Neuromorphic processor for wideband signal analysis

• US 10,162,378 B1
• Filed: 06/23/2017
• Issued: 12/25/2018
• 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: Rao, Shankar R., Petre, Peter, Martin, Charles E.
• Primary Examiner(s): Liu, Shuwang
• Assistant Examiner(s): Hailegiorgis, Fitwi
• Application Number:
  US15/631,307
• Time in Patent Office: 550 Days
• Field of Search: 375350, 342377, 381 66, 367129
• CPC Class Codes: G06F 1/08 Clock generators with changeable or programmable clock frequency
• CPC Class Codes: G06F 18/2134 based on separation criteria, e.g. independent component analysis
• 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 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: G06N 3/065 Analogue means

Abstract

Described is a neuromorphic processor for signal denoising and separation. The neuromorphic processor generates delay-embedded mixture signals from an input mixture of pulses. Using a reservoir computer, the delay-embedded mixture signals are mapped to reservoir states of a dynamical reservoir having output layer weights. The output layer weights are adapted based on short-time linear prediction, and a denoised output of the mixture of input signals us generated. The denoised output is filtered through a set of adaptable finite impulse response (FIR) filters to extract a set of separated narrowband pulses.

Claims

1. A system for wideband signal analysis, the system comprising:

a neuromorphic processor and a non-transitory computer-readable medium having executable instructions encoded thereon such that when executed, the neuromorphic processor performs operations of;

generating delay-embedded mixture signals from a mixture of pulses from a single input;

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

tuning the output layer weights based on short-time linear prediction;

following adaptation of the output layer weights, generating a denoised output of the mixture of input signals; and

extracting separated and denoised pulses by filtering the denoised output through a set of adaptable finite impulse response (FIR) filters,

wherein an adaptation mechanism limits how close the filters in the set of adaptable FIR filters are to one another.

2. The system as set forth in claim 1, wherein tuning the output layer weights comprises iteratively adapting the output layer weights further based on a difference between a predicted signal and an actual signal.

3. The system as set forth in claim 1, wherein filtering the denoised output further comprises adapting a center frequency of each FIR filter using a combination of gradient descent and gradient-free optimization.

4. The system as set forth in claim 1, wherein each FIR filter extracts a unique narrowband pulse.

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

6. The system as set forth in claim 1, wherein each FIR filter in the set of adaptable FIR filters is adapted simultaneously.

7. The system as set forth in claim 1, wherein the one or more processors further perform an operation of determining when a particular FIR filter in the set of adaptable FIR filters is in a process of extracting a source signal.

8. A computer program product for wideband signal analysis, 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;

generating delay-embedded mixture signals from a mixture of pulses from a single input;

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

tuning the output layer weights based on short-time linear prediction;

following adaptation of the output layer weights, generating a denoised output of the mixture of input signals; and

extracting separated and denoised pulses by filtering the denoised output through a set of adaptable finite impulse response (FIR) filters,

wherein an adaptation mechanism limits how close the filters in the set of adaptable FIR filters are to one another.

9. The computer program product as set forth in claim 8, wherein tuning the output layer weights comprises iteratively adapting the output layer weights further based on a difference between a predicted signal and an actual signal.

10. The computer program product as set forth in claim 8, wherein filtering the denoised output further comprises adapting a center frequency of each FIR filter using a combination of gradient descent and gradient-free optimization.

11. The computer program product as set forth in claim 8, wherein each FIR filter extracts a unique narrowband pulse.

12. The computer program product as set forth in claim 8, wherein the input mixture of pulses is denoised and separated in real-time using a constraint that covers a range of electromagnetic and acoustic signals of interest.

13. The computer program product as set forth in claim 8, wherein each FIR filter in the set of adaptable FIR filters is adapted simultaneously.

14. The computer program product as set forth in claim 8, further comprising instructions for causing the one or more processors to further perform an operation of determining when a particular FIR filter in the set of adaptable FIR filters is in a process of extracting a source signal.

15. A computer implemented method for wideband signal analysis, 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;

generating delay-embedded mixture signals from a mixture of pulses from a single input;

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

tuning the output layer weights based on short-time linear prediction;

following adaptation of the output layer weights, generating a denoised output of the mixture of input signals; and

extracting separated and denoised pulses by filtering the denoised output through a set of adaptable finite impulse response (FIR) filters,

wherein an adaptation mechanism limits how close the filters in the set of adaptable FIR filters are to one another.

16. The method as set forth in claim 15, wherein tuning the output layer weights comprises iteratively adapting the output layer weights further based on a difference between a predicted signal and an actual signal.

17. The method as set forth in claim 15, wherein filtering the denoised output further comprises adapting a center frequency of each FIR filter using a combination of gradient descent and gradient-free optimization.

18. The method as set forth in claim 15, wherein each FIR filter extracts a unique narrowband pulse.

19. The method as set forth in claim 15, wherein the input mixture of pulses is denoised and separated in real-time using a constraint that covers a range of electromagnetic and acoustic signals of interest.

20. The method as set forth in claim 15, wherein each FIR filter in the set of adaptable FIR filters is adapted simultaneously.

21. The method as set forth in claim 15, wherein the one or more processors further perform an operation of determining when a particular FIR filter in the set of adaptable FIR filters is in a process of extracting a source signal.