Optimal weights decoding of M-ary suprathreshold stochastic resonance in stochastic pooling network

Suprathreshold stochastic resonance (SSR) is a noise enhancing signal processing phenomenon, occurring in a parallel array of nonlinear elements. In this paper, we investigate the optimal decoding scheme of SSR in stochastic pooling network with a quantization evolution of the output signal-to-quantization-noise ratio (SQNR) by studying the optimal weights and the optimal thresholds. Firstly, we introduce an effective method of weights decoding which makes the better SSR effects than the Wiener linear decoding and is defined as the optimal weights decoding. Moreover, in order to find the optimal thresholds, we select three common forms of thresholds in [0,1] interval the uniform thresholds, the random thresholds and the group thresholds. The result indicates that the group thresholds make a better SSR effect than uniform thresholds, but worse than the random thresholds. Therefore, the random thresholds are the optimal thresholds setting in the M-ary stochastic pooling network. Finally, we discuss the influences of number of elements N and threshold level M on SSR, and find that changing the number of the comparators N in stochastic pooling network is more easier to enhance the performance of SSR than changing the values of M. These works as a complement to the optimal quantification theory will be helpful for the study of optimal thresholds in stochastic pooling network.