Acoustic Charge Transport Implementation of Neural Networks for Communications Applications

High speed, low cost, light weight, and reliable systems that can significantly improve performance are sought for the next generation of military communications systems. The recent advent of broadband communication services has dramatically increased this demand. The opportunity exists to merge two of the most promising advances in cornmunications: acoustic charge transport (ACT) devices and neural networks. ACT devices are available that implement a 128 tap programmable transversal filter operating at 360 MHz on a chip measuring less than 25 square millimiters. Although these devices provide wideband communications signal processing, they are limited to linear operation. It is possible to use these devices to configure neural networks that perform nonlinear filtering operations. Neural network filtering operations for adaptive equalization and signal detection have consistently shown at least a 3dB performance improvement over their linear counterparts. This project will design, implement, test, and analyze an ACT-based implementation of a neural network for communications applications. The resulting ACT-based neural network implementation provides almost an order of magnitude improvement over the faster commercially available neural network devices available today.