US 12,726,387 B2 · Signal processing method and apparatus

Abstract

Embodiments of this application relate to the field of wireless communication technologies, and provide a signal processing method and an apparatus, to improve signal processing performance. In this method, a communication apparatus inputs N1 signal vectors to an input layer of a neural network for dimension increase processing, to obtain N2 high-dimensional signal vectors. The communication apparatus inputs the N2 high-dimensional signal vectors to an interaction layer of the neural network, to obtain a feature between the N2 high-dimensional signal vectors and a feature of each high-dimensional signal vector. The communication apparatus inputs the feature between the N2 high-dimensional signal vectors and the feature of each high-dimensional signal vector to an output layer of the neural network for an operation, to obtain an output signal vector.

Claims20

Dependent claims

A signal processing method for a communication apparatus, the method comprising:

mapping an input signal to N1 resource patches (RPs), each RP of the N1 RPs comprises one or more resource elements (REs), the each RP comprises one signal vector determined based on the input signal, wherein N1 is a positive integer;
inputting N1 signal vectors to an input layer of a neural network for dimension increase processing to obtain N2 high-dimensional signal vectors;
inputting the N2 high-dimensional signal vectors to an interaction layer of the neural network to obtain a feature between the N2 high-dimensional signal vectors and a feature of each high-dimensional signal vector, wherein N2 is a positive integer;
and
inputting the feature between the N2 high-dimensional signal vectors and the feature of each high-dimensional signal vector to an output layer of the neural network for an operation, to obtain an output signal vector.

2 more elements

A communication apparatus, comprising:

a processor;
and
a transceiver coupled to the processor;
the transceiver being configured to obtain an input signal;
and
the processor being configured to:
map the input signal to N1 resource patches (RPs), wherein the each RP of the N1 RPs comprises one or more resource elements (REs), the each RP comprises one signal vector determined based on the input signal;
input N1 signal vectors to an input layer of a neural network for dimension increase processing to obtain N2 high-dimensional signal vectors;
input the N2 high-dimensional signal vectors to an interaction layer of the neural network to obtain a feature between the N2 high-dimensional signal vectors and a feature of each high-dimensional signal vector;
and
input the feature between the N2 high-dimensional signal vectors and the feature of each high-dimensional signal vector to an output layer of the neural network for an operation, to obtain an output signal vector.

8 more elements

Dependencies are inferred from the claim text.

Signal processing method and apparatus

Life of this patent

Signal processing method and apparatus

Life of this patent

Abstract

Claims20