Search

US-12627530-B2 - Data signal processing in multiple sub-panel-based uplink MIMO transmission

US12627530B2US 12627530 B2US12627530 B2US 12627530B2US-12627530-B2

Abstract

According to an example embodiment, a method comprising obtaining received data signals representative of one or more bits of information transmitted by at least one user equipment to a receiver of a radio cellular network, said receiver comprising two or more sub-panels of antennas, obtaining interference and noise covariance matrices individually estimated at at least two of said sub-panels for said received data signals, obtaining equalized signals by performing joint channel and noise equalization of said received data signals at least from said estimated covariance matrices, for at least one bit of said information transmitted by one of said users, obtaining a first set of LLRs, generated at least from said received data signals and said equalized signals, and obtaining a second set of applying the LLRs and the covariance matrices as inputs to a neural network, said neural network having been trained to output corrected LLRs.

Inventors

  • Pavan KOTESHWAR SRINATH

Assignees

  • NOKIA SOLUTIONS AND NETWORKS OY

Dates

Publication Date
20260512
Application Date
20240731
Priority Date
20230804

Claims (10)

  1. 1 . A method comprising: Obtaining received data signals representative of one or more bits of information transmitted by at least one user equipment to a receiver of a radio cellular network, said receiver comprising two or more sub-panels of antennas, obtaining interference and noise, I+N, covariance matrices individually estimated at at least two of said sub-panels for said received data signals, obtaining equalized signals by performing joint channel and noise equalization of said received data signals at least from said estimated I+N covariance matrices, for at least one bit of said information transmitted by one of said users, obtaining a first set of Log Likelihood Ratios, LLRs, generated at least from said received data signals and said equalized signals, obtaining a second set by applying the first set of LLRs and the I+N covariance matrices as inputs to a neural network, said neural network having been trained to output corrected LLRs.
  2. 2 . The method of claim 1 , wherein the method further comprises combining said individually estimated I+N covariance matrices into an overall block-diagonal I+N covariance matrix, said overall I+N covariance matrix being used for obtaining the first set of LLRs.
  3. 3 . The method of claim 2 , wherein, said neural network comprising a first and a second neural networks, the overall I+N covariance matrix is input to one or more convolutional layers of said first neural network, said first neural network being trained to output a vector of covariance elements, said method further comprises concatenating said vector of covariance elements with a vector comprising said first set of LLRs, a concatenated vector being obtained, and applying said concatenated vector as an input to the second neural network, said corrected LLRs being output by said second neural network.
  4. 4 . The method according to claim 3 , wherein the first and second neural networks are trained using a binary cross-entropy loss function.
  5. 5 . A non-transitory computer-readable medium comprising program instructions which when executed by a receiver cause the receiver to perform a method comprising: obtaining received data signals representative of one or more bits of information transmitted by a user equipment to a receiver of a radio cellular network, said receiver comprising two or more sub-panels of antennas, obtaining interference and noise, I+N, covariance matrices individually estimated at each of said sub-panels for said received data signals, obtaining equalized signals by performing joint channel and noise equalization of said received data signals at least from said estimated I+N covariance matrices, for at least one bit of said information transmitted by one of said users, obtaining Log Likelihood Ratios, LLRs, generated for each of two or more sub-panels of said receiver, at least from said received data signals and said equalized signals, applying the LLRs and the I+N covariance matrices as inputs to a neural network, said neural network having been trained to output corrected LLRs, and obtaining the output corrected LLRs.
  6. 6 . An apparatus comprising: at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: obtain received data signals representative of one or more bits of information transmitted by at least one user equipment to a receiver of a radio cellular network, said receiver comprising two or more sub-panels of antennas, obtain interference and noise, I+N, covariance matrices individually estimated at at least two of said sub-panels for said received data signals, obtain equalized signals by performing joint channel and noise equalization of said received data signals at least from said estimated I+N covariance matrices, for at least one bit of said information transmitted by one of said users, obtain a first set of Log Likelihood Ratios, LLRs, generated at least from said received data signals and said equalized signals, obtain a second set by applying the first set of LLRs and the I+N covariance matrices as inputs to a neural network, said neural network having been trained to output corrected LLRs.
  7. 7 . The apparatus of claim 6 , wherein the instructions, when executed by the at least one processor, cause the apparatus to: combine said individually estimated I+N covariance matrices into an overall block-diagonal I+N covariance matrix, said overall I+N covariance matrix being used for obtaining the first set of LLRs.
  8. 8 . The apparatus of claim 7 , wherein said neural network comprising a first and a second neural networks, the overall I+N covariance matrix is input to one or more convolutional layers of said first neural network, said first neural network being trained to output a vector of covariance elements, and wherein the instructions, when executed by the at least one processor, cause the apparatus to: concatenate said vector of covariance elements with a vector comprising said first set of LLRs, a concatenated vector being obtained, and apply said concatenated vector as an input to the second neural network, said corrected LLRs being output by said second neural network.
  9. 9 . The apparatus of claim 8 , wherein the first and second neural networks are trained using a binary cross-entropy loss function.
  10. 10 . The apparatus of claim 6 , wherein the apparatus is incorporated within the receiver, and wherein the receiver further comprises two or more sub-panels of antennas.

Description

RELATED APPLICATION This application claims benefit of priority from European Patent App. No. 23189644.0, filed Aug. 4, 2023, the disclosure of which is hereby incorporated in its entirety by reference herein. TECHNICAL FIELD Various example embodiments relate generally to a method and receiver for processing data signals received from multiple user equipment in multiple sub-panel-based uplink Multi User-Multi Input Multi Output (MU-MIMO) Transmission. BACKGROUND 6G standard is expected to enable a few times more data-rates than 5G. As a result, a base station (BS) or gNB is expected to be equipped with a higher number of antenna elements (AE) that are in the range 512-1024 (against a target of around for 5G). This will also necessitate a larger number of transceivers (TRX), around 256-512 compared to 32-64 in 5G, and the frequency band of interest is expected to be 7-20 GHz. This is called an “extreme MIMO system”. As illustrated by FIG. 1, a larger number of TRXs can be accommodated within the same-sized antenna array by shifting to a larger carrier frequency. When progressing from 64 TRX to 256 or more TRX at a gNB, the following challenge immediately presents itself. At present, there is a limit on the number of TRX that can fit in a single monolithic System-On-Chip (SoC) without dangerously overheating the chip. This number is closer to 64 for safe operations using latest CMOS (Complementary metal-oxide-semiconductor) processors (5 or 7 nm). SUMMARY The scope of protection is set out by the independent claims. The embodiments, examples and features, if any, described in this specification that do not fall under the scope of the protection are to be interpreted as examples useful for understanding the various embodiments or examples that fall under the scope of protection. According to a first aspect, a method comprises: obtaining received data signals representative of one or more bits of information transmitted by at least one user equipment to a receiver of a radio cellular network, said receiver comprising two or more sub-panels of antennas,obtaining interference and noise, I+N, covariance matrices individually estimated at at least two of said sub-panels for said received data signals,obtaining equalized signals by performing joint channel and noise equalization of said received data signals at least from said estimated I+N covariance matrices,for at least one bit of said one or more bits of information transmitted by one of said users, obtaining a first set of Log Likelihood Ratios, LLRs, generated at least from said received data signals and said equalized signals, andobtaining a second set by applying the first set of LLRs and the I+N covariance matrices as inputs to a neural network, said neural network having been trained to output corrected LLRs. The method according to the first aspect may comprise combining said individually estimated I+N covariance matrices into an overall block-diagonal I+N covariance matrix, said overall I+N covariance matrix being used for obtaining the first set of LLRs. In one or more non limiting exemplary embodiments, said neural network comprising a first and a second neural networks, the overall I+N covariance matrix is input to one or more convolutional layers of said first neural network, said first neural network being trained to output a vector of covariance elements, said method further comprises: concatenating said vector of covariance elements with a vector comprising said first set of LLRs, a concatenated vector being obtained, and applying said concatenated vector as an input to the second neural network, said corrected LLRs being output by said second neural network. In one or more non limiting exemplary embodiments, the first and second neural networks are trained using a binary cross-entropy loss function. According to a second aspect, an apparatus comprises means for performing a method comprising: obtaining received data signals representative of one or more bits of information transmitted by at least one user equipment to a receiver of a radio cellular network, said receiver comprising two or more sub-panels of antennas,obtaining interference and noise, I+N, covariance matrices individually estimated at at least two of said sub-panels for said received data signals,obtaining equalized signals by performing joint channel and noise equalization of said received data signals at least from said estimated I+N covariance matrices,for at least one bit of said one or more bits of information transmitted by one of said users, obtaining a first set of Log Likelihood Ratios, LLRs, generated at least from said received data signals and said equalized signals, andobtaining a second set by applying the first set of LLRs and the I+N covariance matrices as inputs to a neural network, said neural network having been trained to output corrected LLRs. The apparatus may comprise means for performing one or more or all steps of the method according to the first aspect.