An Experimental Study of The Multiple Channels and Channel Switching in Wireless Sensor Networks

Transcription

1 An Experimental Study of The Multiple Channels and Channel Switching in Wireless Sensor Networks Haiming Chen 1,2, Li Cui 1, Shilong Lu 1,2 1 Institute of Computing Technology, Chinese Academy of Sciences 2 Graduate University of the Chinese Academy of Sciences {chenhaiming, lcui, lushilong}@ict.ac.cn Abstract So far, a number of multi-channel MAC protocols have been proposed for wireless sensor networks (WSNs). It is imperative to get the following three important parameters for design of multi-channel protocols in WSNs. One is the number of available orthogonal, another is the time required to switch between two channels, and the other is the total energy consumed to switch between two channels. But to the best of our knowledge little practical information is available about these three parameters. This paper describes our approaches to determine the above three parameters for the Mica2 node, which is equipped with a CC1 radio and set to work in the channel of 433 MHz by default. Through extensive experiments, we find that for the Mica2 nodes (1) there are 48 channels available; (2) it requires about 5 ms to switch between two channels; (3) it consumes about 194 nj energy to switch between two channels. At last, we also point out some implications of these results for design of multi-channel protocols. The approaches presented in this paper can be easily extended to determine the parameters for the Zigbee (CC242) nodes, such as MicaZ and TelosB. 1. Introduction To exploit the multiple channels available in the wireless networks, like based ad hoc networks, researchers have proposed some multi-channel MAC protocols [1, 2, 3, 4]. The aim of these protocols is to allow parallel data transmission, and to improve network throughput. Considering the much lower bandwidth for wireless sensor networks (WSNs), more and more researchers try to explore the underlying support of multi-channel communication in WSNs. So far, a number of multi-channel MAC protocols have been proposed for WSNs, e.g. MMSN [5], CMAC [6], TMMAC [7], MCMAC [8], YMAC [9]. However, to the best of our knowledge no one has done dedicated investigation on the multi-channel itself so far. The problem of multi-channel itself is equivalent to three questions, which are listed below. 1. How many orthogonal channels are available in wireless sensor networks? The number of available orthogonal channels determines the benefits that can be achieved from the multi-channel protocols. Intuitively, the more number of orthogonal channels available in the networks, the more benefits expected to be achieved. So it is necessary to know the exact number before designing multichannel protocols. 2. What s the time cost of channel switching? The duration of channel switching has significant impacts on the performance of the designed multi-channel protocols, since it will cause unexpected delay and waste some bandwidth. So we should also care about this question when designing multi-channel protocols. 3. What s the energy cost of channel switching? For the wireless sensor networks, nodes are often powered by unrechargeable batteries. Energy efficiency should be taken as an important property of the proposed multi-channel protocols. So it is unavoidable to consider the energy cost of channel switching when designing multi-channel protocols. In one word, the answers to the above three questions are imperative parameters for design of multichannel protocols, but currently little practical information is available to answer the questions accurately. Motivated by such trouble, we conduct extensive experiments to find the accurate answers to the above three questions. Using the Mica2 [1] nodes as the testing platform, we firstly estimate the exact number of

2 available orthogonal channels, and then we measure the period of switching between two channels. At last, we compute the energy consumption of channel switching. Due to the limited space, we only present our efforts to answer the above three questions for the CC1 [11] nodes in this paper. However, the methodologies presented here can be easily extended to find the answers for the Zigbee (CC242) [12] nodes, such as MicaZ and TelosB [13]. The rest of paper is organized as follows. In section 2, we describe the steps to determine the available orthogonal channels for Mica2 nodes, and present the results drawn from extensive experiments. In section 3, we depict the approach to precisely measure the time spent in switching between two channels. In section 4, we compute the energy consumed in channel switching. In section 5, we give an overview of related work. In section 6, we summarize the results obtained from our measurements, discuss some implications of the results and point out the future work. 2. How many orthogonal channels are available in WSNs? For the Mica2 nodes, which is widely used in low-bandwidth sensor networks, its frequency is set to 433 MHz by default. Referring to the datasheet of CC1 [11], we know that CC1 can work in a band of frequencies from 315 MHz to 915 MHz, with a frequency separation of 64 KHz. Now the question is arisen: Are all the channels from 315 MHz to 915 MHz available for the Mica2 nodes? Intuitively, the answer is no, since the range of available channels is bounded by the peripheral circuit of the nodes. Even if many people can answer the above question confidently, to the best of our knowledge few people can answer the following question accurately. What s the range of channels available for the Mica2 nodes? So we firstly conduct following experiments to determine the exact range of channels available for the Mica2 nodes Range of available channels We set up a simple scenario as Figure 1 shows, in which a Mica2 node is programmed to transmit some messages to a sink node in a sequence of channels. The availability of a specific channel is determined by judging whether the sink node can receive the messages from the Mica2 node in this channel. The application running on the Mica2 node is Figure 1. Experimental scenario to determine the range of channels available for the Mica2 nodes addr type group length channel seq Figure 2. Format of messages used in TestMultiChannel named TestMultiChannel, which implements following two procedures: (1) tuning the frequency of the CC1 radio, and (2) transmitting 1 messages to the sink node in each channel. In the first procedure, we tune the frequency of the CC1 radio to all the channels listed in the Appendix A sequentially. This procedure is implemented by calling the TunePreset command provided by the CC1ControlM module. Frequencies of the channels listed in Table 3 are generated by the channelgen.exe, with frequencies of 1MHz spacing specified as the parameters of the command. Register settings of these channels are stored in the CC1K Params table, which is a two-dimension array defined in the CC1Const.h head file. Reasons why we only examine the channels with 1 MHz spacing are that: (1) The CC1 radio can not be tuned to an arbitrary frequency in the range from 3 KHz to 1GHz, since it uses a digital frequency synthesizer. (2) Although the channel separation can be 64 KHz according the datasheet of CC1, in our preliminary experiments we find that two channels can overhear each other when the spacing is less than 3 KHz. We are also confirmed by our preliminary experiments that for the specified frequencies with spacing between 3 KHz and 1 MHz, the generated channels are the same. After tuning the radio to a specific channel, 1 messages are transmitted by the Mica2. The format of messages is showed in Figure 2. The addr field contains the source address of the message. The type field indicates the type of the message. For the TestMultiChannel application, it can be one of two enumerated numbers. One is AM GETCHANNEL=x64, and the other is AM SETCHANNEL=x65. We refer to these two types of messages as GetChannelMsg and SetChannelMsg respectively. The definitions of the following two fields, namely group and length, are identical with

3 Packet Reception Ratio Channel Index Figure 3. Packet reception ratios in each channel the active messages [14]. The channel field is the index of channels listed in Table 3. For the GetChannelMsg, the channel field indicates the current channel of the radio, while for the SetChannelMsg it indicates the channel to be set. The seq is a uint 16 integer, which indicates the sequence of the packet. In each channel, the Mica2 node sends 99 GetChannelMsg every 2 seconds, followed by a SetChannelMsg. For example, the last two messages sent by the Mica2 node are: FF FF 64 7D FF FF 65 7D When the sink receives the first message, it will not tune its channel and keep listening on channel 1. About 2 seconds later, the sink will receive the second message. Once receiving the SetChannelMsg, the sink node will tune its channel according to the data specified in the channel field, which is channel 2 in the above example. It should be noted that when the Mica2 node finishes sending the message, it also tunes its channel as the message indicates. In such way, these two nodes keep synchronized working in the same channel. Figure 3 shows the ratio of packets successfully received by the sink in each channel. From this figure, we can see that the sink can receive almost all the packets sent by the Mica2 node in all the channels except channel and channel 49. So we conclude that channels from 1 to 48 are available for the Mica2 nodes. Now a new question is arisen: Can all these channels be used simultaneously without interference with each other? In the following subsection, we present the methodology to examine the impact of simultaneous transmission in the adjacent channel and determine the number of non-overlapping channels. Figure 4. Experimental scenario to determine the number of non-overlapping channels 2.2. Number of non-overlapping channels As shown in Figure 4, a more complicated scenario is set up to examine the impact of simultaneous transmission in the adjacent channel. The program running on one Mica2 node (node 1) is the same with the previous scenario described in Section 2.1. Once tuned on, node 1 starts transmitting messages to a sink node (sink 1). For the other Mica2 node (node 2), it listens to the channel when it is tuned on. Node 2 keeps listening to the channel until it receives a SetChannelMsg from node 1. Once receiving a SetChannelMsg, it tunes its radio to the next channel of node 1 and starts transmitting message to the other node (sink 2). In such way, node 1 and node 2 are synchronized to send messages to the corresponding sink nodes in two adjacent channels respectively. Figure 5 shows the ratio of packet successfully received by the sink 1 and sink 2 in each channel. Compared with Figure 1, we can see that the packet reception ratio of sink 1 in this scenario is almost the same with that in the previous scenario. So we can conclude that channels from 1 to 48 are non-overlapping for the Mica2 nodes. In short, through extensive experiments we find that there are 48 orthogonal channels available for the Mica2 nodes. 3. What s the time cost of switching between two channels? The time cost of channel switching is mainly resulting from the self-calibration when the radio is tuned to a different channel. Referring to the datasheet of CC1, the calibration consists of two periods. One is for calibrating the receiving frequency, and the other is for calibrating the transmitting frequency. In the following paragraphs, we refer to these two periods as Calibrating RX and Calibrating TX for short. The main task of this section is to present the method of measur-

4 Packet Reception Ratio Sink 1.2 Sink Channel Index Figure 5. Packet reception ratios of Sink 1 and Sink 2 in each channel - Figure 6. Test circuit to measure the power consumption of the node in different states ing the durations of these two periods and the measured results Methodology Since the radio will turn on different components in different period of calibrating, it is supposed to have different power consumptions in Calibrating RX and Calibrating TX. So we can determine the durations of Calibrating RX and Calibrating TX, by examining the variation of power consumption of the node over time. Figure 6 shows the test circuit to measure the variation of power consumption when the node is in different states. The test circuit is composed of a Mica2 node, a test resistor R and a DC power supply. The Mica2 runs a testing program, which puts the radio tuned on and off alternately in a cycle of 1 ms. When the radio is turned on, it switches channel before listening in it. We know that the current consumption will be varying with the state change of the node, and the variation can be probed by an oscilloscope. For the very low power RF transceiver, like CC1, the variation is too small to be observed. So we connect a 22Ω resistor in series with the Mica2 Figure 7. Measurement setup node, and an instrumentation amplifier (Analog Devices AD62 [15]) with a gain of 5.13 is used to amplify the voltage across the resistor. This amplified voltage is measured by a Tektronix 1MHz digital oscilloscope (Tektronix TDS 112B [16]) and a 2 MHz 1X probe. The oscilloscope provides a sampling resolution up to 1 GHz. The DC power supply used is a Matrix MPS33L-3. It is adjusted to output a voltage of 3.3 V as the supply of V cc in the test circuit. Figure 7 shows the measurement setup, and a screen capture of the oscilloscope is shown in Figure 8. For later analysis, we save the sampled data of the oscilloscope to a file. The plot of the saved file is shown in Figure Result From Figure 9, we can observe the state change of the node clearly. When the node starts up, it firstly switches channel, and then keeps listening until it turns off its radio in the next cycle. To locate the starting time and the ending time of each state, we extract a segment of data from the saved file. The plot of the extracted data is shown in Figure 1. From this figure, we can clearly see the two peri-

5 Voltage (V) Calibrating RX Calibrating TX Starting Listening Time (ms) Figure 8. A screen capture of the oscilloscope Voltage (V) Channel Switching 2 4 Time (ms) Listening Figure 9. State change of the node Power down ods, namely Calibrating RX and Calibrating TX, during channel switching. One may wonder why there is a Starting period after calibrating. It is because in our preliminary experiment we find that it is imperative to restart the radio after calibrating, hence we add such process to complete the channel switching. As shown in Figure 1, the duration of each period of channel switching can be determined if the x- coordinate of the four turning points are known. The duration of each period is computed by the following equations. T cal rx = X 2 X 1, T cal tx = X 3 X 2 T start = X 4 X 3, T total = X 4 X 1 (1) Going through the saved file, we find the corresponding entries of the turning points. Using the time column of the entries, we get the duration of each period of channel switching, which is shown in Table 1. From this table, we can conclude that it costs the node about 5 ms to switch channels. Due to the limitation of the oscilloscope, we can only save 25 samplings to the file. Judging from the previous measurements, we know that the whole period Figure 1. Periods of channel switching State Duration (ms) Calibrating RX 22.8 Calibrating TX Starting 4.32 Total Table 1. Duration of each period of channel switching of channel switching is about 5 ms. So we set the sampling rate of the oscilloscope to 25/5 = 5KHz. In such way, we capture the full period of channel switching with a time precision of.2 ms. This precision is eligible for us to estimate the time cost of channel switching. 4. What s the energy cost of channel switching? In the previous section, we present the details of how to determine the duration of channel switching. The essence of the method is examining the variation of power consumption of the node over time. So in the previous experiments we have got all the data needed to determine the energy cost of channel switching. In this section we show the details of how to compute the power consumption of channel switching, using the measured data of the previous experiment Methodology It should be noted that the measured data shown in Figure 1 is the instantaneous voltage across the test resistor V r (t), which is amplified by a factor of So the instantaneous current of the test resistor is: I r (t) = V r(t) R G, (2)

6 Current (ma) Calibrating RX Calibrating TX Listening Starting Time (ms) State Current (ma) Power (mw) Energy (nj) Calibrating RX Calibrating TX Starting Total Table 2. Energy consumption in each period of channel switching Figure 11. Current consumption in a period of channel switching where R is 22 Ω and G is Since the test resistor is connected in series with the Mica2 node, we can get that I m (t) = I r (t). (3) Figure 11 shows the instantaneous current of the Mica2 node in a cycle of 1 ms. Seeing the small vibration of current consumption during each period of channel switching, we compute the average to indicate the current consumption in each period of channel switching. The computations of the averages are formulated by the following equations. I cal rx = ΣI(t) N r, I cal tx = ΣI(t), I start = ΣI(t) (4) N t N s I(t) is the discrete samplings of current consumption in the corresponding period of channel switching, while N r, N t and N s are the number of samplings. The power consumption of the node in different periods of channel switching can be computed by the following equations. P cal rx =V m I cal rx (t),p cal rx =V m I cal tx (t),p start =V m I start (t) (5) In our experiments, the input voltage of the Mica2 node is adjusted to a constant, i.e., V m = 3V. So the energy cost of channel switching is E cs = (P cal rx T cal rx +P cal tx T cal tx +P start T start ) 1 nj (6) 4.2. Result Table 2 shows the results obtained from the measured data using the above equations. We can see that for the CC1 radio the total energy consumed in channel switching is nj. 5. Related work Recently, some researchers try to implement their proposed multi-channel protocol in the real testbed. An important parameter that they should know in advance is the number of available orthogonal channels in the networks. Mishra et al [17] did some empirical studies to explore this property of channels. Wu et al [18] investigated this property for the nodes equipped with CC242 radio. Although the authors specified the number of available orthogonal channels, they did not make any efforts to answer the other two questions listed in Section 1. Shnayder et al [19] conducted some measurements to profile the node s power consumption in different states. However, they did not give any results on the power consumption of channel switching, or the time cost of channel switching. Le et al [2] considered the channel switching time in protocol design, but did not do any quantitative research on it. 6. Conclusion and future work In this paper, we conduct extensive empirical studies to get three important parameters for design of multi-channel protocols in wireless sensor networks. The three important parameters are (1) number of available orthogonal, (2) time required to switch between two channels, and (3) total energy consumed to switch between two channels. Using the Mica2 node as the testing platform, we present our efforts to determine these three parameters for the CC1 radio. Here we summarize the results drawn from our measurements, and discuss their implications for design of multi-channel protocols in wireless sensor networks. 1. There are 48 channels available for the Mica2 nodes. The frequency separation between two adjacent channels is about 1 MHz. Compared with the 16 non-overlapping channels for the CC242 radio [12], CC1 radio seems to have some ad-

7 vantages in providing supports for multi-channel protocols. However, CC1 has much lower bandwidth compared with CC It requires about 5 ms to switch between two channels. For the CC1 radio, when the data rate is 19.2 Kbps it can send 12 bytes, i.e., at least 3 messages of maximum length, in the period of channel switching. So when designing multichannel protocol for the wireless sensor networks, we have to take this overhead into consideration. 3. It consumes about 194 nj energy to switch between two channels. As shown in Figure 11, the power consumption of channel switching is much higher than that of listening. So it is necessary to avoid switching channels frequently for the purpose of energy efficiency. It is worth noticing that the methodologies presented in this paper can be easily extended to determine these parameters for Zigbee (CC242) nodes, such as MicaZ and TelosB. In the future, we plan to complete these work, and make a comprehensive comparison of the costs of channel switching for different radio modules. A. Channel List Frequencies of all the channels used in the paper is listed in Table 3. References [1] P. Bahl, R. Chandra, and J. Dunagan. SSCH: Slotted Seeded Channel Hopping for Capacity Improvement in IEEE Ad-Hoc Wireless Networks. In Proceedings of ACM MobiCom 4, pages , Pennsylvania, USA, 24. [2] J. So and N. Vaidya. Multi-channel MAC for Ad hoc Networks: Handling Multi-channel Hidden Terminals Using A Single Transceiver. In Proceedings of ACM MobiHoc 4, pages , Tokyo, Japan, 24. [3] J. Mo, H.W. So, and J. Walrand. Comparison of Multi-channel MAC Protocols. In Proceedings of ACM MSWiM 5, pages , Montreal, Canada, 25. [4] K. Xing, X. Cheng, L. Ma, and Q. Liang. Superimposed Code Based Channel Assignment in Multi-Radio Multi- Channel Wireless Mesh Networks. In Proceedings of ACM MobiCom 7, pages 15 26, Montreal, Canada, 27. [5] G. Zhou, C. Huang, T. Yan, T. He, J. A. Stankovic, and T. F. Abdelzaher. MMSN: Multi-Frequency Media Access Control for Wireless Sensor Networks. In Proceedings of IEEE INFOCOM 6, pages 1 13, Barcelona, Spain, 26. Channel Index Specified frequency (MHz) Actual frequency (MHz) Table 3. List of channels generated by the channelgen.exe, with frequencies of 1MHz spacing specified as the parameters of the command

8 [6] K.R. Chowdhury, N. Nandiraju, D. Cavalcanti, and D.P. Agrawal. CMAC - A Multi-channel Energy Efficient MAC for Wireless Sensor Networks. In Proceedings of IEEE WCNC 6, pages , Las Vegas, USA, 26. [7] J. Zhang, G. Zhou, C. Huang, S.H. Son, and J.A. Stankovic. TMMAC: An Energy Efficient Multi- Channel MAC Protocol for Ad Hoc Networks. In Proceedings of IEEE ICC 7, pages , Amsterdam, Netherlands, 27. [8] X. Chen, P. Han, Q.-S. He, S.-L. Tu, and Z.-L. Chen. A Multi-Channel MAC Protocol for Wireless Sensor Networks. In Proceedings of IEEE CIT 6, pages , Washington, DC, USA, 26. [9] Y. Kim, H. Shin, and H. Cha. Y-MAC: An Energy- Efficient Multi-channel MAC Protocol for Dense Wireless Sensor Networks. In Proceedings of ACM IPSN 8, pages 53 63, Missouri, USA, 28. [1] Jason L. Hill and David E. Culler. MICA: A wireless Platform for Deeply Embedded Networks. IEEE Micro, pages 12 24, 22. [11] CC1 Single Chip Very Low Power RF Transceiver. [12] CC GHz IEEE / ZigBee-ready RF Transceiver. [13] J. Polastre, R. Szewczyk, and D. Culler. Telos: Enabling Ultra-low Power Wireless Research. In Proceedings of ACM IPSN 5, page 48, Los Angeles, California, 25. [14] A. Woo S. Hollar D. Culler J. Hill, R. Szewczyk and K. Pister. System Architecture Directions for Networked Sensors. In Proceedings of the ASPLOS, pages 93 14, Cambridge, MA, USA, 2. [15] Analog Devices AD62 Instrumentation Ampifier. [16] Tektronix TDS 112B Digital Storage Oscilloscope. [17] A. Mishra, V. Shrivastava, S. Banerjee, and W. Arbaugh. Partially Overlapped Channels Not Considered Harmful. In Proceedings of ACM SIGMETRICS 6, pages 63 74, Saint Malo, France, 26. [18] Y. Wu, J.A. Stankovic, T. He, and S. Lin. Realistic and Efficient Multi-Channel Communications in Wireless Sensor Networks. In Proceedings of IEEE INFO- COM 8, pages , 28. [19] V. Shnayder, M. Hempstead, B. Chen, G. Werner, and M. Welsh. Simulating the Power Consumption of Largescale Sensor Network Applications. In Proceedings of ACM SenSys 4, pages 188 2, Baltimore, MD, USA, 24. [2] H. K. Le, D. Henriksson, and T. Abdelzaher. A Practical Multi-channel Media Access Control Protocol for Wireless Sensor Networks. In Proceedings of ACM IPSN 8, pages 7 81, Missouri, USA, 28.