1 MAC Layer Design for Wireless Sensor Networks Wei Ye USC Information Sciences Institute

2 Introduction  Wireless sensor network  Large number of densely distributed nodes  Battery powered  Multi-hop ad hoc wireless network  Node positions and topology dynamically change  Self-organization  Sensor-net applications  Nodes cooperate for a common task  In-network data processing

3 Introduction  Important attributes of MAC protocols  Collision avoidance Basic task — medium access control  Energy efficiency  Scalability and adaptivity Number of nodes changes overtime  Latency  Fairness  Throughput  Bandwidth utilization

4  Contention-based protocols  CSMA — Carrier Sense Multiple Access Ethernet Not enough for wireless (collision at receiver)  MACA — Multiple Access w/ Collision Avoidance RTS/CTS for hidden terminal problem RTS/CTS/DATA Overview of MAC protocols AB C Hidden terminal: A is hidden from C’s CS

5 Overview of MAC Protocols  Contention-based protocols (contd.)  MACAW — improved over MACA RTS/CTS/DATA/ACK Fast error recovery at link layer  IEEE Distributed Coordination Function Largely based on MACAW  Protocols from voice communication area  TDMA — low duty cycle, energy efficient  FDMA — each channel has different frequency  CDMA — frequency hopping or direct sequence

6 Example: IEEE DCF  Distributed coordinate function: ad hoc mode  Virtual and physical carrier sense (CS) Network allocation vector (NAV), duration field  Binary exponential backoff  RTS/CTS/DATA/ACK for unicast packets  Broadcast packets are directly sent after CS  Fragmentation support RTS/CTS reserve time for first (frag + ACK) First (frag + ACK) reserve time for second… Give up tx when error happens

7 Example: IEEE DCF  Timing relationship

8 Energy Efficiency in MAC Design  Energy is primary concern in sensor networks  What causes energy waste?  Collisions  Control packet overhead  Overhearing unnecessary traffic  Long idle time bursty traffic in sensor-net apps Idle listening consumes 50—100% of the power for receiving (Stemm97, Kasten) Dominant in sensor nets

9 Energy Efficiency in MAC Design  TDMA vs. contention-based protocols  TDMA can easily avoid or reduce energy waste from all above sources  Contention protocols needs to work hard in all directions  TDMA has limited scalability and adaptivity Hard to dynamically change frame size or slot assignment when new nodes join Restrict direct communication within a cluster  Contention protocols easily accommodate node changes and support multi-hop communications

10 TDMA Protocols  Bluetooth  Clustering (piconet)  FH-CDMA between clusters  TDMA within each cluster Centralized control by cluster head Only master-slave communication Master polling slave and schedule tx  At most 8 active nodes can be in a cluster — scalability problem

11 TDMA Protocols  LEACH: Low-Energy Adaptive Clustering Hierarchy — by Heinzelman, et al.  Similar to Bluetooth  CDMA between clusters  TDMA within each cluster Static TDMA frame Cluster head rotation Node only talks to cluster head Only cluster head talks to base station (long dist.)  The same scalability problem

12  Self-Organaiztion — by Sohrabi and Pottie  Have a pool of independent channels Frequency band or spreading code Potential interfering links select different channels  Talk to each neighbor in different time slots  Sleep during unscheduled time  Example 6 different channels Each node has its own scheduled time slots — superframe TDMA Protocols

13 TDMA Protocols  Self-Organaiztion (Contd.)  Result (if a node has n neighbors) Uses n different channels to talk to each of them Schedules n time slots to send to each of them and n time slots to receive from each of them  Looks like TDMA, but actually FDMA or CDMA Any pair of two nodes can talk at the same time  Low bandwidth utilization

14 Contention-Based Protocols  PAMAS: Power Aware Multi-Access with Signalling — by Singh and Raghavendra  Improve energy efficiency from MACA  Avoid overhearing by putting node into sleep  Control channel separates from data channel RTS, CTS, busy tone to avoid collision Probe packets to find neighbors transmission time  Increased hardware complexity Two channels need to work simultaneously, meaning two radio systems.

15 Contention-Based Protocols  Piconet — by Bennett, Clarke, et al.  Low duty-cycle operation — energy efficient Sleep for 30s, beacon, and listen for a while Sending node needs to listen for receiver’s beacon first, then Carrier sense before sending data  May wait for long time before sending  Tx rate control — by Woo and Culler  Carrier sense + adaptive rate control  Promote fair bandwidth allocation  Helps for congestion avoidance

16 Contention-Based Protocols  Power save (PS) mode in IEEE DCF  Assumption: all nodes are synchronized and can hear each other (single hop)  Nodes in PS mode periodically listen for beacons & ATIMs (ad hoc traffic indication messages)  Beacon: timing and physical layer parameters All nodes participate in periodic beacon generation  ATIM: tell nodes in PS mode to stay awake for Rx ATIM follows a beacon sent/received Unicast ATIM needs acknowledgement Broadcast ATIM wakes up all nodes — no ACK

17 Contention-Based Protocols  Example of PS mode in IEEE DCF

18 Energy Efficiency: Contention  Extend PS mode for Multi-hops — By Tseng, et al. at IEEE Infocom 2002  Nodes do not synchronize with each other  Designed 3 sleep patterns — ensure nodes listen intervals overlap, example: Periodically fully-awake interval: similar to S-MAC Problem on broadcast — wake up each neighbor

19 S-MAC: Introduction  S-MAC — by Ye, Heidemann and Estrin  Tradeoffs  Major components in S-MAC Periodic listen and sleep Collision avoidance Overhearing avoidance Massage passing Latency Fairness Energy

20 Periodic Listen and Sleep  Problem: Idle listening consumes significant energy  Solution: Periodic listen and sleep Turn off radio when sleeping Reduce duty cycle to ~ 10% (150ms on/1.5s off) sleep listen sleep Latency Energy

21 Periodic Listen and Sleep  Schedules can differ Prefer neighboring nodes have same schedule — easy broadcast & low control overhead Border nodes: two schedules broadcast twice Node 1 Node 2 sleep listen sleep listen sleep Schedule 2 Schedule 1

22 Periodic Listen and Sleep  Schedule Synchronization  New node tries to follow an existing schedule  Remember neighbors’ schedules — to know when to send to them  Each node broadcasts its schedule every few periods of sleeping and listening  Re-sync when receiving a schedule update  Periodic neighbor discovery  Keep awake in a full sync interval over long periods

23 Collision Avoidance  Problem: Multiple senders want to talk  Options: Contention vs. TDMA  Solution: Similar to IEEE ad hoc mode (DCF)  Physical and virtual carrier sense  Randomized backoff time  RTS/CTS for hidden terminal problem  RTS/CTS/DATA/ACK sequence

24 Overhearing Avoidance  Problem: Receive packets destined to others  Solution: Sleep when neighbors talk  Basic idea from PAMAS (Singh, Raghavendra 1998)  But we only use in-channel signaling  Who should sleep? All immediate neighbors of sender and receiver  How long to sleep? The duration field in each packet informs other nodes the sleep interval

25 Message Passing  Problem: Sensor net in-network processing requires entire message  Solution: Don’t interleave different messages  Long message is fragmented & sent in burst  RTS/CTS reserve medium for entire message  Fragment-level error recovery — ACK — extend Tx time and re-transmit immediately  Other nodes sleep for whole message time Fairness Energy Msg-level latency

26 Msg Passing vs fragmentation  S-MAC message passing RTS Data 19 ACK 18 CTS 20 Data 17 ACK 16 Data 1 ACK 0 RTS 3... Data 3 ACK 2 CTS 2 Data 3 ACK 2 Data 1 ACK 0  Fragmentation in IEEE No indication of entire time — other nodes keep listening If ACK is not received, give up Tx — fairness

27 Implementation on Testbed Nodes  Compared MAC modules 1.IEEE like protocol w/o sleeping 2.Message passing with overhearing avoidance 3.S-MAC (2 + periodic listen/sleep)  Platform Mica Motes (UC Berkeley) 8-bit CPU at 4MHz, 128KB flash, 4KB RAM 433MHz radio TinyOS: event-driven

28 Experiments: two-hop network  Topology and measured energy consumption on source nodes Source 1 Source 2 Sink 1 Sink 2 Each source node sends 10 messages — Each message has 400B in 10 fragments Measure total energy over time to send all messages

29 Recent Progress: Adaptive Listen  Reduces latency due to periodic sleep  At end of each transmission, nodes wake up for a short time  Next transmission can start right away  Example: data flow A  B  C  A sends RTS, B replies CTS, C overhears CTS  C will wake up when A  B is done  B  C can start right away AB C

30 Recent Progress: 10-hop Experiments  Compare S-MAC in 3 different modes  10-hop linear network with 11 nodes

31 S-MAC Information  URL:  Released S-MAC source code (for TinyOS 0.6.1)  Currently porting to nesC environment (TinyOS 1.0)