Superimposed CSI Feedback Method for LoS-Aware UAV Millimeter-Wave Systems

Abstract

The present invention discloses a superimposed CSI feedback method for a LoS-aware UAV millimeter-wave system, including: obtaining a sensing result through a LoS sensing module according to an estimation matrix of U2G CSI; performing compressed G2U CSI estimation on a received signal to obtain a compressed G2U CSI recovery vector; making a decision on the compressed G2U CSI recovery vector through a LoS decision maker according to the sensing result to obtain a compressed G2U CSI recovery vector with LoS existing or not existing; obtaining a compressed G2U CSI optimization vector with LoS existing through a LoS-assisted network according to the compressed G2U CSI recovery vector with LoS existing; obtaining a U2G data detection vector through superimposed interference cancellation for the compressed G2U CSI optimization vector with LoS existing or the compressed G2U CSI recovery vector with LoS not existing; and then obtaining a G2U CSI reconstruction vector through a CSI recovery network. Compared with non-superimposed CSI feedback, the present invention saves the energy consumption of a UAV transmitter and the bandwidth occupation of a UAV millimeter-wave system; compared with superimposed CSI feedback, the present invention improves the reconstruction accuracy of G2U CSI and U2G data.

Description

Technical Field

[0001] The present invention relates to the technical field of superimposed feedback of a millimeter-wave massive MIMO (multiple input multiple output) system of an unmanned aerial vehicle (UAV) in the frequency division duplex (FDD) mode, and particularly relates to a method for feeding back superimposed channel state information (CSI) based on LoS perception of a UAV millimeter-wave system. Background Art

[0002] As a key technology to meet the high reliability, excellent flexibility and large bandwidth availability of future 5G (the fifth generation wireless communication) networks, the UAV millimeter-wave massive MIMO system makes up for the significant path attenuation of millimeter-wave propagation through hundreds of antennas deployed at the ground base station end. At the same time, many operations that bring performance improvement in the UAV millimeter-wave massive MIMO system (such as the selection of modulation schemes, resource management and beamforming, etc.) rely on the acquisition of accurate ground base station to UAV user terminal (G2U, ground-to-UAV) CSI. In the frequency division duplex (FDD) mode, there is a weak reciprocity between channels in the UAV millimeter-wave massive MIMO system, and G2U CSI can only be fed back from the UAV user terminal to the base station.

[0003] Traditional feedback technologies based on compressed sensing (CS) can reduce the system feedback overhead to a certain extent by utilizing signal sparsity, but there is a large amount of computational overhead in the reconstruction process of compressed signals; feedback technologies based on deep learning (DL) have attracted wide attention due to their simple structure and fast online testing, etc., but still occupy additional spectrum resources in the process of their CSI feedback.

[0004] In recent years, superimposed coding (SC) technology has been widely applied to various fields of wireless communication due to its characteristic of being able to efficiently utilize spectrum resources. However, superimposed coding inevitably introduces superimposed interference. Existing CSI feedback schemes based on SC are not dedicated development methods for the UAV millimeter-wave massive MIMO system, so their applicability in the UAV millimeter-wave massive MIMO system remains to be further verified. In particular, the inherent characteristic of the high probability occurrence of the line-of-sight transmission (LoS, Line of Sight) scenario in the UAV millimeter-wave massive MIMO system has not been exploited. Summary of the Invention

[0005] The object of the present invention is to provide a superimposed CSI feedback method based on a LoS-aware UAV millimeter-wave system. Compared with non-superimposed CSI feedback, the present invention saves the energy consumption of the UAV transmitter and the bandwidth occupancy of the UAV millimeter-wave system; compared with superimposed CSI feedback, the present invention is developed based on the perception of the LoS scenario, greatly improving the reconstruction accuracy of G2U CSI and UAV-to-ground base station (U2G, UAV-to-ground) data, and at the same time significantly improving the reconstruction efficiency of G2U CSI and U2G data.

[0006] The technical solution of the present invention is as follows:

[0007] A superimposed CSI feedback method based on a LoS-aware UAV millimeter-wave system includes the following steps:

[0008] S1. According to the estimation matrix of U2G CSI Obtain the corresponding perception result w through the LoS perception module;

[0009] The U2G CSI refers to the channel state information from the UAV side to the ground base station;

[0010] The LoS refers to the direct path with the strongest energy within the line-of-sight range;

[0011] The estimation matrix of the U2G CSI Is obtained by performing channel estimation on the received signal The estimation method is selected from one or more of LS estimation, MMSE estimation, ML estimation, MAP estimation, and pilot-assisted estimation;

[0012] Wherein, L a Represents the number of cluster paths, N represents the number of antennas of the ground base station, and M represents the length of the U2G data;

[0013] S2. Compress the received signal Y at the base station side to recover the G2U CSI, and obtain the compressed G2U CSI recovery vector

[0014] The process of compressing the G2U CSI recovery includes: first despreading, first compressing the G2U CSI estimation, compressing the G2U CSI interference cancellation, U2G data detection, U2G data interference cancellation, second despreading, and second compressing the G2U CSI estimation;

[0015] The first despreading is the inverse process of spreading at the UAV side for the received signal Y to obtain the first despreading vector;

[0016] The compressed G2U CSI refers to the channel state information from the ground base station to the UAV side after compression;

[0017] The first compressed G2U CSI estimation refers to estimating the first despread signal with respect to the compressed G2U CSI to obtain a preliminary compressed G2U CSI estimation vector, and the estimation method is selected from one or more of LS estimation, MMSE estimation, ML estimation, MAP estimation, and pilot-assisted estimation;

[0018] The compressed G2U CSI interference cancellation refers to subtracting the preliminary compressed G2U CSI estimation vector from the received signal Y to obtain a preliminary U2G data vector;

[0019] The U2G data refers to the user data from the UAV side to the ground base station side in the transmission signal at the ground base station side;

[0020] The U2G data detection refers to detecting the preliminary U2G data vector with respect to the U2G data to obtain a preliminary U2G data detection vector, and the detection method is selected from one or more of ZF detection, MMSE detection, and maximum likelihood detection;

[0021] The compressed U2G data interference cancellation refers to subtracting the preliminary U2G data detection vector from the received signal Y to obtain a compressed G2U CSI vector after interference cancellation;

[0022] The second despreading refers to despreading the compressed G2U CSI vector after interference cancellation to obtain a second despread vector;

[0023] The second compressed G2U CSI estimation refers to estimating the second despread signal with respect to the compressed G2U CSI to obtain a compressed G2U CSI recovery vector The estimation method is selected from one or more of LS estimation, MMSE estimation, ML estimation, MAP estimation, and pilot-assisted estimation;

[0024] The compressed G2U CSI recovery vector refers to the corresponding compressed G2U CSI vector obtained by the ground base station side after recovering according to the compressed G2U CSI vector z in the transmission signal of the UAV side;

[0025] S3. According to the sensing result w, use the LoS discriminator to judge whether there is LoS in the compressed G2U CSI recovery vector to obtain the compressed G2U CSI recovery vector with LoS or the compressed G2U CSI recovery vector without LoS

[0026] S4. Recover the compressed G2U CSI recovery vector with LoS present Obtain the compressed G2U CSI optimization vector with LoS present through the LoS-assisted network

[0027] S5. The compressed G2U CSI optimization vector with LoS present Or the compressed G2U CSI recovery vector without LoS present Recover the U2G data detection vector through superposition interference cancellation

[0028] The superposition interference cancellation refers to subtracting the compressed G2U CSI optimization vector with LoS present Or the compressed G2U CSI recovery vector without LoS present From the equalized signal to obtain the U2G data detection vector

[0029] The equalized signal refers to the signal obtained after channel equalization processing of the received signal Y, and the channel equalization method is selected from one or more of ZF equalization, MMSE equalization, decision feedback equalization, maximum likelihood equalization, and blind equalization;

[0030] S6. The compressed G2U CSI optimization vector with LoS present Or the compressed G2U CSI recovery vector without LoS present Recover the G2U CSI reconstruction vector through the CSI recovery network

[0031] In some specific embodiments, the LoS sensing module in step S1 includes the following sub-steps:

[0032] S11. According to the estimated matrix of U2G CSI Obtain the preliminary sensing result ω using the LoS sensing network;

[0033] S12. According to the preliminary sensing result ω, obtain the sensing result w through a threshold hard decision maker.

[0034] In some preferred embodiments, the LoS sensing network includes:

[0035] An input layer with a linear activation function, a convolutional layer with a ReLU activation function, a max-pooling layer, a flattening layer, and a fully-connected output layer with a Sigmoid activation function; wherein, the size of the convolutional kernel of the convolutional layer is a×a, the number of convolutional kernels is b, the size of the filter of the max-pooling layer is c×c, and the number of neurons in the fully-connected output layer is 1. a, b, and c respectively represent the size of the convolutional kernel, the number of convolutional kernels, and the size of the filter of the max-pooling layer determined according to engineering presets;

[0036] Construct a training data set Train the LoS perception network to obtain the network parameters Θ of the LoS perception network e ;

[0037] The training data of the LoS perception network Obtained by real-valuing the estimated matrix of U2G CSI That is

[0038]

[0039] The training label e of the LoS perception network is obtained by judging whether there is a LoS scenario for the U2G CSI G. That is, if there is a LoS scenario for the U2G CSI G, then e = 1; otherwise, e = 0;

[0040] During online operation, the estimated matrix of U2G CSI After real-valuing, it is input into the LoS perception network to obtain a preliminary perception result ω;

[0041] In a specific application, the input of the LoS perception network is the real-valued form of the estimated matrix of the U2G CSI And the output is the preliminary perception result ω;

[0042] More preferably, the training loss function of the LoS perception network adopts the mean square error loss function.

[0043] In a specific application, the threshold of the threshold hard decision maker is β, and the perception result w obtained by the threshold hard decision maker is 0 or 1. β represents the threshold determined according to engineering presets.

[0044] In some specific embodiments, the LoS decision maker in step S3 further includes:

[0045] Taking the compressed G2U CSI recovery vector As the input of the LoS decision maker, and the decision condition of the LoS decision maker is the perception result w; when the perception result w is 0, the output of the LoS decision maker is the compressed G2U CSI recovery vector without LoS When the sensing result w is 1, the output of the LoS decision maker is the compressed G2U CSI recovery vector with LoS present

[0046] In some preferred embodiments, the LoS-assisted network described in step S4 includes:

[0047] An input layer with a linear activation function, a hidden layer with a Leaky ReLU activation function, and an output layer with a linear activation function; where the number of nodes in the input layer, hidden layer, and output layer are 2N, kN, and 2N respectively, and k represents the hidden layer node coefficient determined according to engineering presets;

[0048] Using the real-valued compressed G2U CSI vector sent from the UAV side as a label to construct a training dataset Train the LoS-assisted network to obtain the network parameters Θ of the LoS-assisted network z ;

[0049] The training data of the LoS-assisted network is obtained by real-valuing the compressed G2U CSI recovery vector with LoS present That is

[0050]

[0051] The real-valued compressed G2U CSI vector z label is obtained by real-valuing the compressed G2U CSI vector z, that is,

[0052] z label = [Re(z), Im(z)];

[0053] During online operation, the compressed G2U CSI recovery vector with LoS present is real-valued and input into the LoS-assisted network to obtain the optimized vector of the compressed G2U CSI recovery with LoS present

[0054] In a specific application, the input of this LoS-assisted network is the real-valued form of the compressed G2U CSI recovery vector with LoS present and the output is the optimized vector of the compressed G2U CSI recovery with LoS present

[0055] More preferably, the training loss function of this LoS-assisted network uses the mean square error loss function.

[0056] In some preferred embodiments, the CSI recovery network described in step S6 includes:

[0057] An input layer with a linear activation function, a hidden layer with a Leaky ReLU activation function, and an output layer with a linear activation function; where the number of nodes in the input layer, hidden layer, and output layer are 2N, qL a N, and 2L a N, where q represents the hidden layer node coefficient determined according to the engineering preset;

[0058] Using the real-valued G2U CSI vector h sent from the UAV side label as the label, there is a compressed G2U CSI optimization vector with LoS or a real-valued compressed G2U CSI recovery vector without LoS as the training input h train , construct a training dataset {h train , h label} to train the CSI recovery network and obtain the network parameters Θ h ;

[0059] The real-valued compressed G2U CSI recovery vector without LoS is obtained by real-valuing the compressed G2U CSI recovery vector without LoS , that is

[0060]

[0061] The real-valued G2U CSI vector h label is obtained by real-valuing the G2U CSI vector h, that is,

[0062] h label = [Re(h), Im(h)];

[0063] where Re(·) represents taking the real part and Im(·) represents taking the imaginary part;

[0064] During online operation, input the compressed G2U CSI optimization vector with LoS or the real-valued compressed G2U CSI recovery vector without LoS into the CSI recovery network to obtain the G2U CSI reconstruction vector

[0065] In a specific application, the input of the CSI recovery network is in the form of the compressed G2U CSI optimization vector with LoS or the real-valued compressed G2U CSI recovery vector without LoS , and the output is the G2U CSI reconstruction vector

[0066] More preferably, the training loss function of the CSI recovery network adopts the mean square error loss function.

[0067] The present invention utilizes the inherent characteristic that the LoS scenario highly probably occurs in the UAV millimeter-wave massive MIMO system. By using the LoS perception module to perceive whether there is LoS in the estimated matrix of the U2G CSI, it is determined whether there is LoS in the compressed G2U CSI recovery vector obtained by using compressed G2U CSI recovery. For the compressed G2U CSI recovery vector with LoS, the LoS-assisted network is used to obtain the compressed G2U CSI optimization vector with LoS, and then the U2G data detection vector is recovered through superposition interference cancellation, and the G2U CSI reconstruction vector is recovered through the CSI recovery network; for the compressed G2U CSI recovery vector without LoS, the U2G data detection vector and the G2U CSI reconstruction vector are directly recovered through superposition interference cancellation and the CSI recovery network respectively. In the UAV millimeter-wave massive MIMO system, the present invention can effectively reduce the mutual interference caused by superposition coding without increasing the energy consumption of the UAV transmitter and the spectrum overhead of the UAV millimeter-wave system, so as to ensure the reconstruction accuracy of the U2G data of the G2U CSI.

[0068] Compared with non-superposed CSI feedback, the present invention saves the energy consumption of the UAV transmitter and the bandwidth occupancy of the UAV millimeter-wave system; compared with superposed CSI feedback, the present invention is developed based on the perception of the LoS scenario, greatly improving the reconstruction accuracy of the G2U CSI and the U2G data, and at the same time significantly improving the reconstruction efficiency of the G2U CSI and the U2G data. Description of the Drawings

[0069] Figure 1 is the overall process schematic diagram of the present invention;

[0070] Figure 2 is the structural schematic diagram of the LoS perception network of the present invention;

[0071] Figure 3 is the structural schematic diagram of the LoS-assisted network of the present invention;

[0072] Figure 4 is the structural schematic diagram of the CSI recovery network of the present invention. Detailed Embodiments

[0073] The present invention will be described in detail below in conjunction with the embodiments and the drawings. However, it should be understood that the embodiments and the drawings are only used for exemplary description of the present invention, and cannot constitute any limitation to the protection scope of the present invention. All reasonable transformations and combinations within the scope of the inventive concept of the present invention fall within the protection scope of the present invention.

[0074] Referring to Figure 1 , a specific superimposed CSI feedback method for a LoS-aware UAV millimeter-wave system includes:

[0075] S1. Obtain the corresponding sensing result w through the LoS sensing module according to the estimated matrix of U2G CSI ;

[0076] Among them, some more specific implementation manners are as follows:

[0077] The estimated matrix of the U2G CSI is obtained by performing channel estimation on the received signal , and the estimation method is selected from one or more of LS estimation, MMSE estimation, ML estimation, MAP estimation, and pilot-assisted estimation;

[0078] Among them, L a represents the number of cluster paths, N represents the number of antennas of the ground base station, and M represents the length of the U2G data.

[0079] The LoS sensing module includes the following sub-steps:

[0080] S11. Obtain the preliminary sensing result ω by using the LoS sensing network according to the estimated matrix of U2G CSI ;

[0081] S12. Obtain the sensing result w through the threshold hard decision device according to the preliminary sensing result ω;

[0082] Referring to Figure 2 , the LoS sensing network includes:

[0083] An input layer containing a linear activation function, a convolutional layer containing a ReLU activation function, a max pooling layer, a flattening layer, and a fully connected output layer containing a Sigmoid activation function.

[0084] The convolutional kernel size of the convolutional layer of the LoS sensing network is a×a, the number of convolutional kernels is b, the filter size of the max pooling layer is c×c, and the number of neurons in the fully connected output layer is 1. a, b, and c respectively represent the convolutional kernel size, the number of convolutional kernels, and the filter size of the max pooling layer determined according to the engineering preset.

[0085] The process of obtaining the preliminary sensing result ω through the LoS sensing network includes:

[0086] Construct a training data set Train the LoS sensing network to obtain the network parameters Θ of the LoS sensing network e ;

[0087] The LoS perception network training data is obtained by real-valuing the estimation matrix of U2G CSI That is

[0088]

[0089] The LoS perception network training label e is obtained by determining whether there is a LoS scenario for the U2G CSI G. That is, if there is a LoS scenario in the U2G CSI G, then e = 1; otherwise, e = 0;

[0090] During online operation, the estimation matrix of U2G CSI is real-valued and then input into the LoS perception network to obtain a preliminary perception result ω;

[0091] In a specific application, the input of the LoS perception network is the real-valued form of the estimation matrix of the U2G CSI and the output is the preliminary perception result ω;

[0092] The training loss function of the LoS perception network uses the mean square error loss function.

[0093] In a specific application, the threshold of the threshold hard decision maker is β, the perception result w obtained by the threshold hard decision maker is 0 or 1, and β represents the threshold determined according to engineering presets.

[0094] S2. Compress the received signal Y at the base station side for G2U CSI recovery to obtain a compressed G2U CSI recovery vector

[0095] Among them, some more specific implementation manners are as follows:

[0096] The compressed G2U CSI recovery process includes: the first despreading, the first compressed G2U CSI estimation, compressed G2U CSI interference cancellation, U2G data detection, U2G data interference cancellation, the second despreading, and the second compressed G2U CSI estimation;

[0097] The first despreading is the inverse process of spreading the received signal Y at the UAV side to obtain a first despreading vector;

[0098] The compressed G2U CSI refers to the channel state information from the ground base station to the UAV side after compression;

[0099] The first compression G2U CSI estimation refers to estimating the first despread signal with respect to the compressed G2U CSI to obtain a preliminary compressed G2U CSI estimation vector, and the estimation method is selected from one or more of LS estimation, MMSE estimation, ML estimation, MAP estimation, and pilot-assisted estimation;

[0100] The compressed G2U CSI interference cancellation refers to subtracting the preliminary compressed G2U CSI estimation vector from the received signal Y to obtain a preliminary U2G data vector;

[0101] The U2G data refers to the user data from the UAV side to the ground base station side in the transmission signal of the ground base station;

[0102] The U2G data detection refers to detecting the preliminary U2G data vector with respect to the U2G data to obtain a preliminary U2G data detection vector, and the detection method is selected from one or more of ZF detection, MMSE detection, and maximum likelihood detection;

[0103] The compressed U2G data interference cancellation refers to subtracting the preliminary U2G data detection vector from the received signal Y to obtain a compressed G2U CSI vector after interference cancellation;

[0104] The second despreading refers to despreading the compressed G2U CSI vector after interference cancellation to obtain a second despread vector;

[0105] The second compression G2U CSI estimation refers to estimating the second despread signal with respect to the compressed G2U CSI to obtain a compressed G2U CSI recovery vector The estimation method is selected from one or more of LS estimation, MMSE estimation, ML estimation, MAP estimation, and pilot-assisted estimation;

[0106] The compressed G2U CSI recovery vector refers to the corresponding compressed G2U CSI vector obtained by the ground base station after recovering according to the compressed G2U CSI vector z in the transmission signal of the UAV side.

[0107] S3. According to the sensing result w, use the LoS discriminator to judge whether there is LoS in the compressed G2U CSI recovery vector to obtain a compressed G2U CSI recovery vector with LoS or a compressed G2U CSI recovery vector without LoS

[0108] Among them, some more specific implementation manners are as follows:

[0109] The compressed G2U CSI recovery vector As the input of the LoS detector, the decision condition of the LoS detector is the sensing result w; when the sensing result w is 0, the output of the LoS detector is the compressed G2U CSI recovery vector without LoS When the sensing result w is 1, the output of the LoS detector is the compressed G2U CSI recovery vector with LoS

[0110] S4. According to the compressed G2U CSI recovery vector with LoS Obtain the compressed G2U CSI optimization vector with LoS through the LoS-assisted network

[0111] Among them, some more specific implementation manners are as follows:

[0112] Refer to Figure 3 , the LoS-assisted network includes:

[0113] An input layer containing a linear activation function, a hidden layer containing a Leaky ReLU activation function, and an output layer containing a linear activation function.

[0114] The number of nodes in the input layer, hidden layer, and output layer of this LoS-assisted network are 2N, kN, and 2N respectively, where k represents the hidden layer node coefficient and can be obtained according to engineering presets.

[0115] The process of obtaining the compressed G2U CSI optimization vector with LoS through this LoS-assisted network includes:

[0116] Taking the real-valued compressed G2U CSI vector sent from the UAV side as a label to construct a training dataset Training the LoS-assisted network to obtain the network parameters Θ of the LoS-assisted network z ;

[0117] The training data of the LoS-assisted network is obtained by real-valuing the compressed G2U CSI recovery vector with LoS , that is

[0118]

[0119] The real-valued compressed G2U CSI vector z label is obtained by real-valuing the compressed G2U CSI vector z, that is,

[0120] z label = [Re(z), Im(z)];

[0121] During online operation, there will be a compressed G2U CSI recovery vector with LoS After being real-valued, it is input into the LoS-assisted network to obtain an optimized vector for the compressed G2U CSI recovery with LoS

[0122] In a specific application, the input of the LoS-assisted network is the real-valued form of the compressed G2U CSI recovery vector with LoS and the output is an optimized vector for the compressed G2U CSI recovery with LoS

[0123] The training loss function of the LoS-assisted network uses the mean square error loss function

[0124] S5. The compressed G2U CSI optimized vector with LoS or the compressed G2U CSI recovery vector without LoS is recovered through superposition interference cancellation to obtain a U2G data detection vector

[0125] Among them, some more specific implementation manners are as follows:

[0126] The superposition interference cancellation refers to subtracting the compressed G2U CSI optimized vector with LoS or the compressed G2U CSI recovery vector without LoS from the equalized signal to obtain a U2G data detection vector

[0127] The equalized signal refers to the signal obtained after performing channel equalization processing on the received signal Y, and the channel equalization method is selected from one or more of ZF equalization, MMSE equalization, decision feedback equalization, maximum likelihood equalization, and blind equalization

[0128] S6. According to the compressed G2U CSI optimized vector with LoS or the compressed G2U CSI recovery vector without LoS a G2U CSI reconstruction vector is recovered through a CSI recovery network

[0129] Among them, some more specific implementation manners are as follows:

[0130] Referring to Figure 4 , the CSI recovery network includes:

[0131] an input layer containing a linear activation function, a hidden layer containing a Leaky ReLU activation function, and an output layer containing a linear activation function

[0132] The number of nodes in the input layer, hidden layer, and output layer of the CSI recovery network are 2N, qL a N, and 2L a N, where q represents the node coefficient of the hidden layer and can be obtained according to engineering presets.

[0133] The process of obtaining the G2U CSI reconstruction vector through the CSI recovery network includes:[[]]

[0134] Using the real-valued G2U CSI vector h label sent from the UAV side as a label, the compressed G2U CSI optimization vector with LoS or the real-valued compressed G2U CSI recovery vector without LoS as the training input h train , constructing a training dataset {h train , h label} to train the CSI recovery network and obtain the network parameters Θ h ;

[0135] The real-valued compressed G2U CSI recovery vector without LoS is obtained by real-valuing the compressed G2U CSI recovery vector without LoS , that is

[0136]

[0137] The real-valued G2U CSI vector h label is obtained by real-valuing the G2U CSI vector h, that is,

[0138] h label = [Re(h), Im(h)];

[0139] where Re(·) represents taking the real part and Im(·) represents taking the imaginary part;

[0140] During online operation, the compressed G2U CSI optimization vector with LoS or the real-valued compressed G2U CSI recovery vector without LoS is input into the CSI recovery network to obtain the G2U CSI reconstruction vector

[0141] In a specific application, the input of the CSI recovery network is the compressed G2U CSI optimization vector with LoS or the real-valued compressed G2U CSI recovery vector without LoS in the form of, the output is the G2U CSI reconstruction vector

[0142] The training loss function of the CSI recovery network adopts the mean square error loss function.

[0143] Embodiment 1

[0144] In step S1, obtain the LoS perception network training data A specific embodiment is as follows:

[0145] Assume: L a = 2, N = 3, the estimation matrix of the U2G CSI for real-valued conversion is:

[0146]

[0147] For the estimation matrix of the U2G CSI perform real-valued conversion, that is the LoS perception network training data can be calculated as:

[0148]

[0149] Embodiment 2

[0150] In step S4, obtain the LoS-assisted network training data A specific embodiment is as follows:

[0151] Assume: N = 3, the compressed G2U CSI recovery vector with LoS is:

[0152]

[0153] For the compressed G2U CSI recovery vector with LoS perform real-valued conversion, that is the LoS-assisted network training data can be calculated as:

[0154]

[0155] Embodiment 3

[0156] In step S6, obtain the real-valued G2U CSI vector h label A specific embodiment is as follows:

[0157] Assume: N = 3, L a = 2, the G2U CSI vector h is:

[0158]

[0159] Perform real-valued processing on the G2U CSI vector h, that is, h label = [Re(h), Im(h)], and the real-valued G2U CSI vector h serving as the label of the CSI recovery network can be calculated label as:

[0160]

[0161] The above embodiments are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. Superimposed CSI feedback method for LoS-aware UAV millimeter-wave systems, characterized in that, Including the following steps: S1. According to the estimation matrix of U2G CSI Obtain the corresponding sensing result w through the LoS sensing module; The U2G CSI refers to the channel state information from the UAV side to the ground base station; The LoS refers to the direct path with the strongest energy within the line-of-sight range; The estimated matrix of the U2G CSI Obtained by performing channel estimation on the received signal ; Among them, L a represents the number of cluster paths, N represents the number of antennas of the ground base station, and M represents the length of U2G data; S2. Compress the received signal Y at the base station side for G2U CSI recovery to obtain a compressed G2U CSI recovery vector The compression G2U CSI recovery process includes: the first despreading, the first compression G2U CSI estimation, compression G2U CSI interference cancellation, U2G data detection, U2G data interference cancellation, the second despreading, and the second compression G2U CSI estimation; The compression G2U CSI refers to the compressed channel state information from the ground base station to the UAV side; The first despreading is the inverse process of spreading the received signal Y at the UAV side to obtain the first despreading vector; The first compression G2U CSI estimation refers to estimating the first despread signal with respect to the compression G2U CSI to obtain a preliminary compression G2U CSI estimation vector, and the estimation method is selected from one or more of LS estimation, MMSE estimation, ML estimation, MAP estimation, and pilot-assisted estimation; The compression G2U CSI interference cancellation refers to subtracting the preliminary compression G2U CSI estimation vector from the received signal Y to obtain a preliminary U2G data vector; The U2G data refers to the user data from the UAV side to the ground base station side in the transmission signal at the ground base station side; The U2G data detection refers to detecting the preliminary U2G data vector with respect to the U2G data to obtain a preliminary U2G data detection vector, and the detection method is selected from one or more of ZF detection, MMSE detection, and maximum likelihood detection; The U2G data interference cancellation refers to subtracting the preliminary U2G data detection vector from the received signal Y to obtain a compression G2U CSI vector after interference cancellation; The second despreading refers to despreading the compression G2U CSI vector after interference cancellation to obtain a second despreading vector; The second compression G2U CSI estimation refers to estimating the second despread signal with respect to the compressed G2U CSI to obtain a compressed G2U CSI recovery vector The estimation method is selected from one or more of LS estimation, MMSE estimation, ML estimation, MAP estimation, and pilot-assisted estimation; The compressed G2U CSI recovery vector refers to the corresponding compressed G2U CSI vector obtained after the ground base station recovers according to the compressed G2U CSI vector z in the signal sent by the UAV side; S3. According to the perception result w, use the LoS detector to determine whether there is a LoS for the compressed G2U CSI recovery vector to obtain the compressed G2U CSI recovery vector with LoS or the compressed G2U CSI recovery vector without LoS S4. Compressed G2U CSI Recovery Vector with LoS Present Obtain the Compressed G2U CSI Optimization Vector with LoS Present through the LoS-Assisted Network S5. Optimize the compressed G2U CSI vector with LoS or recover the compressed G2U CSI vector without LoS to obtain the U2G data detection vector through recovery by superposition interference cancellation The superimposed interference cancellation refers to subtracting the compressed G2U CSI optimization vector with LoS or the compressed G2U CSI recovery vector without LoS from the equalized signal to obtain the U2G data detection vector The equalized signal refers to the signal obtained by performing channel equalization processing on the received signal Y, and the channel equalization method is selected from one or more of ZF equalization, MMSE equalization, decision feedback equalization, maximum likelihood equalization, and blind equalization; S6. Optimize the vector according to the compressed G2U CSI with LoS present or the restored vector of the compressed G2U CSI without LoS present Obtain the reconstructed vector of the G2U CSI by restoring it through the CSI restoration network 2. The superposition CSI feedback method for the LoS-aware UAV millimeter-wave system according to claim 1, wherein The LoS sensing module in step S1 includes the following sub-steps: S11. According to the estimation matrix of U2G CSI Use the LoS perception network to obtain the preliminary perception result ω; The LoS sensing network includes: An input layer with a linear activation function, a convolutional layer with a ReLU activation function, a max pooling layer, a flattening layer, and a fully connected output layer with a Sigmoid activation function; where the size of the convolutional kernel in the convolutional layer is a×a, the number of convolutional kernels is b, the size of the filter in the max pooling layer is c×c, and the number of neurons in the fully connected output layer is 1. a, b, and c respectively represent the size of the convolutional kernel, the number of convolutional kernels, and the size of the filter in the max pooling layer determined according to the engineering preset; Construct a training dataset Train the LoS perception network to obtain the network parameters Θ of the LoS perception network e ; The LoS perception network training data is the estimated matrix of U2G CSI obtained by real-valuing, that is The training label e of the LoS sensing network is obtained by making a decision on whether there is LoS for the U2G CSI G, that is, if there is LoS in the U2G CSI G, then e = 1, otherwise, e = 0; During online operation, the estimated matrix of U2G CSI is made real-valued and then input into the LoS perception network to obtain a preliminary perception result ω; S12. According to the preliminary sensing result ω, obtain the sensing result w through a threshold hard decision maker; Among them, the threshold of the threshold hard decision maker is β, the sensing result w obtained by the threshold hard decision maker is 0 or 1, and β represents the threshold determined according to the engineering preset.

3. The superimposed CSI feedback method for the LoS-aware UAV millimeter-wave system according to claim 1, wherein The LoS decision maker described in step S3 further includes: The compressed G2U CSI recovery vector is used as the input of the LoS detector, and the decision condition of the LoS detector is the sensing result w; when the sensing result w is 0, the output of the LoS detector is the compressed G2U CSI recovery vector without LoS When the sensing result w is 1, the output of the LoS detector is the compressed G2U CSI recovery vector with LoS 4. The superimposed CSI feedback method for the LoS-aware UAV millimeter-wave system according to claim 1, wherein The LoS assistance network described in step S4 further includes: An input layer containing a linear activation function, a hidden layer containing a Leaky ReLU activation function, and an output layer containing a linear activation function; among them, the number of nodes in the input layer, hidden layer, and output layer are 2N, kN, and 2N respectively, and k represents the hidden layer node coefficient determined according to the engineering preset; Use the real-valued compressed G2U CSI vector sent from the UAV side as a label to construct a training dataset Train the LoS-assisted network to obtain the network parameters Θ of the LoS-assisted network z ; The LoS-assisted network training data is obtained by real-valuing the compressed G2U CSI recovery vector with LoS, i.e., is obtained by real-valuing the compressed G2U CSI recovery vector with LoS, i.e., The real-valued compressed G2U CSI vector z label is obtained by real-valuing the compressed G2U CSI vector z, i.e.: z label = [Re(z), Im(z)]; When running online, there will be a compressed G2U CSI recovery vector with LoS After being made real-valued, it is input into the LoS-assisted network to obtain an optimized vector for the compressed G2U CSI recovery with LoS 5. The superimposed CSI feedback method for a LoS-aware UAV millimeter-wave system according to claim 1, characterized in that The CSI recovery network described in step S6 further includes: An input layer with a linear activation function, a hidden layer with a Leaky ReLU activation function, and an output layer with a linear activation function; wherein the number of nodes in the input layer, hidden layer, and output layer are 2N, qL a N, and 2L a N, where q represents the hidden layer node coefficient determined according to the engineering preset; Using the real-valued G2U CSI vector sent from the UAV side as a label, there is a compressed G2U CSI optimization vector with LoS or a real-valued compressed G2U CSI recovery vector without LoS as the training input Construct a training dataset {h train , h label} to train the CSI recovery network and obtain the network parameters Θ h ; The real-valued LoS-absent compressed G2U CSI recovery vector is obtained by real-valuing the LoS-absent compressed G2U CSI recovery vector i.e.: The real-valued G2U CSI vector h label is obtained by real-valuing the G2U CSI vector h, i.e., h label = [Re(h), Im(h)]; Among them, Re(·) represents taking the real part, and Im(·) represents taking the imaginary part; During online operation, the compressed G2U CSI optimization vector with LoS will be present or the real-valued compressed G2U CSI recovery vector without LoS is input into the CSI recovery network to obtain the G2U CSI reconstruction vector

Images