NEURONAL INTEGRATIVE MECHANISMS 15 



Changes in the phasic rather than the tonic level of this extramural in- 

 fluence can lead to still more complicated differential effects, since now 

 we have the several successive phases of after-effects, which are different 

 between large and small ganglion cells and between acceleration and in- 

 hibition. An example of these effects is the paradoxical driving of nearly 

 quiescent units by phasic inhibition, via the postinhibitory rebound 

 (Fig- 2). 



Signal Versus Noise and Some 

 Specifications for a Brain 



The common case of continual discharge in sensory nerve fibers may be 

 partly understood as representing ( 1 ) reception of steady states or/and 

 (2) a state of poise associated with high sensitivity, and also (3) a pro- 

 vision for signaling in one line both positive and negative changes in the 

 stimulating parameter (e.g., increases and decreases in temperature, light, 

 or stretch, and forwards and backwards movements of hairs in statolith 

 organs). But it raises a serious problem in any case where rhythmicity is 

 not perfect. What change in the output of the sensory nerve fiber consti- 

 tutes a signal of environmental change and not a random fluctuation in 

 spontaneity ? 



This problem has some general significance and not alone in certain 

 sense organs but probably in the integration of signals between centers in 

 the brain. Deferring for a moment any contributions to the discrimination 

 between signal and noise provided by a multiplicity of parallel channels 

 (afferent nerve fibers), it is necessary to consider first the alternatives 

 available for analyzing the input in each single channel. Among them we 

 may recognize certain ones, as follows. The signal could be regarded as : 



( 1 ) The actual instantaneous frequency, i.e., the reciprocal of the inter- 

 val since the last impulse (Fig. 6). If this constitutes the significant signal 

 for the central analyzing mechanism the threshold of the mechanism would 

 be very high, since the signal has to be reliably higher than any spon- 

 taneous fluctuation in interval. In the general case the fluctuations in indi- 

 vidual intervals are larger than any fluctuations in averaged or integrated 

 frequency. This possibility therefore seems unlikely or at least maladaptive 

 and wasteful of information. 



(2) Frequency averaged over some period (Fig. 6) . The threshold will 

 be lower than in ( 1 ) but, if the time constant of integration is short, not 

 much lower ; and, if the time constant is long, reaction time, detection of 

 brief stimuli, flicker resolution, and spatial localization of moving stimuli 

 will suffer. Some compromise seems quite possible but not as advantageous 

 as the followins:. 



