EP-4738921-A1 - DATA PROCESSING METHOD AND APPARATUS

EP4738921A1EP 4738921 A1EP4738921 A1EP 4738921A1EP-4738921-A1

Abstract

A data processing method and an apparatus are disclosed. In the method, a first apparatus can associate data streams with identifiers, allowing a data receive end or a data transmit end to distinguish between different data streams based on the identifiers. In addition, a training unit of the first apparatus can distinguish between the different data streams based on different identifiers when performing AI training, so that the data streams can be trained by sharing parameters, for example, reusing a same neural network processing method.

Inventors

XU, CHEN
ZHANG, Gongzheng
WANG, JIAN
LI, RONG

Assignees

Huawei Technologies Co., Ltd.

Dates

Publication Date: 20260506
Application Date: 20230824

Claims (20)

A data processing method, comprising: determining one or more data streams, wherein each data stream corresponds to one or more identifiers; and inputting the one or more data streams and identifiers corresponding to the one or more data streams into a training unit, to obtain output information, wherein the training unit comprises one or more artificial intelligence AI modules for signal processing in a transmitter machine or a receiver machine.
The method according to claim 1, wherein one data stream comprises a plurality of symbols, and each symbol corresponds to one identifier.
The method according to claim 1, wherein one data stream comprises a plurality of symbols, and a part or all of the plurality of symbols correspond to one identifier.
The method according to claim 2 or 3, wherein different data streams correspond to different identifiers.
The method according to any one of claims 1 to 4, wherein the inputting the one or more data streams and the identifiers corresponding to the one or more data streams into the training unit comprises: separately concatenating each data stream with a corresponding identifier to obtain a concatenated vector, wherein a dimension of the concatenated vector is determined based on a quantity of data streams and a quantity of symbols of each data stream; and inputting the concatenated vector into the training unit.
The method according to any one of claims 1 to 4, wherein the inputting the one or more data streams and the identifiers corresponding to the one or more data streams into the training unit comprises: separately concatenating each data stream with a corresponding identifier to obtain a concatenated vector, wherein a dimension of the concatenated vector is determined based on a quantity of data streams and a quantity of symbols of each data stream; performing dimension increase on the concatenated vector to obtain a high-dimensional vector; and inputting the high-dimensional vector into the training unit.
The method according to any one of claims 1 to 4, wherein the inputting the one or more data streams and the identifiers corresponding to the one or more data streams into the training unit comprises: performing dimension increase on each data stream to obtain a high-dimensional vector, wherein a dimension of the high-dimensional vector is determined based on a quantity of data streams and a quantity of symbols of each data stream; and performing a first operation on the high-dimensional vector and the one or more identifiers, and inputting a result of the first operation into the training unit.
The method according to any one of claims 1 to 4, wherein the one or more identifiers are determined based on a pseudo-random sequence, and the pseudo-random sequence is related to one or more of a cell identity, a user identifier, and an index of the data stream.
The method according to claim 8, wherein the index of the data stream is indicated by using downlink control information.
The method according to any one of claims 1 to 4, wherein the one or more identifiers are determined based on a complex sequence or a real sequence, and the complex sequence or the real sequence is determined based on a random sequence and a mapping relationship; the mapping relationship comprises a first mapping relationship between a plurality of complex identifiers and a plurality of index values, and/or a second mapping relationship between a plurality of real identifiers and a plurality of index values; and the random sequence is determined based on a user identifier and/or an index of the data stream, and a value in the random sequence is an index value in the first mapping relationship and/or an index value in the second mapping relationship.
The method according to any one of claims 1 to 4, wherein the one or more identifiers are determined based on a first function relationship, and a parameter of the first function relationship is related to one or more of a quantity of data streams, a maximum quantity of supported streams, a maximum dimension of the identifier, a dimension index of the identifier, and a user identifier.
The method according to claim 11, wherein the first function relationship satisfies the following formula: r l n RNTI i = cos l M i d + 2 π × M i d n RNTI 2 16 + j sin l M i d + 2 π × M i d n RNTI 2 16 , wherein r(l, n RNTI , i) comprises the one or more identifiers, l represents the quantity of data streams, d represents the maximum dimension of the identifier, i represents the dimension index of the identifier, M represents the maximum quantity of supported streams, n RNTI represents the user identifier, cos represents a cosine function, and sin represents a sine function.
The method according to claim 1, wherein the one or more identifiers belong to an identifier sequence of a data transmit end, or the one or more identifiers belong to an identifier sequence of a data receive end; and there is a correspondence between the identifier sequence of the data transmit end and the identifier sequence of the data receive end.
The method according to claim 13, wherein the method further comprises: sending, by the data transmit end, the output information, wherein the output information is input to the data receive end after being transmitted through a channel; or sending, by the data receive end, the output information, wherein the output information of the data receive end is used by the data receive end and the data transmit end to update respective training units.
A communication apparatus, comprising a communication unit and a processing unit, wherein the communication unit and the processing unit are configured to perform the method according to any one of claims 1 to 14.
A communication apparatus, comprising a processor and a memory, wherein the memory is configured to store instructions; and when the instructions are executed by the processor, the communication apparatus is enabled to perform the method according to any one of claims 1 to 14.
A computer-readable storage medium, wherein the computer-readable storage medium stores instructions; and when the instructions are run on a computer, the computer is enabled to perform the method according to any one of claims 1 to 14.
A computer program product, comprising instructions, wherein when the instructions are run on a computer, the computer is enabled to perform the method according to any one of claims 1 to 14.
A chip, wherein the chip comprises a processor, or the chip comprises a processor and an interface, and the processor is configured to execute a computer program, so that the chip implements the method according to any one of claims 1 to 14.
A chip system, wherein the chip system comprises a processor and an interface; and the processor is configured to execute a computer program, so that the chip system implements the method according to any one of claims 1 to 14.

Description

TECHNICAL FIELD This application relates to the field of communication technologies, and in particular, to a data processing method and an apparatus. BACKGROUND Currently, artificial intelligence (artificial intelligence, AI) technologies can be applied to various aspects of mobile communication network's physical layer (such as channel encoding and decoding, channel prediction, receiver machines, and so on). For example, for physical-layer AI research, in multiple-input multiple-output (multiple-input multiple-output) scenarios, convolutional neural networks (convolutional neural network, CNN) can be used to design receiver machines. For example, in pilot-less MIMO transmission, to lower neural network complexity and improve generalization/scalability, same parameters may be reused across different layers of the neural network. However, this prevents a receive end from distinguishing data from different streams, thereby limiting training/inference performance of the neural network. SUMMARY This application provides a data processing method and an apparatus. The method can be used to distinguish between multi-layer data streams processed by reusing a same neural network. According to a first aspect, this application provides a data processing method. The method is performed by a first apparatus. The first apparatus is a data transmit end (for example, a transmitter machine). For example, for uplink transmission, the first apparatus may be a terminal, or may be a component (for example, a processor, a chip, or a chip system) of a terminal, or may be a logical module that can implement all or a part of functions of a terminal. For another example, for downlink transmission, the first apparatus may be a network device, or may be a component (for example, a processor, a chip, or a chip system) of a network device, or may be a logical module that can implement all or a part of functions of a terminal. The first apparatus determines one or more data streams, where each data stream corresponds to one or more identifiers. The first apparatus inputs the one or more data streams and identifiers corresponding to the one or more data streams into a training unit, to obtain output information, where the training unit includes one or more artificial intelligence AI modules for signal processing in the transmitter machine. In the method, the first apparatus is the data transmit end, and can associate data streams with identifiers, so that a data receive end (for example, a receiver machine) can distinguish between different data streams based on the identifiers. In addition, the training unit of the first apparatus can distinguish between the different data streams based on different identifiers when performing AI training, so that the data streams can be trained by sharing parameters (for example, reusing a same neural network processing method). This helps improve generalization and scalability of a network. In a possible implementation, one data stream includes a plurality of symbols, and each symbol corresponds to one identifier. For example, it is assumed that one data stream includes M symbols, the M symbols respectively correspond to M identifiers (for example, the M symbols and the M identifiers are in a one-to-one correspondence), and the M identifiers form an identifier sequence. It may be referred to that each data stream corresponds to one identifier sequence. The M identifiers are M different identifiers, and M is a positive integer. In a possible implementation, one data stream includes a plurality of symbols, and a part or all of the plurality of symbols correspond to one identifier. For example, it is assumed that one data stream includes M symbols, and some (it is assumed that there are N symbols, where N is a positive integer less than M) or all (namely, the M symbols) of the M symbols correspond to one identifier. M is a positive integer. Optionally, some or all of other symbols in the M symbols except the N symbols may correspond to another identifier, and the identifier is different from the identifier corresponding to the N symbols. In a possible implementation, different data streams correspond to different identifiers. In the foregoing method, each symbol of one data stream corresponds to one identifier, or some or all symbols in one data stream may correspond to one identifier, and different data streams correspond to different identifiers. This helps distinguish the data streams. In a possible implementation, the first apparatus separately concatenates each data stream with a corresponding identifier to obtain a concatenated vector, where a dimension of the concatenated vector is determined based on a quantity of data streams and a quantity of symbols of each data stream. The first apparatus inputs the concatenated vector into the training unit. In the method, a method for associating a data stream with an identifier may be specifically directly performing concatenation to obtain a concatenated v