1. Anya Apavatjrut, Katia Jaffres-Runser, Claire Goursaud and Jean-Marie Gorce Combining LT codes and XOR network coding for reliable and energy efficient transmissions in wireless sensor networks Sarnoff Symposium (SARNOFF), 2012 35th IEEE

2. Outline Introduction Reaching reliability Gradient broadcast routing Through coding: LT codes Combining LT codes and gradient broadcasting Improving energy with network coding XOR network coding heuristic XLT-GRAB Performance results

3. Introduction In large scale wireless sensor network A node advertises its data to one or several sink nodes. The transmission is multi-hop between the data source node and the sink. High reliability is important. To increase reliability Introducing redundancy through path diversity Adding a coding layer on top of the routing algorithm

4. Multi-path routing Several copies of a same packet travel on multiple paths in parallel Increasing transmission reliability Also increasing in energy expenditure for redundant transmission Gradient broadcast algorithms Allow several nodes at a time to forward a same packet in broadcast based on pre-defined set of forwarding rules A cost field is set in an initialization stage Nodes are able to adjust locally to instantaneous changes in the network node failure or link failure. More flexible Increasing the number of copies traveling in the network.

5. Adding coding layer Each message m Encoded using a specific coding algorithm Adds redundancy to m to compensate for the losses Still retrieve m at the sink HARQ Fountain codes A source S can potentially generate a limitless number of encoded packets until it receives an acknowledgement from D. D only acknowledges end to end a successful decoding to S This acknowledgement can be merged with the gradient cost field maintenance packets of the protocol.

6. In this paper Adding fountain codes to a gradient broadcast algorithm Perfect reliability To reduce the number of redundant packets travel in the network. Using network coding, relays re-combine the received packet along the multi-hop diffusion We show in a simulation study how our implementation of a XOR network coding solution over an LT code [2] performs over a simple gradient broadcast algorithm: Reliability is maintained at a reduced energy and delay cost.

7. Gradient broadcast routing Broadcast mode Any relay hearing a packet has to decide whether it can forward it or not. Only relays located closer to the sink than the previous hop relay are allowed to forward packets. 1. Cost field setup The nodes distributively build the gradient cost field 2. Forwarding stage [26] F. Ye, A. Chen, S. Liu, and L. Zhang. A scalable solution to minimum cost forwarding in large sensor networks. In International Conference on Computer Communications and Networks: Proceedings, pages 304–309, 2001.

8. Cost field setup If Q p +L < Q A then update Q A= Q p +L L=the link cost A new ADV packet is sent with a new packet cost Q p = Q A Sink : ADV packet containing its own cost Q Q=0 flooding All the other nodes have an initial cost Q = +∞ node A QAQA : ADV packet with packet cost Q p

9. Cost field setup The node with the lowest value of Q A sends its packet first Prevents other nodes with higher costs from forwarding their ADV packet. With this algorithm, only one ADV packet per node is sent in the cost field setup stage. The link cost value can be expressed in various metrics (in hops, in meters, etc..). We consider a simple euclidian distance metric.

10. Forwarding stage Once a sensor S has a packet to send to the sink, it appends its own cost Qs to the packet and broadcasts it. All nodes receiving it decide to forward it if and only if their own cost Q i is lower than Qs. This algorithm is particularly reliable compared to single path routing Having very low control overhead but at the price of a very high packet redundancy.

11. Gradient broadcast routing Following works Creating additional forwarding rules to improve the tradeoff between reliability and energy consumption We control the amount of redundancy by introducing a forwarding probability p f. If the sensor is allowed to forward a packet based on its cost, it will do it with probability p f. [12], [13]. [12] K. Jaffr`es-Runser and C. Comaniciu. A probabilistic interference and energy aware gradient broadcasting algorithm for wireless sensornetworks. In Proceedings of IEEE ISWPC 2008, Santorini, Greece,2008. [13] K. Jaffr`es-Runser, C. Comaniciu, J.-M. Gorce, and R. Zhang. U-GRAB:A Utility- Based Gradient Broadcasting Algorithm for Wireless Sensor Networks. In IEEE Conference on Military Communications (MILCOM 2009), Boston, MA, USA, October 2009.

12. Through coding: LT codes Fountain codes provide both rate-less and universal property Transmission reliability can be assured without requiring channel state information LT code Having lower decoding complexity But at the price of a very high packet redundancy [17] M. Luby. LT Codes. In Foundations of Computer Science - FOCS 2002, pages 271–, Vancouver, BC, Canada, November 2002. IEEE Computer Society.

13. Combining LT codes and gradient broadcasting We have considered a wireless sensor network 50 nodes spatially distributed following a Poisson distribution in a 2 dimensional space of 500m×500m. Average node degree is of about three. The source first encodes the information with LT codes before broadcasting the encoded message. The message propagates in a relaying mesh from S to D following the gradient broadcast routing defined earlier.

14. The following simulation results are obtained using WSNet event- driven simulator [24]. Combining LT codes and gradient broadcasting [24] WSNet. Worldsens simulator. http://wsnet.gforge.inria.fr/.

15. Combining LT codes and gradient broadcasting In the simulations, a message is decomposed into K fragments.

16. LT codes show a higher success rate on average for the same forwarding probabilities compared to the no coding case. Even if LT codes should ensure perfectly reliable transmissions, we do not always obtain a success rate equal to 1. Because of bad transmission conditions, nodes can keep trying (unsuccessfully) to relay the new packets of the source. Combining LT codes and gradient broadcasting

17. XOR network coding heuristic Transmission with network coding More scalable and can lead to the optimization of complexity, throughput, transmission delay and security. In this paper,we applying intra-flow network coding to fountain encoded packets Network coding can play an efficient role to optimize the redundancy The degree d of the packet to be created at the relay node is chosen with respect to the Robust Soliton distribution Buffered packets are then randomly selected and XOR-ed together until degree d is obtained or a MAXROUND value is reached if d ∈ {1, 2} then a combination is only performed with probability p=0.2

18. XOR network coding heuristic [2] A. Apavatjrut, C. Goursaud, K. Jaffres-Runser, C. Comaniciu, and J.-M.Gorce. Toward increasing packet diversity for relaying lt fountain codes in wireless sensor networks. Communications Letters, IEEE, 15(1):52-54, January 2011

19. XLT-GRAB The source sends each message m using an LT-code K=100 This ADV message serves two purposes Namely acknowledging m Updating the costs to account for topology changes in the network. The source keeps transmitting coded packets until an acknowledgement is received.

20. A relay node forwards a packet based on a probability p f if it is located closer to the sink than the previous hop relay. A relay decides with a given XORing probability p xor to apply network coding using to forward a network coded packet instead of the received one. XLT-GRAB

21. Performance results Performance results are averaged over 50 consecutive message transmissions, each message encoded with an LT code. We have first evaluated the impact of p f and p xor on the number of duplicated packets. p f < 0.4 : the network is not reliable at all for whichever XORing probability p f > 0.6 : the network is reliable for whichever XORing probability 0.4 < p f < 0.6 : transitory area

22. Performance results

24. The end-to-end message transmission delay in seconds Performance results

25. The total energy consumed by all nodes of the network for emission and reception actions. Performance results