Deep learning channel estimation method based on spatial perception interpolation

Abstract

The invention provides a deep learning channel estimation method based on spatial perception interpolation, and belongs to the technical field of wireless communication and artificial intelligence fusion. The method comprises the following steps: firstly, constructing a pilot frequency index set according to sparse distribution of pilot frequency points, and mapping the pilot frequency index set to a two-dimensional subcarrier-symbol grid; then generating a centrosymmetric and edge-attenuated space weighting matrix alpha (x, y), and adjusting interpolation weight distribution by introducing a position-related Gaussian weighting function; complex field interpolation is carried out on the pilot frequency observation value in combination with a space weighting radial basis function, and a complete channel initial estimation matrix is constructed; and then decomposing the interpolation result into a real part, an imaginary part and a spatial weighting coefficient, constructing a three-channel tensor as input, sending the three-channel tensor into a convolutional neural network for refined estimation, and outputting a final complex channel estimation result. In a preferred embodiment, the convolutional neural network adopts an SRCNN structure, and inputs a real part, an imaginary part and a spatial weighting matrix alpha (x, y) including an interpolation channel to enhance the spatial perception ability of the model. Simulation results show that under a VehA standard channel model, the method is always superior to a traditional LS method under the sparse pilot frequency condition, the performance of the method is close to that of an MMSE method, higher estimation precision and higher robustness are shown, and the method is suitable for a channel recovery task in a high-speed dynamic wireless communication system.

Description

Technical Field

[0001] The present invention belongs to the technical field of communication systems, and in particular relates to a channel estimation method for wireless communications, specifically a deep learning channel estimation method based on the combination of spatially aware interpolation and convolutional neural networks, which is suitable for channel state recovery tasks under sparse pilot conditions in multi-carrier modulation communication systems. Background Art

[0002] In wireless communication systems, channel estimation is a critical step in accurately recovering transmitted information at the receiver, and its accuracy directly impacts the overall system performance. With the increasing demand for high speed, low latency, and high reliability, modern communication systems are widely adopting technologies such as orthogonal frequency division multiplexing (OFDM) and multiple-input multiple-output (MIMO), which places higher demands on the acquisition of channel state information (CSI). Traditional pilot-assisted channel estimation methods, such as linear interpolation, spline interpolation, and radial basis function (RBF) interpolation, achieve good results when pilots are densely distributed or the channel environment is stable. However, in complex scenarios such as sparse pilots or rapidly changing channels, the estimation accuracy decreases significantly, and the estimation lacks the ability to perceive spatial structure.

[0003] Although deep learning methods have been gradually introduced into channel estimation tasks and have shown superior performance under certain conditions, existing methods generally fail to effectively combine the spatial distribution characteristics of pilot points, find it difficult to maintain stable performance in extremely sparse pilot environments, and have problems such as weak generalization ability and poor robustness.

[0004] To address these technical bottlenecks, this paper proposes a channel estimation method that integrates spatially aware interpolation and convolutional neural networks. By constructing a centrally symmetric, edge-attenuated spatial weighting matrix to guide the distribution of interpolation weights, this method achieves a preliminary estimation of the channel under sparse pilot conditions. This method then integrates a deep learning model to refine and optimize the interpolation results, significantly improving estimation accuracy and system robustness under complex channel conditions. Summary of the Invention

[0005] The present invention aims to solve the problem of low estimation accuracy of traditional channel estimation methods in scenarios with sparse pilots and rapidly changing channels. A deep learning channel estimation method and system based on spatial-aware interpolation is proposed, which integrates spatial distribution information with the feature extraction capabilities of convolutional neural networks, thereby improving the accuracy and robustness of channel estimation in complex wireless environments.

[0006] According to the present invention, a deep learning channel estimation method and system based on spatial perception interpolation is provided, comprising the following steps:

[0007] Generate ideal channel matrix using VehA channel model in MATLAB At the same time, a pilot mask is constructed according to the system settings, and additive Gaussian white noise is injected at the pilot position to generate a noisy pilot measurement matrix , used for training and validation.

[0008] An index set of pilot points is constructed and mapped to a two-dimensional subcarrier-OFDM symbol grid to determine the spatial distribution structure of the pilots.

[0009] Generate a spatial weighting coefficient matrix with central symmetry and edge attenuation characteristics based on the pilot spatial position .

[0010] According to the Euclidean distance between the interpolation target point and each pilot point, the radial basis function response is calculated and combined with Get the weighted coefficient matrix .

[0011] Using the pilot observations and the weight matrix, a spatially weighted RBF interpolation method is used to perform preliminary complex channel estimation on non-pilot points to obtain a complete two-dimensional complex matrix.

[0012] The interpolated complex channel real part, imaginary part and optional It is fed into the SRCNN network as a multi-channel tensor for fine depth estimation.

[0013] The SRCNN network outputs the optimized channel estimation results, improves the estimation accuracy, and completes the entire channel estimation process.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] (1) The present invention introduces a spatial weighting mechanism, which significantly improves the adaptability of the interpolation algorithm to sparse pilot distribution and overcomes the performance defects of traditional methods under non-uniform pilot conditions.

[0016] (2) The present invention uses a deep neural network to perform fine compensation on the interpolation results, effectively improving the overall accuracy of channel estimation, especially under low signal-to-noise ratio (SNR) conditions.

[0017] (3) The method of the present invention has a clear structure and the computational complexity is controlled within a reasonable range, which facilitates rapid deployment in actual communication systems and has high engineering practicality.

[0018] (4) The present invention has been simulated and verified under the VehA channel model. The experimental results show that it outperforms traditional linear interpolation and ordinary CNN channel estimation methods in various SNR and sparse pilot scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0020] Figure 1 Deep learning channel estimation method and system based on spatial perception interpolation

[0021] Figure 2 Channel tensor channel reconstruction and SRCNN input structure schematic diagram

[0022] Figure 3 NMSE performance comparison of different channel estimation algorithms under multiple SNRs DETAILED DESCRIPTION

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0024] It should be noted that in this paper, step codes such as S10, S20 are used, the purpose is to more clearly and briefly express the corresponding content, and it does not constitute substantial limitation on the order, and those skilled in the art may execute S10 before S20 in specific implementation, etc., but these should be within the protection scope of the present application.

[0025] Please refer to Figure 1 The present application provides a deep learning channel estimation method and system based on spatial perception interpolation, comprising the following steps:

[0026] S10: generating complex channel data required for training based on MATLAB tool, including ideal channel and noisy pilot observation;

[0027] S20: extracting the coordinate index of pilot point in two-dimensional OFDM grid (72 subcarriers x 14 time slots), constructing pilot mask matrix , and numbering and managing the pilot point position;

[0028] S30: based on the spatial distribution of pilot points, calculating the minimum Euclidean distance from each position on the grid to its nearest pilot point, and generating a spatial weighting matrix ;

[0029] S40: combining and Euclidean distance, calculate the weighted RBF response weight matrix ;

[0030] S50: The real part, imaginary part and spatial weighting matrix of the interpolation result As three channels, they are concatenated to form an input tensor of dimension [N, 72, 14, 3] for fine estimation of subsequent deep learning models;

[0031] S60: Uses a lightweight super-resolution convolutional neural network (SRCNN) to perform end-to-end fine estimation learning on the input tensor, improving channel estimation accuracy and suppressing noise effects.

[0032] S70: The network outputs the optimized complex channel estimation result , which can be used for data restoration in subsequent communication systems.

[0033] The following further describes the steps of a deep learning channel estimation method and system based on spatially aware interpolation:

[0034] S10. Generation of channel data

[0035] Specifically, channel data is generated by using the 3GPP VehA model to generate the corresponding perfect channel data. Pilot points are inserted according to the system's pilot structure, and Gaussian noise is added to the pilot points to generate noisy data. The specific generation includes:

[0036] Generating a perfect channel matrix , pilot position mask , the noisy channel observation value at the pilot point , where the noise follows a complex Gaussian distribution and the optional pilot , these data serve as input samples for subsequent interpolation estimation and training.

[0037] S20, extraction of pilot position and mask construction

[0038] Construct mask based on pilot structure

[0039] Extract the pilot point coordinate set by the mask , used for subsequent interpolation.

[0040] S30, generating a spatial weighting matrix

[0041] This step proposes a central symmetric weighting mechanism with spatial perception ability to generate space

[0042] Inter-weighting coefficient matrix , which is used to guide the subsequent weighted interpolation operation. It specifically includes the following sub-steps:

[0043] For each 2D grid position ( ), calculate its distance to all pilot points ( ) to form a distance set:

[0044]

[0045] in, Represents a grid point ( ) and the Euclidean distance between the i-th pilot point, where i is the pilot point index.

[0046] To strengthen spatial locality, the minimum distance criterion is introduced to define the influence weight of the position relative to the nearest pilot point. First, the minimum distance from the point to all pilot points is extracted:

[0047]

[0048] in Indicates the total number of pilot points. The distance reflects the degree to which the current position is controlled by the nearest pilot point.

[0049] Based on the Gaussian distribution concept, a weighted function with central symmetry and edge attenuation is constructed and defined as follows:

[0050]

[0051] in, It is an adjustable perception radius that controls the speed of weighted attenuation. The features of this function include:

[0052] when When approaching a certain pilot point position, tends to 0, at this time Tends to 1.

[0053] when When far away from all pilot points, Increase, at this time Tends to 1.

[0054] S40, spatial weighted RBF interpolation module

[0055] Using spatially weighted radial basis function interpolation (RBF-Interpolation), based on pilot observations And the weighted value of the corresponding position to complete the preliminary estimation of the non-pilot point. The estimation formula is as follows:

[0056]

[0057] in is the normalized interpolation weight, given by the following formula:

[0058]

[0059] The radial basis function adopts the Gaussian kernel form:

[0060]

[0061] S50, build input tensor

[0062] Decompose the interpolated complex channel into real and imaginary parts, and construct a multi-channel tensor as the neural network input:

[0063]

[0064] Introducing the weight channel:

[0065]

[0066] The three-channel tensor is constructed by the above formula Preprocessing as neural network input.

[0067] S60 and SRCNN network module precise evaluation

[0068] Decompose the complex channel estimate tensor into its real part , imaginary part and the spatial weighting matrix , construct a three-channel tensor:

[0069]

[0070] Will Input the SRCNN convolutional neural network for channel fine estimation, and the output channel is 2, corresponding to the reconstruction of the real and imaginary parts.

[0071] The above processing process is as follows Figure 2 The figure clearly shows that the complex channel is divided into three branch channels (real part, imaginary part, and weight) through the "channel splitting module", and finally spliced ​​into a three-channel input tensor for use by the neural network.

[0072] S70: Output channel estimation result

[0073] The three-channel tensor is input into the SRCNN convolutional neural network for channel estimation to obtain the output tensor:

[0074]

[0075] Merge the real and imaginary parts of the tensor output by the neural network and reconstruct it into a complex channel:

[0076]

[0077] This result is the final channel estimation result of the method proposed in the present invention, which can be used for subsequent data demodulation and data restoration.

[0078] To verify the performance of the proposed method, multiple simulation experiments were conducted under different signal-to-noise ratio (SNR) conditions, and compared with the traditional least squares (LS) method and the minimum mean square error (MMSE) method. The normalized mean square error (NMSE) was selected as the performance indicator, which is defined as follows:

[0079]

[0080] The simulation results are as follows Figure 3 As shown, it can be seen that:

[0081] This paper proposes a deep learning channel estimation method and system based on spatial-aware interpolation, which exhibits significant performance advantages in the medium and high SNR range. Its NMSE performance is significantly better than the traditional LS method and approaches the MMSE theoretical boundary. At the same time, it maintains excellent estimation accuracy and robustness under low pilot density conditions, verifying the algorithm's strong adaptability to complex channel environments.

[0082] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.

Claims

1. A deep learning channel estimation method based on spatial perception interpolation, characterized in that: The steps include: Collect the location information of the pilot points, construct a pilot index set, and map the pilot points to a two-dimensional grid containing subcarrier dimensions and symbol dimensions; Generate a set of spatial weighting coefficients based on the spatial distribution of the pilot points, which are used to describe the weight relationship of each grid position relative to the pilot point, and the spatial weighting coefficients have the characteristics of central symmetry and edge attenuation; Based on the complex channel observation values ​​of the pilot points and the spatial weighting coefficients, a radial basis function interpolation method is used to perform complex domain interpolation on all non-pilot positions in the target grid to generate a complete initial channel estimation matrix; The interpolated initial channel matrix is ​​input into the convolutional neural network, and the deep learning model is used to further refine the estimation and output the final complete channel estimation result.

2. The method according to claim 1, characterized in that The method is applicable to various pilot sparse structure configurations. The number of pilot points can be 8, 16, 24, 36 or 48, and the distribution of pilot points in the two-dimensional grid can be regular arrangement or irregular sparse arrangement.

3. The method according to claim 1, characterized in that The spatial weighting coefficient is generated by calculating the distance relationship between the target position and the pilot point, and the weight decays as the distance increases, which has spatial perception capability.

4. The method according to claim 1, wherein The convolutional neural network is a shallow feedforward neural network, which includes multiple convolutional layers and activation function layers, and is used to perform feature extraction and nonlinear mapping on the initial channel matrix.

5. The method according to claim 1, wherein The input of the convolutional neural network is a three-dimensional tensor, which includes the real channel and imaginary channel of the complex channel interpolation result, and the spatial weighted channel corresponding to their positions.

6. The method according to claim 1, characterized in that The method uses a VehA channel model that complies with the 3GPP standard to perform performance testing and verification in a simulation environment.

Images