公开(公告)号：US20220004848A1

公开(公告)日：2022-01-06

申请号：US17295296

申请日：2018-11-22

Applicant: Nokia Technologies Oy

Inventor： Jakob HOYDIS ,  Ori SHENTAL

IPC: G06N3/04 ,  H04B1/16

Abstract: An apparatus, method and computer program is described including initialising trainable parameters of a receiver of a transmission system, wherein the receiver includes a demodulation module for demodulating received symbols, a quantization module for generating quantized versions of the demodulated symbols and a decoder for generating a decoded output derived from the quantized versions of the demodulated symbols, wherein the demodulation module has at least some trainable weights and the quantization module has at least some trainable weights; receiving a first training sequence of messages at the receiver; obtaining or generating a loss function; and updating at least some of the trainable parameters of receiver based on the loss function, wherein updating at least some of the trainable parameters of receiver includes updating at least some of the trainable weights of the demodulation module and updating at least some of the trainable weights of the quantization module.

公开(公告)号：US20220060220A1

公开(公告)日：2022-02-24

申请号：US17407698

申请日：2021-08-20

Applicant: Nokia Technologies Oy

Inventor： Olli PIIRAINEN ,  Jaakko VIHRIÄLÄ ,  Ori SHENTAL ,  Sivarama VENKATESAN ,  Juha MYLLYLÄ

IPC: H04B7/0426 ,  H04B7/0456 ,  H04B17/345

Abstract: A method comprising obtaining a noise and interference covariance matrix, performing scaling of the noise and interference covariance matrix, determining a sum of diagonal elements of the scaled noise and interference covariance matrix, performing a first spectral shift by the sum to the scaled noise and interference covariance matrix to obtain a first spectral shifted matrix, performing eigenvalue decomposition to the first spectral shifted matrix to obtain an eigenvalue matrix, performing a second spectral shift by the sum to the eigenvalue matrix, limiting the eigenvalue matrix to a diagonal matrix, and obtaining a new noise and interference covariance matrix adding the scaled noise and interference covariance matrix and eigenvalue pairs that are obtained based, at least partly, on the performed eigenvalue decomposition.