Millimeter wave channel prediction method and system for mobile scenario

Abstract

The application relates to a millimeter wave channel prediction method and system for a mobile scene, and the method comprises the following steps: a single-cell time division duplex MIMO channel model comprising a base station and a user is established; a millimeter wave channel prediction network for the mobile scene is constructed on the basis of a transformer neural network; a position coding layer is improved into a learnable position coding layer; a multi-head self-attention layer is improved into a multi-head convolution self-attention layer by introducing a convolution module; the millimeter wave channel prediction network is trained to adapt to the mobile scene; a channel prediction model comprising a channel estimator and the millimeter wave channel prediction network is established; the channel estimator obtains known channel state information according to a pilot signal; and the millimeter wave channel prediction network predicts future channel state information according to the known channel state information. The application is directed to the mobile scene under the millimeter wave channel, future channel state information is obtained through channel prediction, and the capacity of a communication system can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to wireless communication technology, and in particular to a millimeter wave channel prediction method and system for mobile scenarios. BACKGROUND

[0002] Millimeter wave MIMO technology is one of the key technologies in the field of modern wireless communication. By equipping more antennas at the transmitting end and the receiving end, the system capacity can be improved. In the MIMO system, additional antennas also help to increase the transmission distance and coverage of the signal, and reduce the impact of multipath fading and signal interference on the communication system.

[0003] However, when using millimeter wave MIMO technology for communication, the channel state information is generally obtained by transmitting a pilot signal. However, the millimeter wave channel is complex and variable, and the obtained channel state information is likely to be outdated. In millimeter wave MIMO technology, the quality of pre-coding of the transmitted signal is closely related to the instantaneous channel state information. If the channel state information used for pre-coding is outdated, the quality of pre-coding will decrease, resulting in a loss of system capacity and rate.

[0004] Existing channel prediction methods are mostly unable to adapt to millimeter wave channels, especially in mobile scenarios. Therefore, how to improve the system capacity and communication quality of millimeter wave MIMO channels through channel prediction is a problem that needs to be solved. SUMMARY

[0005] The present application aims to solve the problems existing in the prior art and provide a millimeter wave channel prediction method and system for mobile scenarios. A millimeter wave channel prediction network is constructed on the basis of a neural network to predict future channel states in advance, which can effectively improve the pre-coding quality of millimeter wave MIMO channels, achieve higher system capacity, and improve communication quality.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0007] In a first aspect, the present application provides a millimeter wave channel prediction method for mobile scenarios, comprising the following steps:

[0008] A single-cell time division duplex MIMO channel model is established, including one base station and one user;

[0009] A millimeter wave channel prediction network for mobile scenarios is constructed on the basis of a transformer neural network. The position encoding layer is improved to a learnable position encoding layer, and a convolution module is introduced into the multi-head self-attention layer to improve it to a multi-head convolution self-attention layer.

[0010] The millimeter wave channel prediction network is trained to adapt the millimeter wave channel prediction network to a mobile scenario.

[0011] A channel prediction model is established, including a channel estimator and a millimeter wave channel prediction network connected in series, the channel estimator is used to obtain known channel state information according to a pilot signal, and the millimeter wave channel prediction network is used to predict future channel state information according to the known channel state information.

[0012] In a second aspect, the embodiment of the present application provides a millimeter wave channel prediction system for a mobile scenario, comprising the following modules:

[0013] A MIMO channel model construction module is used to establish a single-cell time division duplex MIMO channel model, including a base station and a user;

[0014] A millimeter wave channel prediction network construction module is used to construct a millimeter wave channel prediction network for a mobile scenario on the basis of a transformer neural network, improve a position encoding layer into a learnable position encoding layer, and introduce a convolution module into a multi-head self-attention layer to improve it into a multi-head convolution self-attention layer;

[0015] A millimeter wave channel prediction network training module is used to train the millimeter wave channel prediction network, so that the millimeter wave channel prediction network adapts to a mobile scenario;

[0016] A channel prediction model establishment module is used to establish a channel prediction model, including a channel estimator and a millimeter wave channel prediction network connected in series, the channel estimator is used to obtain known channel state information according to a pilot signal, and the millimeter wave channel prediction network is used to predict future channel state information according to the known channel state information.

[0017] Compared with the prior art, the technical scheme provided by the present application brings at least the following effective effects:

[0018] The millimeter wave channel prediction method and system for a mobile scenario provided by the present application can provide accurate channel state information under complex and variable millimeter wave channels, improve the precoding quality, realize higher system and rate, effectively alleviate the problem that the channel state information of the millimeter wave channel is easy to be outdated in a mobile scenario, realize higher system capacity, and improve the wireless communication quality. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A flowchart of the millimeter wave channel prediction method for a mobile scenario provided by the embodiment of the present application is shown;

[0020] Figure 2 A structure diagram of the millimeter wave channel prediction network in the embodiment of the present application is shown;

[0021] Figure 3A structural schematic diagram of a multi-head convolution self-attention layer in an embodiment of the present application;

[0022] Figure 4 A structural schematic diagram of a channel prediction model established in an embodiment of the present application;

[0023] Figure 5 A comparison schematic diagram of normalized mean square errors of predicted future channel state information of a millimeter wave channel prediction network and channel predictors based on an RNN model, an LSTM model and a transformer model in an embodiment of the present application;

[0024] Figure 6 A comparison schematic diagram of normalized mean square errors of predicted future channel state information of a millimeter wave channel prediction network and transformer models based on a transformer model, a learnable position encoding layer and a multi-head convolution self-attention layer in an embodiment of the present application;

[0025] Figure 7 A comparison schematic diagram of cumulative distribution functions of a millimeter wave channel prediction network and channel predictors based on an RNN model, an LSTM model and a transformer model in an embodiment of the present application;

[0026] Figure 8 A structural block diagram of a millimeter wave channel prediction system for a mobile scenario provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0028] Embodiment 1

[0029] Please refer to Figure 1 In the present embodiment, a millimeter wave channel prediction method for a mobile scenario is provided, comprising the following steps:

[0030] S1, a single-cell time division duplex MIMO channel model is established, and the MIMO channel model comprises one base station and one user.

[0031] In the single-cell time division duplex MIMO channel model, the base station and the user both adopt a uniform linear array antenna configuration. In the present embodiment, the base station is configured with N b root antennas, and the user is configured with Nu With one antenna, the channel state information between the base station and the user in the i-th frame can be represented as:

[0032]

[0033] Where L is the number of propagation paths, α l Let θ be the attenuation factor for the l-th path. b,l Let θ be the angle of arrival of the base station along the l-th path. u,l f is the departure angle of the user on the l-th path. d Let T be the Doppler frequency of the l-th path. s Let a be the frame length of the l-th path. b (·) and a u (·) represents the array steering vector, [·] H This represents the conjugate transpose of a vector.

[0034] The channel state information is considered to be constant during the coherence time of the channel, but changes between two different coherence times.

[0035] Array guide vector a b (·) and a u (·) are respectively:

[0036]

[0037] in, Let λ be the wavenumber of the carrier wave, λ be the wavelength of the carrier wave, and d be the antenna spacing; [·] T This represents the transpose of a vector.

[0038] S2. Construct a millimeter-wave channel prediction network for mobile scenarios. The millimeter-wave channel prediction network is built on the basis of the transformer neural network. The position coding layer is improved into a learnable position coding layer, and the multi-head self-attention layer is improved into a multi-head convolutional self-attention layer by introducing a convolution module.

[0039] like Figure 2 As shown, the millimeter-wave channel prediction network in this embodiment is an improvement on the transformer network model, specifically including a first input layer, a second input layer, a first position coding layer, a second position coding layer, an encoder, a decoder, and an output layer. The first input layer is connected to the encoder via the first position coding layer, the second input layer is connected to the decoder via the second position coding layer, the output of the encoder is connected to the decoder, and the output layer is located at the output of the decoder.

[0040] The encoder comprises a multi-head convolutional self-attention layer, a feedforward layer, a residual connection and a normalization layer. The decoder comprises a multi-head convolutional self-attention layer, a multi-head cross self-attention layer, a feedforward layer, a residual connection and a normalization layer. The residual connection and the normalization layer are different from the multi-head convolutional self-attention layer, the feedforward layer and the multi-head cross self-attention layer, and the multi-head convolutional self-attention layer has only one input, while the residual connection and the normalization layer have two inputs. The encoder and the decoder are connected through the residual connection and the normalization layer for residual connection and layer normalization.

[0041] The first input layer, the second input layer and the output layer are linear layers, which map low-dimensional channel data to a high-dimensional space through linear transformation to obtain high-dimensional data, and map the high-dimensional data back to a low-dimensional space through linear transformation and output future channel state information, which can be represented as:

[0042] Linear(x)=xW+b

[0043] Wherein, x is the input data of the linear layer, Linear(x) is the output data of the linear layer, W is the matrix used for linear transformation, and b is the added bias.

[0044] The first position encoding layer and the second position encoding layer are learnable position encoding layers. In this embodiment, the position encoding layer is improved so that it can continuously optimize the position encoding during the training of the channel predictor. Specifically, the learnable position encoding process can be represented as:

[0045] LPE (pos,i) =fθ(·)

[0046] Wherein, pos is the position of the channel data, i is the dimension index, θ is the neural network parameter of the millimeter wave channel prediction network, and f θ (·) is the change of the position encoding in the training process of the millimeter wave channel prediction network.

[0047] The multi-head convolutional self-attention layer is improved on the basis of the multi-head self-attention layer, as shown in Figure 3

[0048] The self-attention mechanism is a correlation process, which needs to perform three different linear transformations on the input data, namely query transformation, key transformation and value transformation. The self-attention mechanism first calculates the correlation of the data after query transformation and key transformation, and then multiplies the obtained correlation with the data after value transformation to obtain the output of the self-attention, which can be represented as:

[0049]

[0050] ​where Q, K, V are linear transformation matrices for query transformation, key transformation and value transformation respectively, is a scale factor for avoiding variance increase; softmax(·) is an activation function, and the expression can adopt

[0051] And the multi-head self-attention layer allows the channel predictor to obtain information from multiple different places, which can be expressed as:

[0052] multihead(Q, K, V) = concat(head1, …, head h )W

[0053] where head i indicates the self-attention result of the i-th head, concat(·) indicates a concatenation operation on all self-attention results, and W is a linear transformation on the result.

[0054] The embodiment introduces a convolution module to the multi-head self-attention layer, thereby improving it to a multi-head convolution self-attention layer. Compared with the multi-head self-attention layer, the multi-head convolution self-attention layer needs to use a causal convolution kernel with a kernel size of k and a step size of 1 to process the input data before the data enters the query transformation and key transformation. Wherein, the causal convolution kernel can make the millimeter wave channel prediction network not pay attention to future information when processing data, so that the millimeter wave channel prediction network processes data that fits the actual use scenario.

[0055] In addition, the multi-head cross self-attention layer is similar to the multi-head convolution self-attention layer, which is also obtained by introducing a convolution module to the multi-head self-attention layer, and the difference lies in the input. The input of the multi-head cross self-attention layer has two, which are the output of the encoder and the output of the previous layer (i.e. the multi-head convolution self-attention layer) of the multi-head cross self-attention layer, and the multi-head cross self-attention layer performs key transformation and value transformation on the output of the encoder, and performs query transformation on the output of the previous layer of the multi-head cross self-attention layer, and then performs self-attention calculation.

[0056] The feedforward layer uses 2 linear transformations and 1 nonlinear transformation to perform deeper feature extraction on the channel data, which can be expressed as:

[0057] FFN(x) = max(0, xW1 + b1)W2 + b2

[0058] where x is the input data of the feedforward layer, FFN(x) is the output of the feedforward layer, W1 and W2 are matrices for linear transformation, b1 and b2 are added biases, and max(·) represents a maximum value operation.

[0059] The residual connection operation directly adds the output of a certain layer to the output of the next layer, which can effectively alleviate the problems of gradient vanishing and gradient explosion in the neural network. The layer normalization operation normalizes the input data in each layer, maps the input data to a preset numerical range or distribution, which can accelerate the convergence speed of the neural network and improve the stability of the neural network in the training process. The residual connection operation and the layer normalization operation can be represented as:

[0060] y=Layernorm(x+F(x))

[0061] Wherein, y represents the output after the residual connection and the layer normalization operation, x represents the input data of the neural network; F(·) represents the sub-layer operation process that needs to be connected, such as multi-head convolution self-attention operation or feedforward layer operation; Layernorm(·) represents the layer normalization operation.

[0062] S3, training the millimeter wave channel prediction network, so that the millimeter wave channel prediction network can adapt to the mobile scene.

[0063] The data used to train the millimeter wave channel prediction network is the millimeter wave MIMO channel data in the mobile scene, and the optimization function used in the training process is the normalized mean square error loss function, which can be represented as:

[0064]

[0065] Wherein, NMSE is the normalized mean square error, n is the total number of channel state information, h i is the true value of the channel state information, is the channel state information predicted by the millimeter wave channel prediction network.

[0066] S4, establishing a channel prediction model, including a connected channel estimator and a millimeter wave channel prediction network, the channel estimator is used to obtain known channel state information according to the pilot signal, and the millimeter wave channel prediction network predicts the future channel state information according to the known channel state information.

[0067] In the channel prediction process, for the millimeter wave channel prediction network that has been trained, the known channel state information obtained by the pilot is input into the input end of the millimeter wave channel prediction network, and the future channel state information can be obtained at the output end of the millimeter wave channel prediction network. Using the future channel state information can help the wireless communication system to communicate.

[0068] As Figure 4As shown, in the channel prediction model of the embodiment, first, known channel state information is obtained according to a pilot signal, then p known channel state information is input into the trained millimeter wave channel prediction network, and finally q future channel state information (i.e. channel state information to be predicted) is output.

[0069] S41, according to the pilot sequence sent by the user to the base station, the known channel state information is determined by the channel estimator.

[0070] Specifically, the pilot sequence received in the i th frame is:

[0071] y i = H i s i + w i

[0072] where N p is the length of the pilot sequence, is the transmitted pilot sequence orthogonal between different antennas, is the known channel state information of the i th frame, is a zero-mean additive white Gaussian noise.

[0073] S42, according to the p known channel state information, the q future channel state information is predicted by the millimeter wave channel prediction network.

[0074] The expression for predicting q future channel state information can be expressed as:

[0075]

[0076] where f(·) is the millimeter wave channel prediction network, is the channel state information to be predicted in the i th frame, H i is the known channel state information of the i th frame.

[0077] In this embodiment, in the millimeter wave channel prediction network, the input of the encoder is p known channel state information; the input of the decoder is q channel state information taken from the end of the p known channel state information.

[0078] In this embodiment, the system is mathematically modeled and simulated, the simulation uses CDL-A channel data generated by matlab for experiment, the base station uses a uniform linear array with 4 antennas, the user uses a uniform linear array with 2 antennas, the carrier frequency is 30GHz, the user's moving speed is 60km / h, the causal convolution kernel size of the multi-head convolution self-attention layer is 3, and the simulation uses 25 known channel state information to predict 5 unknown channel state information, as shown in Figures 5-7 .

[0079] from Figure 5 The comparison shows that the normalized mean square error of the millimeter-wave channel prediction network of the present invention is smaller than that of channel predictors based on RNN model, LSTM model and transformer model at various prediction lengths, and can better predict future channel conditions.

[0080] from Figure 6 The comparison shows that, compared to the channel predictor of the transformer model, the channel predictor of the transformer model using a learnable position coding layer or a multi-head convolutional self-attention layer can improve the prediction accuracy of future channel state information. Furthermore, the millimeter-wave channel prediction network of this invention, because it simultaneously employs a learnable position coding layer and a multi-head convolutional self-attention layer, can achieve even better channel prediction results. Therefore, the channel prediction method of this invention has better performance in predicting future channel state information.

[0081] exist Figure 7 In this context, the cumulative distribution function reveals the distribution of the difference between the predicted future channel state information and the true value. From... Figure 7 The comparison shows that the channel state information amplitude error predicted by the millimeter-wave channel prediction network of the present invention is within 0.02 for about 80% of the time, while the channel state information amplitude error predicted by the channel predictor based on the RNN model and LSTM model is only within 0.02 for about 60% of the time. Compared with the channel predictor based on the RNN model and LSTM model, the channel prediction method of the present invention has a smaller prediction error.

[0082] Example 2

[0083] Based on the same concept as the millimeter-wave channel prediction method for mobile scenarios in Embodiment 1 above, this embodiment provides a millimeter-wave channel prediction system for mobile scenarios, which can be used to execute the steps of the millimeter-wave channel prediction method for mobile scenarios described above.

[0084] For ease of explanation, the structural diagram of the embodiment of the millimeter-wave channel prediction system for mobile scenarios only shows the parts related to the embodiments of the present invention. Those skilled in the art will understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0085] Please see Figure 8 In this embodiment, a millimeter-wave channel prediction system 100 for mobile scenarios is provided, which includes the following modules:

[0086] The MIMO channel model construction module 101 is configured to establish a single-cell time division duplex MIMO channel model including one base station and one user, wherein the base station and the user both adopt a uniform linear array antenna configuration.

[0087] The millimeter wave channel prediction network construction module 102 is configured to construct a millimeter wave channel prediction network for a mobile scenario on the basis of a transformer neural network, improve a position encoding layer into a learnable position encoding layer, and introduce a convolution module into a multi-head self-attention layer to improve the multi-head convolution self-attention layer.

[0088] The millimeter wave channel prediction network training module 103 is configured to train the millimeter wave channel prediction network so that the millimeter wave channel prediction network is adapted to the mobile scenario, and the training process adopts channel data in the mobile scenario, and a loss function used in the training process is a normalized mean square error loss function.

[0089] The channel prediction model establishment module 104 is configured to establish a channel prediction model including a connected channel estimator and a millimeter wave channel prediction network, the channel estimator is configured to acquire known channel state information according to a pilot signal, and the millimeter wave channel prediction network is configured to predict future channel state information according to the known channel state information.

[0090] It should be noted that the millimeter wave channel prediction system for a mobile scenario of the present application corresponds to the millimeter wave channel prediction method for a mobile scenario of the present application, and the technical features and advantages described in the embodiment of the millimeter wave channel prediction method for a mobile scenario are applicable to the embodiment of the millimeter wave channel prediction system for a mobile scenario, and the specific content can be referred to the description in the embodiment of the method, which will not be described here again, and hereby declared.

[0091] In addition, in the embodiment of the millimeter wave channel prediction system for a mobile scenario of the above embodiment, the logical division of each program module is only an example, and in actual application, the above functions can be completed by different program modules according to needs, for example, the configuration requirements of the corresponding hardware or the convenience of software implementation, that is, the internal structure of the millimeter wave channel prediction system for a mobile scenario is divided into different program modules to complete all or part of the functions described above.

[0092] In summary, the millimeter wave channel prediction method and system for a mobile scenario provided by the present application can provide accurate channel state information under complex and variable millimeter wave channels, the millimeter wave channel prediction method of the present application can effectively alleviate the problem that the channel state information of the millimeter wave channel in the mobile scenario is easily outdated, realize higher system capacity, and improve the quality of wireless communication.

[0093] It is to be understood that the above-mentioned various method embodiments are all described as a combination of a series of acts for the sake of simple description, but those skilled in the art should know that the present application is not limited to the order of the acts described, because according to the present application, certain steps can be performed in other orders or simultaneously.

[0094] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited to the above embodiments, and any changes, modifications, substitutions, combinations, simplifications, etc. made without departing from the spirit and principles of the present application should be equivalent replacement methods and should be included in the protection scope of the present application.

Claims

1. A method for millimeter wave channel prediction for mobile scenarios, characterized in that, The method comprises the following steps: A single-cell time division duplex MIMO channel model is established, including one base station and one user; A millimeter wave channel prediction network for mobile scenarios is constructed based on a transformer neural network, a position encoding layer is improved into a learnable position encoding layer, and a convolution module is introduced into a multi-head self-attention layer to improve it into a multi-head convolution self-attention layer; The millimeter wave channel prediction network is trained to adapt to the mobile scenario; A channel prediction model is established, including a channel estimator and a millimeter wave channel prediction network connected in series, the channel estimator is used to obtain known channel state information according to a pilot signal, and the millimeter wave channel prediction network is used to predict future channel state information according to the known channel state information; In the single-cell time division duplex MIMO channel model, the base station and the user both adopt the antenna configuration of uniform linear array, the base station configures N b root antennas, and the user configures N u root antennas; the channel state information between the base station and the user in the i-th frame is expressed as: where L is the number of propagation paths, a l is the attenuation factor of the lth path, θ b,l is the angle of arrival at the base station of the lth path, θ u,l is the angle of departure at the user of the lth path, f d is the Doppler frequency of the lth path, T s is the frame length of the lth path, a b (·) and a u (·) are array steering vectors, [·] H denotes the conjugate transpose of a vector. Array steering vector a b (·) and a y (·) are respectively: wherein is the wave number of the carrier, λ is the wavelength of the carrier, d is the antenna separation; T denotes the transpose of a vector; The process of establishing the channel prediction model comprises: According to the pilot sequence sent by the user to the base station, the known channel state information is determined by a channel estimator; the pilot sequence received in the ith frame is: y i = H i s i + w i where N p is the length of the pilot sequence, is a pilot sequence that is orthogonal between different antennas, is the known channel state information of the i-th frame, is a zero-mean additive white Gaussian noise; According to the p known channel state information, q future channel state information is predicted through the millimeter wave channel prediction network, which is represented as: where f(·) is a millimeter wave channel prediction network, is the i-th frame future channel state information, H i is the i-th frame known channel state information.

2. The millimeter wave channel prediction method of claim 1, wherein, The millimeter wave channel prediction network comprises a first input layer, a second input layer, a first position encoding layer, a second position encoding layer, an encoder, a decoder and an output layer; the first input layer is connected with the encoder through the first position encoding layer, the second input layer is connected with the decoder through the second position encoding layer, the output end of the encoder is connected with the decoder, and the output layer is arranged at the output end of the decoder; The first position encoding layer and the second position encoding layer are both learnable position encoding layers, and the learnable position encoding process is represented as: LPE (pos,m) = f θ (·) where pos is the position of the channel data, m is the dimension index, Θ is the neural network parameter of the mmWave channel prediction network, f θ (·) is the change of the position encoding by the mmWave channel prediction network training process; The encoder comprises a multi-head convolution self-attention layer, a feedforward layer, a residual connection and a normalization layer; the decoder comprises a multi-head convolution self-attention layer, a multi-head cross self-attention layer, a feedforward layer, a residual connection and a normalization layer; The multi-head convolution self-attention layer is obtained by introducing a convolution module into the multi-head self-attention layer; before the data of the multi-head convolution self-attention layer enters the query transformation and the key transformation, a causal convolution kernel is used to process the input data; The multi-head cross self-attention layer is obtained by introducing a convolution module into the multi-head self-attention layer; the input of the multi-head cross self-attention layer comprises the output of the encoder and the output of the multi-head convolution self-attention layer; the multi-head cross self-attention layer performs key transformation and value transformation on the output of the encoder, performs query transformation on the output of the multi-head convolution self-attention layer, and then performs self-attention calculation; The first input layer, the second input layer and the output layer are all linear layers; the first input layer and the second input layer map low-dimensional channel data to a high-dimensional space through linear transformation to obtain high-dimensional data; the output layer maps the high-dimensional data back to a low-dimensional space through linear transformation and outputs future channel state information.

3. The millimeter wave channel prediction method of claim 2, wherein, The future channel state information output by the output layer is represented as: Linear(x)=xW+b Wherein, x is the input data of the linear layer, Linear(x) is the output data of the linear layer, W is a matrix used for linear transformation, and b is a bias added.

4. The millimeter wave channel prediction method of claim 2, wherein, The feedforward layer uses 2 times of linear transformation and 1 time of nonlinear transformation to perform deeper feature extraction on the channel data, which is represented as: FFN(y)=max(0,yW1+b1)W2+b2 Wherein, y is the input data of the feedforward layer, FFN(y) is the output of the feedforward layer, W1 and W2 are the matrices for linear transformation, b1 and b2 are the added biases, and max(·) represents the maximum value operation.

5. The millimeter wave channel prediction method of claim 1, wherein, The data used for training the millimeter wave channel prediction network is millimeter wave MIMO channel data in a mobile scenario, and the optimization function used in the training process is a normalized mean square error loss function, which is represented as: wherein NMSE is a normalized mean square error, n is a total number of channel state information, h i is a true value of channel state information, is channel state information predicted by the millimeter wave channel prediction network.

6. A millimeter wave channel prediction system for mobile scenarios, characterized in that, The method comprises the following modules: The MIMO channel model construction module is configured to establish a single-cell time division duplex MIMO channel model comprising one base station and one user. The millimeter wave channel prediction network construction module is configured to construct a millimeter wave channel prediction network for a mobile scenario based on a transformer neural network, improve a position encoding layer into a learnable position encoding layer, and introduce a convolution module into a multi-head self-attention layer to improve the multi-head convolution self-attention layer. The millimeter wave channel prediction network training module is configured to train the millimeter wave channel prediction network to adapt the millimeter wave channel prediction network to the mobile scenario. The channel prediction model establishment module is configured to establish a channel prediction model comprising a channel estimator and a millimeter wave channel prediction network connected in series, the channel estimator being configured to obtain known channel state information according to a pilot signal, and the millimeter wave channel prediction network being configured to predict future channel state information according to the known channel state information. In the single-cell time division duplex MIMO channel model, the base station and the user both adopt the antenna configuration of uniform linear array, the base station configures N b root antennas, and the user configures N u root antennas; the channel state information between the base station and the user in the i-th frame is expressed as: where L is the number of propagation paths, a l is the attenuation factor of the lth path, θ b,l is the angle of arrival at the base station of the lth path, θ u,l is the angle of departure at the user of the lth path, f d is the Doppler frequency of the lth path, T s is the frame length of the lth path, a b (·) and a u (·) are array steering vectors, [·] H denotes the conjugate transpose of a vector; Array steering vector a b (·) and a u (·) are respectively: wherein is the wave number of the carrier, λ is the wavelength of the carrier, d is the antenna separation; T denotes the transpose of a vector; The process of establishing the channel prediction model comprises: Based on the pilot sequence sent by the user to the base station, the known channel state information is determined by a channel estimator; the pilot sequence received in the ith frame is: y i = H i s i + w i where N p is the length of the pilot sequence, is a pilot sequence that is orthogonal between different antennas, is the known channel state information of the i-th frame, is a zero-mean additive white Gaussian noise; The millimeter wave channel prediction network comprises a first input layer, a second input layer, a first position encoding layer, a second position encoding layer, an encoder, a decoder, and an output layer; the first input layer is connected to the encoder via the first position encoding layer, the second input layer is connected to the decoder via the second position encoding layer, an output end of the encoder is connected to the decoder, and the output layer is arranged at an output end of the decoder. where f(·) is a millimeter wave channel prediction network, is the i-th frame future channel state information, H i is the i-th frame known channel state information.

7. The millimeter-wave channel prediction system of claim 6, wherein, The first position encoding layer and the second position encoding layer are both learnable position encoding layers, and the learnable position encoding process is represented as: The encoder comprises a multi-head convolution self-attention layer, a feedforward layer, a residual connection, and a normalization layer; and the decoder comprises a multi-head convolution self-attention layer, a multi-head cross self-attention layer, a feedforward layer, a residual connection, and a normalization layer. LPE (pos,m) = f θ (·) where pos is the position of the channel data, m is the dimension index, Θ is the neural network parameter of the mmWave channel prediction network, f θ (·) is the change of the position encoding by the mmWave channel prediction network training process; The multi-head convolution self-attention layer is obtained by introducing a convolution module into the multi-head self-attention layer; before the data of the multi-head convolution self-attention layer enters the query transformation and the key transformation, the input data is processed using a causal convolution kernel. The multi-head cross self-attention layer is obtained by introducing a convolution module into the multi-head self-attention layer; the input of the multi-head cross self-attention layer comprises an output of the encoder and an output of the multi-head convolution self-attention layer; the multi-head cross self-attention layer performs key transformation and value transformation on the output of the encoder, performs query transformation on the output of the multi-head convolution self-attention layer, and then performs self-attention calculation. ​ The first input layer, the second input layer and the output layer are linear layers, the input layer maps low-dimensional channel data to a high-dimensional space through linear transformation to obtain high-dimensional data, and the output layer maps the high-dimensional data back to a low-dimensional space through linear transformation and outputs future channel state information.

Images