Self-recovery system for an intelligent RF front end amplifier

Abstract

This paper describes a self-recovery algorithm for a neural network-based controller for an intelligent radiofrequency front-end amplifier. The neurocontroller provides autonomous operation, assessment and recovery capabilities. The neurocontroller is designed to reconfigure the input and output matching networks architecture, thereby providing control of the gain performance at an operating frequency within a 10-50 GHz frequency band. The controller system is composed of a sliding scale estimator of the gain dynamics in the input and output networks, and two pairs of multilayer perceptron (MLP) neural networks: one pair for control of the input network, and one pair for control of the output network. Each pair consists of a MLP neural network for extraction of feature parameters in input reflection coefficient (gamma) space from the estimated gain dynamics, and one for classification of the extracted features to configuration codes for the respective network. The neurocontroller can also facilitate autonomous adaptation of system architecture in response to failures and/or drift in MEMS components. Using the self-recovery system, 30 GHz simulation results demonstrate an average 98% percent recovery of the amount of decreased gain relative to recovery achieved using a manual tuning approach. Optimal recovery is achieved in an average 5 iterations