Digital adaptive receiver employing maximum-likelihood sequence estimation with neural networks

Claim 2

A digital receiver for maximum-likelihood sequence estimation of digital signals transmitted over a time-dispersive transmission channel, comprising: a matched filter with an input for receiving signals transmitted over the transmission channel; and a neural network for receiving sampled signals output by said matched filter and for providing output signals of said receiver which are estimates of signals initially transmitted over said transmission channel.

Claim 3

The receiver of claim 1 further including a coefficient estimator for receiving digital signals output by said neural network and for producing coefficient information for input back to said neural network to modify feedback connection strengths of said neural network.

Claim 4

The receiver of claim 2 further including a thresholding device connected between the output of said neural network and said coefficient estimator, said thresholding device being operative to produce as an output thereof a first state or a second state from an analog signal output by said neural network.

Claim 5

The receiver of claim 2 further including a logic circuit connected to the output of said coefficient estimator and to said neural network for providing changes in interconnection strengths within said neural network for time varying channel characteristics.

Claim 6

The receiver of claim 4 wherein said logic circuit comprises an adder and a negate circuit.

Claim 7

The receiver of claim 2 further including a sampler for sampling signals carried by said channel and inputting said sampled signals to said coefficient estimator.

Claim 8

The receiver of claim 1 wherein said neural network comprises an analog circuit having a number of high gain amplifiers.

Claim 9

The receiver of claim 1 wherein said neural network includes M inputs, and further including an N-bit storage register for storing digital information output by said matched filter for input to respective M inputs of said neural network.

Claim 10

The receiver of claim 1 further including a switch connected to the input of said neural network for momentarily switching said input to a reference to precondition said neural network to predefined states.

Claim 11

The receiver of claim 1 further including a multiplier for multipling the digit information applied to the inputs of said neural network.

Claim 12

The receiver of claim 10 wherein said multiplier comprises a multiply-by-two circuit.

Claim 13

A digital receiver for maximum-likelihood sequence estimation of digital signals transmitted over a time-dispersive transmission channel, comprising: a matched filter with an input for receiving signals transmitted over a transmission channel; and a network for carrying out neural type of function for receiving sampled signals output by said matched filter and for minimizing a cost function according to an expression ##EQU25## for generating a sequence of data symbols representing a maximum-likelihood estimation of the transmitted sequence, where a i are the digital signals transmitted, z i are sampled output of the matched filter, and s k are the coefficients for the maximum-likelihood sequence estimation.

Claim 14

A digital receiver for maximum-likelihood sequence estimation of digital signals transmitted over a time-dispersive transmission channel, comprising: a matched filter with an input for receiving signals transmitted over a transmission channel; and a network for carrying out neural type of function for receiving sampled signals output by said matched filter and for minimizing a cost function according to an expression ##EQU25## for generating a sequence of data symbols representing a maximum-likelihood estimation of the transmitted sequence, where a i are the digital signals transmitted, z i are sampled output of the matched filter, and s k are the coefficients for the maximum-likelihood sequence estimation.

Claim 15

The digital receiver of claim 12 wherein said network comprises a digital simulation of an analog neural network.

Claim 16

The digital receiver of claim 12 wherein said network comprises an analog neural network.

Claim 17

The digital receiver of claim 12 wherein said network includes means for performing digital and analog functions.

Claim 18

The digital receiver of claim 15 wherein said means for performing digital functions comprises means for changing interconnection strengths of a neural network.

Claim 19

The digital receiver of claim 15 wherein said means for performing analog functions comprises a nonlinear amplifier.

Claim 20

A digital receiver for maximum-likelihood sequence estimation of digital signals transmitted over a time-dispersive transmission channel, comprising: a matched filter having an input and an output, said input for receiving signals transmitted over the transmission channel; a sampler connected to the output of said matched filter for sampling signals output by said matched filter to provide a sampled signal; a multiple stage serial-to-parallel shift register for receiving a series of said sampled signals and providing a parallel output thereof; a neural network having a number of neurons for receiving the parallel output of said shift register, said neural network having a number of outputs comprising the output of said receiver; and a coefficient estimator for receiving an estimated sequence output by said neural network and for receiving digital signals carried by said channel, and further including an output providing coefficient information coupled back to said neural network for modifying connection strengths of said network.

Claim 21

A digital receiver for maximum-likelihood sequence estimation of digital signals transmitted over a time-dispersive transmission channel, comprising: a matched filter having an input and an output, said input for receiving signals transmitted over the transmission channel; a sampler connected to the output of said matched filter for sampling signals output by said matched filter to provide a sampled signal; a multiple stage serial-to-parallel shift register for receiving a series of said sampled signals and providing a parallel output thereof; a neural network having a number of neurons for receiving the parallel output of said shift register, said neural network having a number of outputs comprising the output of said receiver; and a coefficient estimator for receiving an estimated sequence output by said neural network and for receiving digital signals carried by said channel, and further including an output providing coefficient information coupled back to said neural network for modifying connection strengths of said network.

Claim 22

The receiver of claim 18 wherein said matched filter input defines a first input, and further including a second input connected to said coefficient estimator so that said matched filter is provided with coefficient information representing characteristics of the transmission channel.

Claim 23

The receiver of claim 18 further including a multiplier circuit for multiplying the digital output of said shift register.

Claim 24

The receiver of claim 20 wherein said multiplier circuit comprises a multiply by two circuit.

Claim 25

The receiver of claim 18 further including means connected to the input of said neural network for preconditioning the output of said neural network to desired states.

Claim 26

The receiver of claim 22 further including means for activating said preconditioning means periodically in association with different inputs to said neural network.

Claim 27

The receiver of claim 18 further including a thresholding device connected between the output of said neural network and the input of said coefficient estimator for providing bipolar signals to said coefficient estimator.

Claim 28

The receiver of claim 18 wherein said neural network comprises a connection matrix interconnecting a plurality of analog amplifiers.

Claim 29

The receiver of claim 25 further including means in said coefficient estimator for producing neural network coefficients based upon time varying characteristics of said transmission channel, and means for changing interconnection strengths of said neural network connection matrix based upon said coefficients.

Claim 30

The receiver of claim 26 wherein said means for changing comprises an adder and a negation circuit for adding a number to said coefficients and for changing the resultant sign thereof.

Claim 33

The method of claim 28 further including producing feedback coefficients using the output of said neural network; and coupling the coefficients back to said neural network for changing connection strengths in said neural network.

Claim 34

The method of claim 28 further including presenting the digital signals to said neural network in parallel.

Claim 35

The method of claim 28 further including coupling the digital signals to said neural network in groups of said digital signals, wherein ones of said digital signals in each said group are common and thus overlapped in time.

Claim 36

The method of claim 31 further including presetting said neural network before each group of a plurality of groups of digital signals are coupled to said neural network.

Claim 37

The method of claim 28 further including presetting said neural network to predetermined states before the coupling of the digital signals thereto.

Claim 38

The method of claim 28 further including multiplying the digital signals by a predetermined factor before coupling the signals to said neural network.

Claim 39

The method of claim 28 further including providing said neural network with a connection matrix for interconnecting neurons of said neural network with desired strengths, and changing the strengths of said connection matrix when predetermined electrical characteristics of said transmission channel change to thereby provide an adaptive receiver for time varying transmission channels.

Claim 40

The method of claim 35 further including changing said interconnection strengths based upon said coefficients.

Claim 41

The method of claim 36 further including applying a signal to said connection matrix formed by adding said coefficients to produce a sum and negating said sum.

Claim 42

The method of claim 28 further including coupling said digital signals to said neural network having a plurality of analog amplifiers, and processing outputs of said amplifiers to produce bipolar states.

Claim 43

The method of claim 28 further including processing the digital signals so as to minimize the function ##EQU26## where a i are the signals transmitted, z i are sampled output of the filter, and s k are the coefficients for the estimation.