Device and method for reliable classification of wireless signals

1. A communications device, comprising:

a transmitter configured to generate a physical layer signal to be transmitted across a communications channel;

a physical signal modifier configured to modify the physical layer signal prior to transmission as a function of a first set of signal modification parameters to produce a modified physical layer signal;

a machine learning (ML) agent configured to:

parse a feedback signal from a receiver across the communications channel, the feedback signal being a function of the modified physical layer signal;

determine a present tuning status, the present tuning status being a function of 1) the first set of signal modification parameters and 2) the feedback signal;

generate a second set of signal modification parameters based on the present tuning status and a set of stored tuning statuses; and

update the physical signal modifier to implement the second set of signal modification parameters to produce a subsequent modified physical layer signal to be transmitted across the communications channel.

2. The device of claim 1 , wherein the physical signal modifier is a finite impulse response (FIR) filter configured to modify the physical layer signal by modifying a set of FIR taps.

3. The device of claim 1 , wherein the ML agent is configured to parse the feedback signal to determine a predicted class or a softmax of the modified physical layer signal output by a signal classifier at the receiver.

4. The device of claim 3 , wherein the predicted class identifies at least one of the following characteristics of the modified physical layer signal: communications protocol, modulation format, frequency band, communications technology, and encryption scheme.

5. The device of claim 1 , wherein the ML agent is configured to parse the feedback signal to determine a predicted modulation class or a predicted radio frequency (RF) fingerprint of the modified physical layer signal output by a signal classifier at the receiver.

6. The device of claim 1 , wherein the ML agent is configured to parse the feedback signal to determine an error rate output by the receiver, the error rate indicating a rate of error in decoding the modified physical layer signal.

7. The device of claim 1 , wherein the ML agent is configured to generate the second set of signal modification parameters absent an indication of model parameters or weight values implemented by a signal classifier at the receiver.

8. The device of claim 1 , wherein the ML agent is a twin delayed deep deterministic policy gradients (TD3) learning agent.

9. The device of claim 1 , wherein the ML agent is configured to generate at least some of the stored tuning statuses via a training process, the ML agent being configured to incorporate a noise signal into the physical layer signal during the training process.

10. The device of claim 1 , wherein the ML agent is further configured to:

determine a subsequent tuning status, the subsequent tuning status indicating the second set of signal modification parameters; and

update the stored statuses to include the subsequent tuning status.

11. The device of claim 1 , wherein the tuning status is a tuple including values representing the first set of signal modification parameters and the feedback signal.

12. The device of claim 1 , wherein the feedback signal identifies the transmitter based on a signature evident in a received physical layer signal, the signature being associated with hardware of the transmitter.

13. The device of claim 12 , wherein the physical signal modifier is configured to modify the physical layer signal in a manner emphasizing the signature relative to the signature evident in the physical layer signal.

14. The device of claim 12 , wherein the signature is a property of the physical layer signal emergent from one or more imperfections of the hardware of the transmitter device.

15. The device of claim 1 , wherein the communications channel is at least one of a wireless channel, a wired channel, a fiber optic channel, a free space optical channel, and a fluidic channel.

16. The device of claim 1 , wherein the ML agent is a deep reinforcement learning (DRL) agent.

17. A method of communication, comprising:

generating a physical layer signal to be transmitted across a communications channel;

modifying the physical layer signal prior to transmission as a function of a first set of signal modification parameters to produce a modified physical layer signal;

parsing a feedback signal from a receiver across the communications channel, the feedback signal being a function of the modified physical layer signal;

determining a present tuning status, the present tuning status being a function of 1) the first set of signal modification parameters and 2) the feedback signal;

generating a second set of signal modification parameters based on the present tuning status and a set of stored tuning statuses; and

updating a physical signal modifier to implement the second set of signal modification parameters to produce a subsequent modified physical layer signal to be transmitted across the communications channel.

18. The method of claim 17 , wherein the physical signal modifier is a finite impulse response (FIR) filter configured to modify the physical layer signal by modifying a set of FIR taps.

19. The method of claim 18 , wherein the communications channel is at least one of a wireless channel, a wired channel, a fiber optic channel, a free space optical channel, and a fluidic channel.

20. The method of claim 18 , wherein generating the second set of signal modification parameters is via a deep reinforcement learning (DRL) agent.

21. The method of claim 17 , wherein the feedback signal is based on a predicted class or a softmax of the modified physical layer signal output by a signal classifier at the receiver.

22. The method of claim 21 , wherein the predicted class identifies at least one of the following characteristics of the modified physical layer signal: communications protocol, modulation format, frequency band, communications technology, and encryption scheme.

23. The method of claim 17 , wherein the feedback signal is based on a predicted modulation class or a predicted radio frequency (RF) fingerprint of the modified physical layer signal output by a signal classifier at the receiver.

24. The method of claim 17 , wherein the feedback signal is based on an error rate output by the receiver, the error rate indicating a rate of error in decoding the modified physical layer signal.

25. The method of claim 17 , further comprising generating the second set of signal modification parameters absent an indication of model parameters or weight values implemented by a signal classifier at the receiver.

26. The method of claim 17 , wherein generating the second set of signal modification parameters is via a twin delayed deep deterministic policy gradients (TD3) learning agent.

27. The method of claim 17 , further comprising:

generating at least some of the stored tuning statuses via a training process; and

incorporating a noise signal into the physical layer signal during the training process.

28. The method of claim 17 , further comprising:

determining a subsequent tuning status, the subsequent tuning status indicating the second set of signal modification parameters; and

updating the stored statuses to include the subsequent tuning status.

29. The method of claim 17 , wherein the tuning status is a tuple including values representing the first set of signal modification parameters and the feedback signal.

30. The method of claim 17 , wherein the feedback signal identifies the transmitter based on a signature evident in a received physical layer signal, the signature being associated with hardware of the transmitter.

31. The method of claim 30 , further comprising modifying the physical layer signal in a manner emphasizing the signature relative to the signature evident in the physical layer signal.

32. The method of claim 30 , wherein the signature is a property of the physical layer signal emergent from one or more imperfections of the hardware of the transmitter device.