Multichannel MAC. Abstract. 1. Introduction. 2. Background

Transcription

1 Multichannel MAC Team: Juliet Nilesh Mishra, Moo Ryong Ra, Vivek Suriyanarayanan EE652 Wireless Sensor Network, Fall 2007 Instructor: Bhaskar Krishnamachari Abstract Current MAC protocols use a single static frequency allocation for all the links in the network. The 2.4 GHz ISM band is shared by a number of devices along with protocol. External interference originating from devices other than networks can lead to considerable drop in throughput. Current congestion avoidance protocols cannot differentiate between in-network interference and interference coming from external sources such as an radio. The first scenario is correct for usage of congestion control where as in the latter case one has to look for other mechanisms. In this work our contribution is two folds. We first demonstrate a mechanism for detection of interference from sources other than We propose two methods for interference mitigation in case we detect external interference by first considering the scenario when we can assume presence of a control channel and next when we do not have any control channel. We have implemented the control channel scheme and present the results subsequently. 1. Introduction A sensor network is spread both spatially and temporally. The channel condition in the network need not be uniform across all parts of the network over a given period of time. In today s networks we use a single network wide frequency scheme that simplifies development, at the same time suffers from a number of problems. Since, shares the 2.4 GHz ISM band it can suffer from heavy interference by devices using other protocols and Bluetooth devices (both operate in the 2.4 GHz band along with various proprietary protocols) have moved from niche applications into commodity products and one comes across them more often devices can specially affect the operation of devices as they have order of magnitude higher transmit power (26 dbm) as compared to devices (0 dbm) [4]. Co-location of such networks is also becoming more common as both the protocols provide different benefits. On one hand where gives low power, low rate, small range operations devices can provide high bandwidth, long range communication capabilities (with certain changes in software over COTS hardware). A very good example can be seen in the tired networks such as Tenet. Since both the protocols cannot perform inter protocol communication and being the more powerful transmitter we need to build interference mitigation schemes in network. Another problem suffered by static single frequency usage across the network is that all the links in the network contend for the channel. This effectively reduces the channel throughput as it is possible that different links not sharing a node can do simultaneous transfers. The possibility becomes more attractive when we have multiple sinks (such as in CTP) and hence if we can come up with schemes allowing simultaneous data transmissions we can increase the effective throughput of the network. The goal of our project is to make the current CSMA MAC scheme robust against interference by making links frequency agile, which can allow effective communication within the network and co-existence of and networks. We demonstrate using a scheme using the clear channel assessment mode specified in the specification, the ability to differentiate interference due to network from that of radio. We propose two schemes of frequency changing one using a control channel and the second using a seeded channel hopping scheme. We have implemented the first scheme and verified that it works under interference from data source. There are several related works which give schemes to allow the co-existence of and radio or channel hopping. Most of them do not have any actual implementations and evaluate their schemes and protocols in simulations or give a theoretical analysis. Our contribution is two folds in terms of a working method to distinguish between and interference and a prototype implementation for making link interference agile. 2. Background Both and protocols utilize completely different modulation schemes and both the protocols cannot understand packets originating from other protocol s radio. Correct operation of CSMA MAC schemes require some way to do clear channel assessment (CCA) as to when they can treat the channel free specification provides three different modes for CCA [13].

2 Mode 1: Clear channel when received energy is below threshold. Mode 2: Clear channel when not receiving valid IEEE data. Mode 3: Clear channel when energy is below threshold and not receiving valid IEEE data In mode 1 the radio uses a programmable threshold to compare the incident energy at the antenna to decide whether the channel is busy or not. In mode 2 the CCA shall report a busy medium only upon the detection of a DSSS (Direct Sequence Spread Spectrum) signal originating from an source. In this mode the signal may be above or below the energy threshold used in mode 1 (and 3). In mode 3 it looks for both DSSS signal and comparison of incident energy with the threshold to decide clear channel. Current networks use mode 3 which is also the default mode used in TinyOS a popular open source operating system used widely for application development. The CCA scheme is used only at the transmitter to access the channel condition. Depending on the MAC implementation it can proceed with transmission or perform backoff as in traditional CSMA-CA MACs. From the state machine described in [13] one can infer that at the receiver the radio keeps looking for the preamble byte to decide whether there is incoming traffic. As long as the SINR ratio of signal against the interference power of transmission is above a certain threshold ratio, the receiver can have successful reception. Hence, it is possible for the transmitter to send packets into the medium without backing off even in the presence of interference from traffic. [Figure 1] In the specifications channel are provisioned in the 2.4 GHz band. According to the specifications applicable in US, channel of overlap with the channel 1-11 of (Figure 1). Thus, channel 25 and 26 are relatively interference free from They can suffer from interference from other sources such as microwave and leaky electromagnetic equipment e.g. electric motors. 3. Related work The works closest to our work are [1][3]. In [1] the authors detect channel interference from interferer by a drop in throughput but a high energy level. The nodes facing interference form group and change their channel of operation while the nodes falling on the border of interference region maintain a neighbor table consisting of the frequency of operation of neighbors. Packets traversing through the interference suffering region are received and transmitted on different frequencies using this neighbor tables. The authors fail to address some important system design decisions. First they assume that the nodes are able to communicate with each other in the presence of interference during group formation which need not be true always. They do not mention the scheme used in populating the network table or the switching table used by the nodes when they decide to switch to new frequency. Also they fail to give the strategy followed by the border nodes while doing frequency translation for the traversing traffic i.e. how much time the node spends in the multiple frequencies it needs to operate in. Nor do the authors mention how they arrive at the threshold value used to decide interference. Finally the authors do not provide with a complete implementation and evaluate in a very limited setting. In [3] the authors propose a learning based scheme to find the optimal strategy of locating a new non interfered channel. The authors do a theoretical and simulation scheme without an implementation. In [9] Gumandi et al propose a fast channel hopping scheme based on shared hopping sequence to overcome an interfering adversary. We also use a similar mechanism to use a common seed for a link change channels only on finding interference and not pro-actively as in this work. Multiple channel MAC's have been proposed earlier in many works but all of them were considered for increasing in network throughput by allowing multiple simultaneous transmissions over multiple links. We propose the use of multiple channels (or per link channel) for interference mitigation. SSCH [5] uses seeded channel hopping sequence to increase throughput. In [6][7][8] authors propose use of separate control channel for control packet exchange and data channel for normal data transfers. All the above schemes have been evaluated only in simulation and we give a working prototype for one such scheme. 4. Approach Our work can be broadly divided into two parts Interference detection part Interference mitigation part (Frequency agility)

3 4-1. Interference detection We expect that a significant contribution of our work is to provide a scheme which differentiates the behaviors due to interference of from radio and protocols use different physical layer encoding schemes to transmit data. Hence, they cannot communicate with each other. Transmissions from either protocol's radio will appear as noise for the other protocol's radio, increasing the ambient energy at the antennae. Mode 2 of the CCA modes specified in specifications provides a clear channel only in the case if the radio detects presence of a preamble sequence on the air from an node. This mode disregards the RSSI incident at the mote s antenna. Thus if we find a clear channel and high RSSI value reported by the radio then we can conclude that the radio has interference from or other external sources. In this scheme the next challenge is to come up with a threshold value for deciding whether the interferer is powerful enough to disrupt communication. For this we take insights from the following: 1. For modes 1 and 3 the threshold value to decide clear channel is set to -77 dbm [13]. 2. For successful communication via capture effect on CC2420 radio we need to have SINR ratio of 8 db or more [14], [15]. From the above two we can say that the threshold value can be set at -85dBm. This value must be considered as the worst case scenario. This scheme is passive in terms that we do not perform and packet transmissions. Since, the link(s) used by the node is question can be performing on a better link quality (based on RSSI) we can use an active measure such as throughput, delay and packet reception rate and/or beaconing with acknowledgements (heart beats) to capture channel condition for robustness Interference Mitigation Once interference is detected in a section of the network we need to take steps to mitigate the effect. A number of questions arise in this context. When do we decide that the interference is severe enough to go to a new frequency? An associated question would be to decide upon the time frame after which we start taking interference mitigation steps. Overhead associated with the interference mitigation scheme. Which part of the network shall initiate this process? How much of the network shall see a change? Shall it be centralized or distributed? Is their a control channel present in the system or not which is not affected by interference and hence can be used intermittently to negotiate a new channel of operation? Should we change the channel pro-actively on detecting external interference or wait for some data to be transmitted? To address these questions we follow a two fold strategy. Apart from the scheme for channel interference detection in the case when the data transmission is busty and sporadic we use a periodic beacon system. We send a periodic beacon on a link and receive an acknowledgement for it. If there is a successful acknowledgement, it implies that the channel has no or low interference. If there are a series of acknowledgements missed in a window then the transmitter node uses one of the mitigation scheme mentioned below. The network on the whole can be broken down as aggregation of links used for communication. Thus on a micro scale we can measure the effect of interference on a link and decide upon whether to perform interference mitigation or not. Any scheme above this can suffer from problem deciding who initiates the process. Also this scheme will be distributed and limited only to set of nodes affected by the interference. Channel migration can be done proactively depending on application s data transfer profile. Channel Hopping Strategies. Strategy 1: data transfer in many countries is restricted to operate only in the 11 of the 14 possible channels in the ISM band. This provides two primary clear channels in channel number 25 and 26 of channel space. Also the use of channels 1, 6 and 11 as orthogonal non-interfering channels gives 2 more secondary clear channels numbered 15 and 20. Thus we can assume that in presence of only interference we can have a pre-negotiated control channel. On detection of interference the transmitter moves to the control channel and starts sending frequent beacons (frequency more than regular beaconing frequency of heart beats). As soon as the receiver timeouts it also moves to the control channel and they negotiates over for a free channel based on both the receiver's and transmitter's perception of channel quality (sampling strategy not yet thought of completely and left for future exploration). Strategy 2: If the interfering source is other than the interferer then the assumption about an interference free control channels fails. In this scenario we can use the following scheme. The transmitting and receiving node have a common seed number. When the beacon fail the transmitter moves over to the next free channel using the following scheme x = (x + seed) mod 13 as explained in [5], here x is index of the frequency array deciding the frequency of communication. Here to make the function go over all the channels we need to use a prime number of channels (we can also use 17 for modulus with channel 16 and 0 mapped to channel 26 for example). In the case of using 13 we can skip some channels. In this case when a node goes through a complete cycle and does not come across a channel of

4 rendezvous it can initiate the routing as it might be the case that the receiver or the transmitter has been caught in heavy interference or died off and a new parent in the routing tree might be required. The nodes update the seed every time they successfully rendezvous to a common channel. Keeping a separate Rx and Tx frequency for each node (i.e. a node has a tx frequency for communicating with its parent and a rx frequency to communicate with its children in tree topology) solves the problem of deciding which section of the network need to adapt to the interference. As mentioned above since the basic entities in the network are the links we perform mitigation scheme only on the affected links. Also this scheme will be fully distributed hence scalable. In this scheme the node spends most of the time in the Rx frequency and then periodically moves to its Tx frequency for sending the beacons. Each node's parent knows each of its child's sending and receiving frequency hence if it needs to communicate to the child at any time it can switch to its child's receiving frequency and proceeds with communication (this will although be rare as in most of the sensor applications the data flow is from the edge towards the root sink. E.g. test scenario can be the tenet network where the nodes have tasklets to perform the job and send the data to higher tier nodes while the communication towards nodes is mostly a script describing the sequence of tasks and tasklets to use a much less frequent and sporadic event. Assumptions interferer is a persistent interferer that is it continuously sends packets when it is on. Can be achieved by buffer overflow at the transmitter so that it always has some data to send. 2. The interfering energy of the transmissions is addressed as the noise. Higher value employs higher noise floor. 3. The interferer can overwhelm the for some power setting i.e. no transmission can happen as noise floor due to interfering radio is above the RSSI of receivable packet when no interference happens. 4. We are not taking care of sleeping, power or anything else. Primary concern is to make the nodes communicate. 5. There are lots of parameters in the algorithms which need to be decided by experiments for optimal performance. We currently don't shoot for optimal performance but a working solution. 5. Experiment Design In order to evaluate our work we designed a series of experiments to validate our claims Devices and software [Figure 2] Motes: Tmote-sky, TIP800CM(MAXFOR). Stargate and WLAN card as an interferer. WI-SPY USB spectrum analyzer and its software. Several built-in Tiny-OS tools Interference source We use a Stargate single board computer and a wireless card (SMC 2532W-B) to generate interference. We use hostap driver on linux to set the card in ad-hoc mode and generate a UDP traffic stream over the interface to create data transfer. We change the data transfer mode to 1Mbps and send large enough packets to do buffer overflow. This ensures that the channel is continuously having generated energy and models it as a persistent interferer. [Figure 3] Figure 3 gives the state of 2.4 GHz band as visible by the Wispy tool at the experiment location. Since the minimum interference was observable at Wifi channel 8 (2447 MHz), we performed our experiments around it. [Figure 4] Figure 4 displays the channel condition once the interference source is switched on. The maximum power value was greater than -40dBm. For our experiments we wanted to completely block the channel by putting a persistent sender (putting traffic on channel all the time) but could not achieve it. We want to address this in future Design of interference detection experiments In order to validate the detection and effect of

5 interference on communication we set up a pair of sender and receiver nodes. The transmitter is set to perform transfer at maximum transmit power. The sender sends a series of packets containing the following data payload: 1. Serial number 2. Time stamp for sending packet (serial number modulo 100) == 1 i.e. first packet say, TS Time stamp for sending packet (serial number modulo 100) == 0 i.e. 100 th packet, say TS 2. At the receiver we observe the following: 1. Count of number of packets received per 100 packets sent by the sender, say C. 2. Time stamp for first packet received, say TR Time stamp for 100 th packet received, say TR 2. Thus we get 1. Packet reception rate (PRR) = C 2. Delay at the sender = (TS 2 TS 1 ) 3. Throughput indicator = (TR 2 TR 1 ) (higher value employs lower throughput) PRR reflects the number of packets received from the sender at the receiver and the packets lost due to corruption or high noise. Delay at sender reflects whether the sender was observing channel busy and hence had to backoff at MAC. Throughput indicator when compared with delay at sender reflects whether the low PRR is due to backoff or packet corruption. We perform this experiment for all the three modes of CC2420. The results are described in section 6 below Design of interference mitigation exp. In order to give artificial interference, we used Stargate as an interference source. Procedures are like below. 1. First two motes communicate with each other using channel number We turn on Stargate which was programmed to interfere with our motes communication in channel 7 (maximum interference on channel 12). 3. After some time, our network will detect interference and try to do something for mitigation. 4. Both motes move into the negotiation channel which is After negotiation, both motes will move to the new channel which is free of interference (in this case hard coded to channel 19). 6. Now, our network recovers back to healthy communication status. source. Procedures are like below. 1. First two motes communicate each other using channel numbers which is generated as a result of the random seed. For example if the seed is 2 and the prime number chosen is 13. Then, the common sequence of the channel will be (2,3,6,8,10,12,1,3,5,7,9,11,0/13). This will be the common sequence for both the sender and the receiver. 2. Turn on stargate which was programmed to interfere with our motes communication. 3. After some time, our network will detect interference and try to do something for mitigation. 4. The receiver mote, not receiving the beacon signal, jumps to the next channel in the sequence say 2, and wait for the specified timeout for a beacon signal from the sender. The receiver keeps hopping to the next channel in a loop, similar to scanning mode operation. 5. The sender mote, not receiving the ack message within the specified timeout, jumps to the next channel in the sequence, sends a beacon message and waits for the specified timeout for an ack. If the ack in not received within the timeout, it keeps jumping to the next channel and starts sending beacon signals. 6. The hopping of both the sender and the receiver stop once the beacon signal is received at the receiver and the corresponding ACK is received at the sender. The implementation of this strategy is part of the future work. 6. Evaluations 6-1. Basic Channel Characteristics We did two experiments to see the basic behavior of a channel. It is to strengthen our motivation and gives plausibility to go further steps. First, we try to find noise floor value under the circumstance with no interference. The result is like below. X-axis is sampling count (each one is an average of 128 consecutive sampling values) and Y-axis is dbm. Average RSSI is Actually, we could get reasonable result using above procedure. Details are in the next evaluation section Design of interference mitigation experiment with pseudo-random seeding Although this part has not been implemented, we have come up with a solution for the random channel hopping statergy by keeping the seed constant for both the sender and the receiver. In order to give artificial interference, we used stargate as an interference [Figure 5 dbm/time] Second, in order to compare the result of no interference case, we prepare two more motes and broadcast some data right besides of the sampling

6 mote. The result is like below. You can see the result that after injects interference from nodes, we can easily see there are more fluctuation than before and bigger RSSI level. Figure 7 and 8 give the PRR for all the three modes when the sender is set to transmission power of 0dBm and -25dBm respectively. We can see that the PRR decreases as we initiate interference from source. The average PRR values during interference period are given in Table 1 below. 0dBm txpower -25dBm txpower Mode Mode Mode [Figure 6 dbm/time] 6-2. Effect of Interference and detection We conducted the experiments with the following settings: 1. Both the sender and the receiver are exposed to the same interference source with the same level of interference. 2. Both the sender and the receiver are set in the same CCA mode. [Table 1-average PRR during interference] Figures 9-11 give the delay at the sender and throughput indicator for the receiver when they are set in modes 1-3 respectively. We report these values for transmission power of 0dBm only as the values are the same for -25dBm. [Figure 9 Delay and throughput indicator values (mode 1)] [Figure 7 PRR for different CCA modes at 0dBm transmission power at sender] [Figure 10 Delay and throughput indicator values (mode 2)] [Figure 8 PRR for different CCA modes at -25dBm transmission power at sender]

7 To validate this we tried putting the sender and receiver in different modes and the results are given in figure Here we observe that when the transmitter is agnostic to it will transmit without any change in presence of interference. This can help us in the scenario when the sender and the receiver are receiving asymmetric interference from source. To validate we create a scenario when the difference of signal and interference is greater than 8dB at the receiver. The results are given below (Figure 13-14) for sender in modes 2 and 3 respectively. [Figure 11 - Delay and throughput indicator values (mode 3)] The PRR drop is most when the sender and the receiver are set in mode 2 of CCA. This can be attributed to the fact that the sender is agnostic to the presence of source (no change in delay at sender value in figure 9) and hence is not aware of the situation at the receiver. Due to higher value of the noise (-40 dbm) than the signal (max value is -50dBm) the receiver is not able to demodulate the packet hence the drop in PRR. [Figure 13 - Delay and throughput indicator values (mode 2) (interference -65dBm, signal -41dBm)] [Figure 12 - Delay and throughput indicator values (Sender mode 3, Receiver mode 2)] [Figure 14 - Delay and throughput indicator values (mode 3) (interference -65dBm, signal -41dBm)] In both the scenarios the PRR was 100% but as we can see that the delay at the sender is almost twice when the clear channel assessment mode is 3 and hence an increase in the throughput indicator value. This scenario shows that mode 2 which is agnostic can be used both as an indicator of interference from external source as well as used to mitigate interference in cases where the capture effect is in favor of [Figure 13 - Delay and throughput indicator values (Sender mode 2, Receiver mode 3)] 6-3. First Channel Hopping Strategy. We implemented a channel hopping strategy 1 and did

8 experiment with two motes and one stargate. graph(fig.19). [Figure 15 channel no/time] At figure 15, you can see the network change its communication channel from 12 to 19 in order to mitigate an interference from interference source. During the channel transition, it went negotiation channel which is 26 as shown by small point at figure 15. [Figure 19 PRR/time] Easily we could verify all meaningful communication metrics recovered after a channel transition, it could reveal our mitigation strategy has sufficiently meaningful with respect to the robustness of the network under external interference. For example, although interference still exists until 200 secs, throughput recovered to the amount of original channel s one(fig.18) and PRR also showed similar behavior with throughput(fig.19). 7. Conclusions and future works [Figure 16 sequence number/time] At Figure 16, we can see some holes before mitigation process will come up. We saw feasible possibility to detect interference and to mitigate interference from interference during our experiment. If we can differentiate interference from data, we can choose the most appropriate method to mitigate it. Interference from in-network traffic is classified as congestion and necessitates to choose congestion control mechanism such as IFRC [11], WRCP, RCRT [12]. When the interference comes from external source other than , we can use one of the interference mitigation scheme proposed above. Using the appropriate schemes our network will be more robust than current implementations. For future work we plan to investigate the following: [Figure 17 rssi/time] 1. Integrate detection part with channel hopping parts. 2. Implement second channel hopping strategy. 3. Test the scheme on larger real network such as ENL test-bed. 4. Compare with other centralized schemes in terms of performance, throughput, PRR. On completion, our major contributions will be [Figure 18 throughput/time] We tried to investigate as many aspects as possible during interference mitigation experiments. We extracted sequence number of all packets(fig.16), received signal strength value(fig.17) and timestamp. From timestamps and packet information, we calculated throughput(fig.18). And from timestamp and sequence number of packets, we could get PRR 1. Our study is based on practical experimentation with minimum assumption rather than simulation 2. Differentiating between and interference using the CCA modes of CC2420 radio and its working implementation. 3. Interference mitigation schemes against external interference source explored with real implementation. Acknowledgement Prof. Bhaskar Krishnamachari helped us throughout the

9 semester by shepherding our ideas and implementations. We took help from a number of people from ENL, USC. We discussed and modified our approach with inputs from Prof. Ramesh Govindan, Jeongyeup Paek and Sumit Rangwala. Ki-Young Jang and Vinayak S. Naik (from CENS, UCLA) helped in setting up Stargate platform to generate interference. We would also thank Amitabha Ghosh our cheerful course TA for discussions. References [1] Won et al, Adaptive Radio Channel Allocation for Supporting Coexistence of and b [2] Mishra et al, Wake-on-WLAN, Proceedings of the 15th international conference on World Wide Web, May 23-26, 2006, Edinburgh, Scotland [3] Pollin el al, Distributed cognitive coexistence of with [4] Petrova et al, Interference Measurements on Performance Degradation between Co-located IEEE g/n and IEEE Networks [5] Bahl et al, SSCH: Slotted Seeded Channel Hopping for Capacity Improvement in IEEE Ad-Hoc Wireless Networks [6] Wu, et al, A New Multi-Channel MAC Protocol with On-Demand Channel Assignment for Multi-Hop Mobile Ad Hoc Networks, International Symposium on Parallel Architectures, Algorithms and Networks (ISPAN '00), 2000 [7] Li, et. al., MAC-SCC: Medium Access Control with a Separate Control Channel for Multihop Wireless Networks, IEEE Transactions on Wireless Communications, Vol. 5, No. 7, [8] Zhou, et al, MMSN: Multi-Frequency Media Access Control for Wireless Sensor Networks, Infocom 2006 [9] Gummadi et al, Understanding and Mitigating the Impact of RF Interference on Networks, Proceedings of the ACM SIGCOMM 2007, Kyoto, Japan, Aug [10] TinyOS website, [11] S.Rangwala et al, Interference-Aware Fair Rate Control in Wireless Sensor Networks, Sigcomm [12] J.Paek et al, RCRT: Rate-Controlled Reliable Transport for Wireless Sensor Networks, SenSys [13] CC2420 datasheet, [14] Son et al., Experimental Study of Concurrent Transmission in Wireless Sensor Networks, Sensys [15] EE652 class project SINR, tel.pdf [16] Collection tree protocol 2.x/doc/html/tep123.html