A Downlink Physical-Layer Channel Feedback Transmission Method Based on Wireless Map

Abstract

The present invention discloses a downlink physical layer channel feedback transmission method based on a wireless map, which relates to the field of wireless mobile communication technology. It includes: S1. Obtain the user location information of user k when moving in the spatial domain, and upload the user location information to the edge cloud; S2. The edge cloud inputs the user location information into the complex precoding network based on the wireless map; S3. The complex precoding network generates the precoding decision w of user k k , and transmits the precoding decision w k to the base station; S4. The base station executes the precoding decision w k for realizing physical layer information transmission. By proposing a downlink transmission mechanism for physical layer channel feedback and designing a complex precoding network based on the wireless map, the present invention effectively improves the transmission efficiency of physical layer information in wireless communication.

Description

Technical Field

[0001] The present invention relates to the field of wireless mobile communication technologies, and in particular, to a method for downlink physical layer channel feedback transmission based on a wireless map. Background Art

[0002] With the surge in mobile communication demands, multi-antenna systems have been widely deployed to provide efficient spectrum transmission, so as to meet more stringent service requirements. Currently, the multi-antenna transmission of 5G is based on a codebook, and feedback information is required to enable the transmitter to understand the channel conditions, so as to precode the original signals loaded onto different antennas. However, considering the continuous change of the channel in the actual environment, the feedback information for precoding may become outdated, resulting in performance degradation, especially in mobile scenarios where the channel changes rapidly. In addition, in a fully decoupled radio access network (FD-RAN), the uplink base stations (UBSs) and downlink base stations (DBSs) are physically decoupled. Since users within a wide area share the same control base station, the transmission resources for feedback will be more scarce. Therefore, it is imperative to explore a new large-scale MIMO transmission mechanism.

[0003] Currently, both autoencoder technology and deep learning-based channel state information compression technology have received extensive attention in the academic community. Yin et al. proposed a CSI compression feedback network based on a self-information model. However, these works still rely on real-time feedback of the channel and cannot address the challenge of outdated channel feedback caused by latency and channel changes. Location-based precoding technology has attracted great interest in the academic community due to the relatively stable statistical characteristics of the channels at the locations of users. However, most deep learning-based solutions are based on real-valued neural networks (RVNNs), which may lead to performance degradation because the correlation between In-phase and Quadrature (IQ) components is ignored. Complex-valued neural networks (CVNNs) have been applied to different communication scenarios, including signal modulation classification, channel prediction, OFDM receivers, etc., due to their superior performance and ability to directly process complex-valued data. The lightweight CLNet with a complex-valued input layer can effectively extract the complex-valued features of the input. The precoding vector matrix is complex-valued, and the channel statistics are also usually complex-valued.

[0004] That is, the following problems mainly exist at present: in the fully decoupled network, the uplink base station and the base station are physically decoupled, and users within a wide area share the same control base station, resulting in scarce transmission resources for feedback; many studies are based on real-time channel feedback and cannot solve the problem of outdated channel feedback caused by delay and channel changes; most deep learning-based solutions are based on real-valued neural networks (RVNNs), ignoring the correlation between In-phase and Quadrature (IQ) components, which may lead to performance degradation.

[0005] Therefore, it is an urgent problem for those skilled in the art to propose a downlink physical layer channel feedback transmission method based on a wireless map to solve the difficulties existing in the prior art. Summary of the Invention

[0006] In view of this, the present invention provides a downlink physical layer channel feedback transmission method based on a wireless map, by proposing a downlink transmission mechanism for physical layer channel feedback and designing a complex precoding network based on a wireless map to improve the information transmission efficiency of the physical layer of wireless communication.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A downlink physical layer channel feedback transmission method based on a wireless map includes the following steps:

[0009] S1. Obtain the user location information when user k moves in the spatial domain, and upload the user location information to the edge cloud;

[0010] S2. The edge cloud inputs the user location information into a complex precoding network based on a wireless map;

[0011] S3. The complex precoding network generates a precoding decision w k for user k, and transmits the precoding decision w k to the base station;

[0012] S4. The base station executes the precoding decision w k to achieve physical layer information transmission.

[0013] In the above method, optionally, the user location information in S1 includes location information p k and the channel statistical information obtained corresponding to the location information p k ;

[0014] In the above method, optionally, the channel statistical information refers to the second-order statistical quantity of the channel at any position within the entire concerned area and includes the transmit correlation matrix R t(p), receive the correlation matrix R r (p) and the channel fading matrix μ(p);

[0015] Assume G ∈ C N×M is the beam domain channel matrix, whose elements are zero-mean and independently distributed, then the expression of R t (p) is:

[0016]

[0017] where E is the mathematical expectation operation, R is the real number field, M is the number of base station antennas and also the dimension of the obtained matrix;

[0018] The coherent MIMO channel between the user and the base station is H = UGV H , where U and V are the deterministic unitary matrices of the channel, then the expression of R r (p) is:

[0019]

[0020] where N is the number of user antennas and also the dimension of the obtained matrix;

[0021] The elements of μ(p) represent the channel coefficients between different transmit antennas and different receive antennas. These coefficients reflect the attenuation and phase change of the signal during transmission and are used to describe the fading characteristics of the channel. The expression of μ(p) is:

[0022]

[0023] For the above method, optionally, the complex precoding network in S3 takes the user location information of user k as the input and the precoding decision w of user k k as the output.

[0024] For the above method, optionally, the complex precoding network in S3 includes heterogeneous multi-modal input information and a multi-layer fusion information processing mechanism;

[0025] The heterogeneous multi-modal input information includes p k , R t (p), R r (p), μ(p);

[0026] The multi-layer fusion information processing mechanism includes an input layer, a feature extraction layer stack, a feature fusion stack, and an output layer.

[0027] For the above method, optionally, the specific content of the complex precoding network in S3 to generate the precoding decision w of user k k is as follows:

[0028] S31. Combine the four signals pk , R t (p), R r (p), μ(p) input into the input layer;

[0029] S32. For the feature extraction layer stack, the four signals are respectively input into four independent complex feature extraction subnets to obtain embedding feature vectors of different modalities;

[0030] S33. For the feature fusion layer, use the Flatten layer to fuse the encodings of different modalities, and then use complex batch normalization to perform complex normalization on the fused encodings. After multi-modal information fusion, the fused features are processed through multiple layers of complex processing layers;

[0031] S34. At the output layer, normalize the result of CPReLU, and the complex precoding network outputs the normalized precoding w k .

[0032] In the above method, optionally, the complex processing layer in S33 includes a complex fully connected layer and a complex activation function layer;

[0033] The complex fully connected layer processes the real and imaginary parts of the features respectively, and is used to integrate the output features of the previous layer into each neuron of the current layer;

[0034] The complex activation function layer applies a separate PReLU to the real and imaginary parts of the neurons respectively for processing, and is used to achieve complex activation.

[0035] It can be seen from the above technical solutions that, compared with the prior art, the present invention provides a downlink physical layer channel feedback transmission method based on a wireless map, which has the following beneficial effects:

[0036] (1) Realize multi-antenna transmission based on the spatial domain statistical information of the channel combined with the quasi-real-time position of the user, and successfully avoid the problems brought by a large amount of transmission resources occupied by channel feedback;

[0037] (2) It is particularly suitable for high-speed railway communications in large-scale MIMO communication scenarios with high channel dynamics, or satellite communications where it is difficult to achieve accurate channel feedback in some cases;

[0038] (3) By proposing a downlink transmission mechanism for physical layer channel feedback and designing a complex precoding network based on a wireless map, the information transmission efficiency of the physical layer of wireless communication is effectively improved. Description of the Drawings

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained according to the provided drawings.

[0040] Figure 1 Flowchart of a downlink physical layer channel feedback transmission method based on a wireless map provided by the present invention;

[0041] Figure 2 Schematic diagram of a downlink physical layer channel feedback transmission method based on a wireless map provided by the present invention;

[0042] Figure 3 Overall framework diagram of the complex precoding network provided by the present invention. Detailed implementation manners

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to 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 of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0044] By the user periodically uploading their own location information to the edge cloud, the base station realizes downlink multi-antenna transmission based on the location p of user k k and the channel statistical information at the corresponding location. Realize downlink multi-antenna transmission.

[0045] Referring to Figure 1 and Figure 2 As shown, the present invention discloses a downlink physical layer channel feedback transmission method based on a wireless map, including the following steps:

[0046] S1. Obtain the user location information when user k moves in the spatial domain, and upload the user location information to the edge cloud;

[0047] S2. The edge cloud inputs the user location information into the complex precoding network based on the wireless map;

[0048] S3. The complex precoding network generates the precoding decision w of user k k , and transmits the precoding decision w k to the base station;

[0049] S4. The base station executes the precoding decision w k to realize physical layer information transmission.

[0050] Specifically, when user k moves in the spatial domain, the user location information is periodically uploaded to the edge cloud. The location of user k can change dynamically over time and is a function p of time t k (t).

[0051] Furthermore, the user location information in S1 includes the location information p k and the channel statistical information obtained corresponding to the location information p k

[0052] Furthermore, the channel statistical information refers to the second-order channel statistics at any location within the entire region of interest and includes the transmit correlation matrix R t (p), the receive correlation matrix R r (p), and the channel fading matrix μ(p);

[0053] Assume that G ∈ C N×M is the beam domain channel matrix, whose elements are zero-mean and independently distributed, then the expression of R t (p) is:

[0054]

[0055] In the formula, E represents the mathematical expectation operation, R is the real number field, M is the number of base station antennas and also the dimension of the obtained matrix;

[0056] The coherent MIMO channel between the user and the base station is H = UGV H , where U and V are the deterministic unitary matrices of the channel, then the expression of R r (p) is:

[0057]

[0058] In the formula, N is the number of user antennas and also the dimension of the obtained matrix;

[0059] The elements of μ(p) represent the channel coefficients between different transmit antennas and different receive antennas. These coefficients reflect the attenuation and phase changes of the signal during transmission and are used to describe the fading characteristics of the channel. The expression of μ(p) is:

[0060]

[0061] Specifically, the present invention only considers a single base station providing services to a single user within a sub-band and uses a joint correlation channel model to describe the spatial correlation of each channel;

[0062] R t ​(p) represents the correlation of the channel state between different transmit antennas;

[0063] R r (p) represents the correlation of the channel state between different receive antennas;

[0064] The elements of μ(p) represent the channel coefficients from different transmit antennas to different receive antennas. These coefficients reflect the attenuation and phase change of the signal during transmission and are used to describe the fading characteristics of the channel.

[0065] Furthermore, in S3, the complex precoding network takes the user location information of user k as the input and outputs the precoding decision w of user k k as the output.

[0066] Furthermore, the complex precoding network in S3 includes heterogeneous multi-modal input information and a multi-layer fusion information processing mechanism;

[0067] The heterogeneous multi-modal input information includes p k , R t (p), R r (p), μ(p);

[0068] The multi-layer fusion information processing mechanism includes an input layer, a stack of feature extraction layers, a stack of feature fusion layers, and an output layer.

[0069] Furthermore, as Figure 3 shown, the specific content of the complex precoding network (RMCPNet) in S3 to generate the precoding decision w of user k k is as follows:

[0070] S31. Input the four signals p k , R t (p), R r (p), μ(p) into the input layer;

[0071] S32. For the stack of feature extraction layers, the four signals are respectively input into four independent complex feature extraction subnets to obtain embedding feature vectors of different modalities;

[0072] S33. For the feature fusion layer, use the Flatten layer to fuse the encodings of different modalities, and then use complex batch normalization to perform complex normalization on the fused encodings. After multi-modal information fusion, further process the fused features through multiple layers of complex processing layers;

[0073] S34. At the output layer, normalize the result of CPReLU, and the complex precoding network outputs the normalized precoding w k .

[0074] Further, the complex processing layer in S33 includes a complex fully connected layer and a complex activation function layer;

[0075] The complex fully connected layer processes the real and imaginary parts of the feature separately, and is used to integrate the output features of the previous layer into each neuron of the current layer;

[0076] The complex activation function layer applies a separate PReLU to the real and imaginary parts of the neuron for processing respectively, and is used to achieve complex activation.

[0077] Specifically, the complex processing layer includes a complex fully connected layer (CLinear) and a complex activation function layer (CPReLU), and their processing methods are as follows:

[0078] CLinear is a complex fully connected layer, which processes the real and imaginary parts of the feature separately, and is used to integrate the output features of the previous layer into each neuron of the current layer. Assume that the complex input of this complex layer is Z = Z R +i·Z I , where Z R and Z I represent the real and imaginary parts of Z respectively. Assume that the processing of Z by this complex layer is denoted as ψ(Z), that is, the output feature map. Assume that ψ R and ψ I represent the real part processing channel and the imaginary part processing channel of the neuron in this complex fully connected layer respectively. Then the real part of the output feature map can be expressed as ψ R (z R ) - ψ I (z I ), and the imaginary part of the output feature map can be expressed as (ψ R (z I ) + ψ I (z R ))). Then the output feature map is:

[0079] ψ(Z) = (ψ R (z R ) - ψ I (z I )) + i·(ψ R (z I ) + ψ I (z R ))).

[0080] CPReLU is a complex activation function layer, which applies a separate activation function (PReLU) to the real and imaginary parts of the neuron for processing respectively to achieve complex activation, that is

[0081] CPReLU(z) = PReLU(z R ) + i·PReLU(z I )

[0082] The training of the complex precoding network is carried out with the user throughput as the loss function. Assume that represents an arbitrary input sample, and N B represents the size of each batch of samples during training. Since additional channel information H is required for calculating the loss, the actual batch size during each training is:

[0083]

[0084] Then the formula of the loss function can be expressed as:

[0085]

[0086] In the formula, I is the identity matrix, θ is the network parameter, ρ represents the signal-to-interference-plus-noise ratio (SINR) at the receiver, and the function represents the mapping from position p i and the corresponding channel statistics to the precoding w.

[0087] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0088] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A downlink transmission method without physical layer channel feedback based on wireless map, characterized in that: The following steps are involved: S1. Obtain the user location information of user k when he moves in the spatial domain, and upload the user location information to the edge cloud; S2. The edge cloud inputs the user location information into the complex precoding network based on the wireless map; S3. The complex precoding network generates a precoding decision wk for user k and transmits the precoding decision wk to the base station; S4. The base station performs precoding decision w k , used to realize physical layer information transmission; The complex precoding network in S3 takes the user location information of user k as input and the precoding decision w of user k as input. k is the output; The complex precoding network in S3 includes heterogeneous multimodal input information and a multi-layer fusion information processing mechanism; the heterogeneous multimodal input information includes p k , R t (p), R r (p), μ(p), where p k is the location information, R t (p) is the transmission correlation matrix, R r (p) is the receiving correlation matrix, μ(p) is the channel fading matrix; The multi-layer fusion information processing mechanism includes an input layer, a feature extraction layer stack, a feature fusion stack, and an output layer; The specific content of the complex precoding network generating the precoding decision wk of user k in S3 includes the following steps: S31. The four signals p k , R t (p), R r (p), μ(p) input layer; S32. For the feature extraction layer stack, the four signals are respectively input into four independent complex feature extraction subnetworks to obtain embedded feature vectors of different modalities; S33. For the feature fusion layer, the codes of different modes are fused by using the connection layer, and then the fused codes are normalized by using the complex batch normalization. After the multi-mode information is fused, the fused features are processed by the multi-layer complex processing layer; S34. At the output layer, the result of the complex activation function layer is normalized, and the complex precoding network outputs the normalized precoding w k ; The complex processing layer in S33 includes a complex fully connected layer and a complex activation function layer; The complex fully connected layer processes the real and imaginary parts of the features separately to integrate the output features of the previous layer into each neuron of the current layer; The complex activation function layer applies separate activation functions to the real and imaginary parts of the neuron to achieve complex activation.

2. The method for downlink transmission without physical layer channel feedback based on wireless map according to claim 1, characterized in that: The user location information in S1 includes location information p k and based on the location information p k Corresponding channel statistics obtained 3. The method for downlink transmission without physical layer channel feedback based on wireless map according to claim 2, characterized in that: Channel Statistics The entire area of ​​interest The second-order channel statistics at any position within, including the transmit correlation matrix R t (p), receiving correlation matrix R r (p) and channel fading matrix μ(p); Assume G∈C N×M is the beam domain channel matrix, whose elements are zero mean and independently distributed, then R t The expression of (p) is: Where, E is the mathematical expectation operation, R is the real number domain, M is the number of base station antennas and also the dimension of the obtained matrix; The coherent MIMO channel between the user and the base station is H = UGV H , where U and V are the deterministic unitary matrices of the channel, then R r The expression of (p) is: Where N is the number of user antennas, which is also the dimension of the resulting matrix; The elements of μ(p) represent the channel coefficients from different transmitting antennas to different receiving antennas. These coefficients reflect the attenuation and phase change of the signal during transmission and are used to describe the fading characteristics of the channel. The expression of μ(p) is:

Images