CN-122001720-A - Channel estimation method and system for orthogonal frequency division multiplexing system based on attention mechanism

CN122001720ACN 122001720 ACN122001720 ACN 122001720ACN-122001720-A

Abstract

The invention relates to a wireless communication technology, in particular to a channel estimation method and a channel estimation system of an orthogonal frequency division multiplexing system based on an attention mechanism, which comprises the following steps of S1, obtaining a channel estimation initial value through a least square method, S2, inputting the channel estimation initial value into an attention network to obtain a query matrix, a key matrix and a value matrix, S3, multiplying the query matrix with the key matrix and introducing the key matrix into a unit matrix to obtain correlation according to symbols, then integrally scaling the unit matrix through a scaling factor to obtain a correlation matrix, S4, filtering the value matrix as a weighted value matrix after normalization processing, S5, sending the signal estimation value into an output mapping layer to obtain a refined signal estimation value after secondary filtering, and periodically training data samples to obtain model parameters in the attention network and the output mapping layer. The method has the advantages that the correlation characteristic of the OFDM channel in the time domain and the frequency domain is effectively utilized, and the channel estimation precision is improved.

Inventors

YANG FAN
DING RUIJIE
HUANG JIE
XIANG JU
LAI XINGJUN
LI TENGFEI
LIU HONGYUAN
XU CHENGFENG

Assignees

重庆理工大学
深圳无线电检测技术研究院

Dates

Publication Date: 20260508
Application Date: 20260317

Claims (9)

1. The channel estimation method of the orthogonal frequency division multiplexing system based on the attention mechanism is characterized by comprising the following steps: S1, obtaining a channel estimation initial value of an orthogonal frequency division multiplexing system through a least square method; s2, inputting the initial value of the channel estimation obtained in the step S1 into an attention network to obtain a query matrix, a key matrix and a value matrix, wherein the query matrix is used as a frequency characteristic, the key matrix is used as a time domain characteristic, and the value matrix is combined with signal filtering and residual connection; s3, multiplying the query matrix by the key matrix and introducing the query matrix into the identity matrix to obtain correlation according to symbols, and then integrally scaling the query matrix by a scaling factor to obtain a correlation matrix; S4, carrying out normalization processing on the related matrix, and filtering the value matrix as a weighted value matrix; s5, sending the output of the step S4 into an output mapping layer for secondary filtering to obtain a refined signal estimated value; Model parameters in the attention network and the output mapping layer are obtained through periodic training of data samples.
2. The channel estimation method of an orthogonal frequency division multiplexing system based on an attention mechanism according to claim 1, wherein pilot subcarriers in the orthogonal frequency division multiplexing system are uniformly distributed in a frequency domain, and a certain proportion of known symbols are embedded in each OFDM symbol for channel estimation at a receiving end.
3. The method for channel estimation in an orthogonal frequency division multiplexing system based on an attention mechanism according to claim 2, wherein known symbols generated using Gold sequences are inserted into a time-frequency resource grid for realizing channel estimation.
4. A method for estimating a channel of an ofdm system based on an attention mechanism according to any one of claims 1 to 3, wherein the least square method in step S1 is as follows: Obtaining initial value of channel estimation And constructing a channel estimation matrix according to each channel estimation initial value Where S represents the number of OFDM symbols, As the number of pilot subcarriers to be used, The representation of the plural number of the form, Representing pilot symbol observations obtained by the receiving end at pilot position p, As a result of the known pilot symbols, Indicating that the channel response is to be given, Is channel noise.
5. The method for channel estimation of an orthogonal frequency division multiplexing system based on an attention mechanism as recited in claim 4, wherein in step S2, real parts and imaginary parts of each element of the channel estimation matrix are separated and then spliced to form a real-valued feature matrix And inputting the attention network to perform feature transformation, and obtaining a query matrix according to the following manner Key matrix Sum matrix ; , Wherein, the Representing a real form of the number, A query matrix weight parameter representing the training of the attention network, A key matrix weight parameter representing the training of the attention network, And the weight parameters of the value matrix obtained by training the attention network are represented.
6. The method for channel estimation of an orthogonal frequency division multiplexing system based on an attention mechanism as recited in claim 5, wherein the step S3 is performed as follows Obtaining a correlation matrix Wherein Representing the identity matrix of the cell, Is a scaling factor.
7. The method of channel estimation in an attention-based orthogonal frequency division multiplexing system of claim 1 or 2 or 3 or 5 or 6, wherein the network training process aims to minimize the mean square error between the neural network output and the real channel response.
8. An orthogonal frequency division multiplexing wireless communication system comprises a transmitting end and a receiving end, and is characterized in that the system adopts the method for estimating the channel of the orthogonal frequency division multiplexing system based on the attention mechanism as set forth in any one of claims 1-7.
9. The orthogonal frequency division multiplexing wireless communication system of claim 8 wherein the wireless channels in the system are time-invariant finite multipath channels.

Description

Channel estimation method and system for orthogonal frequency division multiplexing system based on attention mechanism Technical Field The present invention relates to wireless communication technologies, and in particular, to a method and a system for estimating an orthogonal frequency division multiplexing system channel based on an attention mechanism. Background Orthogonal frequency division multiplexing technology is widely used in modern wireless communication systems due to its high spectral efficiency and strong robustness to frequency selective fading. Accurate channel estimation is the key to ensuring reliable equalization and demodulation. The conventional channel estimation methods, such as least square estimation and least mean square error estimation, have the limitation of noise sensitivity (least square estimation LS) or the need for accurate channel prior statistical information (least mean square error estimation MMSE). In recent years, deep learning has been introduced in the field of wireless channel estimation to address the problem of limited performance of LS/MMSE methods in complex channel environments. The existing research mostly adopts LS estimation results as network input, and optimizes the traditional estimation results by utilizing a neural network or convolution cyclic neural network structure based on back propagation, thereby improving estimation accuracy. There are also learners introducing attention mechanisms into the channel estimation framework, assisting MMSE by learning linear filtering parameters through attention weighting to improve estimation performance. However, the existing methods have large model parameter quantity and face many challenges in the practical engineering application process. Disclosure of Invention In view of this, the present invention firstly provides a channel estimation method for an orthogonal frequency division multiplexing system based on an attention mechanism, which uses the channel estimation result of the traditional least square method as the network input, and introduces the self-adaptive modeling time-frequency correlation feature of the attention mechanism, so as to further optimize the LS estimation result. Unlike the end-to-end learning scheme, the method is designed as a post-processing module of the traditional channel estimator, and maintains the compatibility with the existing system architecture while improving the estimation accuracy. In order to achieve the above purpose, the invention adopts the following specific technical scheme; An orthogonal frequency division multiplexing system channel estimation method based on an attention mechanism comprises the following steps: S1, obtaining a channel estimation initial value of an orthogonal frequency division multiplexing system through a least square method; s2, inputting the initial value of the channel estimation obtained in the step S1 into an attention network to obtain a query matrix, a key matrix and a value matrix, wherein the query matrix is used as a frequency characteristic, the key matrix is used as a time domain characteristic, and the value matrix is combined with signal filtering and residual connection; s3, multiplying the query matrix by the key matrix and introducing the query matrix into the identity matrix to obtain correlation according to symbols, and then integrally scaling the query matrix by a scaling factor to obtain a correlation matrix; S4, carrying out normalization processing on the related matrix, and filtering the value matrix as a weighted value matrix; s5, sending the output of the step S4 into an output mapping layer for secondary filtering to obtain a refined signal estimated value; Model parameters in the attention network and the output mapping layer are obtained through periodic training of data samples. Optionally, pilot subcarriers in the orthogonal frequency division multiplexing system are uniformly distributed in a frequency domain, and a certain proportion of known symbols are embedded in each OFDM symbol for channel estimation of a receiving end. Alternatively, known symbols generated using Gold sequences are inserted into a time-frequency resource grid for use in achieving channel estimation. Optionally, the least square method in step S1 is as follows: Obtaining initial value of channel estimation And constructing a channel estimation matrix according to each channel estimation initial valueWhere S represents the number of OFDM symbols,As the number of pilot subcarriers to be used,The representation of the plural number of the form,Representing pilot symbol observations obtained by the receiving end at pilot position p,As a result of the known pilot symbols,Indicating that the channel response is to be given,Is channel noise. Optionally, in step S2, the real part and the imaginary part of each element of the channel estimation matrix are separated and then spliced to form a real-valued feature matrixAnd inputting the attention