Search

CN-122001500-A - Time-varying channel estimation method of clamp antenna system based on deep learning

CN122001500ACN 122001500 ACN122001500 ACN 122001500ACN-122001500-A

Abstract

The invention discloses a time-varying channel estimation method of a clamp antenna system based on deep learning, which comprises the steps of obtaining a received pilot signal generated by the clamp antenna system in downlink transmission and position information of a clamp antenna which is activated currently, carrying out binarization coding processing on the position information to form a space feature vector corresponding to the position dimension of a preset antenna, inputting the received pilot signal and the space feature vector into a pre-trained channel estimation neural network model, wherein the neural network model at least comprises a time sequence encoder for extracting time sequence features, a space encoder for extracting space features, a cross attention module for fusing space-time features and a decoder for outputting channel state information estimated values, and obtaining the channel state information estimated values output by the neural network model.

Inventors

  • YANG RUIZHE
  • LIU TAO
  • LV SUYU
  • LI MENG

Assignees

  • 北京工业大学

Dates

Publication Date
20260508
Application Date
20260113

Claims (10)

  1. 1. A time-varying channel estimation method of a clamp antenna system based on deep learning is characterized by comprising the following steps of S1, obtaining a received pilot signal generated by the clamp antenna system in downlink transmission and position information of a clamp antenna which is activated currently, S2, performing binarization coding processing on the position information to form a space feature vector corresponding to a preset antenna position dimension, S3, inputting the received pilot signal and the space feature vector into a pre-trained channel estimation neural network model, wherein the neural network model at least comprises a time sequence encoder for extracting time sequence features, a space encoder for extracting space features, a cross attention module for fusing space-time features and a decoder for outputting channel state information estimated values, and S4, obtaining the channel state information estimated values output by the neural network model.
  2. 2. The method for estimating time-varying channel of clamp antenna system based on deep learning as claimed in claim 1, wherein in step S2, the binarization encoding process is specifically that an encoding vector with the same dimension as the number of the discrete antenna positions is generated according to a set of discrete antenna positions preset on a dielectric waveguide, and for each position in the set of discrete antenna positions, if an antenna is activated at the position, the position is corresponding to position 1 in the encoding vector, otherwise, the position is set to 0.
  3. 3. The method for estimating time-varying channel of a clamp antenna system based on deep learning according to claim 1 or 2, wherein the time-series encoder is constructed based on a convolutional neural network CNN for extracting time-series dynamic features from the received pilot signals, and the space encoder is constructed based on a multi-layer perceptron MLP for extracting spatial position features from the spatial feature vectors.
  4. 4. The method of claim 3, wherein the timing encoder comprises at least two different sized convolution kernels, wherein a large sized convolution kernel is used to extract long-range dependent features in the received pilot signal and a small sized convolution kernel is used to extract short-range dependent features in the received pilot signal.
  5. 5. The method for estimating time-varying channel of clamp antenna system based on deep learning as claimed in claim 1, wherein the cross attention module uses the spatial features extracted by the spatial encoder as query matrix Q, uses the time sequence features extracted by the time sequence encoder as key matrix K and value matrix V, and calculates by scaling dot product attention mechanism to make the spatial features selectively aggregate information in the time sequence features.
  6. 6. The method of deep learning based clamp antenna system time varying channel estimation of claim 1, further comprising adding position codes to the timing and/or spatial features prior to inputting the timing and spatial features into the cross-attention module.
  7. 7. The method for estimating time-varying channel of a clamp antenna system based on deep learning as claimed in claim 1, wherein the decoder is constructed based on Bi-directional gating cyclic unit Bi-GRU for decoding the time-space integrated characteristic sequence fused by the cross attention module into a channel state information estimated value sequence.
  8. 8. The method for estimating a time-varying channel of a clamp antenna system based on deep learning of claim 1, wherein the training process of the neural network model comprises constructing a training data set, wherein the training data set comprises a plurality of groups of training samples, each group of training samples comprises a received pilot signal sample, a corresponding binarized coded antenna position sample and a real channel state information label, and training the neural network model with a goal of minimizing a loss function between a channel estimated value output by the neural network model and the real channel state information label.
  9. 9. The method of claim 8, wherein the loss function is a mean square error MSE or a normalized mean square error NMSE.
  10. 10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of any one of claims 1 to 9 when executing the computer program.

Description

Time-varying channel estimation method of clamp antenna system based on deep learning Technical Field The invention relates to the technical field of communication, in particular to a time-varying channel estimation method of a clamp antenna system based on deep learning. Background In order to meet performance requirements such as ultra-high speed and ultra-low delay in sixth generation mobile communications (6G), a new antenna technology has been attracting attention, and one of them, a flexible antenna technology called a clip antenna system (PINCHING ANTENNA SYSTEMS, PASS) has been proposed. PASS is built on a rod-shaped transmission line (dielectric waveguide) and functions as a leaky wave antenna by attaching small dielectric particles, called clip antennas (PINCHING ANTENNA, PA), to the dielectric waveguide. PASS, by virtue of its unique implementation, allows for dynamic increase/decrease of the number of antennas or adjustment of the position of antennas to establish a Line of Sight (LoS) link to combat large-scale path loss, as compared to other antenna technologies. Existing work mostly focuses on analysis and optimization of PASS communication performance and assumes availability of perfect accurate Channel State Information (CSI). However, on one hand, due to the high coupling of channels inside and outside the waveguide in the PASS and the architecture of a single radio frequency (Radio Frequency Chain, RFC) driving a plurality of PAs, the traditional channel estimation method cannot effectively recover CSI according to the pilot signal, and on the other hand, due to the mobility of users, the reflection of the scatterer, the dynamic adjustability of the PAs and the like, the channels are generally affected by multiple fading, so that the channel state fluctuates severely, and the difficulty of channel estimation is further increased. In order to achieve accurate channel state tracking, the method adaptively adjusts model parameters according to environmental changes and is suitable for a highly dynamic clamped antenna system, the method aims to provide a PASS-assisted downlink time-varying channel estimation method based on deep learning (DEEP LEARNING, DL), and a system framework is shown in fig. 1. With the pilot-based channel estimation method, as shown in fig. 2, the original data is composed in the form of data blocks, each of which is composed of a number of pilot symbols and data symbols. After the original data is modulated by Quadrature phase shift keying (Quadrature PHASE SHIFT KEYING, QPSK), a received pilot signal is obtained through a PASS channel. In order to accurately model the mapping relation between the received pilot frequency signal and the real channel of the deep neural network, the position of the clamping antenna after binarization encoding processing and the received pilot frequency signal are input into the deep neural network in the form of double data streams, and the estimated value of the channel is obtained after prediction。 Disclosure of Invention The invention aims to combine the requirements of PASS on obtaining accurate CSI, and designs a time-varying channel estimation method of a clamp antenna system based on deep learning. The method comprises the steps of a base station, a dielectric waveguide, a clamping antenna and a target user. The described base station and dielectric waveguide belong to a wired transmission but are not within the estimation of the invention since the intra-waveguide channel is a deterministic component depending on the PA position and the original feed point. The clamping antenna can be activated in the position of the waveguide preset antenna to function as a leaky wave antenna by adopting the mode of discrete activation of the antenna. The clamping antenna and the target user belong to wireless transmission, and are affected by channel fading, so as to estimate a wireless transmission channel between the antenna and the user. The estimated performance may be quantified by a normalized mean square error function (Normalized Mean Square Error, NMSE). PASS operates in near field communication, and the wireless transmission channel is a Line of Sight (LoS) link or a Non-Line of Sight (NLoS) link, which needs to consider a near field spherical wave model. Furthermore, in the near field communication range, the doppler shift does not look like far field communication to the antenna array as a point, but rather has a relationship with each PA in the array, so the doppler frequency is not uniform. The wireless transmission channel is affected by the combination of large-scale path loss, multipath fading and Doppler shift. And a deep neural network estimator needs to be designed to recover the CSI according to the received pilot signals and the antenna position information. Because of the dynamic adjustability of PASS, the dimension of the antenna position vector is uncertain, in order to ensure the spatial information of the antenna pos