Compressive Network Coding for Approximate Sensor Data Gathering

Proceedings of IEEE Globecom'11 |

Published by IEEE Communications Society

This paper presents a joint source and network coding scheme, called compressive network coding (CNC), for approximate data gathering in wireless sensor networks. Injecting the concept of compressive sensing into network coding avoids the all-or-nothing effect of network decoding, allowing CNC to achieve graceful degradation in data precisions when the energy budget is not sufficient. Based on the observation that sensor links are highly unreliable, we formulate the energy constrained multi-source, multi-hop, and multi-path transmission as a network utility maximization (NUM) problem. A practical distributed algorithm is developed to achieve the optimal utility. We carry out simulations over real sensor data. Results show that CNC consistently outperforms conventional network coding, achieving an average gain over 3.7 dB in data reconstruction PSNR. Furthermore, CNC achieves over 10 dB gain for five sensors (out of 53) whose readings contain abrupt changes.