Decoding suprathreshold stochastic resonance with optimal weights

• A weighted summing array of independently noisy binary comparators is investigated.
• We present an optimal linearly weighted decoding scheme for combining the comparator responses.
• We solve for the optimal weights by applying least squares regression to simulated data.
• We find that the MSE distortion of weighting before summation is superior to unweighted summation of comparator responses.
• For some parameter regions, the decrease in MSE distortion due to weighting is negligible.

We investigate an array of stochastic quantizers for converting an analog input signal into a discrete output in the context of suprathreshold stochastic resonance. A new optimal weighted decoding is considered for different threshold level distributions. We show that for particular noise levels and choices of the threshold levels optimally weighting the quantizer responses provides a reduced mean square error in comparison with the original unweighted array. However, there are also many parameter regions where the original array provides near optimal performance, and when this occurs, it offers a much simpler approach than optimally weighting each quantizer's response.