Sensor for use in a neural network

#This invention relates generally to artificial neurons, which simulate the action of biological neurons, and more particularly, to neurons which operate by using a form of delta-sigma modulation to encode the neural states.

#Generally, as shown in FIG. 1, an artificial neuron comprises a number of input signal paths, x.sub.0, x.sub.1, . . . x.sub.n, a signal weighting element w.sub.0, w.sub.1, . . . w.sub.n, with a corresponding numerical weight value associated with each input signal; a plurality of multipliers 101 for combining the input signal x with the weighing element w to provide a weighted input signal xw; an adder 103, which combines the weighted input signals xw to produce an intermediate signal u; and an element which transforms the intermediate signal u (usually by a nonlinear function F) to produce the output signal y of the neuron. A general discussion on neural networks and neurons generally is found in Widrow, et. al., "30 Years of Adaptive Neural Networks: Perceptron, Madaline, and Backpropagation," Proceedings of the IEEE, Vol 78, No. 9, September 1990.

#Artificial neurons may take many forms. Early studies were focused on neurons that operated entirely with continuous-time signals in the analog domain. More recently, neurons have been developed which use digital processing. In these digital neurons, the input and output signals typically are discrete-time, multibit digital values with the weights and computations stored digitally. Regardless of whether the neuron was analog or digital, the computation, including multiplication of the weights, addition of the weighted signals, and transformation by the nonlinear function, required a relatively complex system.

#FIG. 1 shows a block diagram of a generic prior art artificial neuron.

#FIG. 2 shows a block diagram of a generic DS-neuron constructed in accordance with the preferred embodiment.

#FIG. 3 shows a block diagram of a first-order, delta-sigma modulator.

#General Discussion

#Referring now to FIG. 2, a DS-neuron 10 constructed in accordance with the preferred embodiment includes a number (n+1) of input signals x.sub.0, x.sub.1, . . . , x.sub.n on input signal paths 1 to the neuron input unit (NIU) 5. Typically, input signal x.sub.0 is a nonzero constant value, such as 1. After the NIU 5 receives the input signals x, it produces an intermediate signal u on signal path 7. Intermediate signal u then is encoded by a delta-sigma modulator 8 in accordance with the preferred embodiment to produce the output signal y on the output signal path 9 of the DS-neuron 10.

#Generally, NIU 5 produces the modified sum-of-products