Apparatus and method for enhancing transfer function non-linearities in ...

1. A network for operating on a pulse frequency encoded input signal for producing a pulse frequency encoded output signal with a modified pulse rate, the network comprising:

a) an input terminal for inputting the pulse frequency encoded input signal;

b) a delay circuit with an input connected to the input terminal for inputting the pulse frequency encoded input signal and outputting, at an output, a delayed pulse frequency encoded input signal, delayed by at least one minimal pulse interval; and

c) an OR-gate with an output, a first input connected to the input terminal, and a second input connected to the output of the delay circuit, for OR-ing the pulse frequency encoded input signal with the delayed pulse frequency input signal, and for producing a pulse frequency encoded output signal at the output with an increased pulse rate relative to the pulse frequency encoded input signal pulse rate, the increased pulse rate increasing monotonically relative to the pulse frequency encoded input signal pulse rate and smoothly approaching a saturation rate equal to a prescribed maximum allowable pulse frequency encoded signal pulse rate.

2. The network of claim 1 wherein the delay circuit is a tapped delay line with a multiplicity of output taps for producing a multiplicity of variously delayed output signals delayed by at least one minimum pulse interval at the output Ia, the OR-gate means having a corresponding multiplicity of inputs for inputting and combining the multiplicity of variously delayed output signals by logical OR-ing all OR-gate means input signals to produce a pulse frequency encoded output signal representative of the unencoded signal after passing through a circuit with a sigmoidal type transfer characteristic.

3. The network of claim 1 wherein the delay means comprises at least one delay element.

4. The network of claim 1 wherein the delay means is a flip-flop.

5. An artificial pulse frequency type neuron with improved pulse frequency encoded sigmoidal squashing characteristics comprising:

a) a pulse frequency type neuron with a prescribed maximum pulse operating rate, the pulse frequency type neuron including,

i) an excitatory vector input terminal for inputting a pulse frequency encoded excitatory vector input signal with at least one vector element,

ii) an inhibitory vector input terminal for inputting a pulse frequency encoded inhibitory vector input signal with at least one vector element,

iii) a synaptic weighting vector input terminal for inputting a pulse frequency encoded synaptic weighting vector, and

iv) an output terminal or outputting a pulse frequency encoded neuron output signal that is a combined excitatory and inhibitory pulse frequency encoded synaptic weighted signal;

b) a sigmoidal enhancement circuit having an input pulse rate to output pulse rate transfer nonlinear characteristic that monotonically increases the output pulse rate, relative to the input pulse rate, the output rate smoothly approaching a prescribed maximum allowable rate corresponding to the pulse frequency encoded neuron prescribed maximum pulse operating rate, for operating on the pulse frequency encoded neuron output signal, a for producing at an output a pulse frequency encoded signal with a modified pulse rate, the sigmoidal enhancement circuit comprising,

i) an enhancement circuit input terminal connected to the output of the pulse frequency type neuron,

ii) delay circuit with an input connected to the output terminal of the pulse frequency type neuron for inputting the pulse frequency encoded neuron output signal and outputting a delayed pulse frequency encoded neuron output signal, delayed by at least one nominal pulse interval, and

iii) OR-gate means having a first input terminal connected to the enhancement circuit input terminal, a second input terminal connected to the output of the delay means for producing the pulse frequency encoded output signal with a modified pulse rate.

6. The neuron of claim 5 wherein the delay means is a tapped delay line with a multiplicity of output taps and the OR-gate means has a corresponding multiplicity of second input terminals for inputting signals from the multiplicity of output taps for logically OR-ing signals on all OR-gate means input terminals.

7. The neuron of claim 5 wherein the delay means is a flip-flop.

8. The neuron of claim 5 further comprising a serial chain enhancement circuit that includes at least one sigmoidal enhancement circuit, each sigmoidal enhancement circuit connected in serial to form a serially connected chain of sigmoidal enhancement circuits, the serial chain enhancement circuit having a single input connected to the artificial pulse frequency type neuron with improved pulse frequency encoded sigmoidal squashing function characteristics output for providing further improved sigmoidal output characteristics.

9. A method for transforming a pulse frequency encoded input signal and directly producing, without decoding, a pulse frequency encoded output signal with a modified pulse rate, the method steps comprising:

a) inputting a pulse frequency encoded input signal;

b) delaying the pulse frequency encoded input signal by at least one nominal pulse interval; and

c) logically OR-ing the pulse frequency encoded input signal and the delayed pulse frequency encoded input signal and producing a pulse frequency encoded output signal with a modified pulse rate, the modified pulse rate being a nonlinear function of the pulse frequency encoded input signal rate that increases monotonically with the pulse frequency encoded input signal rate and smoothly approaches a maximum pulse rate corresponding to a prescribed maximum pulse frequency encoded signal pulse rate.

10. The method of claim 9 wherein step (b) comprises producing a multiplicity of variously delayed signals by delaying the pulse frequency encoded input signal from step (a), each delayed signal being delayed by at least one nominal pulse interval, and step (c) comprises logically OR-ing the multiplicity of variously delayed signals from step (b) and the pulse frequency encoded input signal from step (a).

11. The method of claim 9 further comprising the following steps:

a) using the pulse frequency encoded output signal as a pulse frequency encoded input signal; and

b) repeating the steps of claim 9 at least once.