2. Z-MAC: a Hybrid MAC for Wireless Sensor Networks Injong Rhee, Ajit Warrier, Mahesh Aia and Jeongki Min Dept. of Computer Science, North Carolina State University SenSys’ 05 Presented by Seung-Min Jung Computer Architecture Laboratory

3. 2 Contents Design of Z-MAC 2 Conclusion 4 Introductions 31 Performance Evaluation 33

4. 3 What is Z-MAC?  A Hybrid MAC  Combine the strengths of CSMA and TDMA while offsetting their weakness  CSMA (Carrier Sense Multiple Access)  High channel utilization and low latency under low contention  Hidden terminal problem (collisions)  TDMA (Time Division Multiple Access)  No hidden terminal problem and high channel utilization under high contention  Not practical due to too many problems

5. 4  Four important sources of wasted energy in WSN:  Idle Listening (required for all CSMA protocols)  Overhearing (since RF is a broadcast medium)  Collisions (Hidden Terminal Problem)  Control Overhead (e.g. RTS/CTS or DATA/ACK) MAC Energy Usage Existing MAC Protocols (S- MAC, B-MAC) Z-MAC Approach

6. 5 Basic Idea of Z-MAC  Each sensor node owns a time slot.  A node may transmit at any time slot.  The owner has the higher priority to transmit data than the non-owners.  When a slot is not in use by its owner, non-owners can steal the slot.  Z-MAC behaves like  CSMA under low contention  TDMA under high contention.

7. 6 # of Contenders Channel Utilization TDMA CSMA IDEAL Effective Throughput

8. 7 Existing Approaches  Hybird (CSMA + TDMA)  S-MAC by Ye, Heidemann and Estrin @ USC Duty cycled Synchronized over macro time scales for neighbor communication  CSMA+Duty Cycle+LPL  B-MAC by Polastre, Hill and Culler @ UC Berkeley Duty cycled, but Low power listen (LPL) –Clever way reducing energy consumption (similar to aloha preamble sampling)

9. 8 Design of Z-MAC  Basic components  Setup phase  Transmission Control  Explicit Contention Notification  Receiving Schedule of Z-MAC  Local Time Synchronization

10. 9 Basic components  Baseline – CSMA  Use Imprecise Topology and Timing Info in a robust way.  Combining CSMA with TDMA  Scalable and Efficient TDMA scheduling

11. 10 Setup Phase  Including  neighbor discovery  slot assignment  local frame exchange  global time synchronization  => Do many things in the setup phase!  High overhead?  It runs only once during the setup phase and does not run until a significant change in the network topology

12. 11 Setup Phase: Neighbor Discovery  Steps  Every node periodically broadcasts a ping to its one-hop neighbors.  A ping message contains the current list of its one-hop neighbors.  Through the process, each node gathers the information of its two-hop neighbors.  Implementation  Every node sends one ping at a random time in each second for 30 seconds.

13. 12 Setup Phase: Slot Assignment  Using DRAND to assign time slots to every node.  DRAND is a distributed implementation of RAND, used for TDMA scheduling or channel assignment for wireless networks.  Ensuring no two nodes within a two-hop communication neighborhood are assigned to the same slot.  The slot number assigned to a node does not exceed the size of its local two-hop neighborhood (δ).  The running time and message complexity are also bounded by O(δ). 012 0 3

14. 13 Setup Phase: Local Framing  Each node needs to decide on the period in which it can use the time slot for transmission.  The period is called the time frame of the node.  Time frame rule  S i : the slot number assigned to node i  F i : the maximum slot number within node i’s two-hop neighborhood  Set node i’s time frame to be 2 a, where a satisfies 2 a-1 ≤ F i ≤ 2 a – 1. That is, node i uses the S i -th slot in every 2 a time slots. i uses the s i -th slot in every 2 a time frame (i's slots are L * 2 a + s i, for all L=1,2,3,...)

15. 14 Example 2 a-1 ≤ F i ≤ 2 a – 1 Node A a = 2 Node C a = 3 Network topology & the slot schedule of all nodes

16. 15 Transmission Control  Two modes: low contention level (LCL) and high contention level (HCL).  Under LCL, non-owners are allowed to compete in any slot with low priority.  Under HCL, a node does not compete in a slot owned by its two-hop neighbors. To avoid being hidden terminal to the owners.  A node is in HCL only when it receives an explicit contention notification (ECN) messages within the last t ECN period.

17. 16 Transmission Rule Node i acquires data to transmit Is node i the owner? Take a random backoff within period T o Is the channel clear? Wait until the channel is not busy Is node i in LCL? Is the current slot owned by its two-hop neighbor? Wait for T o and performs a random backoff within a contention window [T o, T no ] Transmit data!!! Postpone its transmission until the time slot is (1)not owned by a two-hop neighbor or (2) owned by itself NO YES NO

18. 17 Explicit Contention Notification  ECN messages notify neighbors not to act as hidden terminals to the owner of each slot when contention is high.  How to estimate two-hop contention?  According to noise level of the channel High correlation between noise level and two-hop contention. Low noise indicates low contention.

19. 18 Explicit Contention Notification  Steps:  When a transmitting node detects high contention The node sends a unicast message, one-hop ECN, to the destination which is experiencing contention. If there are multiple destinations, it broadcasts a message with information about the multiple destinations.  Assume node j receives one-hop ECN. If node j is the destination, it then broadcasts the ECN, two-hop ECN, to its one-hop neighbors. If not, it simply discards the one-hop ECN.  When a node receives a two-hop ECN, it sets the HCL flags.  ECN suppressing  Take random backoff before the transmission of a one-hop ECN.

20. 19 Example S5, S2, S4 can compete as one-hop neighbors at slot 1

21. 20 Local Time Synchronization  Time synchronization  Among Neighboring senders  Under high contention  Z-MAC adopts a technique from RTP/RTCP (real-time transport protocol).  The control message transmission rate is limited to a small fraction of session bandwidth.  In Z-MAC, a node sends one synchronization packet per every 100 data packets.

22. 21 Performance Evaluations  Implementing Z-MAC in both ns-2 and Mica2/TinyOS.  Comparing the performance of Z-MAC with that of PTDMA(ns-2), Sift(ns-2) and B-MAC(ns-2 and TinyOS).  Three benchmarks  One-hop benchmark  Two-hop benchmark Two clusters, 7 and 8 sending nodes.  Multi-hop benchmark

23. 22 One-hop Benchmark

24. 23 Two-hop Benchmark Sources Sink

25. 24 Multi-hop Benchmark

26. 25 Default settings

27. 26 Throughput  One Hop Utilization; Simulation Z-MAC B-MAC

28. 27 Throughput  One Hop Throughput; Mica2 Experiment Z-MAC B-MAC

29. 28 Throughput  Two Hop Utilization; Simulation Z-MAC B-MAC

30. 29 Throughput  Multi-hop Throughput; Mica2 Experiment Z-MAC B-MAC MULTI-HOP

31. 30 Throughput  Utilization Variation with Time Sync Error Z-MAC

32. 31 Energy Efficiency Z-MAC HCL B-MAC MULTI-HOP

33. 32 Conclusion  Z-MAC combines the strength of TDMA and CSMA  High throughput independent of contention.  Robustness to timing and synchronization failures and radio interference from non-reachable neighbors. Always falls back to CSMA.  Compared to existing MAC  It outperforms B-MAC under medium to high contention.  Achieves high data rate with high energy efficiency.

34. Questions? Email : smjung@camars.kaist.ac.kr Office # : 5578

35. 34 Supplement: Low Power Listening(1) Receive data Carrier sense Receiver Long PreambleData Tx Sender Check Interval  Similar to ALOHA preamble sampling  Wake up every Check-Interval  Sample Channel using CCA  If no activity, go back to sleep for Check-Interval  Else start receiving packet  Preamble > Check-Interval

36. 35 Supplement: Low Power Listening(2) Receive data Carrier sense Receiver Long PreambleData Tx Sender Check Interval  Longer Preamble => Longer Check Interval, nodes can sleep longer  At the same time, message delays and chances of collision also increase  Length of Check Interval configurable by higher layers

37. 36 Time period Time slice 12345 67 TDMA Scheduling : Using DRAND  Two nodes in the interference range assigned to different time slots.  Owners and non-owners C D A F B C D A E B E F Radio Interference Map Input Graph C D A E B F DRAND slot assignment 0 0 1 3 2 1

38. 37 DRAND  Z-MAC requires a conflict-free transmission schedule or a TDMA schedule.  DRAND is a distributed TDMA scheduling scheme. Let G = (V, E) be an input graph, where V is the set of nodes and E the set of edges. An edge e = (u, v) exists if and only if u and v are within interference range. Given G, DRAND calculates a TDMA schedule in time linear to the maximum node degree in G.  DRAND is fully distributed, and is the first scalable implementation of RAND, a famous centralized channel scheduling scheme.

39. 38 Local Time Synchronization  Trust factor(β t ):  R drift : the max clock drift rate of each node  ε clock : the max acceptable clock error  I synch = ε clock / R drift : the min synchronization interval required to achieve the max clock error  α synch : the max weight applying to the new clock value received  S : the avg. rate at which a node receives or sends synchronization messages   How to get new clock value?  C avg : weighted moving avg. clock value  C new : newly received clock value  C avg = (1- β t )C avg + β t C new