An adaptable neuro-control system for a 10--50GHz monolithic microwave integrated circuited-based intelligent front-end amplifier.

摘要

Autonomous, multifunctional systems are of growing interests for implementation into emerging wireless communication applications. Such systems will enable provision of mission-defined services and will benefit system designs in terms of compactness, power consumption, and system life. Multifunctionality implies the ability to reconfigure topologies, whereas, autonomy implies the ability to self control. While such capabilities are available within narrow bandwidths of operation, current technologies inhibit these capabilities over bandwidths required by the emerging applications. Innovative solutions are, therefore, required to increase system bandwidth while minimizing system size, power requirements, and cost, and maintaining high operation performance.;In response to this growing interest, the Defense Advanced Research Project Agency (DARPA) created two research initiatives. The objective of the first initiative is to transform current monolithic microwave integrated circuit (MMIC)-based technologies via the development of new device structures or implementation of new material systems. The objective of the second initiative is to enable the development of intelligent front-end components and system utilizing innovative integrated circuit architectures that exploit advancements in digital and analog technologies.;The research contained in this dissertation addresses the latter initiative by developing neural network-based control systems for front-end amplifiers. The developed neuro-controller provides autonomous operation and recovery capabilities. In operation mode, the control system enables regulation of the gain response of the amplifier over a 10-50GHz bandwidth. Such control is achieved via reconfiguration of the matching network topologies. The control system implements a model reference adaptive control strategy. State dynamics are estimated using a pseudo-sliding mode methodology. Back-propagation neural networks are used to extract state estimates and classify them into configurations that maximize the plant's gain response. The neuro-controller can also facilitate autonomous adaptation of the system's architecture in response to failures in switching components within the matching networks. Under recovery mode, 30GHz simulation results demonstrate an average 94% recovery of the decrease in gain relative to recovery performance using a manual tuning approach. For most failure scenarios investigated, the sliding mode-based control system achieved transient behavior merits as defined by the research goals.