Information in channel-coded systems: correlated receivers

Noise correlation can easily occur in the densely connected systems observed in biological information processing. We study the consequences of noise correlation for a statistically optimal processing of noise-perturbed receptor array outputs. We find a critical importance of the noise correlation length as compared to the receptors' tuning width for both the structure and the performance of the ideal observer. We show the general consistency of our scheme with psychophysical discrimination thresholds obtained in human spatial vision.