Decoding of error correction codes based on reverse diffusion

Patent metadata

Field	Value
Publication number	US-12294387-B2
Application number	US-202318223121-A
Country	US
Kind code	B2
Filing date	Jul 18, 2023
Priority date	Jul 18, 2022
Publication date	May 6, 2025
Grant date	May 6, 2025

Abstract

Disclosed herein are systems and method for training neural network based decoders for decoding error correction codes, comprising obtaining a plurality of training samples comprising one or more codewords encoded using an error correction code and transmitted over a transmission channel where the training samples are subject to gradual interference over a plurality of time steps and associate the encoded codeword(s) with an interference level and a parity check syndrome at each of the plurality of time steps, using the training samples to train a neural network based decoder to decode codewords encoded using an error correction code by (1) estimating a multiplicative interference included in the encoded codeword(s) based on reverse diffusion applied to the encoded codeword(s) across the time steps, (2) computing an additive interference included in the encoded codewords based on the multiplicative interference, and (3) recovering the codeword(s) by removing the additive interference.

First claim

What is claimed is:

1. A method of training a neural network based decoder for decoding error correction codes, comprising:
using at least one processor for:
obtaining a plurality of training samples comprising at least one codeword encoded using an error correction code and transmitted over a transmission channel, the plurality of training samples are subject to gradual interference over a plurality of time steps and associate the at least one encoded codeword with an interference level and a parity check syndrome at each of the plurality of time steps;
using the plurality of training samples to train a neural network based decoder to decode codewords encoded using an error correction code by:
estimating a multiplicative interference included in the at least one encoded codeword based on reverse diffusion applied to the at least one encoded codeword across the plurality of time steps,
computing an additive interference included in the at least one encoded codeword based on the estimated multiplicative interference, and
recovering the at least one codeword by removing the additive interference; and

outputting the trained neural network based decoder for decoding at least one codeword encoded using an error correction code.

2. The method of claim 1 , wherein a distribution of the plurality of time steps over time is selected randomly.

3. The method of claim 1 , further comprising optimizing a distribution of the plurality of time steps over time by applying an iterative process to identify an optimal time step size which minimizes the parity check syndrome for the recovered at least one codeword.

4. The method of claim 3 , further comprising applying grid search to restrict a search space for selecting the distribution of the plurality of time steps over time.

5. The method of claim 1 , wherein the neural network based decoder is implemented using at least one transformer neural network conditioned for the error correction code according to a number of parity errors detected in the at least one codeword, the transformer neural network is conditioned by employing a multi-dimension one hot decoding to a Hadamard product of an initial embedding created for the bits of the at least one codeword.

6. The method of claim 1 , wherein the at least one encoded codeword encodes the zero codeword.

7. A system for training a neural network based decoder for decoding error correction codes, comprising:
at least one processor configured to execute a code, the code comprising:
code instructions to obtain obtaining a plurality of training samples comprising at least one codeword encoded using an error correction code and transmitted over a transmission channel, the plurality of training samples are subject to gradual interference over a plurality of time steps and associate the at least one encoded codeword with an interference level and a parity check syndrome at each of the plurality of time steps;
code instructions to use the plurality of training samples to train a neural network based decoder to decode codewords encoded using an error correction code by:
estimating a multiplicative interference included in the at least one encoded codeword based on reverse diffusion applied to the at least one encoded codeword across the plurality of time steps,
computing an additive interference included in the at least one encoded codeword based on the estimated multiplicative interference, and
recovering the at least one codeword by removing the additive interference; and

code instructions to output the trained neural network based decoder for decoding at least one codeword encoded using an error correction code.

8. A method of using a neural network based decoder trained for decoding error correction codes, comprising:
using at least one processor for:
receiving at least one codeword encoded using an error correction code and transmitted over a transmission channel;
recovering the at least one encoded codeword by applying to it at least one neural network based decoder trained to decode codewords encoded using the error correction code; and
outputting the at least one recovered codeword;

wherein the at least one neural network based decoder is trained using a plurality of training samples, the plurality of training samples comprise at least one codeword encoded using the error correction code and transmitted over a transmission channel, the plurality of training samples are subject to gradual interference over a plurality of time steps and associate the at least one encoded codeword with an interference level and a parity check syndrome at each of the plurality of time steps, the at least one neural network based decoder is trained to decode the at least one encoded codeword by:
estimating a multiplicative interference included in the at least one encoded codeword based on reverse diffusion applied to the at least one encoded codeword across the plurality of time steps,
computing an additive interference included in the at least one encoded codeword based on the estimated multiplicative interference, and
recovering the at least one codeword by removing the additive interference.

Assignees

• Univ Ramot

Inventors

• Choukroun Yoni
• Wolf Lior

Classifications

• H03M13/6597 Primary

  Implementations using analogue techniques for coding or decoding, e.g. analogue Viterbi decoder · CPC title

• H03M13/1108

  Hard decision decoding, e.g. bit flipping, modified or weighted bit flipping · CPC title

• H04L1/0045

  Arrangements at the receiver end · CPC title

• G06N3/084

  Backpropagation, e.g. using gradient descent · CPC title

• H03M13/03

  Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words · CPC title