rajalakshmi, 3D localization technique with mobile robot for improving operability of remote-control devices, in Proc. Int. Conf. Wireless Commun. and Mobile Computing (IWCMC), pp , Aug [9] X. Sheng and Y. Hu, Maximum likelihood multiple-source localization using acoustic energy measurements with wireless sensor networks, IEEE Trans. Signal Process., vol. 53, no. 1, pp , Jan. 25. [1] S. Astapov, J. Preden, and J. Berdnikova, Simplified acoustic localization by linear arrays for wireless sensor networks, in Proc. IEEE 18th Int. Conf. Digital Signal Process. (DSP), 213. [11] P. Kulakowski, J. Vales-Alonso, E. Egea-Lopez, W. Ludwin, and J. Garca-Haro, Angle-of-arrival localization based on antenna arrays for wireless sensor networks, Comput. & Elect. Eng., 21. [12] J. Qiu, D. Chu, X. Meng, and T. Moscibroda, On the feasibility of realtime phone-to-phone 3d localization, in Proc. ACM 9th Conf. Embedded Netw. Sensor Syst., 211. [13] G. Mao, B. Fidan, and B. D. O. Anderson, Wireless sensor network localization techniques, Elsevier/ACM Comput. Netw., 27, pp [14] M. Amarlingam, I. Adithyan, P. Rajalakshmi, Y. Nishimura, M. Yoshida, and K. Yoshihara, Deployment adviser tool for wireless sensor networks, in Proc. IEEE World Forum on Internet of Things (WF-IoT), pp , 6-8 Mar [15] R. J. Kozick and B. M. Sadler, Sensor localization using acoustic doppler shift with a mobile access point, IEEE/SP 13th Workshop on Stat. Signal Process., pp , 17-2 Jul. 25. [16] Y. Nishimura, N. Imai, and K. Yoshishara, A proposal on direction estimation between devices using acoustic waves, in Proc. 8th Int. ICST Conf. Mobile Ubiquitous Syst.: Comput., Netw. Serv., 212, pp [17] W. Huang, Y. Xiong, X. Li, H. Lin, X. Mao, P. Yang, Y. Liu, and X. Wang, Swadloon: Direction finding and indoor localization using acoustic signal by shaking smartphones, IEEE Trans. Mobile Computing, vol. 14, no. 1, pp , Oct [18] Crossbow Technology, Inc. [online]. Available: [19] P. Rajalakshmi. (212). IITH Mote-Wireless Sensor Communication Module. [Online]. Available: raji/downloads/iithmote-webpage.pdf. [2] NDT Resource Center. [Online]. Available: