CN114157331B - Large-scale MIMO channel state information feedback method based on pseudo complex value input - Google Patents
Large-scale MIMO channel state information feedback method based on pseudo complex value input Download PDFInfo
- Publication number
- CN114157331B CN114157331B CN202111563073.8A CN202111563073A CN114157331B CN 114157331 B CN114157331 B CN 114157331B CN 202111563073 A CN202111563073 A CN 202111563073A CN 114157331 B CN114157331 B CN 114157331B
- Authority
- CN
- China
- Prior art keywords
- layer
- encoder
- convolution
- decoder
- matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 21
- 239000011159 matrix material Substances 0.000 claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 13
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 230000009466 transformation Effects 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 26
- 230000006835 compression Effects 0.000 claims description 11
- 238000007906 compression Methods 0.000 claims description 11
- 238000010586 diagram Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 8
- 238000012937 correction Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000000137 annealing Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000007493 shaping process Methods 0.000 claims description 3
- 230000003213 activating effect Effects 0.000 claims description 2
- 238000004891 communication Methods 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 238000012795 verification Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000006978 adaptation Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000002679 ablation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000969 carrier Substances 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000002498 deadly effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/0413—MIMO systems
- H04B7/0417—Feedback systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/0413—MIMO systems
- H04B7/0456—Selection of precoding matrices or codebooks, e.g. using matrices antenna weighting
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station
- H04B7/0613—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission
- H04B7/0615—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission of weighted versions of same signal
- H04B7/0619—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission of weighted versions of same signal using feedback from receiving side
- H04B7/0658—Feedback reduction
- H04B7/0663—Feedback reduction using vector or matrix manipulations
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Radio Transmission System (AREA)
Abstract
The invention discloses a large-scale MIMO channel state information feedback method based on pseudo complex value input, which comprises the following operation steps: (1) Acquiring space frequency domain channel state information at a mobile user terminal and preprocessing to acquire a channel matrix in an angular delay domain; (2) Constructing a feedback model CLPNet comprising an encoder and a decoder; the encoder is deployed at the user end, and the decoder is deployed at the base station; the user inputs the channel matrix into the encoder and feeds back to the base station; the base station obtains a reconstructed channel matrix according to the input to the decoder; (3) Training a feedback model CLPNet to minimize a channel matrix and obtain model parameters; (4) Zero padding and two-dimensional inverse transformation operations are carried out on the matrix output by the feedback model CLPNet to recover the channel matrix of the original space frequency domain, and feedback is completed.
Description
Technical Field
The invention relates to the technical field of communication, in particular to a large-scale MIMO channel state information feedback method.
Background
Large-scale Multiple Input Multiple Output (MIMO) technology is widely recognized as one of the core technologies of the next-generation communication system. After a large number of antennas are configured at the base station end, the base station can greatly improve the channel capacity by utilizing diversity reception and multiplexing technology. The premise of obtaining the potential gain of massive MIMO is that the base station end must receive accurate downlink channel state information. In Frequency Division Duplex (FDD) mode of operation of most cellular systems, there is no reciprocity between channels. Therefore, the device side (UE) must precisely feed back downlink Channel State Information (CSI) to the Base Station (BS). However, when using pilot training downlink for channel estimation, the overhead increases exponentially with the number of antennas, which is a major obstacle for practical deployment of FDD massive MIMO systems. There is a need for CSI compression in the feedback forward to reduce overhead.
In the current research on the feedback of the channel state information of the massive MIMO system, the main idea of most algorithms is to utilize the space-time correlation of the massive MIMO system to reduce the feedback overhead. In particular, the channel matrix may be converted to a vector that may be sparsely represented using the designed orthogonal basis. And then carrying out random compression on the sparse vector at the user end to obtain a low-dimensional measurement value, transmitting the low-dimensional measurement value to the base station end through a feedback link, and reconstructing the original sparse vector by the base station by means of a compressed sensing theory. However, in practical situations, it is difficult to achieve an absolute sparsity effect on the matrix using orthogonal basis. Therefore, the premise of compressed sensing is that sparsity is assumed, and not much practical application is achieved. And the traditional compressed sensing theory has the deadly defects of slow iteration, low efficiency and insufficient utilization of a channel structure.
For these problems, a massive MIMO channel state information feedback method based on deep learning is proposed in patent application No. 201810090648.0. The method takes the pure digital component channel matrix as an image, but does not consider the signal as a complex physical characteristic, most of processing methods are to divide the complex image into virtual and real, the complex phase characteristic cannot be expressed, and the original physical information is lost. In most cases, therefore, the recovery accuracy is insufficient, and particularly, in the case of a low compression ratio, it is too poor. In the large-scale MIMO time-varying channel state information compression feedback and reconstruction method of patent No. 201810833351.9, although deployed at the base station end, the complexity is too high, and the performance is not improved much.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention provides a large-scale MIMO channel state information feedback method based on pseudo complex value input, which can quickly and accurately improve recovery precision and reduce feedback overhead under the condition of low compression ratio.
The purpose of the invention is realized in the following way: a large-scale MIMO channel state information feedback method based on pseudo complex value input comprises the following operation steps:
(1) Obtaining space frequency domain channel state information at mobile user terminal and preprocessing to obtain channel matrix H in angle delay domain a ;
(2) Constructing a feedback model CLPNet comprising an encoder and a decoder; the encoder is deployed at the user end, and the decoder is deployed at the base station; the user matrix H a Inputting an encoder to obtain a code word c with lower dimension and feeding back the code word c to a base station; codeword c obtained by the base station is input to a decoder to obtain a reconstructed channel matrix
(3) Training the feedback model CLPNet to make the channel matrixAnd channel matrix H a Minimizing and obtaining model parameters;
(4) Zero padding and two-dimensional inverse transformation operation are carried out on the matrix output by the feedback model CLPNet, and the channel matrix of the original space-frequency domain is obtainedAnd (5) finishing feedback.
As a further definition of the present invention, the step (2) includes the steps of:
(21) Channel matrix H to be an angular delay domain a As input to the encoder, the output is a compressed codeword c, the encoder comprising in sequence a first convolutional layer, a spatial attention block, a second convolutional layer, and a full-concatenated layer;
(22) The code word c is fed back to the base station end, decoded by a designed decoder, and the decoder takes the code word c as input, outputs and outputs a channel matrix H by the zero filling operation of a convolution kernel a Channel matrix of the same dimensionThe decoder includes a full link layer, two convolutional layers and an MLP block.
As a further limitation of the invention, the input of the encoder is two layers of real matrixes with the size of 32×32, the two layers are the real part and the imaginary part of the complex matrix H, the first layer of the encoder is a convolution layer with the output of 16 channels, the convolution kernel is 1×1, and the function of the encoder is to combine the two layers of the virtual-real separated matrixes into a characteristic diagram, so that the real-virtual fusion is completed, and certain physical information is reserved; the second layer of the encoder is a spatial attention module, and the operation inside the module is to adopt the average pooling and the maximum pooling operation on a channel C of an input F to generate two 2D feature maps; then concatenating the two feature maps to generate a compressed spatial feature descriptor, then convolving it with a standard layer, the convolution kernel size being 3, the output being 1 channel, generating a 2D spatial attention mask, activating the mask with sigmoid, and finally convolving it with the original feature map F i Multiplication results in F with spatial attention o :F o =F i (σ(f c (F avg ;F max ) A) is set forth; the output of the space attention block is connected with a 3X 3 convolution kernel, so that the channels are restored into two channels; the two matrixes are reduced into a one-dimensional matrix, a characteristic diagram with the matrix size of 2048 multiplied by 1 is input into a full-connection layer containing M neurons, and a vector with the size of M multiplied by 1 is output by adopting a linear activation function, namely a codeword c which is compressed by a user to be transmitted to a base station; the compression ratio CR, cr=m/N, N being the number of neurons of the input layer, can be varied by varying the dimension M of the output layer.
As a further limitation of the present invention, the first layer of the decoder is a fully-connected layer whose output contains 2048 neurons, takes the received codeword c as input, and adopts a matrix with size of mx 1, and outputs a vector of 2048×1 by adopting a linear activation function; through the shaping reshape operation, the vector is recombined into two layers of matrixes with the size of 32 multiplied by 32, the matrixes are input into a convolution layer of a second layer, the convolution belongs to head convolution operation, the convolution kernel size is 5 multiplied by 5, and the output size is 32 channels; the third layer is a CLPblock unit which is divided into a main line and a sub line, wherein the main line is formed by three different convolution kernelsThe small convolution layers are formed in parallel, the convolution kernel size distribution is 3*3, 5*5 and 7*7, a batch normalization and linear correction unit LaekyRelu with leakage is used in the convolution layers, and 96 channels are reduced to 32 channels through a 1X 1 convolution kernel after being spliced; finally, according to the residual thought, a sub line is led out from the previous convolution layer, and the sub line and the 1 multiplied by 1 convolution layer output are added and activated by adopting a LeakyRelu function; the input of the second convolution layer is MLPblock output which is a matrix of two layers 32×32 and is activated by HardSigmoid function, and the output of the activation function is used as the final reconstructionReal and imaginary parts of (a) are provided.
As a further definition of the present invention, the model parameters in the step (3) include weights and offsets of the full connection layer and convolution kernels and offsets of the convolution layer.
As a further limitation of the present invention, the step (3) adopts Adam optimization mode and end-to-end learning mode, the learning rate adopts dynamic learning rate, i.e. cosine annealing algorithm, the encoder and decoder are jointly trained to minimize cost function, and the cost function of the whole CLPNet architecture is designed as the channel matrix output by the decoderAnd a real channel matrix H a The cost function is described as follows:
wherein C is the number of samples in the training set, I.I 2 Is the euclidean norm;f e ,f d ,θ e and theta d Represented are encoder, decoder, encoder parameters and decoder parameters, respectively.
Compared with the prior art, the invention has the beneficial effects that:
the channel state information model built by the invention designs a pseudo complex value input layer at the encoder end and adds a space attention mechanism module, compared with CsiNet which takes a channel matrix as an image to carry out virtual-real separation processing, the CLPNet not only reserves phase characteristics, but also distributes different weights for different clusters by utilizing the space attention mechanism characteristics, and gives more attention to resolvable paths with specific delay and arrival angles; the decoder end designs a parallel large convolution channel and a multi-convolution kernel, so that the received image has more characteristics and different receptive fields; under the same compression ratio. CLPNet can achieve smaller reconstruction errors, thereby improving the quality of the reconstructed matrix.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flow chart of the operation of the present invention.
Fig. 2 is a diagram showing the overall structure of CLPNet according to the present invention.
Fig. 3 is a diagram of a spatial attention unit according to the present invention.
Fig. 4 is a block diagram of a CLPblock cell of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Examples
The large-scale MIMO channel state information feedback method based on pseudo complex value input as shown in fig. 1 comprises the following operation steps:
(1) Obtaining space frequency domain channel state information at mobile user terminal and preprocessing to obtain channel matrix H in angle delay domain a The method comprises the steps of carrying out a first treatment on the surface of the In this embodiment, in a massive MIMO communication system, frequency division duplex mode FDD is used, and a base station side is equipped with N t The system uses carrier modulation mode of orthogonal frequency division multiplexing, the number of sub-carriers isUsing COST2100 channel model to produce 150000 space-frequency domain channel matrix samples under 5.3GHz indoor environment by Matlab simulation software, and dividing the samples into training set (100000), verification set (30000) and test set (20000) according to a certain proportion; the produced matrix is directly used as the input of a model, and the user side estimates the downlink channel state information of the large-scale MIMO system according to the training sequence or the pilot frequency sequence sent by the base station to obtain a space-frequency domain channel matrix ++>Matrix size +.>Space-frequency domain channel matrix->Performing two-dimensional discrete Fourier transform to obtain a channel sparse matrix H of an angular delay domain, wherein the transformation is completed through two DFT matrices, and the expression of the coefficient matrix is +.>Wherein->Andbecause the multipath delay is limitedIn order to take the channel matrix in the angular delay domain +.>Rows, get matrix H a Thus, too much information loss is not caused; in this embodiment, 32 rows, N a =32, subcarrier->Taking 1024; thus, based on the number of transmitting antennas and the number of reserved rows, the matrix H after correction is finally obtained a The size is 32 multiplied by 32;
(2) Building a feedback model CLPNet including an encoder and a decoder as shown in fig. 2;
the encoder is deployed at the user side and comprises two convolution layers, a spatial attention block and a full connection layer, wherein the input of the encoder is a real matrix with the size of 32 multiplied by 32, the two layers are the real part and the imaginary part of a complex matrix H, the first layer is a convolution layer with the output of 16 channels, the convolution kernel is 1 multiplied by 1, and the encoder has the function of combining the two layers of the virtual-real separated matrices into a characteristic diagram, thus completing the real-virtual fusion and reserving certain physical information; the second layer is a spatial attention module as shown in fig. 3, and the operations inside the module are to generate two 2D feature maps by adopting average pooling and maximum pooling operations on the channel C of the input F, and the two feature maps are shown in the specificationSubsequently concatenating the two feature maps to generate a compressed spatial property descriptor +.>Then convolving it with standard layer (convolution kernel size 3), outputting 1 channel, generating 2D space attention mask +.>Using sigmoid to activate mask, and finally combining with original feature diagram F i Multiplication results in F with spatial attention o :F o =F i (σ(f c (F avg ;F max ) A) is set forth; the output of the space attention block is connected with a 3X 3 convolution kernel, so that the channels are restored into two channels; the two matrixes are reduced into a one-dimensional matrix, a characteristic diagram with the matrix size of 2048 multiplied by 1 is input into a full-connection layer containing M neurons, and a vector with the size of M multiplied by 1 is output by adopting a linear activation function, namely a codeword c which is compressed by a user to be transmitted to a base station; the compression ratio CR can be changed by changing the dimension M of the output layer, cr=m/N, N being the number of neurons of the input layer;
the decoder is deployed at the base station end; the method comprises a full-connection layer, a CLPblock block and two convolution layers, wherein the first layer outputs the full-connection layer comprising 2048 neurons, the received codeword c is taken as input, a matrix with the size of M multiplied by 1 is subjected to linear activation function, and a vector with the size of 2048 multiplied by 1 is output; through the shaping reshape operation, the vector is recombined into two layers of matrixes with the size of 32 multiplied by 32, the matrixes are input into a convolution layer of a second layer, the convolution belongs to head convolution operation, the highest performance is obtained when the convolution kernel size is 5 multiplied by 5 through ablation verification, the output size is 32 channels, and more characteristic diagrams can be obtained for more convolution channels; the third layer is a CLPblock unit as shown in fig. 4, the unit is divided into a main line and a sub line, the main line is formed by three convolution layers with different convolution kernel sizes in parallel, the convolution kernel sizes are 3*3, 5*5 and 7*7, the large convolution kernel is decomposed into a plurality of 3*3 forms through convolution decomposition, then 5*5 can be replaced by two 3*3, 7*7 can be replaced by three 3*3 convolution kernels, and thus parameters can be effectively reduced; the convolution layer uses a batch normalization and linear correction unit LaekyRelu with leakage, and 96 channels are reduced to 32 channels through a 1X 1 convolution kernel after splicing; finally, according to the residual thought, a sub line is led out from the previous convolution layer, and the sub line and the 1 multiplied by 1 convolution layer output are added and activated by adopting a LeakyRelu function; the input of the second convolution layer is MLPblock output which is a matrix of two layers 32×32 and is activated by HardSigmoid function, and the output of the activation function is used as the final reconstructionReal and imaginary parts of (a) are provided.
(3) The feedback model CLPNet is trained on,matrix the channelsAnd channel matrix H a Minimizing and obtaining model parameters; designing the cost function of the whole CLPNet architecture as the channel matrix of the decoder output +.>And a real channel matrix H a Mean square error of (i.e. cost function)>Wherein C is the number of samples in the training set, I.I 2 Is the euclidean norm; />Wherein f e ,f d ,θ e And theta d Represented are encoder, decoder, encoder parameters and decoder parameters, respectively;
training according to a channel matrix generated by COST2100 in an end-to-end learning mode, selecting Adam for optimization by an optimizer in the training process, wherein 200 samples are used in each batch, a cosine annealing algorithm is adopted for the learning rate, and the whole training set is traversed 1000 times from the initial 0.002 to the final 0.00005; the best-performing model is selected by using the verification set every 10 traversals in the training process, and tested by using the test set
(4) The trained CLPNet model is finally deployed at the user end and the base station for large-scale MIMO channel state information feedback, and a reconstruction matrix is utilizedZero padding operation is carried out, two-dimensional discrete Fourier transform is carried out, and the channel matrix of the original space-frequency domain is obtained
The technical effects of the present invention are explained by experiments as follows:
the verification example is a large-scale MIMO channel state information feedback method based on pseudo complex value input, and an encoder and decoder architecture designed through data driving is adopted; the simulation of the invention and the prior art adopts samples generated by COST2100MIMO channels in a 5.3GHz indoor scene, the transmitting antenna at the base station end is 32, the single antenna at the user end, the system uses the carrier modulation mode of orthogonal frequency division multiplexing, the number of subcarriers is 1024, and the parameter setting is the default parameter selection.
The simulation content: the CLPNet and CsiNet-LSTM of the invention are applied to reconstructed channel matrix under different compression ratiosAnd the original channel matrix H a Performing simulation comparison on the direct normalized mean square error; as in table one, the formula for normalizing the mean square error is described as follows: />Units dB.
List one
From the first table, the error performance of the CLPNet under a plurality of compression ratios in the three simulation methods is greatly improved. The feasibility of the attention mechanism in the communication direction is also demonstrated by the greater weight distribution of the attention mechanism at the encoder side on the resolvable paths and the greater feature patterns generated by the large convolution channels at the decoder side. Therefore, CLPNet has potential application value in FDD massive MIMO channel state information feedback work
The above description of the embodiments is only for aiding in the understanding of the method of the present invention and its core ideas. It should be noted that it will be apparent to those skilled in the art that various modifications and adaptations of the invention can be made without departing from the principles of the invention and these modifications and adaptations are intended to be within the scope of the invention as defined in the following claims.
Claims (4)
1. A large-scale MIMO channel state information feedback method based on pseudo complex value input is characterized by comprising the following operation steps:
(1) Obtaining space frequency domain channel state information at mobile user terminal and preprocessing to obtain channel matrix H in angle delay domain a ;
(2) Constructing a feedback model CLPNet comprising an encoder and a decoder; the encoder is deployed at the user end, and the decoder is deployed at the base station; the user matrix H a Inputting an encoder to obtain a code word c with lower dimension and feeding back the code word c to a base station; codeword c obtained by the base station is input to a decoder to obtain a reconstructed channel matrixThe step (2) comprises the following steps:
(21) Channel matrix H to be an angular delay domain a As input to the encoder, the output is a compressed codeword c, the encoder comprising in sequence a first convolutional layer, a spatial attention block, a second convolutional layer, and a full-concatenated layer; the input of the encoder is two layers of real matrixes with the size of 32 multiplied by 32, the two layers are the real part and the imaginary part of a complex matrix H, the first layer of the encoder is a convolution layer with the output of 16 channels, the convolution kernel is 1 multiplied by 1, and the function of the encoder is to combine the two layers of the real and imaginary matrixes separated into a characteristic diagram to finish real and imaginary fusion; the second layer of the encoder is a space attention block, and the operation inside the module is to adopt the average pooling and the maximum pooling operation on a channel C of an input F to generate two 2D feature maps; then concatenating the two 2D feature maps to generate a compressed spatial feature descriptor, then convolving it with a standard layer convolution kernel of size 3, outputting it as 1-channel, generating a 2D spatial attention mask, activating the mask with sigmoid, and finally convolving it with the original feature map F i Multiplication results in F with spatial attention o :F o =F i (σ(f c (F avg ;F max ) A) is set forth; the output of the space attention block is connected with a 3X 3 convolution kernel, so that the channels are restored into two channels; reducing the two matrixes after average pooling and maximum pooling into a one-dimensional matrix with the size of2048×1, inputting to a full-connection layer containing M neurons, and outputting vector with size of m×1 by using linear activation function, i.e. codeword c compressed by user to base station; the compression ratio CR can be changed by changing the dimension M of the output layer, cr=m/N, N being the number of neurons of the input layer;
(22) The code word c is fed back to the base station end, decoded by a designed decoder, and the decoder takes the code word c as input, outputs and outputs a channel matrix H by the zero filling operation of a convolution kernel a Channel matrix of the same dimensionThe decoder comprises a full connection layer, two convolution layers and a CLPblock block;
(3) Training the feedback model CLPNet to make the channel matrixAnd channel matrix H a Minimizing and obtaining model parameters;
(4) Zero padding and two-dimensional inverse transformation operation are carried out on the matrix output by the feedback model CLPNet, and the channel matrix of the original space-frequency domain is obtainedAnd (5) finishing feedback.
2. The feedback method of massive MIMO channel state information based on pseudo complex-valued input of claim 1, wherein the first layer of the decoder is a fully-connected layer, converts codewords of size mx 1 into 2048 neurons, and outputs a vector of 2048 x 1 by using a linear activation function; through the shaping reshape operation, the vector is recombined into two layers of matrixes with the size of 32 multiplied by 32, the matrixes are input into a convolution layer of a second layer, the convolution belongs to head convolution operation, the convolution kernel size is 5 multiplied by 5, and the output size is 32 channels; the third layer is a CLPblock unit which is divided into a main line and a sub line, the main line is formed by three convolution layers with different convolution kernel sizes in parallel, the convolution kernel sizes are distributed as 3*3, 5*5 and 7*7,the convolution layer uses batch normalization and a linear correction unit LaekyRelu with leakage, and 96 channels are reduced to 32 channels through a convolution kernel of 1 multiplied by 1 after being spliced; finally, according to the residual thought, a sub line is led out from the previous convolution layer, and the sub line and the 1 multiplied by 1 convolution layer output are added and activated by adopting a LeakyRelu function; the input of the second convolution layer is MLPblock output which is a matrix of two layers 32×32 and is activated by HardSigmoid function, and the output of the activation function is used as the final reconstructionReal and imaginary parts of (a) are provided.
3. The massive MIMO channel state information feedback method based on pseudo-complex value input of claim 2, wherein the model parameters in step (3) include weights of full-connection layers, offsets, and convolution kernels of convolution layers, offsets.
4. The method for feeding back the channel state information of the massive MIMO based on the pseudo-complex value input according to claim 3, wherein the step (3) adopts an Adam optimization mode and an end-to-end learning mode, the learning rate adopts a dynamic learning rate, namely a cosine annealing algorithm, the encoder and the decoder are jointly trained to minimize the cost function, and the cost function of the whole CLPNet architecture is designed as a channel matrix output by the decoderAnd a real channel matrix H a The cost function is described as follows:
wherein C is the number of samples in the training set, I.I 2 Is the euclidean norm;f e ,f d ,θ e and theta d Represented are encoder, decoder, encoder parameters and decoder parameters, respectively.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111563073.8A CN114157331B (en) | 2021-12-20 | 2021-12-20 | Large-scale MIMO channel state information feedback method based on pseudo complex value input |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111563073.8A CN114157331B (en) | 2021-12-20 | 2021-12-20 | Large-scale MIMO channel state information feedback method based on pseudo complex value input |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114157331A CN114157331A (en) | 2022-03-08 |
CN114157331B true CN114157331B (en) | 2023-11-24 |
Family
ID=80451667
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111563073.8A Active CN114157331B (en) | 2021-12-20 | 2021-12-20 | Large-scale MIMO channel state information feedback method based on pseudo complex value input |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114157331B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114978264B (en) * | 2022-06-29 | 2023-07-25 | 内蒙古大学 | Mixed precoding method based on terahertz MIMO system |
CN116505989B (en) * | 2023-05-06 | 2024-02-06 | 四川农业大学 | CSI feedback method and system based on complex encoder hybrid neural network |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108390706A (en) * | 2018-01-30 | 2018-08-10 | 东南大学 | A kind of extensive mimo channel state information feedback method based on deep learning |
CN109672464A (en) * | 2018-12-13 | 2019-04-23 | 西安电子科技大学 | Extensive mimo channel state information feedback method based on FCFNN |
CN110311718A (en) * | 2019-07-05 | 2019-10-08 | 东南大学 | Quantization and inverse quantization method in a kind of extensive mimo channel status information feedback |
CN110350958A (en) * | 2019-06-13 | 2019-10-18 | 东南大学 | A kind of more multiplying power compressed feedback methods of CSI of extensive MIMO neural network based |
CN111555781A (en) * | 2020-04-27 | 2020-08-18 | 天津大学 | Large-scale MIMO channel state information compression and reconstruction method based on deep learning attention mechanism |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI687063B (en) * | 2019-01-04 | 2020-03-01 | 財團法人工業技術研究院 | A communication system and codec method based on deep learning and channel state information |
-
2021
- 2021-12-20 CN CN202111563073.8A patent/CN114157331B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108390706A (en) * | 2018-01-30 | 2018-08-10 | 东南大学 | A kind of extensive mimo channel state information feedback method based on deep learning |
CN109672464A (en) * | 2018-12-13 | 2019-04-23 | 西安电子科技大学 | Extensive mimo channel state information feedback method based on FCFNN |
CN110350958A (en) * | 2019-06-13 | 2019-10-18 | 东南大学 | A kind of more multiplying power compressed feedback methods of CSI of extensive MIMO neural network based |
CN110311718A (en) * | 2019-07-05 | 2019-10-08 | 东南大学 | Quantization and inverse quantization method in a kind of extensive mimo channel status information feedback |
CN111555781A (en) * | 2020-04-27 | 2020-08-18 | 天津大学 | Large-scale MIMO channel state information compression and reconstruction method based on deep learning attention mechanism |
Also Published As
Publication number | Publication date |
---|---|
CN114157331A (en) | 2022-03-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112737985B (en) | Large-scale MIMO channel joint estimation and feedback method based on deep learning | |
CN108390706B (en) | Large-scale MIMO channel state information feedback method based on deep learning | |
CN109672464B (en) | FCFNN-based large-scale MIMO channel state information feedback method | |
CN111464220B (en) | Channel state information reconstruction method based on deep learning | |
CN112910807B (en) | Intelligent super-surface channel estimation method and system based on space random sampling | |
CN114157331B (en) | Large-scale MIMO channel state information feedback method based on pseudo complex value input | |
CN109257309B (en) | High-performance large-scale MIMO downlink transmission channel estimation method | |
CN111181671B (en) | Deep learning-based downlink channel rapid reconstruction method | |
CN111555781B (en) | Large-scale MIMO channel state information compression and reconstruction method based on deep learning attention mechanism | |
Yin et al. | Deep CSI compression for massive MIMO: A self-information model-driven neural network | |
CN110289898A (en) | A kind of channel feedback method based on the perception of 1 bit compression in extensive mimo system | |
CN107276646A (en) | Extensive MIMO OFDM up-link iteration detection methods | |
CN114338301B (en) | Channel estimation method of RIS auxiliary millimeter wave system based on compressed sensing | |
Guo et al. | Deep learning for joint channel estimation and feedback in massive MIMO systems | |
CN114884549A (en) | Large-scale MIMO channel state information feedback method based on deep learning | |
CN115865145A (en) | Large-scale MIMO channel state information feedback method based on Transformer | |
CN115208442A (en) | Two-stage beam training method of ultra-large scale array communication system | |
CN113556158B (en) | Large-scale MIMO intelligent CSI feedback method for Internet of vehicles | |
CN114884775A (en) | Deep learning-based large-scale MIMO system channel estimation method | |
CN115021787A (en) | Channel state information feedback method based on complex convolutional neural network | |
CN106130938B (en) | Multi-user joint channel estimation method for TDD large-scale MIMO system | |
CN114567359A (en) | CSI feedback method based on multi-resolution fusion convolution feedback network in large-scale MIMO system | |
CN112769462B (en) | Millimeter wave MIMO broadband channel estimation method based on joint parameter learning | |
US20240089152A1 (en) | Angular domain channel estimation method based on matrix reconstruction for symmetrical nonuniform array | |
CN116505989B (en) | CSI feedback method and system based on complex encoder hybrid neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |