Adaptive satellite channel estimation method based on convolutional neural network

Abstract

The invention relates to an adaptive satellite channel estimation method based on a convolutional neural network, which belongs to the field of channel estimation, and comprises the following steps: S1, constructing a system model of an uplink satellite channel, and generating a channel matrix according to the system model; s2, simulating and generating a receiving signal of a receiving end according to the established system model; s3, according to the pilot frequency, using a least square estimation method to carry out preliminary estimation on the channel; s4, constructing a channel estimation model of an adaptive signal-to-noise ratio based on the convolutional neural network; s5, taking the initial estimation value of the satellite channel as the input of the model, taking the channel characteristic generated according to the uplink satellite channel system model as a label, and taking the channel estimation value as the output of the model; and S6, dividing the label and the input data into a test set and a training set, taking a normalized mean square error as a loss function, training the channel estimation model until the channel estimation model converges, and checking the performance of the channel estimation model by using the test set.

Description

Technical Field

[0001] The invention belongs to the technical field of channel estimation and relates to an adaptive satellite channel estimation method based on a convolutional neural network. Background Art

[0002] In recent years, with the development of broadband wireless communication technology, satellite communication has gradually become an important part of the global communication infrastructure, and its development prospects are broad, covering multiple fields and application scenarios. Satellite communication can cover any place on the earth, including remote areas, oceans and mountainous areas that are difficult to cover with traditional ground communications. In recent years, artificial intelligence technology has developed rapidly, and convolutional neural networks have shown great potential in many fields with their excellent nonlinear fitting ability. Through data-driven methods, neural networks can implicitly model high-dimensional complex models, which shows significant advantages in solving the satellite channel mismatch problem caused by nonlinear signals in the carrier channel and performing channel estimation. Traditional signal processing methods often rely on predefined mathematical models to describe channel characteristics and perform channel estimation. However, when faced with channel mismatch caused by nonlinear signals, these methods are usually difficult to cope with complex satellite channel environments. The introduction of neural networks has changed this situation. By utilizing a large amount of training data, neural networks can automatically learn and capture nonlinear relationships in the channel, generate more accurate channel estimation models, and thus effectively solve the channel estimation problem. The nonlinear fitting ability of neural networks gives them natural model matching advantages in the following aspects: 1. Adaptability to complex channel environments: Neural networks can handle highly complex and changeable channel environments, and capture subtle changes and nonlinear features that are difficult to describe with traditional models. 2. Real-time and robustness: Through online learning and updating, neural networks can adapt to changes in channel states in real time and maintain efficient and stable communication performance. 3. High-dimensional data processing capabilities: Neural networks are able to process high-dimensional and large-scale data, extract useful features from them, and help build accurate channel models. 4. Automated modeling: Unlike traditional models that require human intervention and fine-tuning, neural networks achieve automated modeling through an end-to-end training process, reducing human intervention and complexity.

[0003] Channel estimation is a key link in the satellite communication process. When estimating the channel of a satellite communication link, the rapid movement of the satellite will cause a serious Doppler effect, which will cause the channel to change nonlinearly and have a serious impact on the channel estimation value and the signal-to-noise ratio estimation value. The traditional least squares algorithm alone cannot guarantee the accuracy of the channel estimation value. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide an adaptive satellite channel estimation method based on convolutional neural network. Through the deep learning method, the powerful nonlinear fitting ability of the neural network is utilized, and a signal-to-noise ratio adaptive module is added to the neural network to train an adaptive signal-to-noise ratio neural network model, so that a unified model can be used at the receiving end to estimate the satellite channel.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] An adaptive satellite channel estimation method based on convolutional neural network comprises the following steps:

[0007] S1: construct a system model of the uplink satellite channel and generate a channel matrix based on the system model;

[0008] S2: According to the established system model, simulate and generate the receiving signal of the receiving end;

[0009] S3: Based on the pilot signal, the least square estimation method is used to make a preliminary estimate of the channel;

[0010] S4: Based on convolutional neural network, a channel estimation model with adaptive signal-to-noise ratio is constructed;

[0011] S5: using the preliminary estimation value of the satellite channel obtained in S3 as the input of the channel estimation model, using the channel feature generated according to the uplink satellite channel system model as a label, and using the channel estimation value as the output of the channel estimation model;

[0012] S6: Divide the labels and input data in step S5 into a test set and a training set, use the normalized mean square error as the loss function, train the channel estimation model in step S4 until the channel estimation model converges, and use the test set to verify the performance of the channel estimation model.

[0013] Furthermore, in the system model of the uplink satellite channel, the satellite end is equipped with r s antennas, the user end is equipped with r u The root antenna, the user end is the transmitter, and the satellite end is the receiver, then the system model is expressed as:

[0014]

[0015] in Represents the received signal, P S is the transmission power, S represents the transmitted signal and is the channel matrix that obeys the Rayleigh distribution, f is the Doppler frequency deviation caused by the moving speed of the low-Earth orbit satellite, W∈ is complex Gaussian white noise; β represents the unified parameter of single-beam radio wave propagation, which is expressed as:

[0016]

[0017] L FSL is the free space propagation loss, f c is the carrier frequency, c is the speed of light, d is the distance between the satellite and the user end, K b =1.38×10 -23 , T represents the noise temperature at the user, ω represents the carrier bandwidth, μ=2.07123sin(θ) / sin(θ 3dB ), J1 and J3 are first-order and third-order Bessel functions respectively, θ is the off-axis angle, θ 3dB is the angle corresponding to the 3dB beam width; g max is the maximum antenna gain when the off-axis is 0, A is the antenna area, η represents the antenna efficiency, h, R e are the satellite altitude and the earth radius respectively, and γ is the elevation angle.

[0018] Further, in step S3, the least squares estimation method is used to make a preliminary estimate of the channel, which is expressed as:

[0019] α ls =S H ·(S·S H ) -1

[0020]

[0021] Among them, S is the transmitted signal, R(f) is the received signal with frequency deviation, α ls It is the intermediate process expression of the least squares algorithm.

[0022] Furthermore, the channel estimation model in step S4 is composed of N NetBlocks, each NetBlock is composed of two subnets, subnet one is composed of convolution blocks with different convolution kernels, a signal-to-noise ratio adaptive module (SNR adaptive) and a self-attention module (Sa-block), and subnet two is composed of convolution blocks with different convolution kernels and a signal-to-noise ratio adaptive module (SNR adaptive).

[0023] Furthermore, the subnetwork 1 consists of one 32×3×3 and five 64×3×3 convolution blocks, one signal-to-noise ratio adaptive module SNR-adaptive and one self-attention module Sa-block; each convolution block contains a convolution layer Conv, a batch normalization layer BN and an activation layer ReLU; Sa-block consists of a global average pooling operation Gp, a channel reduction operation Hd, and a channel upsampling operation Hu; SNR-adaptive consists of a pooling layer, an SNR splicing layer, and two multi-layer perceptrons MLP.

[0024] Furthermore, the second subnet is a residual network, which consists of three 64×3×3 convolution blocks, one 64×1×1 fully connected layer and a signal-to-noise ratio adaptive module SNR-adaptive.

[0025] Further, the training and testing steps in step S6 are as follows:

[0026] S61: Set the number of training epochs;

[0027] S62: using the least squares estimation value of the channel as input data of the channel estimation model, and adding the prior information of the signal-to-noise ratio to the signal-to-noise ratio adaptive module in the channel estimation model;

[0028] S63: Use SGD optimizer and gradient descent method;

[0029] S64: The output of the channel estimation model is a detailed estimate of the satellite channel;

[0030] S65: Calculate the normalized mean square error between the input of the channel estimation model and the channel characteristic matrix;

[0031] S66: until the number of training times reaches epoch, record the channel estimation model parameter set of the optimal normalized mean square error;

[0032] S67: Testing the channel estimation model using the test set, where the test indicator is the mean square error between the real channel and the channel estimation model input.

[0033] The beneficial effects of the present invention are as follows: the present invention creatively uses a deep learning method, utilizes the powerful nonlinear fitting ability of a neural network, and adds a signal-to-noise ratio adaptive module to the neural network to train a neural network model with adaptive signal-to-noise ratio, so that a unified model can be used at the receiving end to estimate the satellite channel.

[0034] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0036] Figure 1 This is a flow chart of channel estimation of the present invention;

[0037] Figure 2A satellite communication system model diagram in step 1 of the present invention;

[0038] Figure 3 The overall structure diagram of the neural network model in step 4 of the present invention;

[0039] Figure 4 This is a structural diagram of the self-attention module in step 4 of the present invention;

[0040] Figure 5 This is a structural diagram of the adaptive signal-to-noise ratio module in step 4 of the present invention. DETAILED DESCRIPTION

[0041] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0042] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0043] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0044] The specific process of satellite channel estimation based on adaptive signal-to-noise ratio of convolutional neural network proposed by the present invention is as follows: Figure 1 As shown, it includes a least squares estimator and a NetBlock module. The received signal is inputted into the estimated value of the satellite channel after passing through the least squares estimator and NetBlock.

[0045] S1: Establish a satellite channel system model, such as Figure 2 As shown, the satellite is equipped with r s antenna, the user is equipped with r s antennas, where P Sis the transmission power. Assuming that the process is an uplink process, that is, the user end is the transmitter and the satellite end is the receiver, then S represents the transmitted signal and Represents the received signal, is the channel matrix that obeys the Rayleigh distribution. f is the Doppler frequency deviation caused by the moving speed of the low-Earth orbit satellite. In the low-speed scenario, the moving speed of the user end is much smaller than the moving speed of the low-Earth orbit satellite, and the Doppler frequency deviation generated by the user end can be ignored. is complex white Gaussian noise.

[0046]

[0047] β represents the uniform parameter for single-beam radio wave propagation, which can be determined by the following model:

[0048]

[0049] In expression (2), L FSL is the free space propagation loss, determined by equation (4), f c is the carrier frequency, c is the speed of light, d is the distance between the satellite and the user end, K b =1.38×10 -23 , T represents the noise temperature at the user, ω represents the carrier bandwidth, and in expression (3) μ = 2.07123sin(θ) / sin(θ 3dB ), J1 and J3 are first-order and third-order Bessel functions respectively, θ is the off-axis angle, θ 3dB is the angle corresponding to the 3dB beam width. max is the maximum antenna gain when the off-axis is 0, g max Determined by expression (6), where A is the antenna area and η represents the antenna efficiency. In (5), h, R e are the satellite altitude and the radius of the earth respectively, γ is the elevation angle, and c is the speed of light constant.

[0050] S2: Based on the established system model, simulate and generate the receiving signal at the receiving end. According to the system model pointed out by S1, Matlab generates the receiving signal and the corresponding channel matrix.

[0051] S3: Use the traditional least squares method to estimate the satellite channel. The least squares method is as follows:

[0052] α ls =S H ·(S·S H ) -1 (7)

[0053]

[0054] S4: Determine the neural network model for adaptive signal-to-noise ratio based on convolutional neural network, such as Figure 3 As shown, the entire neural network consists of N NetBlocks, each of which has two subnets. Subnet 1 consists of 1 32×3×3 and 5 64×3×3 convolutional blocks, 1 SNR-adaptive module and 1 self-attention module. Each convolutional block contains a convolutional layer (Conv), a batch normalization layer (BN) and an activation layer (ReLU), as shown in Figure 2. Figure 4 As shown in Figure 1, Sa-block consists of Gp, Hd, and Hu operations, where Gp is ​​a global average pooling operation, Hd is a channel reduction operation, and Hu is a channel upsampling operation. Figure 5 As shown in the figure, SNR-adaptive consists of 1 pooling layer, SNR concatenation layer, and two multi-layer perceptrons (MLP). Subnet 2 is a residual network consisting of 3 64×3×3 convolution blocks, a 64×1×1 fully connected layer, and an SNR-adaptive.

[0055] S5: The preliminary estimate of the satellite channel obtained in S3 is used as the input of the model in S4, and the channel characteristics generated according to the uplink satellite system model are used as labels. The output of the model is the final channel estimate.

[0056]

[0057] S6: The labels and input data in S5 are divided into a test set and a training set in a ratio of 1:4, and the normalized mean square error is used as the loss function for training. According to the above method, the preliminary estimate is obtained by the least squares (LS) method, and it is used as the input of the deep learning network model for training. The specific training process is as follows: First, the training samples are derived from the LS estimate described in formula (8). In order to adapt to the input format of the deep learning network, the real and imaginary parts of the LS estimate are separated into two channels and used as the input of the network. At the same time, the real channel response is used as a label to guide the training of the model. We choose the normalized mean square error (NMSE) as the loss function, as shown in formula (10), and improve the performance of the model by minimizing the NMSE between the true value of the channel response and the LS estimate. Through such a training process, the deep neural network can learn the complex relationship between the LS estimate and the true channel response, thereby effectively reducing the negative impact of the Doppler effect and additive noise on the channel estimation, and finally achieving high-precision channel estimation, and inputting the SNR into the network as information to achieve signal-to-noise ratio adaptation.

[0058]

[0059] Training samples According to the above channel operation, the data flow into It is divided into two parts. The first part has several layers of convolution to fully extract data features, making the data dimension 64×w×h. After SNR-adaptive, part of the data stream goes through the pooling layer and is spliced ​​with the added SNR information. After passing through the MLP, it is multiplied and added with the original data stream. After passing through the Gp layer of Sa-block, a 64×1×1 channel feature is obtained, which is converted to 64 / m×1×1 through the Hd layer, where m is the channel reduction factor. The Hu layer makes the channel feature dimension rise back to 64×1×1 and multiplies it with the original data stream. Finally, as Input, output is

[0060] S7: Train until the model converges and use the test set to verify the model performance. The training parameters are shown in Table 1.

[0061] Table 1

[0062]

[0063] In the above embodiments, the description's reference to "this embodiment" indicates that a particular feature, structure, or characteristic described in conjunction with the embodiment is included in at least some embodiments, but not necessarily all embodiments. Multiple occurrences of "this embodiment" do not necessarily all refer to the same embodiment.

[0064] In the above-described embodiments, although the invention has been described in conjunction with specific embodiments of the invention, many substitutions, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art based on the foregoing description. For example, other storage structures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed. Embodiments of the invention are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of the appended claims.

[0065] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, any one of the methods in this embodiment is implemented.

[0066] This embodiment also provides an electronic terminal, including: a processor and a memory;

[0067] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes any one of the methods in this embodiment.

[0068] The computer-readable storage medium in this embodiment can be understood by ordinary technicians in this field: all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other media that can store program codes.

[0069] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication with each other. The memory is used to store computer programs, the communication interface is used to communicate, and the processor and the transceiver are used to run computer programs so that the electronic terminal executes each step of the above method.

[0070] In this embodiment, the memory may include a random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0071] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0072] The present invention can be used in many general or special computing system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like.

[0073] The present invention may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. An adaptive satellite channel estimation method based on convolutional neural network, characterized in that: The following steps are involved: S1: construct a system model of the uplink satellite channel and generate a channel matrix based on the system model; S2: According to the established system model, simulate and generate the receiving signal of the receiving end; S3: Based on the pilot signal, the least square estimation method is used to make a preliminary estimate of the channel; S4: Based on convolutional neural network, a channel estimation model with adaptive signal-to-noise ratio is constructed; S5: using the preliminary estimation value of the satellite channel obtained in S3 as the input of the channel estimation model, using the channel feature generated according to the uplink satellite channel system model as a label, and using the channel estimation value as the output of the channel estimation model; S6: Divide the labels and input data in step S5 into a test set and a training set, use the normalized mean square error as the loss function, train the channel estimation model in step S4 until the channel estimation model converges, and use the test set to verify the performance of the channel estimation model.

2. The method for adaptive satellite channel estimation based on convolutional neural network according to claim 1, characterized in that: In the system model of the uplink satellite channel, the satellite end is equipped with r s antennas, the user end is equipped with r u The root antenna, the user end is the transmitter, and the satellite end is the receiver, then the system model is expressed as: in Represents the received signal, P S is the transmission power, S represents the transmitted signal and is the channel matrix that obeys the Rayleigh distribution, f is the Doppler frequency deviation caused by the moving speed of the low-Earth orbit satellite, is complex Gaussian white noise; β represents the unified parameter of single-beam radio wave propagation, which is expressed as: L FSL is the free space propagation loss, f c is the carrier frequency, c is the speed of light, d is the distance between the satellite and the user end, K b =1.38×10 -23 , T represents the noise temperature at the user, ω represents the carrier bandwidth, μ=2.07123sin(θ) / sin(θ 3dB ), J1 and J3 are first-order and third-order Bessel functions respectively, θ is the off-axis angle, θ 3dB is the angle corresponding to the 3dB beam width; g max is the maximum antenna gain when the off-axis value is 0, A is the antenna area, η represents the antenna efficiency, h, R e are the satellite altitude and the earth radius respectively, and γ is the elevation angle.

3. The method for adaptive satellite channel estimation based on convolutional neural network according to claim 1, characterized in that: In step S3, the least squares estimation method is used to make a preliminary estimate of the channel, which is expressed as: α ls =S H ·(S·S H ) -1 Among them, S is the transmitted signal, R(f) is the received signal with frequency deviation, α ls It is the intermediate process expression of the least squares algorithm.

4. The method for adaptive satellite channel estimation based on convolutional neural network according to claim 1, characterized in that: In step S4, the channel estimation model is composed of N NetBlocks, each NetBlock is composed of two subnets, subnet one is composed of convolution blocks with different convolution kernels, a signal-to-noise ratio adaptive module SNR adaptive and a self-attention module Sa-block, and subnet two is composed of convolution blocks with different convolution kernels and a signal-to-noise ratio adaptive module SNR adaptive.

5. The method for adaptive satellite channel estimation based on convolutional neural network according to claim 4, characterized in that: The subnetwork 1 consists of 1 32×3×3 and 5 64×3×3 convolution blocks, 1 signal-to-noise ratio adaptive module SNR-adaptive and 1 self-attention module Sa-block; each convolution block contains a convolution layer Conv, a batch normalization layer BN and an activation layer ReLU; Sa-block consists of a global average pooling operation Gp, a channel reduction operation Hd, and a channel upsampling operation Hu; SNR-adaptive consists of 1 pooling layer, an SNR splicing layer, and 2 multi-layer perceptrons MLP.

6. The method for adaptive satellite channel estimation based on convolutional neural network according to claim 4, characterized in that: The second subnet is a residual network, which consists of three 64×3×3 convolution blocks, one 64×1×1 fully connected layer and a signal-to-noise ratio adaptive module SNR-adaptive.

7. The method for adaptive satellite channel estimation based on convolutional neural network according to claim 1, characterized in that: The training and testing steps in step S6 are as follows: S61: Set the number of training epochs; S62: using the least squares estimation value of the channel as input data of the channel estimation model, and adding the prior information of the signal-to-noise ratio to the signal-to-noise ratio adaptive module in the channel estimation model; S63: Use SGD optimizer and gradient descent method; S64: The output of the channel estimation model is a detailed estimate of the satellite channel; S65: Calculate the normalized mean square error between the input of the channel estimation model and the channel characteristic matrix; S66: until the number of training times reaches epoch, record the channel estimation model parameter set of the optimal normalized mean square error; S67: Testing the channel estimation model using the test set, where the test indicator is the mean square error between the real channel and the channel estimation model input.

Images