Learning and deploying compression of radio signals

1. A method of deploying at least one machine-learning network that has been trained to determine compact representations of radio frequency (RF) signals,

determining an encoder machine-learning network and a decoder machine-learning network that have been trained to determine compact representations of RF signals, wherein the encoder machine-learning network and the decoder machine-learning network are trained dependent on one another;

receiving a first RF signal wirelessly at a radio receiver of a radio head system, wherein the radio head system is distinct from a baseband processing device communicably coupled to the radio head system;

using the radio head system to use the encoder machine-learning network to process the first RF signal to generate a first compressed signal that is a compressed version of the first RF signal;

transmitting the first compressed signal over a back-end communications channel to a radio processing system distinct from the radio head system, wherein the radio processing system comprises the baseband processing device;

receiving, over the back-end communications channel, at the radio processing system, based on transmission of the first compressed signal, a second compressed signal that comprises the first compressed signal or an alteration thereof, the second compressed signal being a compressed version of the first RF signal;

using the radio processing system to use the decoder machine-learning network to process the second compressed signal to generate a second RF signal that represents a reconstruction of the first RF signal, the second RF signal being an uncompressed version of the second compressed signal; and

using the radio processing system, determining one or more characteristics of the first RF signal based on the second RF signal.

2. The method of claim 1 , wherein the radio head system and the radio processing system are at different locations, and wherein the back-end communications channel comprises a backhaul channel.

3. The method of claim 1 , comprising:

using the radio head system, converting the first RF signal from an analog format to a digital format,

wherein using the encoder machine-learning network to process the first RF signal comprises using the encoder machine-learning network to process the first RF signal in the digital format.

4. The method of claim 1 , comprising:

storing the first compressed signal on a storage device communicably coupled to the radio head system.

5. The method of claim 1 , comprising:

using the radio processing system, converting the second RF signal from a digital format to an analog format; and

transmitting the second RF signal in the analog format, or an alteration thereof, wirelessly from one or more antennas.

6. The method of claim 1 , comprising:

determining a measure of distance between the second RF signal and the first RF signal; and

based on the measure of distance between the second RF signal and the first RF signal exceeding a threshold, determining an occurrence of an error or an anomaly.

7. The method of claim 6 , wherein determining the occurrence of the error or the anomaly comprises identifying at least one of a snooping event or an error in the first RF signal.

8. The method of claim 7 , wherein determining the occurrence of the error or the anomaly comprises identifying the snooping event.

9. The method of claim 1 , wherein at least one of the encoding machine-learning network or the decoding machine-learning network comprises at least one of a deep dense neural network (DNN) or a convolutional neural network (CNN) comprising a series of parametric multiplications, additions, and non-linearities.

10. The method of claim 1 , wherein using the encoder machine-learning network to process the first RF signal and using the decoder machine-learning network to process the second compressed signal comprise:

determining a plurality of basis signals;

using the plurality of basis signals to process the first RF signal by the encoder machine-learning network and generate the first compressed signal; and

using the plurality of basis signals to process the second compressed signal by the decoder machine-learning network and generate the second RF signal.

11. The method of claim 1 , wherein the encoder machine-learning network and the decoder machine-learning network are jointly trained as an auto-encoder to learn compact representations of RF signals, and

wherein the auto-encoder comprises at least one regularization layer that comprises at least one of: weight regularization on network layer weights, activity regularization on network layer activations, or stochastic impairments on network layer activations or network layer weights.

12. The method of claim 1 , wherein the first compressed signal has a reduced bit-precision compared to the first RF signal.

13. The method of claim 1 , wherein the second RF signal has a higher bit-precision than the second compressed signal.

14. The method of claim 1 , wherein at least one of the encoder machine-learning network or the decoder machine-learning network is trained based on at least one of (i) a measure of distance between a training RF signal and a reconstruction of the training RF signal, or (ii) a measure of compression in compressing the training RF signal.

15. The method of claim 14 , wherein the measure of compression indicates at least one of (i) a relative measure of compression between the training RF signal and a compressed form of the training RF signal or (ii) an absolute measure of complexity in the compressed form of the training RF signal.

16. The method of claim 14 , wherein the measure of distance between the training RF signal and the reconstruction of the training RF signal comprises a cross-entropy between the training RF signal and the reconstruction of the training RF signal.

17. The method of claim 1 , wherein the one or more characteristics comprise at least one of content of the first RF signal or an emitter of the first RF signal.

18. A system comprising:

one or more antennas of a radio head system, wherein the radio head system is distinct from a baseband processing device communicably coupled to the radio head system;

at least one processor of the radio head system, the at least one processor communicably coupled to the one or more antennas; and

at least one computer memory of the radio head system, the at least one computer memory coupled to the at least one processor and having stored thereon instructions which, when executed by the at least one processor, cause the at least one processor to perform operations comprising:

determining an encoder machine-learning network that has been trained to determine compact representations of RF signals, wherein the encoder machine-learning network is trained jointly with a decoder machine-learning network, the encoder machine-learning network and the decoder machine-learning network trained dependent on one another;

receiving, from the one or more antennas, a first RF signal to be compressed, wherein the first RF signal, or an alteration thereof, is received wirelessly at the one or more antennas;

using the encoder machine-learning network to process the first RF signal to generate a first compressed signal that is a compressed version of the first RF signal; and

transmitting the first compressed signal over a back-end communications channel to a radio processing system distinct from the radio head system, wherein the radio processing system comprises the baseband processing device

wherein the encoder machine-learning network is trained based on at least one of (i) a measure of distance between a training RF signal and a reconstruction of the training RF signal, or (ii) a measure of compression in compressing the training RF signal, the training comprising:

updating the encoder machine-learning network based on at least one of (i) the measure of distance between the training RF signal and the reconstruction of the training RF signal, or (ii) the measure of compression in compressing the training RF signal.

19. The system of claim 18 , wherein the operations comprise:

storing, for analysis at a subsequent time, the first compressed signal on a storage device communicably coupled to the at least one processor.

20. The system of claim 18 , wherein the operations comprise:

converting the first RF signal from an analog format to a digital format,

wherein using the encoder machine-learning network to process the first RF signal comprises using the encoder machine-learning network to process the first RF signal in the digital format.

21. The system of claim 18 , wherein the measure of distance between the training RF signal and the reconstruction of the training RF signal comprises a cross-entropy between the training RF signal and the reconstruction of the training RF signal.

22. The system of claim 18 , wherein the operations comprise:

receiving the first compressed signal at the radio processing system; and

using the decoder machine-learning network at the radio processing system to process the first compressed signal to generate a second RF signal that represents a reconstruction of the first RF signal.

23. The system of claim 22 , wherein the encoder machine-learning network and the decoder machine-learning network are jointly trained as an auto-encoder to learn compact representations of RF signals, and

wherein the auto-encoder comprises at least one regularization layer that comprises at least one of: weight regularization on network layer weights, activity regularization on network layer activations, or stochastic impairments on network layer activations or network layer weights.

24. The system of claim 22 , wherein using the encoder machine-learning network to process the first RF signal and using the decoder machine-learning network to process the first compressed signal comprise:

determining a plurality of basis signals;

using the plurality of basis signals to process the first RF signal by the encoder machine-learning network and generate the first compressed signal; and

using the plurality of basis signals to process the first compressed signal by the decoder machine-learning network and generate the second RF signal.

25. The system of claim 18 , wherein the encoding machine-learning network comprises at least one of a deep dense neural network (DNN) or a convolutional neural network (CNN) comprising a series of parametric multiplications, additions, and non-linearities.

26. The system of claim 18 , wherein the first compressed signal has a reduced bit-precision compared to the first RF signal.

27. A method comprising:

determining an encoder machine-learning network and a decoder machine-learning network that have been trained to determine compact representations of RF signals, wherein the encoder machine-learning network and the decoder machine-learning network are trained dependent on one another;

receiving a first RF signal wirelessly at a radio receiver of a radio head system;

using the radio head system to use the encoder machine-learning network to process the first RF signal to generate a compressed signal that is a compressed version of the first RF signal;

transmitting the compressed signal from the radio head system to a baseband processing device distinct from the radio head system over a front-end communications bus using a front-end communications protocol;

receiving, over the front-end communications bus, at the baseband processing device, the compressed signal; and

using the baseband processing device to use the decoder machine-learning network to process the compressed signal to generate a second RF signal that represents a reconstruction of the first RF signal, the second RF signal being an uncompressed version of the compressed signal,

wherein at least one of the encoder machine-learning network or the decoder machine-learning network is trained based on a measure of compression in compressing a training RF signal, the training comprising:

updating at least one of the encoder machine-learning network or the decoder machine-learning network based on the measure of compression in compressing the training RF signal.

28. The method of claim 27 , wherein the front-end communications protocol comprises CPRI or VITA-49.