Method for non-linear distortion immune end-to-end learning with autoencoder—OFDM

1. A method for avoiding non-linear distortion in end-to-end learning communication systems, the communication system comprising a transmitter and a receiver, the method comprising:

mapping transmitted information bits to an input of a first neural network;

controlling, by an output of the neural network, parameters of a complementary sequence (CS) encoder, producing an encoded CS;

transmitting the encoded CS through an orthogonal frequency division multiplexing (OFDM) signal;

processing, by Discrete Fourier Transform (DFT), the encoded CS in a frequency domain, to produce a received information signal; and

processing, by a second neural network, the received information signal.

2. The method of claim 1 , wherein the CS encoder comprises an amplitude encoder, a phase encoder, and a shift encoder.

3. The method of claim 2 , wherein mapping the transmitted information bits to an input of a first neural network further comprises:

manually tuning the shift encoder to adjust a position of non-zero elements of the CS;

tuning the amplitude encoder and the phase encoder using the first neural network to produce tuned parameters; and

mapping the information bits to the tuned parameters.

4. The method of claim 2 , wherein, sets for the amplitude encoder and phase encoder are predetermined and offset by the first neural network in order to be able to transmit a large number of information bits.

5. The method of claim 4 , wherein the receiver further comprises a decoder, the method further comprising:

subtracting, by the second neural network, the offsets from the received information signal to produce a remaining information signal; and

decoding, by the decoder, the remaining information signal.

6. The method of claim 1 , wherein the encoded CS is processed by multiple layers at the transmitter, the layers including at least a Golay layer, the method further comprising:

controlling only the amplitude encoder and the phase encoder of the Golay layer; and

forming an autoencoder which captures transmitter, channel, and receiver behaviors.

7. The method of claim 6 , wherein the layers further include a clipping layer configured to limit the amplitude of the information signal.

8. The method of claim 6 , wherein the layers further include a Polar-to-Cartesian layer configured to convert the coordinate system from Polar coordinates to a Cartesian coordinate system.

9. An end-to-end learning communication system for avoiding non-linear distortion, the system comprising:

a transmitter implemented by processing circuitry, the processing circuitry comprising a processor and a memory containing instructions executable by the processor, the processor of the transmitter configured to:

map transmitted information bits to an input of a first neural network;

control, by an output of the neural network, parameters of a complementary sequence (CS) encoder, producing an encoded CS; and

transmit the encoded CS through an orthogonal frequency division multiplexing (OFDM) signal; and

a receiver implemented by processing circuitry, the processing circuitry comprising a processor and a memory containing instructions executable by the processor, the processor of the receiver configured to:

process, by Discrete Fourier Transform (DFT), the encoded CS in a frequency domain, to produce a received information signal; and

process, by a second neural network, the received information signal.

10. The system of claim 9 , wherein the CS encoder comprises an amplitude encoder, a phase encoder, and a shift encoder.

11. The system of claim 10 , wherein mapping the transmitted information bits to an input of a first neural network further comprises:

manually tuning, by the processor of the transmitter, the shift encoder to adjust a position of non-zero elements of the CS;

tuning, by the processor of the transmitter, the amplitude encoder and the phase encoder using the first neural network to produce tuned parameters; and

mapping, by the processor of the transmitter, the information bits to the tuned parameters.

12. The system of claim 10 , wherein the encoded CS is processed by multiple layers at the transmitter, the layers including at least a Golay layer, the processor of the transmitter further configured to:

control only the amplitude encoder and the phase encoder of the Golay layer; and

form an autoencoder which captures transmitter, channel, and receiver behaviors.

13. The system of claim 12 , wherein the layers further include a clipping layer configured to limit the amplitude of the information signal.

14. The system of claim 12 , wherein the layers further include a Polar-to-Cartesian layer configured to convert the coordinate system from Polar coordinates to a Cartesian coordinate system.

15. The system of claim 10 , wherein sets for the amplitude encoder and phase encoder are predetermined and offset by the first neural network in order to be able to transmit a large number of information bits.

16. The system of claim 15 , wherein the processor of the receiver is further configured to:

subtract, by the second neural network, the offsets from the received information signal to produce a remaining information signal; and

decode, by the decoder, the remaining information signal.