Biologically inspired methods and systems for automatically determining the modulation types of radio signals using stacked de-noising autoencoders

Patent metadata

Field	Value
Publication number	US-10003483-B1
Application number	US-201715586168-A
Country	US
Kind code	B1
Filing date	May 3, 2017
Priority date	May 3, 2017
Publication date	Jun 19, 2018
Grant date	Jun 19, 2018

Abstract

Class types of input signals having unknown class types are automatically classified using a neural network. The neural network learns features associated with a plurality of different observed signals having respective different known class types. The neural network then recognizes features of the input signals having unknown class types that at least partially match at least some of the features associated with the plurality of different observed signals having respective different known class types. The neural network determines probabilities that each of the input signals has each of the known class types based on strengths of the matches between the recognized features of the input signals and the features associated with plurality of different observed signals. The neural network classifies each of the input signals as having one of the respective different known class types based on a highest determined probability.

First claim

What is claimed is:

1. A method for automatically determining class types of input signals having unknown class types, comprising:
a) learning, by a neural network including multiple stacked sparse denoising autoencoders (SSDA) with weighted connections, features associated with a plurality of different observed signals having respective different known class types, wherein step a) comprises adjusting assigned weights of the connections based on the features of the plurality of different observed signals;
b) refining, by a softmax component, the adjusted weights of the connections based on outputs of the SSDA;
c) recognizing, by the SSDA, features of the input signals having unknown class types that at least partially match at least some of the features associated with the plurality of different observed signals having respective different known class types;
d) determining, by the softmax component, probabilities that each of the input signals have each of the known class types based on strengths of matches between recognized features of each of the input signals and the features associated with the plurality of different observed signals; and
e) classifying, by the softmax component, each of the input signals as having one of the respective different known class types based on a highest determined probability for each input signal in a manner that is accurate in noisy environments.

2. The method of claim 1 , wherein the class types are modulation types.

3. The method of claim 1 , wherein adjusting the assigned weights of the connections includes comparing outputs of the SSDA to corresponding inputs and adjusting the assigned weights of the connections automatically based on a difference between the outputs and the corresponding inputs.

4. The method of claim 3 , wherein the assigned weights are repeatedly adjusted to minimize the difference between the outputs and the corresponding inputs.

5. The method of claim 1 , wherein refining the adjusted weights of the connections includes estimating, for each output of the SSDA, a probability that the output has a known class type, determining whether the estimated probability is correct, and repeatedly refining the adjusted weights of the connections until the estimated probability is substantially correct.

6. A system for automatically determining modulation types of input signals having unknown modulation types, comprising:
multiple stacked sparse denoising autoencoders (SSDA) with weighted connections, the SSDA configured to:
during a training phase, learn features associated with a plurality of different observed signals having different respective known modulation types and adjusting assigned weights of the connections based on the features of the plurality of different observed signals; and
during a classification phase, recognize features of the input signals that at least partially match at least some of the features associated with the plurality of different observed signals having different respective known modulation types and produce outputs indicative of strengths of the matches of the recognized features of the input signals with the features associated with the plurality of different observed signals; and

a softmax component configured to:
during the training phase, refine the adjusted weights of the connections based on outputs of the SSDA; and
during the classification phase, determine probabilities that each of the input signals has each of the known modulation types based on outputs of the SSDA and classify each of the input signals as having one of the different respective known modulation types based on a highest determined probability for each input signal in a manner that is accurate in noisy environments.

7. The system of claim 6 , wherein, during the training phase, the SSDA adjusts the assigned weights of the connections by comparing outputs of SSDA to corresponding inputs and adjusting the assigned weights of the connections automatically based on a difference between the outputs and the corresponding inputs.

8. The system of claim 7 , wherein the assigned weights of the connections are repeatedly adjusted until the outputs of the SSDA are substantially the same as the corresponding inputs.

9. The system of claim 6 , wherein the softmax component refines the adjusted weights of the connections by estimating, for each output of the SSDA, a probability that the output has a known modulation type, determining whether the estimated probability is correct, and repeatedly refining the adjusted weights of the connections of the SSDA until the estimated probability is substantially correct.

Assignees

• Spawar Systems Ct Pacific
• Us Navy

Inventors

• Migliori Benjamin J
• Gebhardt Daniel J
• Grady Daniel C
• Zeller-Townson Riley

Classifications

• G06N20/00

  Machine learning · CPC title

• G06N3/049

  Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs · CPC title

• H04B1/0003

  Software-defined radio [SDR] systems, i.e. systems wherein components typically implemented in hardware, e.g. filters or modulators/demodulators, are implented using software, e.g. by involving an AD or DA conversion stage such that at least part of the signal processing is performed in the digital domain (digital baseband systems H04L25/00; digital modulation/demodulation H04L27/00; CDMA H04B1/707; TDMA H04B7/2643; image transmission H04N5/00) · CPC title

• H04L27/0008

  arrangements for allowing a transmitter or receiver to use more than one type of modulation (negotiating modulation type for two-way transmission paths H04L5/1453) · CPC title

• H04L27/14

  Demodulator circuits; Receiver circuits · CPC title