1. Ubiquitous Networks Wakeup Scheduling Lynn Choi Korea University

2. Motivation Most of WSN applications have real-time constraints Sensors in battlefield to detect odorless biochemical weapons Disaster monitoring applications Forest fire alarm, volcano monitoring, seismometer Real-time target tracking Intrusion detection Emergency health application Traffic coordination Existing MAC protocols focus on low energy consumption But, how about the communication latency required for real-time applications? Sleep delay A packet can traverse at most a single hop during each wakeup period

3. DMAC: Synchronous Skewed Wakeup “An Adaptive Energy-Efficient and Low-Latency MAC for Data Gathering in Wireless Sensor Networks” Krishnamachari and Raghavendra (at USC), IPDPS 2004. DMAC calls this staggered wakeup Skew the wakeup period of each node in the path from a source node to a sink node Assume the tree topology starting from the sink as a root The wakeup schedule of each node is determined by the level of the node in the tree

4. Wakeup Patterns “Wakeup Scheduling in Wireless Sensor Networks” Keshavarzian, Lee (at Stanford), Venkatraman, MobiHoc 2006. Fully Synchronized Wakeup Pattern (SMAC) All the nodes wake up at the same time Delay = (#hops – 0.5) * T

5. Wakeup Patterns Shifted Even and Odd Pattern Shift the wakeup period of nodes in even levels by T/2 Delay = 0.5 * (#hops) * T

6. Wakeup Patterns Ladder Pattern (DMAC: staggered wakeup) Skew the wakeup period of nodes in the communication path Forward and backward delays are asymmetric

7. Wakeup Patterns Two-Ladders Pattern To improve the delay in both directions Combine the forward ladder with a backward ladder Nodes in the middle levels wake up twice in every period T

8. Wakeup Patterns Crossed-Ladders Pattern Cross the two ladders at one point so that the same wakeup can be used for both directions

9. Wakeup Patterns Multi-Parent Method Embed multiple trees in the network Each node has multiple paths and multiple parents to the sink Depending on the packet arrival time, a node can choose the fastest path to get to the destination

10. SPEEDMAC: Speedy and Energy Efficient Data Delivery MAC Protocol for Real-Time Sensor Network Applications ICC 2010

11. Motivation Sleep delay is the dominant factor of WSN packet latency A packet can traverse at most a single hop each cycle Minimum packet latency = cycle time *hops Most of WSN applications have real-time characteristics Disaster monitoring, real-time target tracking, intrusion detection, health, etc. However, it is practically impossible to obtain both low latency and low energy communication at the same time Sleep delay exists for both synchronous & asynchronous MAC Synchronous scheduling (S-MAC, A-MAC) A packet can traverse at most a single hop (or 2 with ‘adaptive listening’) each cycle since nodes beyond one-hop from the receiver cannot overhear the data. Asynchronous scheduling (B-MAC, Wise-MAC, XMAC) A packet can traverse at most a single hop each cycle since a sender needs to send the preamble before starting the next-hop communication

12. Motivation Synchronous skewed wakeup (DMAC) may be a solution! Schedule the wakeup time of each node in a pipelined fashion in the direction of packet movement so that No sleep delay during the packet movement Issues with synchronous skewed wakeup May fail to deliver the message when multiple sensors compete for the message delivery A single event is likely to be detected by nearby multiple sensors Multiple events may occur simultaneously, which leads to collisions and contentions More idle listening Since a node must wake up during the entire DATA transmission period instead of RTS period as in SMAC May not be practically possible to use such wakeup scheduling techniques for real applications unless these issues are completely resolved.

13. Synchronous Skewed Wakeup S 1 24 3 Sink Node 1 Node 2 Node 3 ACK DATA ACK DATA ACK DATA Tx state Rx state

14. Synchronous Skewed Wakeup S 1 24 3 Sink Node 1 Node 2 Node 3 ACK DATA Node 4 DATA Tx state Rx state

15. SPEED MAC Ideas Goal: Can we achieve both low-energy and low-latency at the same time? 1. A collision signal to detect multi-source events &for fast event delivery A special control packet called SIGNAL packet is used. It has different electrical characteristics from background noise 2. Separate event report period from data delivery period Faster event report using a short control signal Lower energy consumption for idle period To further reduce both the latency and the energy consumption 3. Adaptive wakeup for multi-source events Fast pipelined data delivery for a single-source event Full wakeup and CSMA-based data delivery for a multi-source event Full duty-cycle operation for high-bandwidth transmission Use RTS/CTS for busy periods

16. Synchronous Skewed Wakeup

17. Issues with Synchronous Skewed Wakeup Assumptions Stationary sensor nodes and stationary sinks Many to one communication pattern from multiple sources to the sinksIssues Contention Only a single source can transmit the data and other sources may have to wait Collision When multiple nodes transmit at the same time, the packets will eventually collide in an upper layer and no packet can be transmitted Transmission error When a transmission error occurs, the sender needs to wait for the next cycle For single-source event No contention, no collision, only need to consider error For multiple-source events Need to consider contention, collision, and error

18. SPEED-MAC Event announcement period: Fast Event Announcement In this period, nodes announce the presence of an event by sending a small control packet called a SIGNAL packet. SIGNAL packet: consists of receiver address and collision bit There is NO ACK packet for the signal packet. Collision detection for multi-source events The collision bit tells that the event is a multi-source event. Need to distinguish transmission errors from collision All the senders overhear the signal transmission from its parent To distinguish a single source event from a multi-source event Data transmission period: Adaptive Wakeup In this period, nodes transfer messages by sending DATA packets For a single-source event, the period consists of DATA and ACK Fixed scheduled data transmission for single-source events (not a CSMA) For a multi-source event, the period consists of RTS/CTS/DATA/ACK Contention-based data transmission for multi-source events (CSMA/CA)

19. SPEED-MAC: Single Source Event No traffic Nodes wakeup only during a signal rx slot. Single source traffic: single-packet data Nodes wake up during signal rx/tx/rx slots and data slot

20. SPEED-MAC: Multi-Packet & Multi-Source Event Single source traffic: multi-packet data Nodes wake up during signal rx/tx/rx slots and multiple data slots Multi-source traffic Nodes wake up during signal rx/tx/rx slots and several RTS/CTS/DATA/ACK slots

21. SPEED-MAC with Multiple Sinks We can handle sink-to-sensor, sensor-to-sensor, and many sensors-to-many sinks scenarios

22. Collision/Error Differentiation Transmission error can occur due to two reasons Noise (Error) Unwanted electrical signals interfering with the desired signal The strength of the signal is irregular and variable Collision Multiple simultaneous transmission collide at the receiver The strength of the signal is regular and stronger Can be differentiated at the physical layer by tracking RSSI In case of collision, the SIGNAL control packet is already destroyed. COLLISION SIGNAL does not contain the receiver address anymore. COLLISION SIGNAL packet is broadcast to the nodes in the upper layers False-positive delivery: Nodes in the upper layers after the collision may unnecessarily wakeup

23. Collision/Error Differentiation

24. NS-2 Simulation Parameters # of nodes: 400 grid nodes + 1 sink node Power Tx : 30mW, Rx : 15mW, Idle : 15mW Bandwidth: 20Kbps Packet size Data packet: 100B Signal packet: 6B Control packet: 10B Tx & Rx slot length Data: 103ms, Signal: 22ms Simulation time: 10 min Total number of event: 20 events # of source nodes: 1, 2, 4, 8, 16 nodes Basic cycle time SMAC: 1.44s SPEED-MAC, D-MAC: 2.88s

25. Single Source – Latency SMAC SMAC suffers from the sleep delay and the additional buffering delay when the message generation interval is small. SPEED-MAC vs. DMAC Due to the signaling wakeup period, SPEED-MAC’s data latency is slightly higher than that of DMAC. Signal delivery latency of SPEED-MAC is almost close to the minimum delay achievable and is much smaller than DMAC’s data delivery latency

26. Single Source - Energy SMAC As the packet generation interval decreases SMAC spends more energy in repeated wakeups and buffering. SPEED-MAC vs. DMAC SPEED-MAC can achieve an order of magnitude reduction in the energy consumption compared to DMAC By reducing the idle listening overhead and By removing unnecessary wakeups during idle periods

27. Multiple Sources - Latency SMAC Latency increases substantially as the number of source nodes increases. This is due to the increased contention and buffering for multiple transactions. SPEED-MAC vs. DMAC Constant and faster signal delivery latency even in multi-source events Noticeably higher data packet delay due to its adaptive wakeups and increased control packet (RTS and CTS) overhead for multi-source events. For DMAC we use their assumption that an interference range of a node is twice larger than its transmission range to avoid collision for multi-source events.

28. SMAC SMAC spends more energy due to its higher duty cycle operations SPEED-MAC vs. DMAC Like the single-source case, SPEED-MAC can substantially reduce the energy consumption by reducing the idle listening and removing unnecessary wakeups. Multiple Sources - Energy

29. MICA-2 Mote Implementation Packet size: control packet: 10B, data packet: 100B Contention window: SYNC packet: 15 slots, Data packet: 31 slot SINGLE SOURCE RESULTSMULTIPLE SOURCE RESULTS