Performance evaluation of source extraction in wireless sensor networks

In this paper, the performance of source extraction in bandwidth constrained wireless sensor networks are evaluated. The sensor observations are assumed to be linear instantaneous mixtures of the sources in a sensing field. Two sensor network models are adopted, namely, cluster-based and cluster-free. In cluster-based sensor networks, the cluster members quantize their observations and send the quantized data to the cluster head. Then, the cluster head performs local source extraction, quantizes the extracted signal, and sends the quantized data to the sink. Finally, the sink performs global source extraction. Both single-cluster and multi-cluster cases are considered. In cluster-free sensor networks, all sensors quantize their observations and send the quantized data to the sink. Then, the sink performs global source extraction. The results show that sources can be effectively extracted in both kinds of sensor networks. The performance of source extraction in sensor networks is comparable to that of the benchmarking case where the sensor observations are perfectly gathered at the sink. The impact of the mixing matrix and the amount of available observations on the performance are also discussed.