Sensor selection for communications and signal processing

Sensor selection is an important topic for both communications and signal processing systems. In wireless communication systems one can select from a set of available antennas to reduce the number of processing chains needed. In signal processing systems one may select from a set of available sensors to reduce the number of transmissions and, ultimately, energy consumption. In either case it has been shown recently that system complexity or energy consumption can be reduced without serious performance degradation.;We investigate multiple-input multiple-output (MIMO) systems with receive antenna selection in a spatially correlated channel with mutual coupling. We first assume the training overhead is negligible. We find that MIMO systems with receive antenna selection can outperform the receiver using all the antennas. We then consider a practical situation where the training overhead comes into play. The effective, after training, achievable payload rate of the receiver using the traditional (brute force) channel estimation method is affected greatly by the training overhead. Based on this finding, we propose a channel estimation method to reduce the training overhead. Extensive Monte Carlo simulation shows the proposed reducedcomplexity receiver effectively increases the achievable payload rate.;Assuming spatially correlated Rayleigh fading channels, we investigate the performance of a MIMO receive antenna selection system where only one receive antenna, which maximizes the channel capacity, is selected. We provide closed-form analytical expressions for the system outage probability and an upper bound on the ergodic capacity. Using these results, we also prove analytically that the full diversity order of a system simultaneously using all receive antennas is achieved with the considered antenna selection system. Extensive Monte Carlo simulations are used to validate our analytical findings.;We also propose a suboptimum joint transmit and receive antenna selection algorithm which performs very close to the capacity-optimal exhaustive search method while outperforming the existing suboptimum algorithms. Meanwhile the computational complexity of the proposed algorithm is much less than those of the exhaustive search and the existing suboptimum algorithms. Sensor selection based on observed data is studied for a binary hypothesis testing problem.;Decentralized processing approaches are desired which will save energy, without significant performance loss, by limiting the number of sensor transmissions to a fusion center. The results indicate that proper selection approaches can provide performance relatively close to that of the optimum approach using all the sensors and significant improvement over random selection. One particular distributed approach is shown to provide reasonably large gains over random selection in all the cases considered.