104
because it reflects the geometry of the photoreceptor, which is circular. Irregularities in the
frequency response come from the imperfect roundness of the digital representation of each
receptor (figure 5.10 A, B). As expected, both the uncoupled and coupled networks show an
identical response to whole field stimuli (ʃæ = fy = 0), because whole field stimuli ’’uncou-
ple” the network. The coupled network does not show radial symmetry. This is due to the
fact that the length constant for signal spread is different for one dimensional ʌɪp and two
dimensional stimuli A2d∙ Therefore the frequency response to a ID grating (ʌ or fy = 0),
shown in figure 5.10 D, is different from a two dimensional grating of the same frequency
(Λ = Λ)∙
Figure 5.11 shows that the coupled network has a faster falloff in its response to spatial
frequencies than the uncoupled network. However, only a small part of the plots in fig-
ure 5.11 is actually relevant. Because photoreceptor spacing is 16 microns, according to the
Nyquist sampling rule, any spatial wavelengths shorter than 32 microns will not be able to
be represented by the retina. This corresponds to a frequency of 0.03 cycles per micron (or
a wavelength of 32 microns), or 0.5 cycles per rod. Any frequencies higher than 0.03 cycles
per micron will be aliased by the photoreceptors- perceived as frequencies lower than the
stimulus. Figure 5.12 A shows an example of aliasing occurs in the coupled and uncou-
pled network. Note that both networks in 5.12 A have adjacent photoreceptors that are not
shown.
Noise due to sources internal to the rod is another sort of stimulus the network must
contend with. In the case of the uncoupled network, internal noise currents lead directly