Rapid Gain Estimation for Multi-User Software Defined Radio Applications

Abstract

This paper proposes a machine learning algorithm to estimate the peak of a signal generated as a sum of modulated signals. Each signal in the sum may have different modulation format, carrier frequency, data rate, and power level. A dataset of summed signals with varying parameters of individual signals was simulated to train an artificial neural network. The neural network estimates the peak voltage, central moments, variance, skewness, and kurtosis of the summed signal. Once trained, the neural network can rapidly and accurately predict the peak voltage, and statistics of the summed signal, based on the individual signal parameters, and does not have to generate or observe the signal itself. This can be used in an automatic gain control system, to prevent clipping when multiple modulators share a digital to analog converter or amplifier chain.