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 to process a received communication signal using at least one machine-learning network, the method comprising:
obtaining, at a first device, a received communication signal that includes data information, wherein one or more elements of the data information each correspond to a particular time and a particular frequency in a time-frequency spectrum, the received communication signal corresponding to a transmitted communication signal having been modified by transmission over a communications channel, the transmitted communication signal generated by a second device that is communicably coupled to the first device;
generating, by the first device, data extracted from the received communication signal, the extracted data corresponds to both time and frequency in the time-frequency spectrum;
processing, by the first device, the data extracted from the received communication signal using a machine-learning network that is trained to process communication signals;
in response to processing the data extracted from the received communication signal using the machine-learning network, obtaining, by the first device from the machine-learning network, an output result corresponding to the data extracted from the received communication signal, wherein the output result represents at least one of (i) an estimate of the communications channel communicably connecting the first device and the second device or (ii) an estimate of symbols transmitted in the received communication signal transmitted through the communications channel; and
updating, by the first device, the machine-learning network based on the output result and the received communication signal.
2. The method of claim 1 , wherein the received communication signal includes one or more of a physical uplink shared channel (PUSCH) or physical downlink shared channel (PDSCH).
3. The method of claim 1 , wherein the first device includes multiple antennas or receives signals simultaneously from two or more users.
4. The method of claim 1 , wherein the first device includes a baseband unit and radio unit, wherein the baseband unit performs one or more operations of high-PHY processing and the radio unit performs one or more operations of low-PHY processing.
5. The method of claim 4 , wherein the radio unit of the first device performs analog-to-digital conversion of the received communication signal.
6. The method of claim 4 , wherein the machine-learning network is deployed in the baseband unit of the first device.
7. The method of claim 4 , wherein the machine-learning network is deployed in the radio unit of the first device.
8. The method of claim 1 , wherein the received communication signal includes one or more transmitted symbols represented by a combination of amplitude and phase variations and the first device compares the one or more transmitted symbols with one or more constellation points of a quadrature amplitude modulation (QAM) scheme.
9. The method of claim 1 , comprising:
generating, using the received communication signal, ground truth data.
10. The method of claim 9 , wherein generating, using the received communication signal, the ground truth data comprises:
generating, using reference signals of the received communication signal, the ground truth data.
11. The method of claim 9 , wherein generating, using the received communication signal, the ground truth data comprises:
computing one or more of an error vector magnitude (EVM), bit error rate (BER), frame error rate (FER), code block error rate (BLER), or soft log-likelihood (LLR) error.
12. The method of claim 9 , further comprising:
computing, by the first device, an error term by comparing the output result and the generated ground truth data; and
updating, by the first device, the machine-learning network based on the error term.
13. The method of claim 12 , wherein the generated ground truth data include values of one or more of:
equalized data symbols, bit values or soft log-likelihood (LLR) values, channel estimates determined from one or more of a process of generating the data information, a decision feedback process, or pilot subcarriers, or an out-of-band communication.
14. The method 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 values of one or more weights in the machine-learning network; and
performing an optimization process using the rate of change to update the values of the one or more weights in the machine-learning network.
15. The method of claim 14 , wherein the optimization process comprises one or more of gradient descent, stochastic gradient descent (SGD), or forms of gradient descent with additional optimization terms,
wherein the forms of gradient descent with additional optimization terms include one or more of Adaptive Moment Estimation (Adam), rectified Adam (RAdam), weight decay Adam (AdamW), or Lookahead neural network optimization.
16. The method of claim 14 , wherein the optimization process involves minimizing a loss value between predicted and actual values of subcarriers, bits, soft log-likelihood (LLR) values, or channel responses.
17. The method of claim 1 , wherein the machine-learning network performs operations corresponding to one or more of estimation, interpolation, equalization of pilot symbols, or equalization of data symbols.
18. The method of claim 1 , wherein the machine-learning network is at least partially trained prior to processing the extracted data.
19. The method of claim 1 , wherein the machine-learning network is partially trained by transmitting a second data signal that includes second data information.
20. The method of claim 1 , comprising:
training the machine-learning network using one or more simulated communications channel including 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, statistical channel models, time-delay line (TDL) 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.
21. The method of claim 1 , wherein the output result obtained from the machine-learning network is one of a channel response of the communications channel, an inverse channel response of the communications channel, values of the data information prior to modification by transmission over the communications channel, or bit or log-likelihood bit values of bits encoded by symbols over the communications channel.
22. The method of claim 1 , wherein the machine-learning network includes at least one of a convolutional layer or a fully connected layer.
23. The method of claim 1 , wherein the data information corresponds to data information modulated by the second device using a modulator corresponding to an orthogonal frequency-division multiplexing (OFDM) system.
24. The method of claim 23 , 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), Discrete Fourier Transform-spread-OFDM (DFT-s-OFDM), or elements of other variants of OFDM.
25. The method of claim 1 , wherein the output result includes a prediction by the machine-learning network corresponding to the received communication signal transmitted through the communications channel.
26. The method of claim 1 , wherein generating the extracted data based at least on the received communication signal comprises:
generating a set of one or more subcarriers based at least on the received communication signal.
27. The method of claim 1 , wherein processing the data extracted from the received communication signal using the machine-learning network that is trained to process communication signals comprises:
processing the data extracted from the received communication signal using the machine-learning network that is trained to estimate at least one of (i) the communications channel or (ii) symbols transmitted through the communications channel.
28. A non-transitory computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:
obtaining, at a first device, a received communication signal that includes data information, wherein one or more elements of the data information each correspond to a particular time and a particular frequency in a time-frequency spectrum, the received communication signal corresponding to a transmitted communication signal having been modified by transmission over a communications channel, the transmitted communication signal generated by a second device that is communicably coupled to the first device;
generating, by the first device, data extracted from the received communication signal, the extracted data corresponds to both time and frequency in the time-frequency spectrum;
processing, by the first device, the data extracted from the received communication signal using a machine-learning network that is trained to process communication signals;
in response to processing the data extracted from the received communication signal using the machine-learning network, obtaining, by the first device from the machine-learning network, an output result corresponding to the data extracted from the received communication signal, wherein the output result represents at least one of (i) an estimate of the communications channel communicably connecting the first device and the second device or (ii) an estimate of symbols transmitted in the received communication signal transmitted through the communications channel; and
updating, by the first device, the machine-learning network based on the output result and the received communication signal.
29. 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, cause the one or more processors to perform operations comprising:
obtaining, at a first device, a received communication signal that includes data information, wherein one or more elements of the data information each correspond to a particular time and a particular frequency in a time-frequency spectrum, the received communication signal corresponding to a transmitted communication signal having been modified by transmission over a communications channel, the transmitted communication signal generated by a second device that is communicably coupled to the first device;
generating, by the first device, data extracted from the received communication signal, the extracted data corresponds to both time and frequency in the time-frequency spectrum;
processing, by the first device, the data extracted from the received communication signal using a machine-learning network that is trained to process communication signals;
in response to processing the data extracted from the received communication signal using the machine-learning network, obtaining, by the first device from the machine-learning network, an output result corresponding to the data extracted from the received communication signal, wherein the output result represents at least one of (i) an estimate of the communications channel communicably connecting the first device and the second device or (ii) an estimate of symbols transmitted in the received communication signal transmitted through the communications channel; and
updating, by the first device, the machine-learning network based on the output result and the received communication signal.