Signal processor/analyzer with a neural network coupled to an acoustic charge transport (act) device (act)

Claim 1

A signal processor/analyzer for real-time analysis of analog signals, comprising an acoustic charge transport device (ACT) and an artificial neural network processor, combined on a single, monolithic GaAs substrate; wherein the acoustic charge transport device comprises a channel, a surface acoustic wave driver coupled to the channel, an input contact coupled to a signal input line for an analog signal, a plurality of taps for sensing the charge of a signal in the channel, and an output contact coupled to a signal output line at which the analog signal can exit; each tap of the acoustic charge transport device being coupled via an attenuator to means for summing and to an input of the artificial neural network processor, a summed output being created in the summing means, the artificial neural network processor having an output which can be fed back to the artificial neural network processor for training and other purposes.

Claim 3

A method of signal processing and analysis for real-time analysis of analog signals, comprising the steps of: a) sampling a signal using a surface acoustic wave (SAW) device at an input contact (source); b) transporting charge packets in an acoustic charge transport device (ACT) channel, there being a plurality of taps over the channel; c) non-destructively sensing the charge in packets passing the taps; d) attenuating the signal at outputs from the taps, to normalize the signal to provide normalized data before input to an artificial neural network having a plurality of "neurons" organized in levels; e) multiplying the normalized data entering the "neurons" by a weighting variable, then summing all inputs to each neuron; a resulting sum being passed through a thresholding function and a resulting value being passed to a next level of "neurons"; f) the charge packets continuing in the ACT channel until said charge packets are extracted at an output.

Claim 4

A method of signal processing and analysis for real-time analysis of analog signals, comprising the steps of: a) sampling a signal using a surface acoustic wave (SAW) device at an input contact (source); b) transporting charge packets in an acoustic charge transport device (ACT) channel, there being a plurality of taps over the channel; c) non-destructively sensing the charge in packets passing the taps; d) attenuating the signal at outputs from the taps, to normalize the signal to provide normalized data before input to an artificial neural network having a plurality of "neurons" organized in levels; e) multiplying the normalized data entering the "neurons" by a weighting variable, then summing all inputs to each neuron; a resulting sum being passed through a thresholding function and a resulting value being passed to a next level of "neurons"; f) the charge packets continuing in the ACT channel until said charge packets are extracted at an output.