DIGITAL NEURON MODEL USING DIGITAL PHASE-LOCKED LOOP

We propose a new type of the digital neuron model by using multi-input multilevel-quantized digital phase-locked loop (MM-DPLL), where the input is represented by the phase modulated signal. It is shown that this model has the characteristics of the neuron; spatial summation, temporal summation and thresholding. We applied our model to the pattern recognition and to the Hopefield type associative memory, in order to verify that the network by this model can operate properly. In the pattern recognition, we used the perceptron convergence procedure (delta rule), and confirm the possibility of learning by modifying the connection weights. In the associative memory, we confirm that the network can learn five digit patterns of the fundamental memories, and also can recall the correct pattern for the noisy input pattern.