Learning and deployment of adaptive wireless communications

1 . A method of information transmission through a communication channel, the method comprising:

obtaining, at an encoder device, first information for transmission;

generating, by the encoder device, a first communications signal based on encoding the first information using an encoder machine learning network that is trained by determining a rate of change of an objective function relative to variations in the encoder machine learning network upon processing one or more signals;

causing transmission, by the encoder device, the first communications signal through the communication channel using at least one transmitter;

in response to the transmission of the first communications signal through the communication channel, receiving, by the encoder device, feedback information that indicates at least one of (i) a measure of distance between the first information and second information generated by a decoder device as a reconstruction of the first information, or (ii) channel state information regarding the communication channel,

wherein the second information is generated by the decoder device as the reconstruction of the first information upon reception of a second communications signal corresponding to the first communications signal altered by transmission through the communication channel; and

updating, by the encoder device, one or more parameters of the encoder machine learning network based on the feedback information.

2 . The method of claim 1 , wherein the feedback information comprises channel state information (CSI), and wherein updating one or more parameters of the encoder machine learning network comprises:

determining, based on the channel state, one or more of communications cost, latency, or channel capacity; and

modifying the encoder machine learning network in accordance with the determining.

3 . The method of claim 1 , wherein the encoder machine learning network is trained to learn a plurality of basis functions for communicating over the communication channel, each basis function of the plurality of basis functions corresponding to a different value of channel response, wherein encoding the first information using the encoder machine learning network comprises:

encoding the first information using a first basis function corresponding to a first channel response associated with the communication channel.

4 . The method of claim 3 , wherein the feedback information comprises channel state information (CSI), and wherein updating one or more parameters of the encoder machine learning network comprises one of:

adjusting one or more parameters of the encoder machine learning network for the first basis function based on the CSI, or

applying, to the encoder machine learning network, a different second basis function corresponding to a second channel response associated with the communication channel, the second channel response determined based on the CSI.

5 . The method of claim 1 , wherein updating one or more parameters of the encoder machine learning network based on the feedback information comprises:

determining a channel mode of a plurality of channel modes associated with the communication channel based on the feedback information; and

updating one or more parameters of the encoder machine learning network corresponding to the channel mode of the communication channel.

6 . The method of claim 5 , wherein the plurality of channel modes corresponds to a plurality of channel conditions that comprise one or more of noise level, signal to noise ratio (SNR), delay spread, or time scale of channel variations.

7 . The method of claim 1 , wherein generating the first communications signal based on encoding the first information using the encoder machine learning network comprises:

generating one or more lookup tables based on training the encoder machine learning network, the one or more lookup tables corresponding to different encoding mappings;

accessing, by the encoder, a lookup table of the one or more lookup tables; and

encoding the first information to the first communications signal using the accessed lookup table.

8 . The method of claim 1 , wherein the communication channel comprises one of a radio communication channel, a cellular communication channel, a satellite communication channel, an acoustic communication channel, an optical communication channel, a Wi-Fi channel, or a Bluetooth channel.

9 . The method of claim 1 , wherein generating the second information by the decoder device as the reconstruction of the first information upon reception of a second communications signal corresponding to the first communications signal altered by transmission through the communication channel comprises:

obtaining, at the decoder device, the second communications signal; and

generating, by the decoder device, the second information based on decoding the second communications signal using a decoder machine learning network.

10 . The method of claim 9 , wherein the encoder machine learning network and the decoder machine learning network are jointly trained as an auto-encoder to learn communication over one or more communication channels, and

wherein the auto-encoder comprises at least one channel-modeling layer representing effects of the one or more communication channels on transmitted waveforms.

11 . The method of claim 10 , wherein the at least one channel-modeling layer represents one of (i) additive Gaussian thermal noise in the one or more communication channels, (ii) delay spread caused by time-varying effects of the one or more communication channels, (iii) phase noise caused by transmission and reception over the one or more communication channels, (iv) offsets in phase, frequency, or timing caused by transmission and reception over the one or more communication channels, or (v) unknown time and rate of arrival.

12 . The method of claim 1 , wherein the encoder machine learning network is trained to perform one or more signal processing functions, comprising filtering, modulation, demodulation, equalization, analog-to-digital conversion, digital-to-analog conversion, mapping, or error correction.

13 . The method of claim 1 , wherein the encoder machine learning network comprises an artificial neural network, and wherein updating the encoder machine learning network comprises one or more of:

updating encoding network weights in one or more neural network layers of the encoder machine learning network,

updating network connectivity in one or more neural network layers of the encoder machine learning network,

selecting one or more machine learning models for the encoder machine learning network, or

selecting a particular network architecture for the encoder machine learning network.

14 . The method of claim 1 , wherein the measure of distance between the first information and the second information comprises at least one of

cross-entropy,

mean squared error (MSE),

clipped MSE,

a geometric distance metric, or

an aggregate loss expression combining a plurality of geometric distance metric, entropy-based distance metric, or other classes of distance metrics.

15 . A system comprising:

one or more processors; and

memory storing instructions that, when executed, cause the one or more processors to perform operations comprising:

obtaining, at an encoder device, first information for transmission;

generating, by the encoder device, a first communications signal based on encoding the first information using an encoder machine learning network that is trained by determining a rate of change of an objective function relative to variations in the encoder machine learning network upon processing one or more signals;

causing transmission, by the encoder device, the first communications signal through the communication channel using at least one transmitter;

in response to the transmission of the first communications signal through the communication channel, receiving, by the encoder device, feedback information that indicates at least one of (i) a measure of distance between the first information and second information generated by a decoder device as a reconstruction of the first information, or (ii) channel state information regarding the communication channel,

wherein the second information is generated by the decoder device as the reconstruction of the first information upon reception of a second communications signal corresponding to the first communications signal altered by transmission through the communication channel; and

updating, by the encoder device, one or more parameters of the encoder machine learning network based on the feedback information.

16 . The system of claim 15 , wherein the feedback information comprises channel state information (CSI), and wherein updating one or more parameters of the encoder machine learning network comprises:

determining, based on the channel state, one or more of communications cost, latency, or channel capacity; and

modifying the encoder machine learning network in accordance with the determining.

17 . The system of claim 15 , wherein the encoder machine learning network is trained to learn a plurality of basis functions for communicating over the communication channel, each basis function of the plurality of basis functions corresponding to a different value of channel response, wherein encoding the first information using the encoder machine learning network comprises:

encoding the first information using a first basis function corresponding to a first channel response associated with the communication channel.

18 . The system of claim 17 , wherein the feedback information comprises channel state information (CSI), and wherein updating one or more parameters of the encoder machine learning network comprises one of:

adjusting one or more parameters of the encoder machine learning network for the first basis function based on the CSI, or

applying, to the encoder machine learning network, a different second basis function corresponding to a second channel response associated with the communication channel, the second channel response determined based on the CSI.

19 . One or more non-transitory computer-readable media storing instructions that, when executed, cause one or more processors to perform operations comprising:

obtaining, at an encoder device, first information for transmission;

generating, by the encoder device, a first communications signal based on encoding the first information using an encoder machine learning network that is trained by determining a rate of change of an objective function relative to variations in the encoder machine learning network upon processing one or more signals;

causing transmission, by the encoder device, the first communications signal through the communication channel using at least one transmitter;

in response to the transmission of the first communications signal through the communication channel, receiving, by the encoder device, feedback information that indicates at least one of (i) a measure of distance between the first information and second information generated by a decoder device as a reconstruction of the first information, or (ii) channel state information regarding the communication channel,

wherein the second information is generated by the decoder device as the reconstruction of the first information upon reception of a second communications signal corresponding to the first communications signal altered by transmission through the communication channel; and

updating, by the encoder device, one or more parameters of the encoder machine learning network based on the feedback information.

20 . The one or more non-transitory computer-readable media of claim 19 , wherein the feedback information comprises channel state information (CSI), and wherein updating one or more parameters of the encoder machine learning network comprises:

determining, based on the channel state, one or more of communications cost, latency, or channel capacity; and

modifying the encoder machine learning network in accordance with the determining.

21 . The one or more non-transitory computer-readable media of claim 19 , wherein the encoder machine learning network is trained to learn a plurality of basis functions for communicating over the communication channel, each basis function of the plurality of basis functions corresponding to a different value of channel response, wherein encoding the first information using the encoder machine learning network comprises:

encoding the first information using a first basis function corresponding to a first channel response associated with the communication channel.

22 . The one or more non-transitory computer-readable media of claim 21 , wherein the feedback information comprises channel state information (CSI), and wherein updating one or more parameters of the encoder machine learning network comprises one of:

adjusting one or more parameters of the encoder machine learning network for the first basis function based on the CSI, or

applying, to the encoder machine learning network, a different second basis function corresponding to a second channel response associated with the communication channel, the second channel response determined based on the CSI.