Cognitive denoising of nonstationary signals using time varying reservoir computer

• US 10,712,425 B1
• Filed: 08/24/2018
• Issued: 07/14/2020
• 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: Rao, Shankar R., Petre, Peter, Fong, Bryan H.
• Primary Examiner(s): Hailegiorgis, Fitwi Y
• Application Number:
  US16/112,071
• Time in Patent Office: 690 Days
• Field of Search: 375130, 342377, 704207, 704203
• CPC Class Codes: G01S 13/88 Radar or analogous systems specially adapted for specific applications electromagnetic prospecting or detecting of objects, e.g. near-field detection, G01V3/00
• CPC Class Codes: G01S 7/2813 Means providing a modification of the radiation pattern for cancelling noise, clutter or interfering signals, e.g. side lobe suppression, side lobe blanking, null-steering arrays specially adapted to secondary radar systems G01S13/762; aerials or aerials systems H01Q21/29, H01Q25/00
• CPC Class Codes: G01S 7/354 Extracting wanted echo-signals Doppler systems G01S13/50
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/047 Probabilistic or stochastic networks
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

Described is a system for signal denoising using a cognitive signal processor having a time-varying reservoir. The system receives a noisy input signal of a time-series of data points from a mixture of waveform signals. The noisy input signal is linearly mapped into the time-varying reservoir. A high-dimensional state-space representation of the mixture of waveform signals is generated by combining the noisy input signal with a plurality of reservoir states. The system then generates a denoised signal corresponding to the noisy input signal.

Claims

1. A system for signal denoising using reservoir computing, the system comprising:

a cognitive signal processor having a reservoir computer (RC) and a non-transitory computer-readable medium having executable instructions encoded thereon such that when executed, the cognitive signal processor performs operations of;

receiving a nonstationary, time-varying noisy input signal comprising a time-series of data points from a mixture of waveform signals;

using the RC, linearly mapping the noisy input signal into a time-varying reservoir, wherein the time-varying reservoir is a recurrent neural network;

using the time-varying reservoir, generating a high-dimensional state-space representation of the mixture of waveform signals by combining the noisy input signal with a plurality of reservoir states, wherein each reservoir state corresponds to a response to a time-varying filter in a set of time-varying filters;

applying a phase delay embedding technique to each reservoir state to obtain a history of reservoir state dynamics, resulting in a plurality of delay-embedded states,

wherein the time-varying reservoir is obtained by applying a distinct reservoir state transition matrix for each delay-embedded state; and

generating a denoised signal corresponding to the nonstationary, time-varying noisy input signal.

2. The system as set forth in claim 1, wherein the cognitive signal processor further performs an operation of generating a real-time denoised time-varying spectrogram of the noisy input signal.

3. The system as set forth in claim 1, wherein the cognitive signal processor further performs an operation of adapting the delay-embedded states for producing a predicted input signal a time-step into the future.

4. The system as set forth in claim 1, wherein the cognitive signal processor further performs operations of:

determining an error function between a predicted input signal and the noisy input signal; and

adapting output layer weights of the time-varying reservoir in an iterative process using the error function.

5. The system as set forth in claim 4, wherein the error function is used to adapt the output layer weights via a gradient learning algorithm.

6. The system as set forth in claim 1, wherein each reservoir state transition matrix comprises a block diagonal structure.

7. The system as set forth in claim 1, wherein the nonstationary, time-varying noisy input signal is a chirp signal.

8. The system as set forth in claim 1, wherein the denoised signal is used in an automotive radar system for detection of radar pulses.

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

causing a cognitive signal processor having a reservoir computer to execute instructions encoded on a non-transitory computer-readable medium, such that upon execution, the cognitive signal processor perform operations of;

receiving a nonstationary, time-varying noisy input signal comprising a time-series of data points from a mixture of waveform signals;

using the RC, linearly mapping the noisy input signal into a time-varying reservoir, wherein the time-varying reservoir is a recurrent neural network;

using the time-varying reservoir, generating a high-dimensional state-space representation of the mixture of waveform signals by combining the noisy input signal with a plurality of reservoir states, wherein each reservoir state corresponds to a response to a time-varying filter in a set of time-varying filters;

applying a phase delay embedding technique to each reservoir state to obtain a history of reservoir state dynamics, resulting in a plurality of delay-embedded states

wherein the time-varying reservoir is obtained by applying a distinct reservoir state transition matrix for each delay-embedded state; and

generating a denoised signal corresponding to the nonstationary, time-varying noisy input signal.

10. The method as set forth in claim 9, wherein the cognitive signal processor further performs an operation of generating a real-time denoised time-varying spectrogram of the noisy input signal.

11. The method as set forth in claim 9, wherein the cognitive signal processor further performs an operation of adapting the delay-embedded states for producing a predicted input signal a time-step into the future.

12. The method as set forth in claim 9, wherein the cognitive signal processor further performs operations of:

determining an error function between a predicted input signal and the noisy input signal; and

adapting output layer weights of the time-varying reservoir in an iterative process using the error function.

13. The method as set forth in claim 12, wherein the error function is used to adapt the output layer weights via a gradient learning algorithm.

14. 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 comprising a cognitive signal processor having a reservoir computer for causing the cognitive signal processor to perform operations of;

receiving a nonstationary, time-varying noisy input signal comprising a time-series of data points from a mixture of waveform signals;

using the RC, linearly mapping the noisy input signal into a time-varying reservoir, wherein the time-varying reservoir is a recurrent neural network;

using the time-varying reservoir, generating a high-dimensional state-space representation of the mixture of waveform signals by combining the noisy input signal with a plurality of reservoir states, wherein each reservoir state corresponds to a response to a time-varying filter in a set of time-varying filters;

applying a phase delay embedding technique to each reservoir state to obtain a history of reservoir state dynamics, resulting in a plurality of delay-embedded states

wherein the time-varying reservoir is obtained by applying a distinct reservoir state transition matrix for each delay-embedded state; and

generating a denoised signal corresponding to the nonstationary, time-varying noisy input signal.

15. The computer program product as set forth in claim 14, further comprising instructions for causing the cognitive signal processor to further perform an operation of generating a real-time denoised time-varying spectrogram of the noisy input signal.

16. The computer program product as set forth in claim 14, further comprising instructions for causing the cognitive signal processor to perform an operation of adapting the delay-embedded states for producing a predicted input signal a time-step into the future.

17. The computer program product as set forth in claim 14, further comprising instructions for causing the cognitive signal processor to further perform operations of:

determining an error function between a predicted input signal and the noisy input signal; and

adapting output layer weights of the time-varying reservoir in an iterative process using the error function.

18. The computer program product as set forth in claim 17, wherein the error function is used to adapt the output layer weights via a gradient learning algorithm.