112
molecules [90]. Both of these phenomenon have temporal spectra that are largely confined
to frequencies below 1 Hz [105, 62], so according to the model frequency response (figure
5.8), the retinas affect on these noise sources will be described by the steady-state response.
Our analysis shows that the standard deviation of the noise in the coupled salamander
retina is halved compared to a hypothetical isolated rod. This means that the signal-to-noise
ratio for whole-field stimuli is double what it would be in a rod network without coupling.
The decay in the coupled retina’s response to increasing spatial frequencies means that at a
feature wavelength of 50 microns, the SNR is equal to that of the uncoupled retina. How-
ever, the minimum perceptible wavelength before image aliasing occurs is twice the receptor
spacing, or 32 microns (figure 5.12 A). So for most perceptible spatial frequencies, coupling
between rods leads to an increase in the signal-to-noise over the uncoupled network (figure
5.12 B, C), and for higher frequencies, the SNR is at least 80% of the uncoupled network.
Rod bipolar cells are thought to have synaptic inputs from several rods, as their dendritic
field diameter is ≈ 60 microns [109]. While their receptive fields are larger [113], this is
thought to be due to coupling between bipolar cells, as this is too far to have occurred from
rod-rod coupling. This means that, uncoupled, each bipolar cell may average the responses
of 12-13 rods. The averaging of rod responses by bipolar cells would reduce both the quantal
noise from the synapse and intrinsic noise from photoreceptors. This would seem to make
the averaging of rod signals via rod-rod coupling redundant— if the synapse was linear.
However, current evidence shows that the rod bipolar cell synapse is not linear^ [111]. The
Jlt has been shown that the rod-DBC synapse can be approximated as linear for a given light intensity [110]