ADPCM environment with a neural network predictor engine

Abstract

Presented in this paper is an innovative technique to improve analog-to-digital converters (ADCs). The methodology utilizes a built-in neural network engine to first learn, and then predict, the quantization step size. This produces a completely adaptable ADC, consisting of ADPCM samples fed back with the output of the neural network, which is capable of generating a better quantized output. The algorithmic solution, simulation methodology and results are also presented in this writing.