Cognitive architecture for wideband, low-power, real-time signal denoising

• US 10,153,806 B1
• Filed: 03/07/2017
• Issued: 12/11/2018
• Est. Priority Date: 03/19/2015
• Status: Active Grant

• Current Assignee: HRL Laboratories LLC (The Boeing Co.)
• Original Assignee: HRL Laboratories LLC (The Boeing Co.)
• Inventors: Petre, Peter, Rao, Shankar R.
• Primary Examiner(s): Fan, Chieh M
• Assistant Examiner(s): Hailegiorgis, Fitwi
• Application Number:
  US15/452,412
• Time in Patent Office: 644 Days
• Field of Search: 375130, 342377, 704207
• CPC Class Codes: G06F 18/2134 based on separation criteria, e.g. independent component analysis
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06F 2218/04 Denoising
• CPC Class Codes: G06F 2218/22 Source localisation; Inverse modelling
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: H04B 1/71632 Signal aspects H04B1/7172 and H04B1/7176 take precedence
• CPC Class Codes: H04B 1/719 Interference-related aspects

Abstract

Described is a cognitive signal processor that can denoise an input signal that contains a mixture of waveforms over a large bandwidth. Delay-embedded mixture signals are generated from a mixture of input signals. The delay-embedded mixture signals are mapped with a reservoir computer to reservoir states of a dynamical reservoir having output layer weights. The output layer weights are adapted based on short-time linear prediction. Finally, a denoised output of the mixture of input signals is generated.

Claims

1. A system for signal denoising, the system comprising:

one or more processors and a non-transitory computer-readable medium having executable instructions encoded thereon such that when executed, the one or more processors perform operations of;

generating delay-embedded mixture signals from an input signal comprising a mixture of one or more source signals;

mapping, with a reservoir computer, the delay-embedded mixture signals to reservoir states of a dynamical reservoir having output layer weights, wherein the reservoir computer is a recurrent neural network;

for each input signal, determining a prediction of the input signal using a function of the input signal, current values of the output layer weights, and current and past values of the reservoir states;

determining a prediction error between the prediction of the input signal and the input signal;

for each input signal, iteratively tuning the output layer weights based on the prediction error;

for each input signal, iteratively updating the reservoir states, the prediction error, and the output layer weights; and

generating a denoised output of each input signal.

2. The system as set forth in claim 1, wherein at least one wideband analog-to-digital converter provides input to the dynamical reservoir.

3. The system as set forth in claim 1, wherein the reservoir computer generates outputs by continually mapping reservoir states through a set of distinct linear functions, with one linear function per output.

4. The system as set forth in claim 1, wherein the input signal is denoised in real-time using a constraint that covers a range of electromagnetic and acoustic signals of interest.

5. The system as set forth in claim 1, wherein an online gradient descent algorithm is used in determining the prediction of the input signal.

6. The system as set forth in claim 1, wherein the denoised output is generated from an input signal containing a mixture of waveforms over a 30 Gigahertz (GHz)) bandwidth.

7. A computer implemented method for signal denoising, the method comprising an act of:

causing one or more processers to execute instructions encoded on a non-transitory computer-readable medium, such that upon execution, the one or more processors perform operations of;

generating delay-embedded mixture signals from an input signal comprising a mixture of one or more source signals;

mapping, with a reservoir computer, the delay-embedded mixture signals to reservoir states of a dynamical reservoir having output layer weights, wherein the reservoir computer is a recurrent neural network;

for each input signal, determining a prediction of the input signal using a function of the input signal, current values of the output layer weights, and current and past values of the reservoir states;

determining a prediction error between the prediction of the input signal and the input signal;

for each input signal, iteratively tuning the output layer weights based on the prediction error;

for each input signal, iteratively updating the reservoir states, the prediction error, and the output layer weights; and

generating a denoised output of each input signal.

8. The method as set forth in claim 7, wherein at least one wideband analog-to-digital converter provides input to the dynamical reservoir.

9. The method as set forth in claim 7, wherein the reservoir computer generates outputs by continually mapping reservoir states through a set of distinct linear functions, with one linear function per output.

10. The method as set forth in claim 7, wherein the input signal is denoised in real-time using a constraint that covers a range of electromagnetic and acoustic signals of interest.

11. The method as set forth in claim 7, wherein an online gradient descent algorithm is used in determining the prediction of the input signal.

12. The method as set forth in claim 7, wherein the denoised output is generated from an input signal containing a mixture of waveforms over a 30 Gigahertz (GHz)) bandwidth.

13. A computer program product for signal denoising, the computer program product comprising:

computer-readable instructions stored on a non-transitory computer-readable medium that are executable by a computer having one or more processors for causing the processor to perform operations of;

generating delay-embedded mixture signals from an input signal comprising a mixture of one or more source signals;

mapping, with a reservoir computer, the delay-embedded mixture signals to reservoir states of a dynamical reservoir having output layer weights, wherein the reservoir computer is a recurrent neural network;

for each input signal, determining a prediction of the input signal using a function of the input signal, current values of the output layer weights, and current and past values of the reservoir states;

determining a prediction error between the prediction of the input signal and the input signal;

for each input signal, iteratively tuning the output layer weights based on the prediction error;

for each input signal, iteratively updating the reservoir states, the prediction error, and the output layer weights; and

generating a denoised output of each input signal.

14. The computer program product as set forth in claim 13, wherein at least one wideband analog-to-digital converter provides input to the dynamical reservoir.

15. The computer program product as set forth in claim 13, wherein the reservoir computer generates outputs by continually mapping reservoir states through a set of distinct linear functions, with one linear function per output.

16. The computer program product as set forth in claim 13, wherein the input signal is denoised in real-time using a constraint that covers a range of electromagnetic and acoustic signals of interest.

17. The computer program product as set forth in claim 13, wherein an online gradient descent algorithm is used in determining the prediction of the input signal.