1 Sensor MAC Design Requirements:  Energy efficiency  Simple operations  Working with a large number of sensors  Fair share of the channel among competing sensor nodes

2 Introduction to MAC The role of medium access control (MAC)  Controls when and how each node can transmit in the wireless channel Why do we need MAC?  Wireless channel is a shared medium  Radios transmitting in the same frequency band interfere with each other – collisions  Other shared medium examples: Ethernet

3 Where Is the MAC? Network model from Internet A sublayer of the Link layer  Directly controls the radio  The MAC on each node only cares about its neighborhood Application layer Transport layer Network layer Link/MAC layer Physical layer End-to-end reliability, congestion control Routing Per-hop reliability, flow control, multiple access Packet transmission and reception

4 Key Questions How should Media Access Control (MAC) protocols be designed for sensor networks? What are the metrics for MAC in a multi-hop scenario? How can we achieve these metrics with local algorithms over limited resources?

5 What’s New in Sensor Networks? A special multi-hop wireless network  Large number of nodes  Battery powered  Topology and density change due to node’s sleeping or failure  In-network data processing Sensor-net applications  Sensor-triggered bursty traffic  Can often tolerate some delay  Speed of a moving object places a bound on network reaction time

6 Primary MAC Attributes Collision avoidance  Basic task of a MAC protocol Energy efficiency  One of the most important attributes for sensor networks, since most nodes are battery powered Scalability and adaptivity  Network size, node density and topology change

7 Other MAC Attributes Channel utilization  How well is the channel used? Also called bandwidth utilization or channel capacity Latency  Delay from sender to receiver; single hop or multi-hop Throughput  The amount of data transferred from sender to receiver in unit time Fairness  Can nodes share the channel equally?

8 Energy Efficiency in MAC Design Energy is primary concern in sensor networks What causes energy waste? 1. Collisions 2. Control packet overhead 3. Overhearing unnecessary traffic 4. Long idle time 1. bursty traffic in sensor-net apps 2. Idle listening consumes 50—100% of the power for receiving (Stemm97, Kasten) Dominant factor

9 Classification of MAC Protocols Schedule-based protocols  Schedule nodes onto different sub-channels  Examples: TDMA, FDMA, CDMA Contention-based protocols (our focus)  Nodes compete in probabilistic coordination  Example: DCF

10 CSMA — Carrier Sense Multiple Access  Listening (carrier sense) before transmitting  Send immediately if channel is idle  Backoff if channel is busy  non-persistent, 1-persistent and p-persistent Contention Protocols: CSMA

11 Hidden terminal problem  CSMA is not enough for multi-hop networks (collision at receiver) CSMA/CA (CSMA with Collision Avoidance)  RTS/CTS handshake before sending data  Node c will backoff when it hears b’s CTS Contention Protocols: CSMA/CA ab c Node a is hidden from c’s carrier sense

12 Contention Protocols: IEEE IEEE ad hoc mode (DCF)  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 (fragment + ACK)  First (fragment + ACK) reserve time for second…  Give up transmission when error happens

13 Case Study 1: S-MAC By Ye, Heidemann and Estrin Tradeoffs Latency Fairness Energy

14 S-MAC Overview Design goals  Energy efficiency  Self-configuration and flexibility to node changes Approaches  Contention-based MAC with various energy-conserving features Major components in S-MAC  Periodic listen and sleep  Collision avoidance  Overhearing avoidance  Massage passing

15 Coordinated Sleeping Problem: Idle listening consumes significant energy Solution: Periodic listen and sleep Turn off radio when sleeping Reduce duty cycle to ~ 10% (120ms on/1.2s off) sleep listen sleep Latency Energy

16 Coordinated Sleeping Schedules can differ Prefer neighboring nodes have same schedule — easy broadcast & low control overhead Border nodes: two schedules or broadcast twice Node 1 Node 2 sleep listen sleep listen sleep Schedule 2 Schedule 1

17 Coordinated Sleeping 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

18 Coordinated Sleeping Adaptive listening  Reduce multi-hop latency due to periodic sleep  Wake up for a short period of time at end of each transmission CTS RTS CTS  Reduce latency by at least half listen t1 t2

19 Collision Avoidance S-MAC is based on contention 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

20 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

21 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

22 Implementation and Experiments Platform: Mica Motes Topology: 10-hop linear network S-MAC saved a lot of energy compared with a MAC without sleep Message inter-arrival period (S) Energy consumption (J) 10% duty cycle without adaptive listen No sleep cycles 10% duty cycle with adaptive listen Energy consumption at different traffic load

23 Evaluation Metric: Aggregate Bandwidth Traditional MAC metric channel capacity is a precious resource Maximize total delivered bandwidth from every node in the network to the base station

24 Evaluation Metric: Energy Efficiency Observation:  Energy is the precious resource Goal:  Minimize energy per unit of successful communication to base station while sustaining reasonable channel utilization  Turn off radio whenever possible  Avoid over commit the network Total energy spent in data propagation over a network Total packets received by the base station E =

25 Limitations of SMAC Look at the assumptions made  Broadcast assumes every one can hear the message  Bursty data applications versus periodic applications  Power consumption model  In-band signaling? Let us revisit the design using the link measurement findings

26 Fairness Challenge Challenge: want roughly equal data coverage Observation: originated traffic competes with route-thru traffic at odd with energy efficiency and aggregate bandwidth Goal: minimize variance in bandwidth delivered to base station

27 Case Study 2: Berkeley MAC Design Carrier Sense Multiple Access (CSMA)  no extra control packets (energy efficient) Save energy:  Shorten listening period as much as possible  turn radio off during backoff  Trade bandwidth for battery life Provide feedback to applications to desynchronize  Backoff should signal application to shift phase of sampling Random delay before each transmission  break close synchronization

28 Hidden Nodes in Multi-hop Networks Occurs between every other pair of levels CSMA fails to detect Traditionally addressed with contention-based protocols, but  Control packets (e.g. RTS/CTS/ACKs) induce high overhead given data packets are small  ACKs can be free in multi-hop networks  By hearing your parent forwards your packets  Data aggregation is application specific  Not adequate to solve hidden node problem in multi-hop case (Related Work: V. Bharghavan et al. MACAW)

29 Avoid Hidden Node Corruption A B CD exploits application characteristics “A” refrains from sending for a packet time after parent transmits Hidden node cases like this may be avoided without use of control packets.

30 Platform of Study Rene  4MHz, 8KB flash, 512B RAM  916MHz RF transceiver  10kbps  feet range  Sensors: temperature, light, magnetic field, acceleration Operating System: TinyOS  tiny network stack and other communication support  Small packets size (tens of bytes)

31 Multi-hop Extensions Rate control module inserted between MAC and application Adapts data sampling rate to available bandwidth Balances demand for upstream bandwidth between local, originating traffic and route-thru traffic by adjusting transmission rate  Multihop  Merging traffic flow Provides a mechanism for progressive feedback deep down into the network

32 Rate Control Mechanism snoop on route-thru traffic to estimate children (n) Open parameters  , Apply for forwarding route-thru traffic  Progressive feedback deep down into the network  Packet loss rate provides natural damping effect x  +  /n R x p if fails if success x  +  /n x  +  /n Estimate n based on route-thru traffic Route-thru Traffic

33 Other ARC Results Proposed CSMA with ARC scheme:  Aggregate bandwidth:  ~ 60% of proposed CSMA without ARC scheme  Energy efficiency:  ~ 50% of proposed CSMA at a low   Fairness:  5 – 10 times lower variance

34 Summary of B-MAC Sensor networks characteristics differ from traditional settings enough to require revisiting the basic protocols MAC design  fairness and energy efficiency goals  modified CSMA shown effective Transmission control  local adaptive scheme on originating traffic effective  Implemented and evaluated on simulation and real networked sensors  Each node achieves 20% of multi-hop channel capacity

35 Summary for MAC Design in Sensor Networks MAC protocols can be classified as scheduled and contention-based Major considerations  Energy efficiency  Scalability and adaptivity to number of nodes Major ways to conserve energy  Low duty cycle to reduce idle listening  Effective collision avoidance  Overhearing avoidance  Reducing control overhead

36 What is next: New Sensor Link Protocols Look at the experimental results !  These new findings are not addressed yet Learn from the topology  With static sensor settings, link protocols can be more efficient Application-aware link protocol  Get hints from applications about traffic patterns, rate, etc.  New forms of cross-layer design