A CML also shows some interesting behaviour for step-like and periodic
inputs. In these simulations, 
was not used and an input function was
applied directly to 
. The weights used were again 
and
the nonlinearity parameters were all 
.
Figure 6: The left-hand figures show various step inputs of different
severity and the right-hand figures, the corresponding outputs.
The y-values for the outputs have all had 0.67742 (
) subtracted
from them and are scaled differently: in (d), the y-axis has been
scaled by 
; in (e), by 
; and in (f), by 
.
The effect of increasingly sharp edges is illustrated in
figure 6. The inputs, 
are shown in figures (a)--(c), and
are, respectively, a 
function, a linear ramp and a step. The responses,
, are shown in figures (d)--(f). Note that the vertical scaling
for the outputs is different in each case. It is seen that less severe edges
give rise to a very weak response. Increasing N makes the response
vanishingly small in all cases except the step input.
We see the CML in this case acting as an edge detector.