System for parallelized cognitive signal denoising

• US 10,404,299 B1
• Filed: 03/02/2018
• Issued: 09/03/2019
• Est. Priority Date: 03/07/2016
• Status: Active Grant

• Current Assignee: HRL Laboratories LLC (The Boeing Co.)
• Original Assignee: HRL Laboratories LLC (The Boeing Co.)
• Inventors: Petre, Peter, Fong, Bryan H., Rao, Shankar R.
• Primary Examiner(s): Fan, Chieh M
• Assistant Examiner(s): Hailegiorgis, Fitwi
• Application Number:
  US15/910,922
• Time in Patent Office: 550 Days
• Field of Search: 375295, 375296
• CPC Class Codes: G06F 2218/04 Denoising
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: H04B 1/10 Means associated with receiver for limiting or suppressing noise or interference

Abstract

Described is a cognitive signal processor (CSP) for signal denoising. In operation, the CSP receives a noisy signal as a time-series of data points from a mixture of both noise and one or more desired waveform signals. The noisy signal is linearly mapped to reservoir states of a dynamical reservoir. A high-dimensional state-space representation is then generated of the noisy signal by combining the noisy signal with the reservoir states. A delay-embedded state signal is generated from the reservoir states. The reservoir states are denoised by removing noise from each reservoir state signal, resulting in a real-time denoised spectrogram of the noisy signal. A denoised waveform signal is generated combining the denoised reservoir states. Additionally, the signal denoising process is implemented in software or digital hardware by converting the state-space representation of the dynamical reservoir to a system of delay difference equations and then applying a linear basis approximation.

Claims

1. A cognitive signal processor for signal denoising, the cognitive signal processor comprising:

one or more processors configured to perform operations of;

receiving a noisy signal, the noisy signal being a time-series of data points from a mixture of both noise and one or more desired waveform signals;

linearly mapping the noisy signal to reservoir states of a dynamical reservoir;

generating a high-dimensional state-space representation of the noisy signal by combining the noisy signal with the reservoir states;

generating a delay-embedded state signal from the reservoir states;

denoising the reservoir states by removing noise from each reservoir state signal, resulting in a real-time denoised spectrogram of the noisy signal for further processing; and

generating a denoised waveform signal by combining the denoised reservoir states; and

wherein the signal denoising process is implemented in software or digital hardware by converting the state-space representation of the dynamical reservoir to a system of delay difference equations and then applying a linear basis approximation.

2. The cognitive signal processor as set forth in claim 1, wherein the dynamical reservoir includes, for each reservoir state, an error function that is used to create the denoised reservoir state, the error function being a difference between a current reservoir state signal and a delayed denoised reservoir state signal.

3. The cognitive signal processor as set forth in claim 2, wherein the error function updates weights used in generating the denoised reservoir states, the weights being updated with the error function based on gradient descent.

4. The cognitive signal processor as set forth in claim 1, further comprising an operation of displaying at least one of the denoised waveform signal and the real-time denoised spectrogram on a Display Device.

5. The cognitive signal processor as set forth in claim 1, wherein the dynamical reservoir is a recurrent neural network with a plurality of nodes.

6. The cognitive signal processor as set forth in claim 1, wherein the dynamical reservoir includes a connectivity matrix having a block diagonal structure.

7. The cognitive signal processor as set forth in claim 1, wherein the dynamical reservoir includes an output layer having a set of summing nodes, each summing node receiving a weighted output from each delay-embedded state signal and summing the weighted output to generate the denoised waveform signal.

8. A cognitive signal processor for parallelized signal denoising, the cognitive signal processor comprising:

one or more processors configured to perform operations of;

receiving a noisy signal, the noisy signal being a time-series of data points from a mixture of both noise and one or more desired waveform signals;

linearly mapping the noisy signal to reservoir states of a dynamical reservoir;

generating a high-dimensional state-space representation of the noisy signal by combining the noisy signal with the reservoir states;

generating a delay-embedded state signal from the reservoir states;

denoising the reservoir states by removing the noise from each reservoir state signal independently in parallel, resulting in a real-time denoised spectrogram of the noisy signal for further processing; and

generating a denoised waveform signal based on a weighted combination of the denoised reservoir states; and

displaying least one of the denoised waveform signal and the real-time denoised spectrogram on a Display Device;

wherein the signal denoising process is implemented in software or digital hardware by converting the state-space representation of the dynamical reservoir to a system of delay difference equations and then applying a linear basis approximation.

9. The cognitive signal processor as set forth in claim 8, further comprising an operation of displaying at least one of the denoised waveform signal and the real-time denoised spectrogram on a Display Device.

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

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

receiving a noisy signal, the noisy signal being a time-series of data points from a mixture of both noise and one or more desired waveform signals;

linearly mapping the noisy signal to reservoir states of a dynamical reservoir;

generating a high-dimensional state-space representation of the noisy signal by combining the noisy signal with the reservoir states;

generating a delay-embedded state signal from the reservoir states;

denoising the reservoir states by removing noise from each reservoir state signal, resulting in a real-time denoised spectrogram of the noisy signal for further processing;

generating a denoised waveform signal by combining the denoised reservoir states; and

wherein the signal denoising process is implemented in software or digital hardware by converting the state-space representation of the dynamical reservoir to a system of delay difference equations and then applying a linear basis approximation.

11. The computer program product as set forth in claim 10, wherein the dynamical reservoir includes, for each reservoir state, an error function that is used to create the denoised reservoir state, the error function being a difference between a current reservoir state signal and a delayed denoised reservoir state signal.

12. The computer program product as set forth in claim 11, wherein the error function updates weights used in generating the denoised reservoir states, the weights being updated with the error function based on gradient descent.

13. The computer program product as set forth in claim 10, further comprising an operation of displaying at least one of the denoised waveform signal and the real-time denoised spectrogram on a Display Device.

14. The computer program product as set forth in claim 10, wherein the dynamical reservoir is a recurrent neural network with a plurality of nodes.

15. The computer program product as set forth in claim 10, wherein the dynamical reservoir includes a connectivity matrix having a block diagonal structure.

16. The computer program product as set forth in claim 10, wherein the dynamical reservoir includes an output layer having a set of summing nodes, each summing node receiving a weighted output from each delay-embedded state signal and summing the weighted output to generate the denoised waveform signal.

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

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

receiving a noisy signal, the noisy signal being a time-series of data points from a mixture of both noise and one or more desired waveform signals;

linearly mapping the noisy signal to reservoir states of a dynamical reservoir;

generating a high-dimensional state-space representation of the noisy signal by combining the noisy signal with the reservoir states;

generating a delay-embedded state signal from the reservoir states;

denoising the reservoir states by removing noise from each reservoir state signal, resulting in a real-time denoised spectrogram of the noisy signal for further processing;

generating a denoised waveform signal by combining the denoised reservoir states; and

wherein the signal denoising process is implemented in software or digital hardware by converting the state-space representation of the dynamical reservoir to a system of delay difference equations and then applying a linear basis approximation.

18. The method as set forth in claim 17, wherein the dynamical reservoir includes, for each reservoir state, an error function that is used to create the denoised reservoir state, the error function being a difference between a current reservoir state signal and a delayed denoised reservoir state signal.

19. The method as set forth in claim 18, wherein the error function updates weights used in generating the denoised reservoir states, the weights being updated with the error function based on gradient descent.

20. The method as set forth in claim 17, further comprising an operation of displaying at least one of the denoised waveform signal and the real-time denoised spectrogram on a Display Device.

21. The method as set forth in claim 17, wherein the dynamical reservoir is a recurrent neural network with a plurality of nodes.

22. The method as set forth in claim 17, wherein the dynamical reservoir includes a connectivity matrix having a block diagonal structure.

23. The method as set forth in claim 17, wherein the dynamical reservoir includes an output layer having a set of summing nodes, each summing node receiving a weighted output from each delay-embedded state signal and summing the weighted output to generate the denoised waveform signal.