NEURONAL INTEGRATIVE MECHANISMS 17 



maintained high frequency would gradually raise the denominator until 

 the ratio fell to unity. These modalities or the steady-state aspect of such 

 input must use (2), above. 



A further improvement in the extraction of information from such 

 fluctuating signals arising from phasic input would result if the receiver 

 could place a greater value upon changes in the ratio which occur rapidly, 

 since these are more likely to be real stimulus-signals and less likely to be 

 spontaneous fluctuations. This likelihood is based on the randomness of the 

 successive intervals in spontaneous fluctuations ; several short intervals 

 in succession are highly unlikely. One possible formulation is the following. 



(4) [Ratio (3) ] X [Rate of change of ratio (3) integrated over RC3] . 

 RC3 would probably be intermediate between RCi and RC2. If F suddenly 

 changes, Frci will be increased soon, the ratio will increase, the rate of 

 increase of the ratio will be maximal for a short while, and a sudden stimu- 

 lus will be signalled thereby. The signal meaning "strong stimulus" will 

 then begin to decline even if the high F is maintained — first, because of the 

 decrease over time constant RC3 in the extra multiplying factor, and then 

 because of the gradual fall in the ratio, as the high F gradually raises the 

 denominator. 



The possibility that, instead of a ratio of short-term integrated frequency 

 to long-term integrated frequency, the latter should simply be subtracted 

 from the former is available but seems less likely because of its dispropor- 

 tional sensitivity at low intensities. 



It is possible by one further step to extract still more information from 

 the same pattern of impulses. If the central nervous system can subtract 

 a fixed frequency from the incoming frequency and then estimate the ratio 

 of frequency integrated over the short term to that over the longer period, 

 and increase the value if there has recently been a sudden rise, the etTects 

 of random fluctuations in successive intervals in the background spon- 

 taneity could be largely eliminated. 



[F^^^^^^^~^n..a]mtegr^tedoyerRC. rRate of change of this T 



[Factual — Fnxed] integrated over RC, ^ Lratio during RC3 J 



If the fixed frequency is higher than the extreme range of random fluc- 

 tuations, no ratio is necessary since the numerator is zero, but sensitivity 

 will not be maximal. Sensitivity will be maximal if only a few spontaneous 

 impulses exceed the fixed level ; the ratio will be more quickly altered by 

 a real stimulus and the time constant RCi can now be shorter than in the 

 last two formulae, improving reaction time, flicker, and brief stimulus de- 

 tection while still smoothing enough to eliminate nearly all false signals. 



Thus far we have considered the desiderata for a valid signal of en- 



