B-MAC: Versatile Low Power Media Access for Wireless Sensor Networks

1. B-MAC: Versatile Low Power Media Access for Wireless Sensor Networks SenSys ’04

2. Reminder: Proposal Presentation on Monday, March 5 • Prepare powerpoint slides to motivate your project and show the overall approach to tackling the problem • Come at least 10 minutes before the class begins • Give your slides to the tech staff • Lean how to use the presentation tool etc. • Teams & presentation order: • Mike & Vic (+ Brian) • Chris & Yan • Dahee & Chao • Surabh • Mehmet

3. Design goals of B-MAC • Low power operation • Effective collision • Simple implementation, small code & RAM size • Efficient channel utilization at low & high data rates • Reconfigurable by network protocols • Tolerant to changing RF/Networking conditions • Scalable to large numbers of nodes

4. Small, configurable MAC • Export control to higher services to support wide variety of WSN workloads • WSNs are supposed to support various applications • S-MAC (discussed in the previous class) is more than a link layer protocol • Drawbacks of S-MAC • Scalability: A node may have to remember many schedules and wake up accordingly • MAC layer may not be the best place for sleep scheduling & synchronization

5. Adaptive, reconfigurable MAC • Adaptive bidirectional interface for WSN applications • Reconfigure the MAC protocol based on the current workload • Identify the best parameters for an arbitrary low power WSN applications at compile or run time & estimate the application’s lifetime

6. B-MAC Design • Really simple • CSMA via CCA (Clear Channel Assessment) & backoff • Low power listening via Preamble • Acknowledgment • Enable/disable anything above and allow to build anything on top of the configured B-MAC!

7. B-MAC Interfaces

8. Clear Channel Assessment • Effective collision avoidance • Find out whether the channel is idle • If too pessimistic: waste bandwidth • If too optimistic: more collisions • Key observation • Ambient noise may change significantly depending on the environment • Packet reception has fairly constant channel energy • Software approach to estimating the noise floor

9. Take a signal sample when the channel is assumed to be free • Right after a packet is transmitted or when no valid data is received • Take exponential moving average (EMA) of the median signal strength • Works as a low pass filter • Smoothed idle signal level Sm(t)= a * S(t) + (1 - a) * Sm(t-1) • Sm(t): EMA at time t • S(t-1): Signal strength of ambient noise at t • Sm(t-1): EMA at time t-1 • It contrasts to common threshold-based methods in which only a single sample is taken • Resilient to time-varying ambient noise

10. CCA vs. Threshold techniques Idle CCA dynamically adjusts threshold CCA finds channel busy/idle status with high accuracy • Threshold: waste channel utilization • CCA: Fully utilize the channel since a valid packet could have no outlier significantly below the noise floor

11. CCA can be turned on/off • If turned off, a schedule-based protocol, e.g., S-MAC, can be implemented atop B-MAC • If turned on, initial channel backoff when sending a message • B-MAC does not set the backoff time, but signals an event to the higher service that sent the packet • The higer level service may return an initial backoff time or ignore the event • If ignored, use a short random delay

12. Low Power Listening: Preamble Sampling • Preamble is not a packet but a physical layer RF pulse • Minimize overhead Send data Preamble Sender S R Receiver Active to receive a message Preamble sampling |Preamble| ≥ Sampling period

13. Optional link layer ACK • If enabled, ACK is sent immediately after receiving a unicast packet • Overall, B-MAC is easier to implement than S-MAC • No RTS/CTS • Is this always good?  or ? • We know RTS/CTS can reduce hidden/exposed node problem • You may have to implement RTS/CTS on your own... • Simple but not very friendly • No synchronization • No need for a schedule table in S-MAC • But periodic sleep & wake-up is a good approach to energy saving

14. Modeling Lifetime • Monitoring applications • E = Esleep + Elisten + Ed + Erx + Etx • Given #nodes in the neighborhood, BMAC can estimate the network lifetime • Lifetime tl = 1/E * Cbatt * V * 60 * 60

15. Derivation of Lifetime • Ed = td * cdata * V where td = tdata * r • Etx = ttx * ctxb * V where ttx = r * (Lpreamble + Lpacket) * ttxb • Erx = trx * crxb * V where trx ≤ n * r * (Lpreamble + Lpacket) * trxb • r * sum_i=1^n ( children (i) + 1 ) • Lpreamble ≥ ti / trxb

16. Derivation of Lifetime (Cont’d) • Esample = 17.3 uJ • Elisten ≤ Esample * 1/ti • tlisten = (tr_init + tr_on + trx/tx + tsr) * 1/ti • tsleep = 1 – trx– ttx– td– tlisten • Esleep = tsleep * csleep * V • Lifetime tl = 1/E * Cbatt * V * 60 * 60

17. Network Parameters • Scientists may determine the physical location of the nodes & ideal sampling rate • Compute the parameters to get the best lifetime that B-MAC can achieve • Best LPL check interval is the lowest line at a given network density in the following graph If n=60, 25ms is best If n = 20, 50ms check interval is optimal

18. Experiments • Compare BMAC to S-MAC & T-MAC • T-MAC is similar to S-MAC, but a receiver goes to sleep if it does not receive any message • B-MAC & S-MAC: implemented in TinyOS • TMAC: simulated in Matlab

19. TinyOS Implementation • B-MAC does not need timestamp

20. Packet transmission time vs. Checking frequency • More frequent checking of the radio • Shorter transmission time • More energy consumption LPL check interval

21. Channel utilization • Place n nodes equidistant from a receiver • Increase n to increase load • BMAC relies on higher level services to send data according to the traffic pattern • Consider hidden node problem too • For example, after a packet is sent to the parent, nodes in the same cell wait for a certain amount of time for the parent to forward the packet up the tree Always work? More difficulty for developing sensing applications?

22. Energy per byte

23. End-to-end latency

24. Contention-Free Protocols

25. Classic Protocols • TDMA (Time Division Multiple Access) • A node can sleep when it is not its turn to send or receive • FDMA (Frequency Division Multiple Access) • CDMA (Code Division Multiple Access) • No node within two hops can use the same slot to avoid the hidden node problem

26. Optimal channel assignment • Achieve contention-free communication using the minimum number of channels • The problem of assigning a minimum number of channels for an arbitrary graph is NP-hard • Develop efficient heuristics • Centralized approaches do not scale

27. Stationary MAC and Startup • Local synchronization only • Starting phase • Handshaking on a common control channel • Each link utilizes a unique random frequency or CDMA frequency hopping code • Assume there are sufficiently many frequencies or codes • Periodically use the slot

28. BFS/DFS-based scheduling • Breadth-first or depth-first traversals of a data gathering tree • Every single node is given a slot • BFS might provide more chances for aggregation • DFS may transmit individual data more quickly • Global synchronization required

29. Reservation-based synchornized MAC (ReSync) • TDMA is not flexible enough to allow the traffic from each node to change over time • ReSync provides more flexibility • Each node maintains an epoch based on its local time (or can be synchronized with nearby neighbors) • Select a regular time in each epoch to send a short intent message • Probability of collisions is low since the intent is very short • Listen long enough to learn when the neighbor is sending the intent • The intended receiver wakes up at the corresponding time to receive the message • No RTS/CTS • Data transmissions are scheduled randomly

30. Traffic-adaptive medium access (TRAMA) • Distributed TDMA for flexible & dynamic scheduling of time slots • Divide time epochs into a set of short signaling slots followed by a set of longer transmission slots • Key components • Neighbor protocol (NP) • Schedule exchange protocol (SEP) • Adaptive election algorithm (AEP)

31. Neighbor Protocol • Nodes exchange one-hop neighbor info during the random access signaling slots • Ensure the slots are long enough to allow all nodes to get consistent two-hop neighbor info

32. Schedule exchange protocol • Each node publishes its schedule during the last winning slot in each epoch • Use bitmaps to indicate the intended unicast or multicast recipients • Sleep when not required to transmit or receive

33. Adaptive election algorithm • Hash function based on node IDs and time • Ensure there’s a unique ordering of node priorities within any two-hop region at each time • A node transmits iff it has the highest priority among the two hop neighbors at the moment • Sophisticated slot reuse protocol

34. Summary • MAC protocols in WSNs • Arbitration for access to the wireless channel • Energy conservation • B-MAC is a nice building block for diverse applications and workloads • TDMA, for example, can be built on top of it • Energy savings • Preamble • S-MAC like periodic sleep & wake-up • Adaptive listening in S-MAC & T-MAC • D-MAC & DESS try to minimize the E2E delay, while using sleep modes • TDMA • No idle listening • Higher complexity distributed algorithms or tight synchronization required

35. Questions?