CN-121984642-A - Multi-user detection method and device based on extended factor graph and graph annotation network

CN121984642ACN 121984642 ACN121984642 ACN 121984642ACN-121984642-A

Abstract

The invention discloses a multi-user detection method and device based on an extended factor graph and a graph attention network, wherein the method comprises the steps of carrying out multi-time oversampling on an asynchronous uplink signal of a receiving end, obtaining a plurality of sampling observations in each symbol period, constructing the extended factor graph based on the sampling observations, carrying out feature extraction on the oversampled received signal, inputting a real part and an imaginary part as a deep neural network to obtain node features, introducing the graph attention network on the extended factor graph to carry out information interaction and model training, and judging the received signal of each user by utilizing a model obtained by training, so as to realize multi-user joint detection. The invention realizes the effective detection of multi-user signals in an asynchronous SCMA system by modeling the oversampled signals as an extended factor graph structure and introducing a graph attention mechanism, and solves the problems of high multi-user detection complexity and insufficient anti-asynchronous interference capability in an uplink asynchronous SCMA scene.

Inventors

LI LIYAN
Zhu keli
ZHAO RONG

Assignees

浙江大学

Dates

Publication Date: 20260505
Application Date: 20260403

Claims (9)

1. A multi-user detection method based on an extended factor graph and a graph meaning network is characterized by being applied to an asynchronous uplink Sparse Code Multiple Access (SCMA) system and comprising the following steps: controlling each user to map an input bit sequence into a sparse codeword dedicated to the user, performing non-orthogonal modulation on a plurality of orthogonal resource blocks, and then transmitting the sparse codeword to enable a plurality of user signals to be transmitted in a superposition manner on the same resource block, and obtaining an asynchronous uplink receiving signal with symbol-level time dislocation by a receiving terminal; performing multiple oversampling processing on the asynchronous uplink received signal to obtain an oversampled received signal, acquiring a plurality of sampling observations corresponding to different time alignment positions in each symbol period, and constructing an extended factor graph according to the sampling observations; Extracting features of the oversampled received signal, taking a real part and an imaginary part of the oversampled received signal as input of a deep neural network respectively, extracting a feature representation for multi-user detection through multi-layer nonlinear mapping, and taking the feature representation as initial features of nodes in the extended factor graph; And executing multi-user detection based on the graph attention network on the extended factor graph, judging the modulation symbols of each user according to the output result of the graph attention network, and completing multi-user joint detection in an asynchronous sparse code multiple access system.
2. The method of claim 1 wherein performing multiple oversampling on the asynchronous upstream received signal to obtain a plurality of sample observations corresponding to different time aligned positions in each symbol period comprises: And carrying out time sampling on the asynchronous uplink received signal for at least two times in each symbol period to obtain a plurality of sampling observations corresponding to different symbol alignment positions.
3. The method of claim 1, wherein the constructing an extended factor graph comprises: expanding the resource nodes in the traditional sparse code multiple access factor graph into a plurality of resource-sampling sub-nodes which are in one-to-one correspondence with the sampling positions; and establishing a connection relation between the user node and the resource-sampling sub-node according to the non-zero codeword position of the user on the resource block and the arrival time delay of the user signal, and forming an extension factor graph reflecting the asynchronous superposition structure.
4. A method according to claim 3, characterized in that the feature extraction of the oversampled received signal comprises: Respectively taking a real part and an imaginary part of the oversampled received signal as input features of a deep neural network; carrying out nonlinear mapping on the input features through a plurality of full-connection layers and nonlinear activation functions, and extracting high-dimensional feature representation for representing the superposition characteristics of asynchronous symbols; the high-dimensional feature representation is used as an initial feature of a corresponding node in the graph attention network.
5. The method of claim 4, wherein performing graph-attention-network-based multi-user detection on the extended factor graph comprises: Carrying out weighted modeling on information interaction between the user node and the resource-sampling sub-node by adopting a graph attention mechanism; And the contribution degree of different adjacent nodes to the detection result in the information aggregation process is adaptively adjusted according to the attention weight, so that the influence of multiple access interference caused by asynchronous transmission on the detection result is reduced.
6. The method of claim 5, wherein the graph-annotation network is trained in an end-to-end manner and a loss function is constructed based on detection errors, and network parameters are jointly optimized to improve accuracy of multi-user detection in an asynchronous SCMA system.
7. The multi-user detection device based on the extended factor graph and the graph meaning network is characterized by being applied to an asynchronous uplink Sparse Code Multiple Access (SCMA) system and comprising the following components: The first module is used for controlling each user to map an input bit sequence into a sparse codeword dedicated to the user, performing non-orthogonal modulation on a plurality of orthogonal resource blocks and then transmitting the sparse codeword, enabling a plurality of user signals to be transmitted in a superposition manner on the same resource block, and obtaining an asynchronous uplink receiving signal with symbol-level time dislocation by a receiving end; the second module is used for carrying out multiple oversampling processing on the asynchronous uplink received signal to obtain an oversampled received signal, acquiring a plurality of sampling observations corresponding to different time alignment positions in each symbol period, and constructing an extended factor graph according to the sampling observations; The third module is used for extracting the characteristics of the oversampled received signal, taking the real part and the imaginary part of the oversampled received signal as the input of a deep neural network respectively, extracting the characteristic representation for multi-user detection through multi-layer nonlinear mapping, and taking the characteristic representation as the initial characteristics of each node in the extended factor graph; And a fourth module, configured to perform multi-user detection based on a graph attention network on the extended factor graph, and perform decision on modulation symbols of each user according to an output result of the graph attention network, so as to complete multi-user joint detection in an asynchronous sparse code multiple access system.
8. An electronic device comprising a processor and a memory; Wherein the processor runs a program corresponding to executable program code stored in the memory by reading the executable program code for implementing the method according to any one of claims 1-6.
9. A non-transitory computer readable storage medium, on which a computer program is stored, characterized in that the program, when executed by a processor, implements the method according to any one of claims 1-6.

Description

Multi-user detection method and device based on extended factor graph and graph annotation network Technical Field The invention relates to the technical field of wireless communication, in particular to a multi-user detection method and device based on an extended factor graph and a graph annotation network. Background With the rapid development of applications such as internet of things, internet of vehicles and large-scale machine type communication, a wireless access system needs to support simultaneous access of more users under the condition of limited spectrum resources. Non-orthogonal multiple access techniques are considered as an effective means to improve system access capability and spectral efficiency by allowing multiple users to transmit in superposition on the same time-frequency resource. The sparse code multiple access is used as a typical code domain non-orthogonal multiple access technology, and the sparse mapping characteristic of users on a resource block is utilized to map multi-user signals to limited resource units, so that multi-user concurrent transmission is realized under an overload scene. Under ideal conditions, SCMA systems generally assume that user signals remain strictly synchronized at the receiving end, and at this time, the multi-user superposition structure can be modeled by using a factor graph, and a message passing Algorithm (MESSAGE PASSING Algorithm, MPA) based on the factor graph is used to implement near-optimal multi-user detection. However, in an actual wireless communication environment, due to factors such as scattered geographical locations of users, propagation delay differences, inconsistent terminal hardware, etc., it is often difficult to achieve strict time synchronization of user signals, so that received signals exhibit asynchronous superposition characteristics. In an asynchronous SCMA system, sparse code words of different users are misplaced on a time axis, an original factor graph structure constructed based on synchronous assumptions is destroyed, and signals on resource nodes no longer meet an ideal sparse superposition relation. The time mismatch can introduce extra multi-user interference, so that the traditional MPA detection algorithm based on a factor graph structure is difficult to accurately describe the interference relation between users in the message updating process, the message reliability is reduced, and the system error code performance is obviously degraded. Most of the existing researches are developed aiming at synchronous SCMA scenes, and the researches on the structural changes of factor graphs under asynchronous conditions and the influence on detection performance are still limited. How to effectively model the relation between multi-user signals under the asynchronous transmission condition of users, and improve the detection robustness on the premise of keeping reasonable calculation complexity is still a problem to be solved in the design of the receiving end of the asynchronous SCMA system. Disclosure of Invention The invention mainly aims at solving the problems of poor detection performance, insufficient robustness, higher complexity and the like of a traditional message transmission algorithm based on a factor graph under an asynchronous condition in an uplink asynchronous sparse code multiple access system, and provides a multi-user detection method based on an extended factor graph and a graph-meaning network so as to realize stable and low-complexity multi-user detection under a symbol timing mismatch condition. Another object of the present invention is to propose a multi-user detection device based on an extended factor graph and a graph-based network. A third object of the present invention is to propose an electronic device. A fourth object of the present invention is to propose a non-transitory computer readable storage medium. To achieve the above objective, an embodiment of a first aspect of the present invention provides a multi-user detection method based on an extended factor graph and a graph meaning network, which is applied to an asynchronous uplink sparse code multiple access SCMA system, and includes: controlling each user to map an input bit sequence into a sparse codeword dedicated to the user, performing non-orthogonal modulation on a plurality of orthogonal resource blocks, and then transmitting the sparse codeword to enable a plurality of user signals to be transmitted in a superposition manner on the same resource block, and obtaining an asynchronous uplink receiving signal with symbol-level time dislocation by a receiving terminal; performing multiple oversampling processing on the asynchronous uplink received signal to obtain an oversampled received signal, acquiring a plurality of sampling observations corresponding to different time alignment positions in each symbol period, and constructing an extended factor graph according to the sampling observations; Extracting features of the oversa