Low overhead procedures for two-sided model monitoring

Patent metadata

Field	Value
Publication number	US-12574092-B2
Application number	US-202418900238-A
Country	US
Kind code	B2
Filing date	Sep 27, 2024
Priority date	Sep 29, 2023
Publication date	Mar 10, 2026
Grant date	Mar 10, 2026

Abstract

The disclosure describes a method of evaluating performance of a two-sided model used for performing CSI compression. The method includes receiving, by a network, from a UE, a compressed channel H t , a second eigenvector EV t+1 corresponding to a channel H t+1 , and a first similarity score determined between a first eigenvector EV t and the second eigenvector EV t+1 . A channel H t is reconstructed by processing the compressed channel H t using a decoder part of the two-sided AI/ML model. Eigen decomposition of the channel H t is performed to obtain an eigenvector (EV) t . A second similarity score is determined between the eigenvectors (EV) t and EV t+1 . Alternatively, the first similarity score is determined between an estimated channel H and a codebook based precoder W, and a second similarity score is determined between a reconstructed channel H and the codebook based precoder W.

First claim

We claim:

1 . A method of determining performance of a two-sided artificial intelligence/machine learning (AI/ML) model, the method comprising:
receiving, by a network, from a user equipment (UE), a compressed form of channel H t , a second eigenvector EV t+1 corresponding to a channel H t+1 , and a first similarity score determined between a first eigenvector EV t and the second eigenvector EV t+1 ; reconstructing, by the network, a reconstructed channel Ĥ t by processing the compressed form of channel H t using a decoder part of the two-sided AI/ML model for Channel State Information (CSI) compression; performing, by the network, eigen decomposition of the reconstructed channel Ĥt to obtain an eigenvector ÊV t ; determining, by the network, a second similarity score between the eigenvectors ÊV t and EV t+1 ; and comparing, by the network, the first similarity score and the second similarity score for evaluating performance of the two-sided AI/ML model.

2 . The method as claimed in claim 1 , wherein the first similarity score is determined by the UE by:
performing a first estimation of the channel H t at time t; performing eigen decomposition of the channel H t to obtain the first eigenvector EV t at time t; predicting the channel H t+1 at time t+1; performing an eigen decomposition of the channel H t+1 to obtain the second eigenvector EV t+1 at time t+1; and comparing the first eigenvector EV t and the second eigenvector EV t+1 for obtaining the first similarity score.

3 . The method as claimed in claim 1 , further comprising transmitting, to the network, by the UE, a ground-truth CSI report in a periodic, aperiodic, or semi-persistent manner.

4 . A method of determining performance of a two-sided artificial intelligence/machine learning (AI/ML) model, the method comprising:
receiving, by a network, from a user equipment (UE), a compressed channel H, a Precoding Matrix Indicator (PMI) corresponding to a precoder W, and a first similarity score determined between the compressed channel H and the precoder W; processing, by the network, the compressed channel H using a decoder part of the two-sided AI/ML model for Channel State Information (CSI) compression, to obtain a reconstructed channel Ĥ; identifying, by the network, the precoder W using the PMI; determining, by the network, a second similarity score between the reconstructed channel Ĥ and the precoder W; and comparing, by the network, the first similarity score and the second similarity score for determining performance of the two-sided AI/ML model.

5 . The method as claimed in claim 4 , wherein the first similarity score is determined by the UE by:
performing a first estimation of the channel Hat time t; identifying the precoder W, from a codebook, for determining best Signal to Interference and Noise Ratio (SINR); identifying the PMI corresponding to the precoder W; and comparing the channel H and the precoder W for determining the first similarity score.

Assignees

• Indian Inst Tech Madras

Inventors

• Ganti Radhakrishna
• Pirati Venkata Siva Sai Prasad
• Yerrapragada Anil Kumar
• Sattianarayanin Jeeva Keshav

Classifications

• H04B17/3913 Primary

  Predictive models, e.g. based on neural network models · CPC title

• H04B7/0456

  Selection of precoding matrices or codebooks, e.g. using matrices antenna weighting · CPC title

• H04B7/0626 Primary

  Channel coefficients, e.g. channel state information [CSI] · CPC title