Neural network-based decoder architecture for 5G

Patent metadata

Field	Value
Publication number	US-2025286656-A1
Application number	US-202519073480-A
Country	US
Kind code	A1
Filing date	Mar 7, 2025
Priority date	Mar 7, 2024
Publication date	Sep 11, 2025
Grant date	—

Abstract

The invention discloses systems and methods for decoding sequence based feedback signaling in wireless communication. The system comprises a plurality of User Equipment (UE) and the base station configured for establishing wireless communication links. The base station includes a multi-label neural network classifier configured to decode the signals. The method for decoding the wireless communication includes the steps of: providing the multi-label neural network classifier, generating training datasets and training the network using same equipment as used for communication or any other system. The method involves receiving the input PUCCH Format 0 signal, and predicting the NUE phase rotations a using the neural network. The α values are used to map back with the UCI-specific cyclic shift mcs for each UE. The system and method outperform conventional DFT-based decoders across all SNR ranges and dopplers and are robust enough to identify false transmissions, thus eliminating the need for thresholds.

First claim

We claim:

1 . A system for feedback signaling in wireless communication, comprising:
at least one User Equipment (UE) configured to receive at least one downlink transmission that requires acknowledgement; and a base station, configured to receive the Uplink Control Information (UCI) carried by the PUCCH Format 0 for decoding, the base station comprising a multi-label neural network classifier.

2 . The system of claim 1 , wherein the base station is configured for downlink transmission which is to be received by a UE.

3 . The system of claim 1 , the multi-label neural network classifier comprising:
an input layer containing 12 or more neurons configured to receive 12 or more real or complex sequence inputs or 24 neurons configured to receive 24 real sequence inputs and an additional optional input for metadata; one or more dense layers, containing 64 or more neurons; and an output layer containing 12 or more neurons representing up to 12 phase rotation (α) values of the multiplexed users.

4 . A method for decoding UCI wireless communication, comprising:
providing a multi-label neural network classifier to serve as a generalized PUCCH Format 0 decoder; generating a dataset for training the decoding model; initializing the decoding model; training the decoding model; testing the decoding model; receiving the input PUCCH Format 0 signal and associated metadata of a number of users N UE ; feeding the signal to the decoding model; predicting N UE phase rotation values corresponding to at least some; and obtaining α values.

5 . The method of claim 4 , wherein providing the multi-label neural network classifier comprises providing:
an input layer containing 12 or more neurons configured to receive 12 or more real or complex sequence inputs or 24 neurons configured to receive 24 real sequence inputs and an additional optional input for metadata; one or more dense layers, containing 64 or more neurons; and an output layer containing 12 or more neurons representing up to 12 phase rotation (α) values of the multiplexed users.

6 . The method of claim 4 , wherein providing the multi-label neural network classifier comprises providing:
an input layer containing 25 neurons configured to receive 24 real sequence inputs and the metadata as 25 th input; three dense layers, each containing 256 neurons; and an output layer containing 12 neurons configured to receive up to 12 phase rotation (α) values of the multiplexed users.

7 . The method of claim 4 , wherein generating the dataset for training includes one or both of:
collecting and storing waveform samples generated in a simulation software environment; or collecting and storing real time received signal waveform samples from a live communication link.

8 . The method of claim 6 , wherein training the multi-label neural network classifier comprises running multiple epochs of training wherein each epoch is 1 pass of the entire training dataset through the decoding model.

9 . The method of claim 8 , wherein training the multi-label neural network classifier comprises running 150 or more epochs of training wherein each epoch is 1 pass of the entire training dataset through the decoding model.

10 . The method of claim 4 , wherein training the decoding model includes:
running multiple epochs of training wherein each epoch is 1 pass of the entire training dataset through the decoding model; using a summation of waveforms resulting from multiplexed transmission by multiple transmitting entities; using selected data conforming to a single (example: median) or multiple value(s) of SNR; and dropping out randomly selected neurons and their connections with a certain probability during training.

11 . The method of claim 10 , wherein using a summation of waveforms includes the scenario where no entity transmits a signal.

12 . The method of claim 4 , comprising performing the training on a same hardware on which the method is deployed, or on a different system.

13 . The method of claim 4 , wherein testing the decoding model includes providing upper bound value of the number of multiplexed users as the metadata input which is offset from the true value.

14 . The method of claim 5 , comprising pre-processing the signal prior to feeding.

15 . The method of claim 4 , wherein predicting comprises determining the N UE phase rotation values α 0 , α 1 . . . αN UE −1 applied to the base sequence for classification.

16 . The method of claim 15 , wherein the obtaining includes obtaining either a single α value, multiple α values, or zero α values.

17 . The method of claim 1 , wherein receiving at the base station includes:
Hybrid Automatic Repeat Request (HARQ) acknowledgements for prior downlink transmissions; Scheduling Request (SR) for the subsequent allocation of uplink transmission resources; and Channel State Information (CSI) reports including channel quality metrics that facilitate link adaptation, precoding, and downlink resource allocation.

18 . The method of claim 7 , generating the dataset for training includes operating the transmitter or the receiver or both in a specific data collection mode to generate specific training sequences.

19 . The method of claim 14 , wherein the preprocessing comprises using one of Fourier Transform, correlation, scaling or absolute value squared to the input signal.

Assignees

• Indian Inst Tech Madras

Inventors

• Ganti Radhakrishna
• Sattianarayanin Jeeva Keshav
• Yerrapragada Anil Kumar

Classifications

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/084

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