Deep Learning Eigen-Beamforming

According to one example embodiment, a method is provided including initializing a first recurrent neural network with a first vector; iteratively training a weight matrix for one or more layers of the first recurrent neural network, wherein W(t) is the weight matrix of the layer of the first recurr...

Full description

Saved in:
Bibliographic Details
Main Author SHENTAL, Ori
Format Patent
LanguageEnglish
Published 09.12.2021
Subjects
CALCULATING
COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
COMPUTING
COUNTING
ELECTRIC COMMUNICATION TECHNIQUE
ELECTRICITY
PHYSICS
TRANSMISSION
Online AccessGet full text

Cover

More Information
Summary:According to one example embodiment, a method is provided including initializing a first recurrent neural network with a first vector; iteratively training a weight matrix for one or more layers of the first recurrent neural network, wherein W(t) is the weight matrix of the layer of the first recurrent neural network corresponding to iteration t, and wherein W(t) is based at least on a channel covariance matrix; the first vector; and an output from a previous layer of the first recurrent neural network corresponding to iteration, t−1: and determining a first eigenvector based on a converged output from the first recurrent neural network, wherein the first eigenvector is used to perform beamforming for multiple input multiple output reception and/or transmission.
Bibliography:Application Number: US201817284526