Radio signal identification, identification system learning, and identifier deployment

• US 10,643,153 B2
• Filed: 04/24/2018
• Issued: 05/05/2020
• Est. Priority Date: 04/24/2017
• Status: Active Grant

• Current Assignee: Virginia Tech Intellectual Properties, Inc.
• Original Assignee: Virginia Tech Intellectual Properties, Inc.
• Inventors: O'Shea, Timothy James
• Primary Examiner(s): Nilsson, Eric
• Application Number:
  US15/961,465
• Publication Number:
  US 20180308013A1
• Time in Patent Office: 742 Days
• CPC Class Codes: G06F 18/217 Validation; Performance evaluation; Active pattern learning techniques
• CPC Class Codes: G06F 2218/00 Aspects of pattern recognition specially adapted for signal processing
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/006 based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/048 Activation functions
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 3/082 modifying the architecture, e.g. adding, deleting or silencing nodes or connections
• CPC Class Codes: G06N 3/086 using evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: H04B 17/30 of propagation channels
• CPC Class Codes: H04W 24/08 Testing, supervising or monitoring using real traffic

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training and deploying machine-learned identification of radio frequency (RF) signals. One of the methods includes: determining an RF signal configured to be transmitted through an RF band of a communication medium; determining first classification information that is associated with the RF signal, and that includes a representation of a characteristic of the RF signal or a characteristic of an environment in which the RF signal is communicated; using at least one machine-learning network to process the RF signal and generate second classification information as a prediction of the first classification information; calculating a measure of distance between (i) the second classification information that was generated by the at least one machine-learning network, and (ii) the first classification information associated with the RF signal; and updating the at least one machine-learning network based on the measure of distance.

Claims

1. A method of training at least one machine-learning network to classify radio frequency (RF) signals, the method performed by at least one processor executing instructions stored on at least one computer memory coupled to the at least one processor, the method comprising:

determining an RF signal that is configured to be transmitted through an RF band of a communication medium;

extracting one or more features of the RF signal using prior knowledge about the RF signal;

determining first classification information associated with the RF signal based on the RF signal and the extracted one or more features of the RF signal, the first classification information comprising a representation of at least one of a characteristic of the RF signal or a characteristic of an environment in which the RF signal is communicated;

using at least one machine-learning network to process the RF signal and generate second classification information as a prediction of the first classification information;

calculating a measure of distance between (i) the second classification information that was generated by the at least one machine-learning network as the prediction of the first classification information, and (ii) the first classification information that was associated with the RF signal; and

updating the at least one machine-learning network based on the measure of distance between the second classification information and the first classification information.

2. The method of claim 1, wherein the measure of distance between the second classification information and the first classification information comprises at least one of (i) a cross-entropy between the second classification information and the first classification information, or (ii) a geometric distance metric between the second classification information and the first classification information.

3. The method of claim 1, wherein the first classification information and the second classification information indicate at least one of (i) a type of modulation utilized to transmit the RF signal, (ii) a type of traffic, protocol, or standard that is carried by or that is utilized to transmit the RF signal, (iii) a type of hardware apparatus that transmits or receives the RF signal, (iv) a type of RF emission or a type of RF emission pattern occurring in the communication medium, or (v) a presence of one or more types of communications signals.

4. The method of claim 1, wherein using the at least one machine-learning network to process the RF signal and generate the second classification information comprises:

determining a plurality of basis signals;

processing the RF signal using the plurality of basis signals to generate a plurality of basis coefficients for the RF signal; and

generating the second classification information based on the plurality of basis coefficients of the RF signal.

5. The method of claim 4, wherein updating the at least one machine-learning network based on the measure of distance between the second classification information and the first classification information comprises:

updating the plurality of basis signals that are utilized to process RF signals, based on the measure of distance between the second classification information and the first classification information.

6. The method of claim 1, wherein updating the at least one machine-learning network based on the measure of distance between the second classification information and the first classification information comprises:

determining an objective function that comprises the measure of distance between the second classification information and the first classification information;

calculating a rate of change of the objective function relative to variations in the at least one machine-learning network;

selecting, based on the calculated rate of change of the objective function, a variation for the at least one machine-learning network; and

updating the at least one machine-learning network based on a selected variation for the at least one machine-learning network.

7. The method of claim 6, wherein the objective function comprises a weighted combination of the measure of distance between the second classification information and the first classification information with at least one of:

(i) a computational throughput or delay achieved by the at least one machine-learning network in classifying RF signals, (ii) a computational dwell time associated with the at least one machine-learning network in classifying RF signals, or (iii) a computational duty cycle associated with the at least one machine-learning network in classifying RF signals.

8. The method of claim 1, wherein updating the at least one machine-learning network comprises:

updating at least one network weight or at least one connection in one or more layers of at least one artificial neural network that forms part of the at least one machine-learning network.

9. The method of claim 1, wherein using the at least one machine-learning network to process the RF signal and generate the second classification information further comprises:

determining, based on the at least one machine-learning network, a first time scale for processing the RF signal; and

using the at least one machine-learning network to processes the RF signal based on the first time scale,

wherein updating the at least one machine-learning network further comprises updating the first time scale to a second time scale based on the measure of distance between the second classification information and the first classification information.

10. The method of claim 1, wherein the RF signal comprises at least one of an acoustic RF signal, an optical RF signal, or other analog RF signal that is configured to be transmitted and received by man-made RF communication systems.

11. The method of claim 1, wherein the prior knowledge about the RF signal comprises features about the RF signal obtained using one or more of expert hand crafted techniques, or automated feature-extracting techniques.

12. The method of claim 1, wherein the extracted one or more features of the RF signal are trained based on results of classification performance.

13. The method of claim 1, further comprising:

determining a type of the RF signal using the extracted one or more features of the RF signal.

14. A method of deploying at least one machine-learning network that has been trained to classify radio frequency (RF) signals, the method performed by at least one processor executing instructions stored on at least one computer memory coupled to the at least one processor, the method comprising:

determining at least one machine-learning network has been trained to classify RF signals configured to be transmitted through an RF band of a communication medium, wherein the at least one trained machine-learning network has been trained based on a measure of distance between (i) first classification information associated with an RF training signal, wherein the first classification information was generated based on the RF training signal and one or more features of the RF training signal extracted using prior knowledge about the RF training signal, and (ii) second classification information for the RF training signal that was generated by the at least one trained machine-learning network as a prediction of the first classification information;

setting at least one parameter of an RF receiver based on the at least one trained machine-learning network;

using the RF receiver to receive an analog RF waveform from an RF spectrum of the communication medium, and to process the analog RF waveform to generate a discrete-time representation of the analog RF waveform as a received RF signal; and

using the at least one trained machine-learning network to process the received RF signal and generate predicted RF signal classification information,

wherein the predicted RF signal classification information comprises a representation of at least one of a characteristic of the received RF signal or a characteristic of an environment in which the received RF signal was communicated.

15. The method of claim 14, wherein setting the at least one parameter of the RF receiver based on the at least one machine-learning network comprises:

setting at least one of a tuning parameter, a filtering parameter, a scheduling of operations, or a channelization parameter of the RF receiver according to the at least one trained machine-learning network, or according to updates to the at least one trained machine-learning network.

16. The method of claim 14, wherein using the at least one trained machine-learning network to process the received RF signal and generate the predicted RF signal classification information comprises:

determining a plurality of basis signals;

processing the received RF signal using the plurality of basis signals to generate a plurality of basis coefficients of the received RF signal; and

perform classification on the plurality of basis coefficients of the received RF signal; and

output a result of the classification as the predicted RF signal classification information.

17. The method of claim 14, wherein the predicted RF signal classification information indicates at least one of (i) a type of modulation utilized to transmit the received RF signal, (ii) a type of traffic, protocol, or standard that is carried by or that is utilized to transmit the received RF signal, (iii) a type of hardware apparatus that transmits or receives the received RF signal, (iv) a type of RF emission or a type of RF emission pattern occurring in the communication medium, or (v) a presence of one or more types of communications signals.

18. The method of claim 14, wherein the at least one trained machine-learning network comprises at least one artificial neural network that has been trained using an objective function that comprises the measure of distance between (i) the first classification information associated with the RF training signal, and (ii) the second classification information for the RF training signal that was generated by the at least one trained machine-learning network as the prediction of the first classification information.

19. A system comprising:

at least one processor; and

at least one computer memory that is operably connectable to the at least one processor and that has stored thereon instructions which, when executed by the at least one processor, cause the at least one processor to perform operations comprising;

determining an RF signal that is configured to be transmitted through an RF band of a communication medium;

extracting one or more features of the RF signal using prior knowledge about the RF signal;

determining first classification information associated with the RF signal based on the RF signal and the extracted one or more features of the RF signal, the first classification information comprising a representation of at least one of a characteristic of the RF signal or a characteristic of an environment in which the RF signal is communicated;

using at least one machine-learning network to process the RF signal and generate second classification information as a prediction of the first classification information;

calculating a measure of distance between (i) the second classification information that was generated by the at least one machine-learning network as the prediction of the first classification information, and (ii) the first classification information that was associated with the RF signal; and

updating the at least one machine-learning network based on the measure of distance between the second classification information and the first classification information.

20. The system of claim 19, wherein the measure of distance between the second classification information and the first classification information comprises at least one of (i) a cross-entropy between the second classification information and the first classification information, or (ii) a geometric distance metric between the second classification information and the first classification information.

21. The system of claim 19, wherein the first classification information and the second classification information indicate at least one of (i) a type of modulation utilized to transmit the RF signal, (ii) a type of traffic, protocol, or standard that is carried by or that is utilized to transmit the RF signal, (iii) a type of hardware apparatus that transmits or receives the RF signal, (iv) a type of RF emission or a type of RF emission pattern occurring in the communication medium, or (v) a presence of one or more types of communications signals.

22. The system of claim 19, wherein using the at least one machine-learning network to process the RF signal and generate the second classification information comprises:

determining a plurality of basis signals;

processing the RF signal using the plurality of basis signals to generate a plurality of basis coefficients for the RF signal; and

generating the second classification information based on the plurality of basis coefficients of the RF signal.

23. The system of claim 22, wherein updating the at least one machine-learning network based on the measure of distance between the second classification information and the first classification information comprises:

updating the plurality of basis signals that are utilized to process RF signals, based on the measure of distance between the second classification information and the first classification information.

24. The system of claim 19, wherein updating the at least one machine-learning network based on the measure of distance between the second classification information and the first classification information comprises:

determining an objective function that comprises the measure of distance between the second classification information and the first classification information;

calculating a rate of change of the objective function relative to variations in the at least one machine-learning network;

selecting, based on the calculated rate of change of the objective function, a variation for the at least one machine-learning network; and

updating the at least one machine-learning network based on a selected variation for the at least one machine-learning network.

25. The system of claim 24, wherein the objective function comprises a weighted combination of the measure of distance between the second classification information and the first classification information with at least one of:

(i) a computational throughput or delay achieved by the at least one machine-learning network in classifying RF signals, (ii) a computational dwell time associated with the at least one machine-learning network in classifying RF signals, or (iii) a computational duty cycle associated with the at least one machine-learning network in classifying RF signals.

26. The system of claim 19, wherein updating the at least one machine-learning network comprises:

updating at least one network weight or at least one connection in one or more layers of at least one artificial neural network that forms part of the at least one machine-learning network.

27. The system of claim 19, wherein using the at least one machine-learning network to process the RF signal and generate the second classification information further comprises:

determining, based on the at least one machine-learning network, a first time scale for processing the RF signal; and

using the at least one machine-learning network to processes the RF signal based on the first time scale,

wherein updating the at least one machine-learning network further comprises updating the first time scale to a second time scale based on the measure of distance between the second classification information and the first classification information.

28. The system of claim 19, wherein the RF signal comprises at least one of an acoustic RF signal, an optical RF signal, or other analog RF signal that is configured to be transmitted and received by man-made RF communication systems.

29. The system of claim 19, wherein the prior knowledge about the RF signal comprises features about the RF signal obtained using one or more of expert hand crafted techniques, or automated feature-extracting techniques.

30. The system of claim 19, wherein the extracted one or more features of the RF signal are trained based on results of classification performance.

31. The system of claim 19, wherein the operations further comprise:

determining a type of the RF signal using the extracted one or more features of the RF signal.

32. A system comprising:

at least one processor; and

at least one computer memory that is operably connectable to the at least one processor and that has stored thereon instructions which, when executed by the at least one processor, cause the at least one processor to perform operations comprising;

determining at least one machine-learning network has been trained to classify RF signals configured to be transmitted through an RF band of a communication medium, wherein the at least one trained machine-learning network has been trained based on a measure of distance between (i) first classification information for an RF training signal, wherein the first classification information was generated based on the RF training signal and one or more features of the RF training signal extracted using prior knowledge about the RF training signal, and (ii) second classification information for the RF training signal that was generated by the at least one trained machine-learning network as a prediction of the first classification information;

setting at least one parameter of an RF receiver based on the at least one trained machine-learning network;

using the RF receiver to receive an analog RF waveform from an RF spectrum of the communication medium, and to process the analog RF waveform to generate a discrete-time representation of the analog RF waveform as a received RF signal; and

using the at least one trained machine-learning network to process the received RF signal and generate predicted RF signal classification information,

wherein the predicted RF signal classification information comprises a representation of at least one of a characteristic of the received RF signal or a characteristic of an environment in which the received RF signal was communicated.

33. The system of claim 32, wherein setting the at least one parameter of the RF receiver based on the at least one machine-learning network comprises:

setting at least one of a tuning parameter, a filtering parameter, or a channelization parameter of the RF receiver according to the at least one trained machine-learning network, or according to updates to the at least one trained machine-learning network.

34. The system of claim 32, wherein using the at least one trained machine-learning network to process the received RF signal and generate the predicted RF signal classification information comprises:

determining a plurality of basis signals;

processing the received RF signal using the plurality of basis signals to generate a plurality of basis coefficients of the received RF signal; and

perform classification on the plurality of basis coefficients of the received RF signal; and

output a result of the classification as the predicted RF signal classification information.

35. The system of claim 32, wherein the predicted RF signal classification information indicates at least one of (i) a type of modulation utilized to transmit the received RF signal, (ii) a type of traffic, protocol, or standard that is carried by or that is utilized to transmit the received RF signal, (iii) a type of hardware apparatus that transmits or receives the received RF signal, (iv) a type of RF emission or a type of RF emission pattern occurring in the communication medium, or (v) a presence of one or more types of communications signals.

36. The system of claim 32, wherein the at least one trained machine-learning network comprises at least one artificial neural network that has been trained using an objective function that comprises the measure of distance between (i) the first classification information associated with the RF training signal, and (ii) the second classification information for the RF training signal that was generated by the at least one trained machine-learning network as the prediction of the first classification information.