Reduced complexity joint decoding for turbo-coded wireless sensor networks

We consider a data-gathering wireless sensor network, modeled by a Chief Executive Officer (CEO) problem. The Fusion Centre (FC) decodes data that are separately encoded at each sensor node by turbo codes. The CEO model introduces a correlation model between sensors' data. As shown in the literature, this correlation can be employed at the FC to perform a joint graph-based decoding. The optimal decoding is provided by the well-known sum-product algorithm; however, the sum-product summations impose a computational complexity that exponentially grows by increasing the number of sensors. In this paper, we propose a suboptimal joint decoding algorithm in which we first perform a reliability sorting and then we use a set of most-reliable nodes to update extrinsic information for other nodes. This algorithm exponentially reduces the decoding complexity compared to the sum-product algorithm and achieves BERs impressively close to the ones achieved by the sum-product decoding. We show by simulations that applying reliability sorting substantially improves the performance of the proposed algorithm. We also show that, by fixing the number of most-reliable nodes and increasing the total number of sensors, we could further reduce the average BER of the system, while keeping the same decoding complexity for joint decoding.