1. 1 Wireless Sensor Networks Akyildiz/Vuran Chapter 6: Error Control

2. 2 Wireless Sensor Networks Akyildiz/Vuran Error Control PURPOSE:  Error-free  In-sequence  Duplicate-free  Loss-free

3. 3 Wireless Sensor Networks Akyildiz/Vuran Wireless Errors  Causes: Fading, interference, loss of bit synchronization  Results in bit errors, sometimes bursty  For low power wireless communication  sometimes quite high average bit error rates (BERs) (10 -2 … 10 -4 possible!)

4. 4 Wireless Sensor Networks Akyildiz/Vuran Error Control Approaches  Automatic Repeat reQuest (ARQ)  Forward Error Correction (FEC)  Hybrid ARQ (Type I and II)

5. 5 Wireless Sensor Networks Akyildiz/Vuran ARQ (Quick Recap)  Basic procedure  Put header information before the payload  Compute a checksum and add it to the end of the packet  Typically: Cyclic redundancy check (CRC),  quick, low overhead, low residual error rate  Provide feedback from receiver to sender  Send positive or negative acknowledgement (ACK/NACK)

6. 6 Wireless Sensor Networks Akyildiz/Vuran ARQ  Sender uses timer to detect that ACKs have not arrived  Assumes packet has not arrived  Optimal timer setting?  If sender infers that a packet has not been received correctly, sender can retransmit it  What is maximum number of retransmission attempts?

7. 7 Wireless Sensor Networks Akyildiz/Vuran Standard ARQ protocols  Alternating bit – at most one packet outstanding, single bit sequence number  Go-back-N – send up to N packets, if a packet has not been ACKed when timer goes off, retransmit all packets after the first unacknowledged packet  Selective Repeat – when timer goes off, only send the unacknowledged packet(s)

8. 8 Wireless Sensor Networks Akyildiz/Vuran How to use ACKs ?  Be careful about ACKs from different layers  A MAC ACK (e.g., S-MAC) does not necessarily imply buffer space in the link layer  Do not (necessarily) ACK every packet – use cumulative ACKs  Tradeoff against buffer space  Tradeoff against number of negative ACKs to send

9. 9 Wireless Sensor Networks Akyildiz/Vuran When to Retransmit ?  Assuming sender has decided to retransmit a packet – when to do so?  For a good channel, any time is as good as any  For fading channels, try to avoid bad channel states – postpone transmissions  Instead (e.g.): send a packet to another node if in queue (exploit multi-user diversity)  REMARK: If the previous packet was corrupted then the channel was bad

10. 10 Wireless Sensor Networks Akyildiz/Vuran When to Retransmit ?  How long to wait?  Example solution: Probing protocol  Idea: reflect channel state by two protocol modes, “normal” and “probing”  When error occurs, go from normal to probing mode  In probing mode, periodically send short packets (ACKed by receiver) – when successful, go to normal mode

11. 11 Wireless Sensor Networks Akyildiz/Vuran Forward Error Control (FEC)  Idea: Endow symbols in a packet with additional redundancy to withstand a limited amount of random permutations  Additionally: interleaving – change order of symbols to withstand burst errors Header Payload Redundant Bits  l DATA n-k k n

12. 12 Wireless Sensor Networks Akyildiz/Vuran Forward Error Control

13. 13 Wireless Sensor Networks Akyildiz/Vuran Popular Block Code FECs  Reed-Solomon Codes (RS)  Bose-Chaudhuri-Hocquenghem Codes (BCH)

14. 14 Wireless Sensor Networks Akyildiz/Vuran Popular Block Codes  Energy Consumption:  Encoding: Negligible overhead (linear-feedback shift register)  Decoding: depends on block length (n) and Hamming distance (t) Eng. Consumption for multiplication Eng. Consumption for addition

15. 15 Wireless Sensor Networks Akyildiz/Vuran Convolutional Codes  Code rate: ratio of k user bits mapped onto n coded bits  Constraint length k determines coding gain  Energy  Encoding: cheap  Decoding: Viterbi algorithm, energy & memory depends exponentially (!) on constraint length

16. 16 Wireless Sensor Networks Akyildiz/Vuran Hybrid ARQ (HARQ)  ARQ  Efficient when channel is good - No redundant bits are sent  Inefficient when channel is bad – whole packet is sent again  FEC  Inefficient when channel is good – Redundant bits sent anyway  Efficient when channel is bad – No retransmissions  Hybrid ARQ  Marry ARQ and FEC  Get advantages from both techniques

17. 17 Wireless Sensor Networks Akyildiz/Vuran Hybrid ARQ  Send uncoded or lightly coded packet first  If in error, retransmit with a stronger code  Two types based on retransmission technique  HARQ Type I  HARQ Type II

18. 18 Wireless Sensor Networks Akyildiz/Vuran Hybrid ARQ Type I (HARQ-I) Header + Payload Redundant Bits e.g. BCH(128,106,3) Packet in error Send NACK Send uncoded packet NACK Send coded packet No errors No ACK sent

19. 19 Wireless Sensor Networks Akyildiz/Vuran Hybrid ARQ Type I (HARQ-I) Header + Payload e.g. BCH(128,78,7) Packet in error Send NACK Send lightly coded packet NACK Send more powerful coded packet No errors No ACK sent Redundant Bits e.g. BCH(128,106,3)

20. 20 Wireless Sensor Networks Akyildiz/Vuran Hybrid ARQ Type II (HARQ-II) Header + Payload Redundant Bits e.g. BCH(128,106,3) Packet in error Send NACK Send uncoded packet NACK Send redundant bits No errors No ACK sent

21. 21 Wireless Sensor Networks Akyildiz/Vuran Hybrid ARQ  Type I – Does not require previous packet to be stored  Type II – Reduces bandwidth usage

22. 22 Wireless Sensor Networks Akyildiz/Vuran Error Control through Transmit Power Control  Higher transmission power reduces the packet error rate by improving SNR.  Energy consumption and interference is increased  Radio should support different power levels (may not be applicable in many WSNs (low cost nodes)

23. 23 Wireless Sensor Networks Akyildiz/Vuran Cross-Layer Comparison: FEC, ARQ, HARQ  Investigate the tradeoffs between ARQ, FEC, and HARQ in terms of  Energy consumption  Latency  Packet error rate  The first work that jointly considers  Broadcast wireless channel  Multi-hop communication  Realistic channel model  2-D topology  Realistic hardware models (Mica2 and MicaZ) M. C. Vuran and I. F. Akyildiz, “Cross-layer Analysis of Error Control in Wireless Sensor Networks, IEEE SECON ‘06, September 2006. M. C. Vuran and I. F. Akyildiz, “Error Control in Wireless Sensor Networks: A Cross Layer Analysis”, IEEE Trans. on Networking, vol. 17, no. 4, pp. 1186-1199, April 2009.

24. 24 Wireless Sensor Networks Akyildiz/Vuran Cross-Layer Effects  Multi-hop communication affects  End-to-end packet error rate  Energy consumption  Latency  Broadcast wireless channel  Nodes other than communicating parties are affected (overhearing)  Wireless channel  Channel characteristics drastically influence BER performance Cross-layer analysis of error control is necessary

25. 25 Wireless Sensor Networks Akyildiz/Vuran Motivation  ARQ  Reactive error control  Send redundancy in case of errors (lack of ACK)  FEC  Proactive error control  Incorporate redundancy to correct errors  Improves the error resiliency of a single packet

26. 26 Wireless Sensor Networks Akyildiz/Vuran Error Resiliency  Packet Error Rate (PER) is a function of signal to noise ratio (SNR), i.e., channel quality  ARQ vs. FEC: FEC codes provide the same PER with lower SNR, i.e., lower channel quality  For PER *, SNR ARQ >SNR FEC  As error correction capability, t, increases, error resiliency improves

27. 27 Wireless Sensor Networks Akyildiz/Vuran Error Resiliency  How can error resiliency be exploited in multi-hop WSN? d ARQ P t,ARQ -P n SNR ARQ  PER * Path Loss ARQ case Transmit Power Noise Power

28. 28 Wireless Sensor Networks Akyildiz/Vuran Error Resiliency  How can this be exploited in multi-hop WSN?  Hop-length Extension (increase hop length for FEC)  P t,FEC = P t,ARQ, d FEC > d ARQ d FEC P t,ARQ -P n SNR FEC  PER * Path Loss FEC case

29. 29 Wireless Sensor Networks Akyildiz/Vuran Error Resiliency  How can this be exploited in multi-hop WSN?  Hop-length Extension (increase hop length for FEC)  P t,FEC = P t,ARQ, d FEC > d ARQ  Transmit Power Control (decrease transmit power)  P t,FEC < P t,ARQ, d FEC = d ARQ d ARQ P t,FEC -P n SNR FEC  PER * Path Loss FEC case

30. 30 Wireless Sensor Networks Akyildiz/Vuran Cost  FEC codes improve error resiliency at the cost of  Encoding/decoding (energy + latency)  Tx/Rx longer packets (energy + latency) Header Payload Redundant Bits a l DATA n-k

31. 31 Wireless Sensor Networks Akyildiz/Vuran Network Model  Channel-aware routing  Next hop j is selected if received SNR,  j >  Th (SNR threshold)  Contention-based medium access  RTS-CTS-DATA(-ACK) exchange  Duty cycle operation  Nodes are active δ fraction of the time

32. 32 Wireless Sensor Networks Akyildiz/Vuran Cross-layer Analysis  Channel model Transmit Power Received Power Path loss exponent Log-normal Shadow fading

33. 33 Wireless Sensor Networks Akyildiz/Vuran Cross-layer Analysis  Channel model then, Received SNR SNR threshold Variance of log-normal fading (usually ~3.5) Noise power

34. 34 Wireless Sensor Networks Akyildiz/Vuran  Total energy consumed as a flow-based energy consumption analysis  Expected number of hops from a source to sink with distance D E[d next-hop ] is the expected hop distance. R inf is the approximated transmission range. Energy Consumption

35. 35 Wireless Sensor Networks Akyildiz/Vuran Effect of Duty Cycle  Consumed energy per hop in one hop for transmitting a packet Detailed derivations for ARQ, FEC, and HARQ in the paper Eng. consumption of the transmitter node Eng. consumption of the receiver node Eng. consumption of the neighbor nodes

36. 36 Wireless Sensor Networks Akyildiz/Vuran Latency Analysis  Expected end-to-end latency Expected number of hops Expected latency per hop

37. 37 Wireless Sensor Networks Akyildiz/Vuran Decoding Latency and Energy  Encoding for block codes is negligible  Decoding latency for a (n,k,t) block code  Both Mica2 and MicaZ use 8-bit microcontrollers  Decoding energy Processor cycle latency (250ns) Processor current (8mA) Voltage (3V)

38. 38 Wireless Sensor Networks Akyildiz/Vuran Bit Error Rate  Mica2 nodes use frequency shift keying (FSK)  MicaZ nodes use offset quadrature phase shift keying (O- QPSK) with direct sequence spread spectrum (DSSS) Bandwidth Bit rate # of chips per bit (16) =2 for MicaZ

39. 39 Wireless Sensor Networks Akyildiz/Vuran Numerical Evaluations  Energy consumption, latency and PER are found for Mica2 and MicaZ, and ARQ, FEC, and HARQ schemes

40. 40 Wireless Sensor Networks Akyildiz/Vuran Numerical Evaluations  FEC Codes  BCH (n,k,t) - (128,50,13), (128,78,7), (128,106,3)  RS (n,k,t) – (7,3,2), (15,9,3), (31,19,6)  Hybrid ARQ (I & II) Notation:  HARQ-I (t 1,t 2 ) means:  Hybrid ARQ type I  t 1 : error correction capability for 1 st transmission (t 1 = 0  uncoded packet)  t 2 : 2 nd transmission  BCH codes are used for coded transmission

41. 41 Wireless Sensor Networks Akyildiz/Vuran Hop Length Extension  Energy consumption vs. SNR threshold  Lower SNR threshold leads to longer hop length  BCH and RS achieve the same PER for lower SNR threshold compared to ARQ  BCH (t=7) consumes less energy than ARQ  RS (t=3) consumes more energy than ARQ Energy Consumption (MicaZ) PER ARQ =10 -2 PER BCH =10 -2 PER RS =10 -2

42. 42 Wireless Sensor Networks Akyildiz/Vuran Hop Length Extension  Compared to ARQ, BCH and RS achieve lower latency for the same PER  Lower SNR threshold  longer hops  less number of hops  lower latency Latency (MicaZ) PER ARQ =10 -2 PER BCH =10 -2 PER RS =10 -2

43. 43 Wireless Sensor Networks Akyildiz/Vuran Hop Length Extension  Taxonomy function  Received bit per consumed resources (time*energy) i.e., resource efficiency

44. 44 Wireless Sensor Networks Akyildiz/Vuran Hop Length Extension  For MicaZ nodes, FEC is more resource efficient than ARQ  BCH is 30% more efficient than RS MicaZ

45. 45 Wireless Sensor Networks Akyildiz/Vuran Hop Length Extension  For Mica2, ARQ is more resource efficient  Why?  Lower bit rate (16 kbps vs. 250 kbps – MicaZ)  Packet transmission duration is more important than hop length extension in Mica2  ARQ outperforms FEC since shorter packets are sent Taxonomy function, T (Mica2)

46. 46 Wireless Sensor Networks Akyildiz/Vuran Transmit Power Control  Significant reduction in energy consumption with transmit power control for BCH  RS codes are not energy efficient with transmit power control  BCH is more energy efficient than ARQ Energy Consumption (Mica2)

47. 47 Wireless Sensor Networks Akyildiz/Vuran Transmit Power Control  Trade off latency  Decrease in transmit power  increase in # of hops  ARQ has lower latency when transmit power is decreased for FEC codes Latency (Mica2)

48. 48 Wireless Sensor Networks Akyildiz/Vuran Hybrid ARQ  HARQ-II outperforms ARQ and FEC schemes  HARQ-I is less energy efficient – whole packet is retransmitted  For HARQ-II, energy efficiency improves if error correction capability (t) increases  HARQ-II implementation cost is also lower than HARQ-I Energy Consumption (MicaZ)

49. 49 Wireless Sensor Networks Akyildiz/Vuran Hybrid ARQ  HARQ-II, BCH, and RS provide lower latency compared to ARQ  HARQ-I latency is significantly higher – Retransmission + Encoding/Decoding overhead  HARQ-II and RS most suitable for delay sensitive traffic

50. 50 Wireless Sensor Networks Akyildiz/Vuran Conclusions  A cross-layer analysis of ARQ, FEC, and HARQ is performed  Developed framework enables a generic analysis with various parameters (e.g. BCH, RS, Mica2, MicaZ)  Careful selection of FEC parameters improve latency without hampering energy consumption (HARQ and RS suitable for multimedia traffic)  Hardware support for FEC will improve performance