Channel estimation simulation method based on multi-feature input GRU

Abstract

The invention relates to the field of communication technology research, in particular to a channel estimation simulation method based on multi-feature input GRU, which comprises the following steps of: collecting channel estimation at a pilot frequency under a real scene (such as a high-speed scene), arranging a pilot frequency signal in an OFDM (Orthogonal Frequency Division Multiplexing) time-frequency domain resource grid, and obtaining a channel estimation result; the time, the frequency, the antenna subscript and the channel estimation value at the pilot frequency serve as input data of model training; inputting the training data into a multi-feature GRU model for training to obtain a pre-trained GRU channel estimation model; applying the trained channel estimation model on an online simulation platform, and when the MSE is increased, performing online fine adjustment on GRU model parameters through a Reptile meta-learning algorithm to obtain a real-time GRU channel estimation model; the whole channel estimation system has channel estimation and real-time adaptation capabilities by combining a Reptile algorithm and a GRU model; the structure is simpler and clearer, and parameters are fewer, so that model training is more efficient, and the method is easier to implement on a simulation platform with limited resources.

Description

Technical Field

[0001] The present invention relates to the field of communication technology research, and specifically to a channel estimation simulation method based on multi-feature input GRU. Background Art

[0002] In recent years, with the continuous development of China's economy and technology, the emergence of 4G and 5G networks has improved people's communication capabilities and is still evolving. More and more services are also more dependent on high-quality and high-efficiency wireless mobile communication systems. However, in some special scenarios such as high-speed mobile scenarios and urban dense scenarios, frequency offset and phase change caused by the Doppler effect and multipath effect still pose significant challenges to the performance of communication systems.

[0003] In the traditional channel estimation process, model-based methods such as the least squares method and the maximum likelihood estimation method are usually used to predict and compensate for signal distortion. These methods rely on accurate modeling of channel characteristics, but in the face of complex scenarios such as high-speed movement, multipath effect, and Doppler frequency shift, traditional models may not be able to fully capture the dynamic changes of the channel, resulting in a decrease in estimation accuracy.

[0004] By simulating the connection and information transmission of neurons and training the network model with input data, neural networks can learn and recognize complex patterns. This method can better adapt to the non-linear and time-varying characteristics of the channel, especially in complex environments where traditional methods are difficult to handle. Through deep learning, neural networks can be trained to recognize patterns and structures in the channel, thus achieving more efficient and accurate channel estimation in various environments. Patent No. CN113285899A discloses a time-varying channel estimation method and system based on deep learning. By constructing a channel model, using historical channel information and received pilot signals, combining the least squares estimation (LS) and first-order autoregressive (AR) models, and the backpropagation neural network (BP neural network) for channel estimation. In the offline training stage, this method optimizes network parameters through the quantized conjugate gradient descent method to achieve relatively efficient online channel estimation; through deep learning techniques, especially the BP neural network, to capture channel change characteristics to improve estimation accuracy. However, in this method, the input of training data uses the received complex signals, and the complex signals at the receiving end generally contain time, frequency, and user data information or pilot information. If the same complex signal is received in different time-frequency domain resource grids, the same channel estimation value will be obtained. Therefore, bringing user data into the training process will lead to inaccurate channel estimation models; moreover, this method is only applicable to high-speed mobile scenarios and has great limitations. Patent No. CN110266620A discloses a channel estimation method for 3D MIMO-OFDM systems based on convolutional neural networks. This method trains data based on the CECNN convolutional neural network to obtain a channel estimation model, and finally inputs data and predicts the complete channel response value of the 3D MIMO-OFDM system. The CECNN convolutional neural network used in this method has three convolutional layers, which will lead to increased training overhead and longer training time. In addition, this method is also only applicable to the 3D MIMO scenario and also has certain limitations. Therefore, based on these problems, it is particularly necessary to simplify the network structure to adapt to the massive pilot information in the communication process and reduce the training overhead; then, it is also necessary to optimize the model input to remove the influence of user data on channel estimation and improve the accuracy of simulation; finally, it is particularly necessary to improve the generalization ability of the neural network to adapt to different scenarios. Summary of the Invention

[0005] The main objective of the present invention is to improve the accuracy of LTE physical channel estimation and the adaptability of the algorithm, accelerate the training speed of the channel estimation model, reduce the training overhead, improve the real-time performance and versatility of the model, and provide a channel estimation simulation method based on multi-feature input GRU to solve the problems proposed in the above background technology.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] A channel estimation simulation method based on GRU with multi-feature input, the method comprising:

[0008] S100. Design an organization method for training data at pilots, shield the influence of received pilot data on the channel estimation value, and reduce the factors affecting the channel estimation result to time, frequency, and the antenna where the pilot is located;

[0009] S200. Design a GRU model based on multi-feature input, and simultaneously process three key features affecting the channel estimation value to more comprehensively capture the dynamic characteristics of channel changes;

[0010] S300. When performing online channel estimation, use the Reptile algorithm to learn the best estimate of the current model parameters online, and obtain a GRU network model with the best parameters and the channel estimation at other positions except the pilots.

[0011] Preferably, in the implementation process of organizing training data at pilots in S100:

[0012] First, determine the channel estimation scenario, use the MATLAB simulation software to obtain pilot data through the LS method or collect DMRS pilot data within a certain time period in a specific scenario. Screen and clean the collected pilot data, and fill in the missing values and outliers with the mean value to ensure the effectiveness of the training data.

[0013] Further, organize the time, frequency, and antenna information at the pilots. Denote the DMRS pilot signal on the i-th antenna at time slot i and frequency f as P i t,f , then the training task formed at this pilot is:

[0014]

[0015] where t P represents the time slot where this pilot is located, f P represents the frequency domain position where this pilot is located, i P represents the antenna subscript where this pilot is located, H P represents the channel estimation value measured at the receiving end for this pilot, which is in complex form.

[0016] In the simulation environment, since the pilot signal obtained at the receiving end is represented in complex form, it is necessary to split the channel estimation of pilot P i t,f to obtain the real part representation and the imaginary part representation of the channel estimation at this pilot, denoted as:

[0017]

[0018] Divide the training data into two groups, where the training data for each step size is represented as:

[0019]

[0020] Since the DMRS signals in different time slots are arranged at equal intervals, and the interval between DMRS signals in two consecutive time slots is 7 OFDM symbols, i.e., 0.5 ms, taking 0.5 ms as the step size of the input time series of the GRU model, the real part and imaginary part input matrices are respectively:

[0021]

[0022] Preferably, in the GRU model based on multi-feature input designed in S200:

[0023] First, determine the algorithm structure. The GRU algorithm sets two gated units, an update gate and a reset gate, and controls the influence degree of historical pilot information on the current pilot information through these two gated units.

[0024] Next, design the update gate. The update gate is used to control how much of the previous pilot's state information is retained in the current hidden transformation and how much new pilot state information is introduced. The goal is to output a value between 0 and 1 through the sigmoid function, and this value determines the weight of retaining the hidden state of the previous time step. If the value of the update gate is close to 1, it means more historical information is retained; if it is close to 0, it means more historical pilot information is forgotten and new pilot information is introduced. By referring to the sigmoid function, the transition between weights is smooth. This smooth transition helps the model find a balance between retaining historical pilot information and introducing new pilot information, avoiding drastic changes. Due to the use of non-linear decision-making, this will help the model capture more complex channel change characteristics and patterns, and improve the adaptability of the model in different environments. By introducing the update gate, the GRU model can more flexibly handle the relationship between historical pilot information and new pilot information when dealing with channel estimation problems. The calculation formula of the update gate is:

[0025] z t =σ(W z ·[h t-1 ,x t,f,i )

[0026] where, x t,f,i represents the pilot signal input at the current time step (i.e., the t-th time slot), f and i respectively represent the frequency and antenna subscript where the current pilot is located. σ represents the sigmoid activation function, W z is the weight matrix of the update gate, and h t-1 is the pilot hidden state of the previous time step. x t,f,iThe input dimension is three, indicating that under this model, the result of channel estimation is only affected by three factors: time, frequency, and the antenna where it is located, eliminating the interference caused by using the received pilot signal carrying pilot data as training data in traditional convolutional neural networks, and further improving the accuracy of the network model.

[0027] Furthermore, the reset gate of the GRU model is designed. Similar to the update gate, the calculation formula of the reset gate is as follows:

[0028] r t = σ(W r · [h t-1 , x t,f,i )

[0029] Among them, W r is the weight matrix of the reset gate. The output of the reset gate is element-wise multiplied with the previous hidden state h t-1 to weight the pilot historical information and control the influence of historical pilot information on the current output. In traditional RNNs, errors accumulate over time, leading to the long-term dependence problem. The reset gate reduces this error propagation by allowing the model to reset historical pilot information when necessary, improving the generalization ability of the model.

[0030] Furthermore, the candidate hidden state is designed, and the formula is as follows:

[0031]

[0032] Among them, W is the weighted matrix of the candidate hidden state, and * represents element-wise multiplication. The candidate hidden state provides a proposal for a new hidden state at the current time step. If the reset gate has forgotten most of the pilot historical information, then the candidate hidden state will be mainly determined by the current input.

[0033] Furthermore, the model will calculate the final hidden state. The current final hidden state h t is the weighted sum of the hidden state h t-1 at the previous time step and the candidate hidden state at the current time step. The weight is determined by the output z t of the update gate. The formula is as follows:

[0034]

[0035] This weighted sum allows the GRU to dynamically adjust the degree of dependence on historical pilot information and new pilot information, thus better handling the long-term dependence problem in sequence data. In this way, the GRU can effectively capture the dynamic changes in channel estimation at the pilot and improve the performance of the model.

[0036] Further, input the real - part input matrix Input re and the imaginary - part input matrix Input im into the above - mentioned GRU model respectively for training to obtain optimized model parameters and a trained GRU model. These model parameters include the weight matrices and bias terms of the update gate, reset gate, and candidate hidden state. The trained GRU model can effectively capture the temporal dependencies in the sequence based on the input real - part and imaginary - part data and accurately estimate the channel state or other target variables. When performing online channel estimation, first, form three - dimensional data of the time slot, frequency, and antenna subscript where the received pilot signal is located and input it into the trained GRU model to obtain the channel estimation value at the pilot. Then, obtain the channel estimation value of the PUSCH channel through an interpolation method.

[0037] Preferably, when performing online channel estimation in S300, the Reptile algorithm is used to efficiently update the parameters of the pre - trained GRU network model:

[0038] First, determine the structure of the Reptile algorithm: The Reptile algorithm consists of a base learner and a meta - learner. The goal of the base learner is to quickly learn from M online - received pilot signals and adapt to the current pilot signals through a small number of gradient updates. After the task training of the base learner is completed at the previous moment, the learned parameters are fed back to the meta - learner for meta - level updates. The meta - learner updates the parameters of the GRU neural network by collecting the parameter update information of the base learner on multiple tasks. Such an update strategy enables the model to still have good real - time performance and channel estimation performance in the face of a complex channel environment. Further, design the parameter estimation of the base learner, and the formula is as follows:

[0039]

[0040] where is the loss function of the base learner on task T j , θ is the initial value of the parameters provided by the meta - learner to the base learner, is the parameter estimation value obtained after M iterations of updating the parameters in the base learner. Finally, update the parameters in the meta - learner:

[0041]

[0042] Through the Reptile algorithm, the parameters in the GRU network can be fine-tuned in real time. Moreover, due to the simpler structure of the Reptile algorithm, compared with the traditional MAML algorithm, the Reptile algorithm avoids the calculation of second-order derivatives and reduces the computational complexity. Especially for large-scale signal transceiver systems with 8 to 64 antennas, Reptile can significantly reduce the calculation time and enable the GRU channel estimation model to have better generalization ability.

[0043] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0044] First, the present invention collects the channel estimation and its time-frequency domain information at the pilot in a real scenario (such as a high-speed scenario), and classifies and labels the pilot signals in the OFDM time-frequency domain resource grid to obtain the time, frequency, antenna subscript at this pilot, and channel estimation value of the DMRS pilot signal as the input data for model training. Then, the training data is input into the model for training to obtain an optimized multi-feature GRU channel estimation model. Finally, the trained channel estimation model is applied on an online simulation platform, and the Reptile meta-learning algorithm is used to perform online fine-tuning on the model to obtain an optimized multi-feature GRU channel estimation model. This model has the ability of real-time update and adaptive adjustment. Compared with the original channel estimation algorithms such as LS, the system of the present invention can significantly improve the accuracy and effectiveness of channel estimation in a specific environment and improve the generalization ability of the model. By using the multi-feature GRU network and the Reptile meta-learning algorithm, this system can not only quickly adapt to new tasks with a small number of samples, but also achieve efficient online fine-tuning in practical applications to further optimize the performance of channel estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0046] Figure 1 is the algorithm flowchart of a channel estimation simulation method based on multi-feature input GRU of the present invention;

[0047] Figure 2 is the flowchart of a channel estimation simulation method based on multi-feature input GRU of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] Please refer to Figure 1 - Figure 2 , the present invention provides a technical solution:

[0050] Embodiment 1:

[0051] As Figure 1 shown, the present invention includes the following steps:

[0052] S100. Design an organization method for training data at the pilot, which effectively shields the influence of received pilot data on the channel estimation value, reduces the factors affecting the channel estimation result to time, frequency, and the antenna where the pilot is located, simplifies data training, and improves the accuracy of the subsequent GRU model.

[0053] Preferably, first, select DMRS as the pilot signal, and the channel to be estimated is the PUSCH channel. Since DMRS is evenly distributed in each subframe with a step size of 0.5 ms, compared with selecting pilots at other positions, it can help simplify the screening of training data. Then, label and sort the time-frequency domain, antenna subscript where the current pilot is located, and the channel estimation value at this pilot to obtain the original training input data. Subsequently, split the channel estimation value to obtain the real and imaginary parts of the channel estimation at the pilot, which are respectively composed of two sets of training data. This organization method of training data effectively shields the interference of pilot data on the model and avoids the situation where the receiving end calculates the same channel estimation result when receiving the same complex pilot signal at different time-frequency domain resources, thereby improving the accuracy and effectiveness of channel estimation.

[0054] Taking a specific example, in the MATLAB simulation environment, for an 8-antenna MIMO system, the receiving end receives a certain DMRS pilot signal, expressed as 1 + 6i, and the channel estimation at this pilot is calculated as 9 + 2i by the LS method. Since the channel estimation characterizes the characteristics of the channel and has nothing to do with the data at the pilot. And the pilot signal 1 + 6i carries three kinds of information: pilot data, frequency information, and time information. If directly used as training data input into the model, it will greatly reduce the accuracy of model training.

[0055] Taking this pilot as an example, the present invention obtains the position where the current pilot is located through the function input parameter as

[0056] subframe = 3, slot = 1, frequency = 15 kHz, and antennaIndex = 1. Through subframe and slot, it can be calculated that the current pilot is at the 7th time slot. Then the original training data at this pilot can be expressed as: T j = [7, 1, 15, 1, 9 + 2i] T , where j represents the j-th group of data in the training task dataset. By this method, the complete original training dataset is obtained as T = [T 1 , T 2 , … T j , … T N T . Compared with directly using the received pilot as training data originally, the training data organization method of the present invention removes the interference of received data on calculating channel estimation. In addition, adding the antenna dimension to the training data can also expand the generality of the subsequent model.

[0057] S200. Design a GRU model based on multi-feature input. Different from the traditional time single-feature LSTM algorithm, the GRU model based on multi-feature input can simultaneously process three key features affecting the channel estimation value, so as to more comprehensively capture the dynamic characteristics of channel changes.

[0058] Preferably, a GRU model based on multi-features is designed. The output and input of the model are both real signals. The specific implementation scheme is as follows: First, split the collected complex channel estimation values to obtain real part data and imaginary part data; subsequently, input the real part data and imaginary part data into the model for training respectively to obtain a real part GRU model and an imaginary part GRU model; during online verification, by inputting the time coordinate and frequency coordinate of the current RE into the real part GRU model and the imaginary part GRU model simultaneously, real part channel estimation values and imaginary part channel estimation values are obtained; finally, splice the obtained real part and imaginary part channel estimation values to obtain the complete channel response value.

[0059] Taking a specific example: For the above original pilot training data T = [T 1 , T 2 , … T j , … T N T , in MATLAB, use the real function and imag function to obtain the real part training data and imaginary part training data at the pilot respectively: T re = real(T), T im = imag(T), and input T re and T im ​​Two GRU models with the same structure and multiple features are trained separately as inputs to obtain a real - part GRU model and an imaginary - part GRU model respectively. During online verification, by simultaneously inputting the time, frequency, and antenna sub - script of the current RE into the real - part GRU model and the imaginary - part GRU model, real - part channel estimation values and imaginary - part channel estimation values are obtained. Finally, the obtained real - part and imaginary - part channel estimation values are concatenated to obtain a complete channel response value. By splitting the channel estimation, the simulation complexity of the model is effectively reduced, the model structure is simplified, and the calculation efficiency is improved. In a MATLAB simulation environment with limited memory resources, since the channel estimation at the original pilot is split, the output dimension of the model is reduced, the memory consumption during training is reduced, and the feasibility of the simulation is improved.

[0060] Preferably, a GRU model based on multiple features is designed, and its specific implementation scheme includes:

[0061] 1) Determine the number of model input features and the output. Since the channel to be estimated is a PUSCH channel, the input of the model is the time slot, frequency, and antenna sub - script where the DMRS pilot is located, and the output of the model is the channel estimation value of this pilot.

[0062] 2) Control the influence degree of historical pilot information on current pilot information by setting an update gate and a reset gate.

[0063] 3) Calculate the candidate hidden state, which provides a proposal for a new hidden state at the current time step, improving the generalization ability of the model.

[0064] 4) Perform weighted fusion on the candidate hidden state and the hidden state of the previous time step, calculate the final hidden state, and obtain the output of the model.

[0065] Taking a specific example, for a MIMO system with 8 antennas, the original training data set at the pilot is: T = [T 1 ,T 2 ,…T j ,…T N T , then the parameter settings of the GRU model are: input_size = 3, indicating that the dimension of the input features is 3, hidden_size = 10, indicating that the dimension of the hidden layer is 10, and the output dimension is default set to 1. Considering the complex channel environment under real conditions and the multi - feature input of the model, setting a higher hidden_size can provide a richer feature representation and improve the generalization ability of the multi - feature GRU model. According to 2), the designed update gate is:

[0066] z t =σ(W z ·[h t-1 ,x t,f,i )​

[0067] The designed reset gate is as follows:

[0068] r t = σ(W r · [h t-1 , x t,f,i )

[0069] According to 3), calculate the hidden candidate state:

[0070]

[0071] According to 4), calculate the final candidate state:

[0072]

[0073] The multi-feature GRU model designed through 1)-4) has one less gate compared to the original LSTM model, with a simple structure and low complexity. When processing large-scale pilot data, the training speed is greatly improved. Since the GRU model has fewer parameters, it also reduces the memory consumption in the simulation environment, improving the efficiency and scalability of the simulation.

[0074] S300. When performing online channel estimation, the best estimate of the current model parameters is learned online through the Reptile algorithm, realizing the online fine-tuning of the GRU model parameters, obtaining a GRU network model with the best parameters, and obtaining the channel estimation at other positions except for the pilots.

[0075] Preferably, an algorithm for online fine-tuning the GRU network model parameters is designed, and its specific implementation scheme is as follows:

[0076] First, a base learner is set up to quickly learn the M pilot signals received online and adapt to the current pilot signals through a small number of gradient updates. Then, a meta-learner is set up to update the parameters of the current model by collecting the parameter update information of the base learner on M tasks. Through the online learning mechanism of the Reptile algorithm, the real-time adjustment and optimization of the GRU network model parameters are realized, thereby improving the model's ability to quickly adapt to new tasks. The Reptile algorithm performs rapid training on M tasks based on the current channel to update the parameters of the GRU model trained offline in real time, enabling the GRU model to quickly reach better performance through a small number of gradient updates when the current channel state changes rapidly.

[0077] Taking a specific example, when performing online update: First, when it is detected that the slot where the current pilot is located is greater than a certain threshold, collect the pilot data of the previous M steps of the current pilot as the support set, denoted as T M = [T 1 , T2 ,…T m T , the parameter estimation value obtained by the base learner through M iterations can be calculated as follows:

[0078]

[0079] θ is the parameter to be optimized in the current GRU model. Then, update the initial parameter values in the meta-learner:

[0080]

[0081] After each training task is completed, the meta-learner will input φ i back into the base learner for optimization. After completing the training of M groups of data in the support set, a pre-trained Reptile model is obtained. The network parameters output by this model contain the channel change characteristics before the (M + 1)-th pilot, and have good real-time performance. Further, for each group of pilot data that arrives, the parameter estimation values of the base learner and the meta-learner are updated in real time through the above two-step algorithm. When it is monitored that the mean square error (MSE) index of the existing GRU network model exceeds the preset threshold level, indicating that the current model performance has declined and cannot meet the predetermined accuracy requirements, the system will automatically trigger the Reptile algorithm to intervene. At this time, the Reptile algorithm will use its optimized network parameters to perform online fine-tuning on the GRU network model, in order to effectively reduce the bit error rate and enhance the real-time performance of the network while maintaining the network stability. Compared with the original MAML meta-learner, since Reptile simplifies the parameter update process through a first-order optimization method and reduces the dependence on the second derivative, it has a simple structure, more efficient training, reduces the hardware requirements for the simulation environment, and is easier to implement and debug. And through experiments, it shows that compared with MAML, the final MSE obtained by Reptile is almost the same, indicating that the effects obtained by the two algorithms are almost the same.

[0082] Embodiment 2:

[0083] The computer-readable storage medium of this embodiment stores a computer program, and when this program is executed by a processor, it implements the steps in a channel estimation simulation method based on a multi-feature input GRU in Embodiment 1.

[0084] The computer-readable storage medium of this embodiment can be the internal storage unit of the terminal, such as the hard disk or memory of the terminal; the computer-readable storage medium of this embodiment can also be the external storage device of the terminal, such as the plug-in hard disk, smart memory card, secure digital card, flash card, etc. equipped on the terminal; further, the computer-readable storage medium can also include both the internal storage unit and the external storage device of the terminal.

[0085] ​The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0086] Embodiment 3:

[0087] The computer device of this embodiment includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in a channel estimation simulation method based on multi-feature input GRU in Embodiment 1.

[0088] In this embodiment, the processor can be a central processing unit, or can also be other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.; the memory can include a read-only memory and a random access memory, and provides instructions and data to the processor. A part of the memory can also include a non-volatile random access memory. For example, the memory can also store information about the device type.

[0089] Those skilled in the art should understand that the content disclosed in the embodiments can be provided as a method, a system, or a computer program product. Therefore, this solution can be implemented in the form of a hardware embodiment, a software embodiment, or a form combining software and hardware embodiments. Moreover, this solution can be implemented in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories and optical memories, etc.) that contain computer-usable program codes.

[0090] This solution is described with reference to the flowcharts and / or block diagrams of the methods and computer program products according to the embodiments of this solution. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented; these computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, so that the instructions executed by the processors of the computer or other programmable data processing devices generate a device for implementing the specified functions in one Figure 1 one flow or multiple flows and / or block diagram Figure 1 one block or multiple blocks.

[0091] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device that implements the operations in the process Figure 1 one process or multiple processes and / or the functions specified in the schematic Figure 1 of one block or multiple blocks.

[0092] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one process or multiple processes and / or the schematic Figure 1 of one block or multiple blocks.

[0093] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a magnetic disk, an optical disc, a read-only memory (ROM), or a random access memory (RAM), etc.

[0094] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A channel estimation simulation method based on multi-feature input GRU, characterized in that: The method comprises: S100, designing a method for organizing training data at the pilot, shielding the influence of the received pilot data on the channel estimation value, and reducing the factors affecting the channel estimation result to time, frequency and the antenna where the pilot is located; S200. Design a GRU model based on multi-feature input to simultaneously process three key features that affect channel estimation and more comprehensively capture the dynamic characteristics of channel changes. S300, when performing online channel estimation, the best estimation of the current model parameters is learned online through the Reptile algorithm, and the GRU network model with the best parameters and the channel estimation of other positions except the pilot are obtained.

2. A channel estimation simulation method based on multi-feature input GRU as claimed in claim 1, characterized in that: The method for organizing training data at the pilot frequency in S100 includes: The PUSCH channel is selected as the simulation channel, and its pilot is DMRS; The signal at the pilot is marked to obtain the time, frequency, antenna index and channel estimation value of the pilot signal, which are used as input data for model training; The channel estimation value represented by a complex number is split to obtain the real part representation and the imaginary part representation of the channel estimation at the pilot signal, which respectively form two groups of training data.

3. A channel estimation simulation method based on multi-feature input GRU as claimed in claim 1, characterized in that: The S200 includes: When the trained model is applied online for channel estimation, the original pilot data is split to obtain real part training data and imaginary part training data; The real part training data and the imaginary part training data are respectively input into the multi-feature GRU model for training to obtain a real part GRU model and an imaginary part GRU model; During online verification, the real channel estimation value and the imaginary channel estimation value are obtained by inputting the time, frequency and antenna index information of the current pilot into the real GRU model and the imaginary GRU model respectively; The obtained real and imaginary channel estimation values ​​are concatenated to obtain a complete channel response value.

4. A channel estimation simulation method based on multi-feature input GRU as claimed in claim 3, characterized in that: The GRU model with multiple feature inputs in S200 includes: The input of the model is the time slot, frequency, antenna index and channel estimation value at the pilot. The channel estimation value is determined by the time slot, frequency and antenna index of the pilot. The influence of historical pilot information on current pilot information is controlled by setting update gate and reset gate; Compute candidate hidden states, providing a proposal for a new hidden state for the current time step; The candidate hidden state and the hidden state of the previous time step are weightedly fused to calculate the final hidden state. By completing a specified number of training tasks, an offline pre-trained GRU model with better parameters is obtained.

5. A channel estimation simulation method based on multi-feature input GRU as claimed in claim 1, characterized in that: In S300, the best estimation of the current model parameters is learned online by the Reptile algorithm, including: Set up a base learner to quickly learn the M groups of pilot data before the current moment, and adapt to the current pilot signal through a small amount of gradient updates; at the same time, set up a meta learner to update the global parameter initialization by collecting parameter update information of the base learners on multiple tasks; When the channel estimation capability of the existing GRU model parameters is poor, the parameters of the GRU model are updated online by activating the output of the Reptile algorithm.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in a channel estimation simulation method based on multi-feature input GRU as described in any one of claims 1 to 5 are implemented.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the steps in the channel estimation simulation method based on multi-feature input GRU as described in any one of claims 1-5 are implemented.

Images