Processing communications signals using a machine-learning network
Methods, systems, and apparatus, including computer programs encoded on computer-storage media, for processing communications signals using a machine-learning network are disclosed. In some implementations, pilot and data information are generated for a data signal. The data signal is generated using a modulator for orthogonal frequency-division multiplexing (OFDM) systems. The data signal is transmitted through a communications channel to obtain modified pilot and data information. The modified pilot and data information are processed using a machine-learning network. A prediction corresponding to the data signal transmitted through the communications channel is obtained from the machine-learning network. The prediction is compared to a set of ground truths and updates, based on a corresponding error term, are applied to the machine-learning network.
1. A method performed by at least one processor to train at least one machine-learning network to process a received communication signal, the method comprising:
generating one or more of pilot and data information for a data signal, wherein one or more elements of the pilot and data information each correspond to a particular time and a particular frequency in a time-frequency spectrum;
generating the data signal by modulating the pilot and data information using a modulator for an orthogonal frequency-division multiplexing (OFDM) system;
transmitting the data signal through a communications channel to obtain modified pilot and data information;
processing the modified pilot and data information using a machine-learning network;
in response to the processing using the machine-learning network, obtaining, from the machine-learning network, a prediction corresponding to the data signal transmitted through the communications channel;
computing an error term by comparing the prediction to a set of ground truths; and
updating the machine-learning network based on the error term.
2. The method of claim 1 , wherein the machine-learning network performs operations corresponding to pilot estimation, interpolation, and equalization.
3. The method of claim 1 , wherein the communications channel is a simulated channel that includes one or more of an Additive White Gaussian Noise (AWGN) or Rayleigh fading channel model, International Telecommunication Union (ITU) or 3 rd Generation Partnership Project (3GPP) fading channel models, emulated radio emissions, propagation models, ray tracing within simulated geometry or an environment to produce channel effects, or a second machine-learning network trained to approximate measurements over a real channel.
4. The method of claim 1 , wherein the communications channel is a real communications channel between a first device and a second device, and wherein transmitting the data signal through the communications channel comprises transmitting the data signal from the first device to the second device and obtaining the modified pilot and data information comprising a version of the data signal received by the second device.
5. The method of claim 1 , wherein the prediction obtained from the machine-learning network is one of a channel response of the communications channel, an inverse channel response of the communication channel, or values of the pilot and data information prior to transmitting the data signal through the communications channel.
6. The method of claim 1 , wherein the set of ground truths are values of equalized data symbols or channel estimates determined from one or more of a process of generating the pilot and data information, a decision feedback process, pilot subcarriers, or an out-of-band communication.
7. The method of claim 1 , wherein updating the machine-learning network based on the error term comprises:
determining, based on a loss function, a rate of change of one or more weight values within the machine-learning network; and
performing an optimization process using the rate of change to update the one or more weight values within the machine-learning network.
8. The method of claim 7 , wherein the optimization process comprises one or more of gradient descent, stochastic gradient descent (SGD), Adam, RAdam, AdamW, or Lookahead neural network optimization.
9. The method of claim 7 , wherein the optimization process involves minimizing a loss value between predicted and actual values of subcarriers or channel responses.
10. The method of claim 1 , wherein the machine-learning network is a fully convolutional neural network or a partially convolutional neural network.
11. The method of claim 1 , wherein the OFDM system includes one or more elements of cyclic-prefix orthogonal frequency division multiplexing (CP-OFDM), single carrier frequency division multiplexing (SCFDM), filter bank multicarrier (FBMC), or elements of other variants of OFDM.
12. A non-transitory computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:
generating one or more of pilot and data information for a data signal, wherein one or more elements of the pilot and data information each correspond to a particular time and a particular frequency in a time-frequency spectrum;
generating the data signal by modulating the pilot and data information using a modulator for an orthogonal frequency-division multiplexing (OFDM) system;
transmitting the data signal through a communications channel to obtain modified pilot and data information;
processing the modified pilot and data information using a machine-learning network;
in response to the processing using the machine-learning network, obtaining, from the machine-learning network, a prediction corresponding to the data signal transmitted through the communications channel;
computing an error term by comparing the prediction to a set of ground truths; and
updating the machine-learning network based on the error term.
13. The non-transitory computer-readable medium of claim 12 , wherein the machine-learning network performs operations corresponding to pilot estimation, interpolation, and equalization.
14. The non-transitory computer-readable medium of claim 12 , wherein the communications channel is a simulated channel that includes one or more of an Additive White Gaussian Noise (AWGN) or Rayleigh fading channel model, International Telecommunication Union (ITU) or 3rd Generation Partnership Project (3GPP) fading channel models, emulated radio emissions, propagation models, ray tracing within simulated geometry or an environment to produce channel effects, or a second machine-learning network trained to approximate measurements over a real channel.
15. The non-transitory computer-readable medium of claim 12 , wherein the communications channel is a real communications channel between a first device and a second device, and wherein transmitting the data signal through the communications channel comprises transmitting the data signal from the first device to the second device and obtaining the modified pilot and data information comprising a version of the data signal received by the second device.
16. The non-transitory computer-readable medium of claim 12 , wherein the prediction obtained from the machine-learning network is one of a channel response of the communications channel, an inverse channel response of the communication channel, or values of the pilot and data information prior to transmitting the data signal through the communications channel.
17. The non-transitory computer-readable medium of claim 12 , wherein the set of ground truths are values of equalized data symbols or channel estimates determined from one or more of a process of generating the pilot and data information, a decision feedback process, pilot subcarriers, or an out-of-band communication.
18. The non-transitory computer-readable medium of claim 12 , wherein updating the machine-learning network based on the error term comprises:
determining, based on a loss function, a rate of change of one or more weight values within the machine-learning network; and
performing an optimization process using the rate of change to update the one or more weight values within the machine-learning network.
19. The non-transitory computer-readable medium of claim 18 , wherein the optimization process comprises one or more of gradient descent, stochastic gradient descent (SGD), Adam, RAdam, AdamW, or Lookahead neural network optimization.
20. The non-transitory computer-readable medium of claim 12 , wherein the machine-learning network is a fully convolutional neural network or a partially convolutional neural network.
21. The non-transitory computer-readable medium of claim 12 , wherein the OFDM system includes one or more elements of cyclic-prefix orthogonal frequency division multiplexing (CP-OFDM), single carrier frequency division multiplexing (SCFDM), filter bank multicarrier (FBMC), or elements of other variants of OFDM.
22. A system, comprising:
one or more processors; and
machine-readable media interoperably coupled with the one or more processors and storing one or more instructions that, when executed by the one or more processors, perform operations comprising:
generating one or more of pilot and data information for a data signal, wherein one or more elements of the pilot and data information each correspond to a particular time and a particular frequency in a time-frequency spectrum;
generating the data signal by modulating the pilot and data information using a modulator for an orthogonal frequency-division multiplexing (OFDM) system;
transmitting the data signal through a communications channel to obtain modified pilot and data information;
processing the modified pilot and data information using a machine-learning network;
in response to the processing using the machine-learning network, obtaining, from the machine-learning network, a prediction corresponding to the data signal transmitted through the communications channel;
computing an error term by comparing the prediction to a set of ground truths; and
updating the machine-learning network based on the error term.
23. The system of claim 22 , wherein the machine-learning network performs operations corresponding to pilot estimation, interpolation, and equalization.
24. The system of claim 22 , wherein the communications channel is a simulated channel that includes one or more of an Additive White Gaussian Noise (AWGN) or Rayleigh fading channel model, International Telecommunication Union (ITU) or 3rd Generation Partnership Project (3GPP) fading channel models, emulated radio emissions, propagation models, ray tracing within simulated geometry or an environment to produce channel effects, or a second machine-learning network trained to approximate measurements over a real channel.
25. The system of claim 22 , wherein the communications channel is a real communications channel between a first device and a second device, and wherein transmitting the data signal through the communications channel comprises transmitting the data signal from the first device to the second device and obtaining the modified pilot and data information comprising a version of the data signal received by the second device.
26. The system of claim 22 , wherein the prediction obtained from the machine-learning network is one of a channel response of the communications channel, an inverse channel response of the communication channel, or values of the pilot and data information prior to transmitting the data signal through the communications channel.
27. The system of claim 22 , wherein the set of ground truths are values of equalized data symbols or channel estimates determined from one or more of a process of generating the pilot and data information, a decision feedback process, pilot subcarriers, or an out-of-band communication.
28. The system of claim 22 , wherein updating the machine-learning network based on the error term comprises:
determining, based on a loss function, a rate of change of one or more weight values within the machine-learning network; and
performing an optimization process using the rate of change to update the one or more weight values within the machine-learning network.
29. The system of claim 28 , wherein the optimization process comprises one or more of gradient descent, stochastic gradient descent (SGD), Adam, RAdam, AdamW, or Lookahead neural network optimization.
30. The system of claim 22 , wherein the machine-learning network is a fully convolutional neural network or a partially convolutional neural network and wherein the OFDM system includes one or more elements of cyclic-prefix orthogonal frequency division multiplexing (CP-OFDM), single carrier frequency division multiplexing (SCFDM), filter bank multicarrier (FBMC), or elements of other variants of OFDM.