ProDrnNet-based channel estimation method in OFDM (Orthogonal Frequency Division Multiplexing) system

Abstract

The invention relates to a channel estimation method based on a deep residual learning network (ProDrnNet) in an OFDM (Orthogonal Frequency Division Multiplexing) system. The channel estimation method comprises the following steps of: firstly, carrying out channel estimation on the basis of the ProDrnNet; according to the method, firstly, channel estimation at a pilot frequency is preliminarily obtained at a receiving end through a least square method (LS), the information serves as a low-resolution noisy image to be input into a channel estimation model, the model learns a mapping relation from the low-resolution noisy image to high-resolution denoising information, and therefore a restored high-resolution denoising image is output from the model, and a channel estimation result is obtained. And accurate channel state information is obtained. Simulation results show that the provided ProDrnNet model has more advantages on the accuracy of restoring the channel state information compared with a traditional estimation method, and the channel information can still be accurately reconstructed in various channel environments.

Description

Technical Field

[0001] The present invention belongs to the technical field of channel estimation in an orthogonal frequency division multiplexing (OFDM) system, and is a channel estimation method based on deep residual learning network (ProDrnNet). Background Art

[0002] Orthogonal frequency division multiplexing (OFDM) can effectively resist multipath interference and frequency selective fading by distributing data to multiple orthogonal subcarriers. Its main advantages include high spectrum efficiency, strong anti-interference ability and good adaptability. It has been widely used in wireless communications and broadband networks in recent years. In 5G communication systems, due to the complex and changeable wireless scenarios, the signals obtained at the receiving end will be severely interfered, and it is necessary to accurately estimate the channel information to complete the signal recovery work.

[0003] The pilot-based channel estimation scheme in the OFDM system mainly uses the receiver to estimate the channel based on the received information and the pre-inserted pilot. There are two commonly used estimation methods, namely the least square method and the minimum mean square error. The least square method is simple to implement and is widely used in engineering, but it ignores the influence of noise and interference between subcarriers, and the estimation of channel information is not accurate enough. The MMSE algorithm has excellent channel estimation performance, but the algorithm requires prior information of the channel and is difficult to implement in actual engineering. In addition, other traditional linear channel estimation methods are based on statistical models, with high implementation complexity and poor robustness in complex wireless channel environments. The channel estimation model based on deep learning learns channel characteristics from a large amount of data and can accurately restore the signal. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a channel estimation method based on deep residual learning in an OFDM system. This method requires offline training of the ProDrnNet model first, which learns the complex mapping relationship between the channel information at the input pilot and the complete channel information in the 5G channel. In the online estimation stage, the channel information at the pilot estimated by the least squares method is directly input into the trained ProDrnNet model to obtain the estimated information of the entire channel to complete the channel estimation.

[0005] To achieve the above object, the present invention provides the following technical methods:

[0006] First, the ProDrnNet model is trained offline using the dataset generated by the 5G channel. The specific structure of the model includes the following steps:

[0007] ① Upsampling is performed at the input end, and then the pilot channel features are extracted through a multi-scale feature extraction module. The multi-scale feature module uses 32 1×1 convolution kernels in the first layer, and 32 3×3 convolution kernels in the next two layers. The relu activation function is used after each convolution layer to convert linear information into nonlinear information. After completing the multi-scale feature extraction, the features are fused through a concat layer and the dimension is reduced through a 1×1 convolution layer.

[0008] ②After step ①, it passes through 3 downsampling encoding modules. In the downsampling encoding module, the input data first enters the average pooling layer with a step size of 2 and a size of 2×2 for dimensionality reduction and increasing the receptive field, and then increases the network depth through 2 residual structures. Finally, the output size is adjusted through a convolutional layer. The 3 downsampling encoding modules form a contraction network.

[0009] ③ Jump-connect the output of each downsampling encoding module in step ② to the input of the upsampling decoding module, add it to the serial data and input it into the upsampling decoding module. In the upsampling decoding module, the input data first extracts deep features through a residual structure, and then gradually restores the image size through a deconvolution layer. The three upsampling decoding modules completely restore the image size.

[0010] ④ Add the output of step ③ to the output of step ① and pass it through a convolution layer to obtain the restored image.

[0011] Then, in the online estimation stage, the channel information at the pilot is preliminarily estimated using the received pilot information and the least squares method:

[0012] The channel estimation information at the pilot signal is then used as a low-resolution noisy image input into the trained ProDrnNet model to restore the complete denoised channel information and complete the channel estimation.

[0013] The beneficial effects of the present invention are:

[0014] The present invention relates to a channel estimation method based on ProDrnNet. The method can fully extract the channel response characteristics at the pilot in the OFDM channel estimation system, and it is easier to learn the complex mapping relationship between the pilot channel response estimation value and the entire channel response, so as to obtain a better channel estimation effect. Simulation results show that the method is similar to the channel estimation method based on the ChannelNet model in the literature [1] "SOLTANI M, POURAHMADI V, MIRZAEI A, et al. Deep learning-based channel estimation [J]. IEEE Communications Letters, 2019, 23 (4): 652-655." and the literature [2] "LI L, CHEN H, CHANG HH, et al. Deep presidual learning meets OFDM channel estimation [J]. IEEE Wireless Communications Letters, 2020, 9(5): 615-618." and the channel estimation method based on NE-MCRNNet in the literature [3] "Yuan Jianguo, Zuo Zhihang, Yu Yiran, Su Jie. A channel estimation scheme based on NE-MCRNNet in OFDM system [J / OL]. Optoelectronics·Laser. https: / / link.cnki.net / urlid / 12.1182.O4.20240402.1138.002" have better channel estimation performance. Therefore, in practical applications, the method of the present invention has high utilization value and practical significance for the design of channel estimation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to make the purpose, technical method and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration:

[0016] Figure 1 It is a technical roadmap of the method of the present invention;

[0017] Figure 2 This is a network model diagram of the method proposed in the present invention;

[0018] Figure 3 A supplementary diagram for the downsampling encoding module in the network model;

[0019] Figure 4 A supplementary diagram for the upsampling decoding module in the network model;

[0020] Figure 5 This is a performance comparison chart of the method of the present invention and other deep learning methods; DETAILED DESCRIPTION

[0021] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] 1. Combine with attachment Figure 1 , Attachment Figure 2 , Attachment Figure 3 With attached Figure 4 It is shown that a channel estimation method based on ProDrnNet in OFDM system is completed by the following steps:

[0023] Step 1: Use the dataset generated by the 5G channel offline to train the ProDrnNet model. The specific composition of the model is as follows: Figure 2 , Figure 3 and Figure 4 Instructions, the specific steps are as follows:

[0024] ① Upsampling is performed at the input end, and then the pilot channel features are extracted through a multi-scale feature extraction module. The multi-scale feature module uses 32 1×1 convolution kernels in the first layer, and 32 3×3 convolution kernels in the next two layers. The relu activation function is used after each convolution layer to convert linear information into nonlinear information. After completing the multi-scale feature extraction, the features are fused through a concat layer and the dimension is reduced through a 1×1 convolution layer.

[0025] ②After step ①, it passes through 3 downsampling encoding modules. In the downsampling encoding module, the input data first enters the average pooling layer with a step size of 2 and a size of 2×2 for dimensionality reduction and increasing the receptive field, and then increases the network depth through 2 residual structures. Finally, the output size is adjusted through a convolutional layer. The 3 downsampling encoding modules form a contraction network.

[0026] ③ Jump-connect the output of each downsampling encoding module in step ② to the input of the upsampling decoding module, add it to the serial data and input it into the upsampling decoding module. In the upsampling decoding module, the input data first extracts deep features through a residual structure, and then gradually restores the image size through a deconvolution layer. The three upsampling decoding modules completely restore the image size.

[0027] ④ Add the output of step ③ to the output of step ① and pass it through a convolution layer to obtain the restored image.

[0028] Step 2: In the online estimation stage, the pilot information and the least squares method are used to preliminarily estimate the channel information at the pilot position. The channel estimate at the pilot position in the least squares method is expressed as:

[0029]

[0030] In formula (1), X p is the pilot information of the transmitter, Y p is the pilot information of the receiving end, is the channel estimation information at the pilot, which is expressed in the following form after mathematical transformation:

[0031]

[0032] Step 3: The channel estimation information at the pilot obtained in step 2 is used as a low-resolution noisy image and input into the ProDrnNet model trained in step 1 to restore the complete denoised channel information. The denoised channel information is expressed as:

[0033]

[0034] In formula (3), f(·) is the ProDrnNet model function, and θ is the hyperparameter set of the network model.

[0035] Step 4: The received information is restored using the channel information obtained in step 3, and equalization, demodulation and other operations are performed to finally obtain the denoised information.

[0036] 2. Combine with attachment Figure 5 In order to verify that the invented OFDM channel estimation method has a better estimation performance, the model is simulated using MATLAB software. The simulation parameters are shown in Table 1:

[0037] Table 1 Simulation parameter settings

[0038] Parameters Set Size Number of effective subcarriers 72 Carrier frequency 2.1GHz Channel bandwidth 1.6MHz Modulation QPSK Number of OFDM symbols 14 Number of pilots 48 Channel ETU

[0039] The channel estimation performance of the ProDrnNet model is compared with the channel estimation method based on the ChannelNet model in the literature [1] "SOLTANI M, POURAHMADI V, MIRZAEIA, et al. Deep learning-based channel estimation [J]. IEEE Communications Letters, 2019, 23(4): 652-655." and the channel estimation method based on the ChannelNet model in the literature [2] "LI L, CHEN H, CHANG HH, et al. Deep residual learning meets OFDM channel estimation [J]. IEEE Wireless Communications The performance of the channel estimation method based on the ReEsNet model in the literature [2020, 9(5): 615-618] is compared with that of the channel estimation method based on the NE-MCRNNet in the literature [3] "Yuan Jianguo, Zuo Zhihang, Yu Yiran, Su Jie. A channel estimation scheme based on NE-MCRNNet in OFDM system [J / OL]. Optoelectronics·Laser. https: / / link.cnki.net / urlid / 12.1182.O4.20240402.1138.002".

[0040] Under the same parameter conditions, the minimum mean square error comparison between the signal estimated by different channel estimation models and the actual signal is shown in the figure below: Figure 3 As shown in Figure 2, the ProDrnNet model established in the present invention has superior performance than the ChannelNet model, the ReEsNet model and the NE-MCRNNet model in the range of signal-to-noise ratio of 0 to 30 dB.

[0041] Based on the above simulation verification, it can be seen that the present invention proposes a channel estimation method based on ProDrnNet in OFDM system to address the problem that the current channel estimation system of OFDM system is not accurate enough, is easily affected by channel noise and has poor robustness. The simulation results show that under the same conditions, the channel estimation scheme of the present invention can fully extract the channel response characteristics at the pilot, and it is easier to learn the complex mapping relationship between the pilot channel response estimation value and the entire channel response, so as to obtain a better channel estimation effect.

[0042] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A channel estimation method based on ProDrnNet in an orthogonal frequency-division multiplexing (OFDM) system; this method first uses the data set generated by the 5G channel offline to train the ProDrnNet model, and then uses the pilot information at the signal receiving end to preliminarily estimate the pilot channel using the least squares method when receiving the signal. The channel estimation information at the pilot is then used as a low-resolution noisy image input into the trained ProDrnNet model for denoising and reconstructing the channel information. Finally, the received signal is equalized and demodulated to restore it to the original signal; The specific steps are as follows: Step 1: Use the data set generated by the 5G channel offline to train the ProDrnNet model. The specific structure of the model includes the following steps: ① Upsampling is first performed at the input end, and then the pilot channel features are extracted through a multi-scale feature extraction module. The multi-scale feature module uses 32 1×1 convolution kernels in the first layer, and 32 3×3 convolution kernels in the next two layers. The relu activation function is used after each convolution layer to convert linear information into nonlinear information. After completing the multi-scale feature extraction, the features are fused through a concat layer and the dimension is reduced through a convolution layer of size 1×1; ②After step ①, the data is input into the downsampling coding module through three downsampling coding modules. First, it enters the average pooling layer with a step size of 2 and a size of 2×2 to reduce the dimension and increase the receptive field. Then, the network depth is increased through two residual structures. The residual structures are all composed of convolutional layers with a size of 3×3. Finally, the output size is adjusted through a convolutional layer. The three downsampling coding modules form a contraction network. ③ Jump connect the output of each downsampling encoding module in step ② to the input of the upsampling decoding module, add it to the serial data and input it into the upsampling decoding module. In the upsampling decoding module, the input data first extracts deep features through a residual structure, and then gradually restores the image size through a deconvolution layer. The three upsampling decoding modules completely restore the image size; ④ Add the output of step ③ to the output of step ① and pass it through a convolution layer to obtain the restored image; Step 2: In the online estimation stage, the pilot information and the least squares method are used to preliminarily estimate the channel information at the pilot position. The channel estimate at the pilot position in the least squares method is expressed as: In formula (1), X p is the pilot information of the transmitter, Y p is the pilot information of the receiving end, is the channel estimation information at the pilot, which is expressed in the following form after mathematical transformation: Step 3: The channel estimation information at the pilot obtained in step 2 is used as a low-resolution noisy image and input into the ProDrnNet model trained in step 1 to restore the complete denoised channel information. The denoised channel information is expressed as: In formula (3), f() is the ProDrnNet model function, and θ is the hyperparameter set of the network model; Step 4: The received information is restored using the channel information obtained in step 3, and equalization, demodulation and other operations are performed to finally obtain the denoised information.

Images