Neural network architecture based on summation of phase-coherent alternating current signals

Claim 2

An alternating current (AC) neuron system having a number n, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', where k is an index which ranges from 1 to n, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in , said system comprising: a weighting means for giving to said signal of amplitude v k in a weight w ik respectively for each k, thereby providing a weighted signal; a summing means for forming an AC sum signal S i in the form ##EQU29## and, a nonlinear processing means for converting said AC sum signal S i into a neuron output v i out , said nonlinear processing means including an alternating current (AC) to direct current (DC) processing means for converting said AC sum signal S i into a DC signal and a threshold means connected to said AC to DC processing means for determining whether said DC signal exceeds a threshold T i , wherein said threshold means produces an output as a function of the difference D i between said DC signal and said threshold T i .

Claim 3

An alternating current (AC) neuron system in accordance with claim 1, wherein: v k in ' and v k in are 180 degrees out of phase with each other, so that v k in '=-v k in .

Claim 4

The AC neuron system of claim 2 further comprising: a source of AC to be gated; and, AC gate means connected to said threshold means and to said AC source for producing an AC output signal, said output signal consisting of a signal of amplitude +v i out and a second signal of amplitude -λ i out v i out , wherein λ i out ≧0 is a scale factor of arbitrary magnitude, wherein the AC output signal has the form of v i out =θ(D i ), wherein θ is a function of its argument D i which is 1, in arbitrary output units, for D i larger than a positive threshold t + , and where θ is -λ, wherein λ≧0 is a scale factor of arbitrary magnitude in the same output units, when D i is smaller than a negative threshold -t - , and where θ varies smoothly from -λ to 1 as D i varies from -t - to t + .

Claim 5

The AC neuron system of claim 3, wherein said means for producing an AC output signal further comprises: phase-shifting means, having an input connected to said source of AC to be gated, and an output connected to said AC gate means, wherein said phase shifting means increments the phase of said output signal v i out by a constant phase shift.

Claim 6

The AC neuron system of claim 4, wherein said phase shifting means further comprises: a first AC terminal connected directly, and a second AC terminal connected through an inverting op-amp, to said source AC to be gated; a double-pole double-throw switch connected to said first and second AC terminals in direct order for a first throw and in reversed order for a second throw; a resistor R i having a first and second terminal, with said first terminal connected to one pole of said double-pole double-throw switch; a capacitor C i having a first and second terminal, with said first terminal of said capacitor connected to the second pole of said double-pole double throw switch; and, with the second terminal of said capacitor C i connected to the second terminal of said resistor R i , and wherein said common second terminal is connected, as the source of AC to be gated, to said AC gate means.

Claim 7

The AC neuron system of claim 2, wherein said weighting means further comprises: a first capacitor C ik + , and a second capacitor C ik - , said capacitors having a first terminal connected to the respective input terminals receiving signals +v k in and -λ k in v k in , and their second terminals connected together to form a summing node i, wherein the weight w ik given to the input v k in is proportional to the difference C ik + -λ k in C ik - .

Claim 8

The AC neuron system of claim 2, wherein said weighting means further comprises: phase-shifting and signal dividing means, having an input coupled to said input terminals receiving input signals +v k in and -λ k in v k in , and an output coupled to said AC to DC processing means, wherein said phase-shifting means increments the phase of the input amplitude v k in by a constant phase shift.

Claim 9

The AC neuron system of claim 7, wherein λ k in =1 and said phase-shifting and signal dividing means further comprise: a phase-shifter capacitor C ik p and resistor R ik p , and a voltage divider capacitor C ik , wherein said input terminals receiving input signals of amplitudes +v k in and -v k in are connected to a first terminal of said capacitor C ik p and resistor R ik p respectively, and the second terminal of capacitor C ik p and resistor R ik p are connected in common together to a first terminal of capacitor C ik , with the second terminals of capacitors C ik for the different values of k connected to form a summing node i.

Claim 10

The AC neuron system of claim 8 further comprising: a double pole, double throw switch connected to said input terminals receiving input signals of amplitudes +v k in and -v k in , in direct and reversed orders for the two throws, with the two switch poles connected to said capacitor C ik p and resistor R ik p .

Claim 11

The AC neuron system of claim 2, wherein said AC to DC processing means comprises: a rectifier receiving as an AC input said sum signal S i ; and, filtering means connected to said rectifier and to said threshold means.

Claim 12

The AC neuron system of claim 10 further comprising: a high input impedance follower having an input receiving said sum signal S i , and an output connected to said rectifier, wherein said follower isolates said summing means from said AC to DC processing means.

Claim 13

The AC neuron system of claim 10 further comprising: a high input impedance amplifier having an input receiving said sum signal S i , and an output connected to said rectifier, wherein said amplifier isolates said summing means from said AC to DC processing means, and provides signal amplification with a gain of at least unity as required by said AC to DC processing means.

Claim 14

The AC neuron system of claim 10 wherein said threshold means further comprises: a comparator having a first input, a second input and an output, wherein said first input of said comparator is connected to said filtering means, and said second input of said comparator is connected to said threshold T i .

Claim 15

The AC neuron system of claim 10 wherein said AC to DC processing means further comprises: an offset means for providing an offset u i to said sum S i , producing a modified sum S i +u i which comprises the rectifier input signal.

Claim 16

The AC neuron system of claim 14 wherein said weighting means and said offset means further comprise: a first capacitor C ik + and a second capacitor C ik - , said capacitors having a first terminal connected to the respective input signal terminals receiving input signals +v k in and -λ k in v k in , and their second terminals connected together to form a summing node i; an AC offset signal means for providing a first signal +U i , and a second signal -λ i off U i , wherein λ i off ≧0 is a scale factor of arbitrary magnitude; and, a first capacitor C ui + , and a second capacitor C ui - , said capacitors having a first terminal connected to the respective offset signals +U i and -λ i off U i , and their second terminals connected together to said summing node i, wherein said offset u i is proportional to (C ui + -λ i off C ui - )U i .

Claim 17

The AC neuron system of claim 2, wherein said AC to DC processing means comprises: a source providing an AC signal of amplitude u i which is phase-coherent with the signal of amplitude v k in ; a heterodyne mixer means having a first and a second input, and a DC-coupled output, said first input of said heterodyne mixer receiving said sum signal S i , and said second input of said heterodyne mixer receiving said AC signal of amplitude u i ; and, filtering means connected to the output of said heterodyne mixer and to said threshold means.

Claim 18

The AC neuron system of claim 16 further comprising: a high input impedance follower having an input receiving said sum signal S i , and an output connected to said heterodyne mixer, wherein said follower isolates said summing means from said AC to DC processing means.

Claim 19

The AC neuron system of claim 16 further comprising: a high input impedance amplifier having an input receiving said sum signal S i , and an output connected to said heterodyne mixer, wherein said amplifier isolates said summing means from said AC to DC processing means, and provides signal amplification with a gain of at least unity as required by said AC to DC processing means.

Claim 20

The AC neuron system of claim 16 wherein said threshold means further comprises: a comparator having a first input, a second input and an output, with said first input of said comparator connected to said filtering means, and with said second input of said comparator connected to said threshold T i .

Claim 21

An alternating current neuron system as in claim 2, further comprising: means to split said sum signal S i into two identical signals S 1i =S i and S 2i =S i ; a source of AC signals of amplitudes u i and -u i which are phase-coherent with the signal of amplitude v k in received on the first input terminal; two sets of AC to DC processing means, the first of which processes said signal S 1i and said AC signal of amplitude u i to give a DC output X 1i , and the second of which processes said signal S 2i and said AC signal of amplitude -u i , to give a DC output X 2i ; comparison means to determine the difference X 1i -X 2i thereby providing a difference signal D i ; a source of AC to be gated; and, AC gate means connected to said comparison means and to said AC source for producing an AC output signal, said output signal consisting of a first signal +v i out and a second signal -λ i out v i out wherein λ i out ≧0 is a scale factor of arbitrary magnitude, wherein the AC output signal has the form of v i out =θ(D i ), wherein θ is a function of its argument D i which is 1, in arbitrary output units, for D i larger than a positive threshold t + , and where θ is -λ, wherein λ≧0 is a scale factor of arbitrary magnitude in the same output units, when D i is smaller than a negative threshold -t - , and where θ varies smoothly from -λ to 1 as D i varies from -t - to t + .

Claim 22

An AC neuron system having a number N, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', with k an index which ranges from 1 to N, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in said system comprising: a multiple weighting means for giving to said signal of amplitude v k in multiple weights w lik respectively for each k, with l ranging from 1 to L, where 2≦L≦N, thereby providing multiple weighted signals; a summing means for forming L sum signals S li in the form ##EQU30## and, a nonlinear processing means for converting said L sum signals S li into a neuron output v i out including an AC source providing a set of phase-coherent AC output reference signals r m , with m ranging from 1 to M, with M at least 1, wherein said nonlinear processor means is connected to said AC source to produce an AC output signal v i out as a nonlinear function of said L sum signals S li and said reference signals r m .

Claim 23

An AC neuron system having a number N, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', with k an index which ranges from 1 to N, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in said system comprising: a multiple weighting means for giving to said signal of amplitude v k in multiple weights w lik respectively for each k, with l ranging from 1 to L, where 2≦L≦N, thereby providing multiple weighted signals; a summing means for forming L sum signals S li in the form ##EQU30## and, a nonlinear processing means for converting said L sum signals S li into a neuron output v i out including an AC source providing a set of phase-coherent AC output reference signals r m , with m ranging from 1 to M, with M at least 1, wherein said nonlinear processor means is connected to said AC source to produce an AC output signal v i out as a nonlinear function of said L sum signals S li and said reference signals r m .

Claim 24

An AC neuron system in accordance with claim 21, wherein: v k in ' and v k in are 180 degrees out of phase with each other, so that v k in '=-v k in .

Claim 25

An alternating current (AC) neuron system having a number n, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', where k is an index which ranges from 1 to n, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in , said system comprising: a weighting means for giving to said signal of amplitude v k in a weight w ik respectively for each k, thereby providing a weighted signal, said weighting means including a first capacitor C ik , and a second capacitor C' ik , said capacitors having a first terminal connected to the respective input terminal receiving said signals of amplitude v k in and λ k in v k in ', and their second terminals connected together to form a summing node i, wherein said weight w ik is proportional to the combination C ik +λ k in (v k in '/v k in )C' ik ; a summing means for forming an AC sum signal S i in the form ##EQU31## and, a nonlinear processing means for converting said sum signal S i into a neuron output v i out .

Claim 26

An alternating current (AC) neuron system having a number n, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', where k is an index which ranges from 1 to n, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in , said system comprising: a weighting means for giving to said signal of amplitude v k in a weight w ik respectively for each k, thereby providing a weighted signal, said weighting means including a first capacitor C ik , and a second capacitor C' ik , said capacitors having a first terminal connected to the respective input terminal receiving said signals of amplitude v k in and λ k in v k in ', and their second terminals connected together to form a summing node i, wherein said weight w ik is proportional to the combination C ik +λ k in (v k in '/v k in )C' ik ; a summing means for forming an AC sum signal S i in the form ##EQU31## and, a nonlinear processing means for converting said sum signal S i into a neuron output v i out .

Claim 27

An alternating current (AC) neuron system in accordance with claim 23, wherein: v k in ' and v k in are 180 degrees out of phase with each other, so that v k in '=-v k in .

Claim 28

An AC neuron system having a number N, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', with k an index which ranges from 1 to N, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in , said system comprising: a multiple weighting means for giving to said signal of amplitude v k in multiple weights w lik respectively for each k, with l ranging from 1 to L, where 2≦L≦N, thereby providing multiple weighted signals, wherein said multiple weighting means includes a first capacitor C lik , and a second capacitor C' lik , said capacitors having a first terminal connected to the respective input terminal receiving said signals of amplitude v k in and λ k in v k in ', and their second terminals connected together to form a summing node li, wherein said multiple weight w lik is proportional to the combination C lik +λ k in (v k in '/v k in )C' lik ; a summing means for forming L sum signals S li in the form ##EQU32## and, a nonlinear processing means for converting said L sum signals S li into a neuron output v i out .

Claim 29

An AC neuron system having a number N, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', with k an index which ranges from 1 to N, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in , said system comprising: a multiple weighting means for giving to said signal of amplitude v k in multiple weights w lik respectively for each k, with l ranging from 1 to L, where 2≦L≦N, thereby providing multiple weighted signals, wherein said multiple weighting means includes a first capacitor C lik , and a second capacitor C' lik , said capacitors having a first terminal connected to the respective input terminal receiving said signals of amplitude v k in and λ k in v k in ', and their second terminals connected together to form a summing node li, wherein said multiple weight w lik is proportional to the combination C lik +λ k in (v k in '/v k in )C' lik ; a summing means for forming L sum signals S li in the form ##EQU32## and, a nonlinear processing means for converting said L sum signals S li into a neuron output v i out .

Claim 30

An AC neuron system in accordance with claim 25, wherein: v k in ' and v k in are 180 degrees out of phase with each other, so that v k in '=-v k in .

Claim 31

An alternating current (AC) neuron system having a number n, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', where k is an index which ranges from 1 to n, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in , said system comprising: a weighting means for giving to said signal of amplitude v k in a weight w ik respectively for each k, thereby providing a weighted signal, wherein for at least two of the signals of amplitude v k in said weighting means includes a voltage divider including a capacitance coupled to one of the terminals of the input signal terminal pair for said signal of amplitude v k in ; a summing means for forming an AC sum signal S i in the form ##EQU33## and, a nonlinear processing means for converting said AC sum signal S i into a neuron output v i out .

Claim 32

An alternating current (AC) neuron system having a number n, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', where k is an index which ranges from 1 to n, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in , said system comprising: a weighting means for giving to said signal of amplitude v k in a weight w ik respectively for each k, thereby providing a weighted signal, wherein for at least two of the signals of amplitude v k in said weighting means includes a voltage divider including a capacitance coupled to one of the terminals of the input signal terminal pair for said signal of amplitude v k in ; a summing means for forming an AC sum signal S i in the form ##EQU33## and, a nonlinear processing means for converting said AC sum signal S i into a neuron output v i out .

Claim 33

An alternating current (AC) neuron system in accordance with claim 27, wherein: v k in ' and v k in are 180 degrees out of phase with each other, so that v k in '=-v k in .

Claim 34

An AC neuron system having a number N, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', where k is an index which ranges from 1 to N, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in , said system comprising: a multiple weighting means for giving to said signal of amplitude v k in multiple weights w lik respectively for each k, with l ranging from 1 to L, where 2≦L≦N, thereby providing multiple weighted signals, wherein for at least two of the signals of amplitude v k in said multiple weighting means includes a voltage divider including a capacitance coupled to one of the terminals of the input signal terminal pair for said signal of amplitude v k in ; a summing means for forming L sum signals S li in the form ##EQU34## and, a nonlinear processing means for converting said L sum signals S li into a neuron output v i out .

Claim 35

An AC neuron system having a number N, which is greater than 1, of input signal terminal pairs for receiving phase-coherent alternating signal inputs, a first terminal of each input pair receiving a signal of amplitude v k in , and a corresponding second terminal of each input pair receiving a signal of amplitude λ k in v k in ', where k is an index which ranges from 1 to N, wherein λ k in ≧0 is a scale factor of arbitrary magnitude and v k in ' is an amplitude phase displaced with respect to v k in , said system comprising: a multiple weighting means for giving to said signal of amplitude v k in multiple weights w lik respectively for each k, with l ranging from 1 to L, where 2≦L≦N, thereby providing multiple weighted signals, wherein for at least two of the signals of amplitude v k in said multiple weighting means includes a voltage divider including a capacitance coupled to one of the terminals of the input signal terminal pair for said signal of amplitude v k in ; a summing means for forming L sum signals S li in the form ##EQU34## and, a nonlinear processing means for converting said L sum signals S li into a neuron output v i out .

Claim 36

An AC neuron system in accordance with claim 29, wherein: v k in ' and v k in are 180 degrees out of phase with each other, so that v k in '=-v k in .

Claim 37

An AC neuron system having a number n, which is greater than 1, of input signal terminal groups, each group containing a plurality of terminals, for receiving phase-coherent alternating signal inputs, a first terminal of each input group receiving a signal of amplitude v k in , with k an index which ranges from 1 to n, and with each further terminal t of aforesaid input group receiving a signal of amplitude λ k ,t in v k ,t in , wherein λ k ,t in ≧0 is a scale factor of arbitrary magnitude and v k ,t in is an amplitude phase displaced with respect to v k in , said system comprising: a weighting means for giving to said signal of amplitude v k in a weight w ik respectively for each k, thereby providing a weighted signal, wherein for at least two of the signals of amplitude v k in said weighting means includes a voltage divider including a capacitance coupled to one of the terminals of the input signal terminal group for said signal of amplitude v k in ; a summing means for forming an AC sum signal S i in the form ##EQU35## and, a nonlinear processing means for converting said AC sum signal S i into a neuron output v i out .

Claim 38

An AC neuron system having a number n, which is greater than 1, of input signal terminal groups, each group containing a plurality of terminals, for receiving phase-coherent alternating signal inputs, a first terminal of each input group receiving a signal of amplitude v k in , with k an index which ranges from 1 to n, and with each further terminal t of aforesaid input group receiving a signal of amplitude λ k ,t in v k ,t in , wherein λ k ,t in ≧0 is a scale factor of arbitrary magnitude and v k ,t in is an amplitude phase displaced with respect to v k in , said system comprising: a weighting means for giving to said signal of amplitude v k in a weight w ik respectively for each k, thereby providing a weighted signal, wherein for at least two of the signals of amplitude v k in said weighting means includes a voltage divider including a capacitance coupled to one of the terminals of the input signal terminal group for said signal of amplitude v k in ; a summing means for forming an AC sum signal S i in the form ##EQU35## and, a nonlinear processing means for converting said AC sum signal S i into a neuron output v i out .

Claim 39

An AC neuron system having a number N, which is greater than 1, of input signal terminal groups, each group containing a plurality of terminals, for receiving phase-coherent alternating signal inputs, a first terminal of each input group receiving a signal of amplitude v k in , with k an index which ranges from 1 to N, and with each further terminal t of aforesaid input group receiving a signal of amplitude λ k ,t in v k ,t in , wherein λ k ,t in ≧0 is a scale factor of arbitrary magnitude and v k ,t in is an amplitude phase displaced with respect to v k in , said system comprising: a multiple weighting means for giving to said signal of amplitude v k in multiple weights w lik respectively for each k, with l ranging from 1 to L, where 2≦L≦N, thereby providing multiple weighted signals, wherein for at least two of the signals of amplitude v k in said multiple weighting means includes a voltage divider including a capacitance coupled to one of the terminals of the input signal terminal group for said signal of amplitude v k in ; a summing means for forming L sum signals S li in the form ##EQU36## and, a nonlinear processing means for converting said L sum signals S li into a neuron output v i out .

Claim 40

An AC neuron system having a number N, which is greater than 1, of input signal terminal groups, each group containing a plurality of terminals, for receiving phase-coherent alternating signal inputs, a first terminal of each input group receiving a signal of amplitude v k in , with k an index which ranges from 1 to N, and with each further terminal t of aforesaid input group receiving a signal of amplitude λ k ,t in v k ,t in , wherein λ k ,t in ≧0 is a scale factor of arbitrary magnitude and v k ,t in is an amplitude phase displaced with respect to v k in , said system comprising: a multiple weighting means for giving to said signal of amplitude v k in multiple weights w lik respectively for each k, with l ranging from 1 to L, where 2≦L≦N, thereby providing multiple weighted signals, wherein for at least two of the signals of amplitude v k in said multiple weighting means includes a voltage divider including a capacitance coupled to one of the terminals of the input signal terminal group for said signal of amplitude v k in ; a summing means for forming L sum signals S li in the form ##EQU36## and, a nonlinear processing means for converting said L sum signals S li into a neuron output v i out .