Compressive linear network coding for efficient data collection in wireless sensor networks

We address the problem of data collection in a wireless sensor network. Network coding is used for data delivery. The correlation between the measurements is exploited to recover the data at the sink, even in case of rank-deficient network matrix. The network coding operations are seen as lossy source compression, achieved by a finite-field random code generated during transmission. Decoding is performed using belief propagation on a factor graph which accounts for the correlation between the sensor measurements. Experimental results illustrate the performance of this technique for various field sizes and correlation levels.