CN115085782B - Intelligent reflecting surface joint feedback and mixed precoding method based on deep learning - Google Patents
Intelligent reflecting surface joint feedback and mixed precoding method based on deep learning Download PDFInfo
- Publication number
- CN115085782B CN115085782B CN202210681193.6A CN202210681193A CN115085782B CN 115085782 B CN115085782 B CN 115085782B CN 202210681193 A CN202210681193 A CN 202210681193A CN 115085782 B CN115085782 B CN 115085782B
- Authority
- CN
- China
- Prior art keywords
- precoding
- matrix
- reflecting surface
- precoding matrix
- analog
- 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
Images
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
- H04B7/0456—Selection of precoding matrices or codebooks, e.g. using matrices antenna weighting
-
- 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
- H04B7/06—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
- H04B7/0621—Feedback content
- H04B7/0626—Channel coefficients, e.g. channel state information [CSI]
-
- 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)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses an intelligent reflecting surface joint feedback and mixed precoding method based on deep learning. In the online prediction stage, a user side inputs a real-time channel into a channel compression network and feeds back compression information to a base station side, the base station side inputs feedback information into a mixed precoding network to obtain an analog precoding matrix and a digital precoding matrix, then digital precoding is carried out on a sending signal according to the digital precoding matrix, the digital precoding signal is transmitted to an intelligent reflecting surface through a feed source, and the intelligent reflecting surface carries out phase shift processing on the digital precoding signal according to the analog precoding matrix and sends the digital precoding signal to the user side. The invention utilizes the deep learning technology to realize the channel feedback of multiple users and the hybrid precoding based on the intelligent reflecting surface, thereby reducing the power consumption of the system.
Description
Technical Field
The invention relates to the technical field of wireless communication, in particular to an intelligent reflecting surface joint feedback and mixed precoding method based on deep learning.
Background
With the rapid development of wireless technology, traffic data and traffic have assumed a surge situation. Conventional MIMO systems typically employ all-digital precoding, however, as the number of antennas increases, the number of RF chains in the all-digital system increases substantially, which not only increases the overall power consumption of the system, but also increases the complexity and cost of hardware. The current hybrid precoding architecture achieves performance similar to that of an all-digital architecture by using fewer RF chains, and can achieve balance between spectrum efficiency and system power consumption. However, in higher frequency bands, such as Terahertz (THz) scenarios, the number of antennas may further increase, and the number of phase shifters implementing analog precoding in the hybrid precoding architecture may become large, which greatly increases the complexity and power consumption of the system as a whole.
Recently, in exploring the next generation (6G) communication technology, smart reflective surfaces (Intelligent Reflecting Surface, IRS) have been receiving attention from the academia and industry for their advantages of low cost, low power consumption, ease of deployment, etc. The intelligent reflecting surface is a plane formed by a large number of low-cost passive reflecting elements, each element can independently change the amplitude or phase of an incident signal so as to enable the incident signal to be sent in one or more expected directions, and the function of analog precoding can be realized under lower power consumption. However, the complexity of co-optimization of the intelligent reflective surface with the base station side becomes very high due to the large number of passive reflective elements.
In the frequency division duplex mode, after a user acquires downlink channel state information (Channel State Information, CSI), the downlink CSI needs to be fed back to the base station through a feedback link, and the base station performs precoding design according to the fed back CSI to improve overall performance of the system. In THz scenario, the number of antennas is very large, the channel dimension increases, and feedback overhead becomes huge, which greatly increases the burden of the system, and affects the performance of the hybrid precoding of the base station.
In order to realize multi-user channel information feedback and intelligent reflection surface-based hybrid precoding at a base station end, reduce feedback overhead and reduce power consumption and computational complexity of hybrid precoding, the invention provides an intelligent reflection surface joint feedback and hybrid precoding method based on deep learning.
Disclosure of Invention
The invention aims to provide an intelligent reflecting surface joint feedback and mixed precoding method based on deep learning so as to solve the problems in the background technology.
In order to achieve the above purpose, the present invention provides the following technical solutions: the intelligent reflecting surface joint feedback and mixed precoding method based on deep learning is realized by an off-line training stage and an on-line prediction stage; the off-line training phase is characterized by comprising the following steps:
s11: the user terminal receives the channel state information, and calculates a digital precoding matrix of the base station and an analog precoding matrix of the intelligent reflecting surface according to the channel;
s12: constructing a combined feedback and mixed precoding deep learning model of the intelligent reflecting surface, preparing a digital precoding matrix of a channel and a base station and an analog precoding matrix of the intelligent reflecting surface into a data set, and training the combined feedback and mixed precoding deep learning model of the intelligent reflecting surface;
s13: splitting a trained combined feedback and mixed precoding deep learning model of an intelligent reflecting surface into a mixed precoding network and K channel compression networks, wherein the mixed precoding network is 4-10 layers of the combined feedback and mixed precoding deep learning model of the intelligent reflecting surface, the kth channel compression network is the kth network in the K networks of 1-3 layers in the combined feedback and mixed precoding deep learning model of the intelligent reflecting surface, k=1, K is the number of users, the K channel compression networks and weights thereof are deployed in sequence on the K user terminals, and the mixed precoding network and weights thereof are deployed on a base station terminal;
the online prediction stage comprises the following steps:
s21: k user terminals acquire real-time channel state information, input real-time channels into a channel compression network to acquire compressed channel state information, and feed back the compressed channel state information to a base station terminal;
s22: the base station end inputs feedback information of K users into a mixed precoding network, obtains a digital precoding matrix and an analog precoding matrix according to the output of the mixed precoding network, and transmits the analog precoding matrix to the intelligent reflecting surface through a feed source;
s23: the base station end performs digital precoding on the transmission signal according to the digital precoding matrix to obtain a digital precoding signal;
s24: and transmitting the digital pre-coded signals to the sub-surface of the intelligent reflecting surface through the feed source, performing phase shift processing on the digital pre-coded signals by the passive reflecting elements on the sub-surface according to the analog pre-coding matrix, and transmitting the phase shifted signals to the user side.
Preferably, in the step S11, the step of calculating the digital precoding matrix of the base station and the analog precoding matrix of the intelligent reflection surface according to the channel is as follows:
t1, initializing probability vector
Where i=1,.. ite ，N ite For the number of iterations->
Vector element p in (b) n Represents the probability of +1 selected from { +1, -1}, 1 N The vector is an all-1 vector with dimension of N multiplied by 1, and N is the number of passive elements of the intelligent reflecting surface;
t2, using probability vectors
Generating C analog precoding vectors of dimension nx1 with set { +1, -1}>
Will be
Analog precoding matrix as intelligent reflector>
Diagonal element of>
Other elements of 0 according to
Calculating C digital precoding matrix +.>
Wherein->
ρ is the transmit power, +.>
Is the channel between the intelligent reflecting surface and the K users. />
A link matrix from the radio frequency to the intelligent reflecting surface;
t3 according to
Calculating spectral efficiency R for the ith iteration (i) Wherein
For the signal-to-interference-and-noise ratio of the kth user, wherein +.>
Digital precoding matrix->
Is the kth column vector, h k For intelligent reflection surface to user k channel, sigma 2 Is the noise power; sequencing the C calculated frequency spectrum efficiencies from high to low, and selecting the first N elite Spectral efficiency, N elite The analog precoding vectors corresponding to the frequency spectrum efficiency are combined into N elite Matrix of XN->
T4, calculation
N-th column vector->
Number S of +1 in n Where n=1,..n, N, will +.>
Probability vector +.1 as the i+1 th iteration>
The value of the nth element;
t5, will
Repeating the steps T2-T5 as probability vector of the (i+1) th iteration, and repeating the iteration N ite Secondary times;
t6, N ite In +1 iteration, use the N ite Generated probability vectors for multiple iterations
Repeating step T2 according to ∈ ->
G calculating spectral efficiency, wherein->
And->
Respectively the N th ite A digital precoding matrix and an analog precoding generated by +1 iterations; sequencing the C calculated spectral efficiencies from high to low, selecting the highest spectral efficiency, converting the corresponding analog precoding vector into an analog precoding matrix, and taking the analog precoding matrix as an optimal analog precoding matrix F a The corresponding digital precoding matrix is used as the optimal digital precoding matrix F b 。
Preferably, in the step S12, the joint feedback and hybrid precoding deep learning model of the intelligent reflection surface is composed of K identical channel compression networks and hybrid precoding networks, where K is the number of users; the kth channel compression network comprises three fully connected layers, k=1, K, the first and second layers having dimensions 1024 and 512 respectively, the activation function is Relu, the dropoff layer is arranged after the activation function, the third layer dimension is Q, wherein Q is the dimension of channel compression, and the activation function and the dropoff layer are not arranged; the first layer in the mixed precoding network is a full-connection layer, the dimension is 2048, the second layer is a Reshape layer, vectors are reconstructed into a three-dimensional matrix of 32 x 2, the third layer and the fourth layer are convolution layers, the number of convolution kernels is 16, the size of the convolution kernels is 2 x 2, the convolution layers are subjected to batch normalization, an activation function is Relu, the fifth layer and the sixth layer are full-connection layers, the dimension is 1024, the activation function is Relu, the seventh layer is the full-connection layer, the dimension is T, T=N+2MK, N is the number of passive reflection elements of the intelligent reflection surface, and M is the number of radio frequency links.
Preferably, in the step S12, in the data set of the joint feedback and hybrid precoding deep learning model of the smart reflector, the input of the kth channel compression network is the channel h from the smart reflector to the user k k The output is
Where k=1,.. a ,F b Respectively analog and digital pre-coding matrixes vec T () Representing vectorizing and transpose the matrix, diag () representing taking the diagonal elements of the matrix,/->
The real part and the imaginary part of the complex number are respectively represented, and the angle represents the transformation of the complex number into an angle.
Preferably, in the step S12, when training the deep learning model of the joint feedback and hybrid precoding, the loss function is set as the mean square error
Wherein the method comprises the steps of
As a tag of the data set,
for outputting data, where F a ,F b Analog and digital precoding matrices of the tag, respectively,>
analog and digital precoding matrices, vec, respectively, of the estimates of the model outputs during training T () Representing vectorizing and transpose the matrix, diag () representing taking the diagonal elements of the matrix,/->
The real part and the imaginary part of the complex number are respectively represented, the angle represents that the complex number is converted into an angle, the learning rate is 0.0003, the attenuation factor is 0.95, and the epoch is set to 400.
Preferably, in the step S22, the output of the network is
Wherein->
Analog and digital pre-coding matrixes, vec respectively output by on-line prediction stage models T () Represents vectorizing and transpose the matrix, diag () represents taking the diagonal elements of the matrix,
respectively representing the real part and the imaginary part of a complex number, and transforming the complex number into an angle and +.>
Is converted into complex form, which is multiplied by N x N identity matrix to obtain analog precoding matrix +.>
Will be
Transforming into matrix form of MxK, combining real part and imaginary part to obtain digital precoding matrix +.>
Wherein N is the number of passive reflecting elements of the intelligent reflecting surface, M is the number of radio frequency links, and K is the number of users.
Preferably, in the step S24, the digital pre-encoded signal at the mth radio frequency is transmitted to the mth sub-surface via the feeder line, and the q-th reflective element on the mth sub-surface is used for performing the analog pre-encoding according to the analog pre-encoding F a The mth diagonal matrix of (a)
Is +.q diagonal element->
To set a phase shift, where m=1..m, q=1..q, M is the number of radio frequency links and the number of sub-surfaces,/-for each sub-surface>
For the number of passive reflective elements on the sub-surface, N is the total number of passive reflective elements on the intelligent reflective surface,
is an analog precoding matrix of the intelligent reflecting surface,
diag { } represents transforming a vector into a diagonal matrix, ++>
θ q,m A phase shift of the q-th passive reflective element for the m-th subsurface; the signals transmitted to the intelligent reflecting surface are subjected to phase shift processing through the passive reflecting element, and the signals subjected to phase shift processing are transmitted to K users.
Compared with the prior art, the invention has the beneficial effects that:
(1) The invention utilizes the intelligent reflecting surface to realize the digital precoding of the base station and the analog precoding of the intelligent reflecting surface, thereby greatly reducing the complexity and the power consumption of the system while ensuring the overall performance of the system.
(2) The invention utilizes the deep learning technology to carry out the mixed pre-coding design according to the channel state information fed back by multiple users, and compared with the traditional algorithm, the invention greatly reduces the calculation complexity under the condition of close performance.
(3) The invention compresses the channel state information estimated by multiple users and feeds back the channel state information to the base station by utilizing the deep learning technology, and compared with the method for directly feeding back the channel state information with high dimension, the feedback cost is greatly reduced.
(4) The invention saves a great amount of calculation expenditure through offline training and online prediction, is suitable for various indoor and outdoor scenes, and has good robustness.
Drawings
FIG. 1 is a schematic diagram of a hybrid precoding system based on intelligent reflecting surfaces in accordance with the present invention;
FIG. 2 is a schematic diagram of a deep learning model of joint feedback and hybrid precoding of an intelligent reflector of the present invention;
FIG. 3 is a block diagram of a channel compression neural network according to the present invention;
fig. 4 is a block diagram of a hybrid precoding neural network according to the present invention;
fig. 5 is a flowchart of the intelligent reflecting surface joint feedback and hybrid precoding method based on deep learning.
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.
As shown in fig. 1, in the millimeter wave massive MIMO system, a base station end has M radio frequency links, K data streams, a user is a single antenna, the number of the user is K, the intelligent reflection surface is composed of N passive reflection elements and is divided into M sub-surfaces, each sub-surface contains q=n/M elements, and considering the situation that the number of users and the number of radio frequency links are equal, that is, k=m. Radio frequency link and subThe surfaces are connected by a feed source, each radio frequency link is connected with only one corresponding sub-surface, and signals are sent
Through digital precoding matrix F b Processing the transmission to a radio frequency link, the radio frequency link transmitting the transmission signal via a feeder line to each passive reflection element on the corresponding sub-surface, the passive reflection elements on the sub-surface being in accordance with an analog precoding matrix F a And adjusting the signal phase and then transmitting the signal phase to K user terminals. The received signals y of the K users are:
y＝HF a GF b s+n
wherein the method comprises the steps of
Digital precoding matrix for base station, +.>
Is the channel between the intelligent reflecting surface and the K users. />
For a link matrix of radio frequency to intelligent reflecting surface 1 Q Is a full 1 vector of Q x 1,
an analog precoding matrix for the intelligent reflecting surface, < >>
Phase shift of the q-th passive reflecting element for the m-th sub-surface,/th sub-surface>
H channel modeling Using Saleh-Valenzuela model, +.>
L k,m For the number of multipaths between the kth user and the mth subsurface,/for the number of multipaths between the kth user and the mth subsurface>
For complex gain of the first path, +.>
Is a direction vector, wherein->
θ is azimuth and pitch, respectively. In this system model, spectral efficiency can be expressed as:
wherein SINR k The signal-to-interference-and-noise ratio for the kth user is expressed as:
wherein f k Digital precoding matrix F b Is the k-th column vector, sigma 2 Is the noise power.
In each coherence time of the channel, the kth user side estimates the downlink channel h from the intelligent reflecting surface to the user k k Where k=1,..k, h will be k Channel compressor for inputting user k
To divide channel h k Compression into a codeword s of low dimension k I.e.
The K users will compress codeword s= [ s ] 1 ,s 2 ,...,s k ]To the base station end, the base station end willCodeword s input hybrid precoder
Obtaining a digital precoding matrix F of the base station b And an analog precoding matrix F of the intelligent reflecting surface a I.e.
The base station end performs the precoding according to the digital precoding matrix F b Carrying out digital precoding on the transmitted signals to obtain digital precoding signals, transmitting the digital precoding signals to the sub-surfaces of the intelligent reflecting surface through the feed source, and enabling the passive reflecting elements on the sub-surfaces to be in accordance with the analog precoding matrix F a And performing phase shift processing on the digital pre-coded signal, and transmitting the phase-shifted signal to a user side.
Based on the above system, the method for intelligent reflecting surface joint feedback and hybrid precoding based on deep learning provided by the invention has the specific steps shown in fig. 5, and the steps are described in detail below.
The method is realized by an offline training stage and an online prediction stage; the offline training phase comprises the following steps:
s11: the user terminal receives the channel state information, and calculates a digital precoding matrix of the base station and an analog precoding matrix of the intelligent reflecting surface according to the channel;
s12: constructing a combined feedback and mixed precoding deep learning model of the intelligent reflecting surface, preparing a digital precoding matrix of a channel and a base station and an analog precoding matrix of the intelligent reflecting surface into a data set, and training the combined feedback and mixed precoding deep learning model of the intelligent reflecting surface;
s13: splitting a trained combined feedback and mixed precoding deep learning model of an intelligent reflecting surface into a mixed precoding network and K channel compression networks, wherein the mixed precoding network is 4-10 layers of the combined feedback and mixed precoding deep learning model of the intelligent reflecting surface, the kth channel compression network is the kth network in the K networks of 1-3 layers in the combined feedback and mixed precoding deep learning model of the intelligent reflecting surface, k=1, K is the number of users, the K channel compression networks and weights thereof are deployed in sequence on the K user terminals, and the mixed precoding network and weights thereof are deployed on a base station terminal;
the online prediction stage comprises the following steps:
s21: k user terminals acquire real-time channel state information, input real-time channels into a channel compression network to acquire compressed channel state information, and feed back the compressed channel state information to a base station terminal;
s22: the base station end inputs feedback information of K users into a mixed precoding network, obtains a digital precoding matrix and an analog precoding matrix according to the output of the mixed precoding network, and transmits the analog precoding matrix to the intelligent reflecting surface through a feed source;
s23: the base station end performs digital precoding on the transmission signal according to the digital precoding matrix to obtain a digital precoding signal;
s24: and transmitting the digital pre-coded signals to the sub-surface of the intelligent reflecting surface through the feed source, performing phase shift processing on the digital pre-coded signals by the passive reflecting elements on the sub-surface according to the analog pre-coding matrix, and transmitting the phase shifted signals to the user side.
The steps of calculating a digital precoding matrix of the base station and an analog precoding matrix of the intelligent reflecting surface according to the channel are as follows:
t1, initializing probability vector
Where i=1,.. ite ，N ite For the number of iterations->
Vector element p in (b) n Represents the probability of +1 selected from { +1, -1}, 1 N The vector is an all-1 vector with dimension of N multiplied by 1, and N is the number of passive elements of the intelligent reflecting surface;
t2, using probability vectors
Generating C analog precoding vectors of dimension nx1 with set { +1, -1}>
Will be
Analog precoding matrix as intelligent reflector>
Diagonal element of>
The other elements of (2) are 0 according to +.>
Calculating C digital precoding matrix +.>
Wherein->
ρ is the transmit power, +.>
Is the channel between the intelligent reflecting surface and the K users. />
A link matrix from the radio frequency to the intelligent reflecting surface;
t3 according to
Calculating spectral efficiency R for the ith iteration (i) Wherein
For the signal-to-interference-and-noise ratio of the kth user, wherein +.>
Digital precoding matrix->
Is the kth column vector, h k For intelligent reflection surface to user k channel, sigma 2 Is the noise power; sequencing the C calculated frequency spectrum efficiencies from high to low, and selecting the first N elite Spectral efficiency, N elite The analog precoding vectors corresponding to the frequency spectrum efficiency are combined into N elite Matrix of XN->
T4, calculation
N-th column vector->
Number S of +1 in n Where n=1,..n, N, will +.>
Probability vector +.1 as the i+1 th iteration>
The value of the nth element;
t5, will
Repeating the steps T2-T5 as probability vector of the (i+1) th iteration, and repeating the iteration N ite Secondary times;
t6, N ite In +1 iteration, use the N ite Generated probability vectors for multiple iterations
Repeating step T2 according to ∈ ->
H,/>
G calculating spectral efficiency, wherein->
And->
Respectively the N th ite A digital precoding matrix and an analog precoding generated by +1 iterations; sequencing the C calculated spectral efficiencies from high to low, selecting the highest spectral efficiency, converting the corresponding analog precoding vector into an analog precoding matrix, and taking the analog precoding matrix as an optimal analog precoding matrix F a The corresponding digital precoding matrix is used as the optimal digital precoding matrix F b 。
The structure of the joint feedback and hybrid pre-coding deep learning model of the intelligent reflecting surface:
as shown in fig. 2, the joint feedback and hybrid precoding deep learning model of the intelligent reflection surface is composed of K identical channel compression networks and hybrid precoding networks, where K is the number of users; as shown in fig. 3, the kth channel compression network includes three full connection layers, k=1,..k, the dimensions of the first layer and the second layer are 1024 and 512, respectively, the activation function is Relu, the activation function is dropout layer, the third layer dimension is Q, where Q is the dimension of channel compression, and there is no activation function and dropout layer; as shown in fig. 4, in the hybrid precoding network, the first layer is a fully connected layer, the dimension is 2048, the second layer is a Reshape layer, the vector is reconstructed into a three-dimensional matrix of 32×32×2, the third layer and the fourth layer are convolution layers, the number of convolution kernels is 16, the convolution kernels are 2×2, the convolution layers are subjected to batch normalization, the activation function is Relu, the fifth layer and the sixth layer are fully connected layers, the dimension is 1024, the activation function is Relu, the seventh layer is a fully connected layer, the dimension is T, wherein t=n+2mk, N is the number of passive reflection elements of the intelligent reflection surface, and M is the number of radio frequency links.
Data set preparation of intelligent reflecting surface combined feedback and mixed pre-coding deep learning model:
in the data set of the joint feedback and hybrid precoding deep learning model of the intelligent reflection plane, the input of the kth channel compression network is the channel h from the intelligent reflection plane to the user k k The output is
Where k=1,.. a ,F b Respectively analog and digital pre-coding matrixes vec T () Representing vectorizing and transpose the matrix, diag () representing taking the diagonal elements of the matrix,/->
The real part and the imaginary part of the complex number are respectively represented, and the angle represents the transformation of the complex number into an angle.
Training of a joint feedback and hybrid pre-coded deep learning model of intelligent reflection surfaces:
when training the combined feedback and mixed pre-coding deep learning model of the intelligent reflecting surface, the loss function is set as the mean square error
Wherein->
Tag for data set->
For outputting data, where F a ,F b Analog and digital precoding matrices of the tag, respectively,>
analog and digital precoding matrices, vec, respectively, of the estimates of the model outputs during training T () Representing vectorizing and transpose the matrix, diag () representing taking the diagonal elements of the matrix,/->
The real part and the imaginary part of the complex number are respectively represented, the angle represents that the complex number is converted into an angle, the learning rate is 0.0003, the attenuation factor is 0.95, and the epoch is set to 400.
The process of obtaining the analog and digital precoding matrix by calculation in the online prediction stage:
the output of the network is
Wherein->
Analog and digital pre-coding matrixes, vec respectively output by on-line prediction stage models T () Representing vectorizing and transpose the matrix, diag () representing taking the diagonal elements of the matrix,/->
Respectively representing the real part and the imaginary part of a complex number, and transforming the complex number into an angle and +.>
Is converted into complex form, which is multiplied by N x N identity matrix to obtain analog precoding matrix +.>
Will->
Transforming into matrix form of MxK, combining real part and imaginary part to obtain digital precoding matrix +.>
Wherein N is the number of passive reflecting elements of the intelligent reflecting surface, M is the number of radio frequency links, and K is the number of users.
The intelligent reflection carries out the process of analog precoding on the signals:
the digital pre-coded signal on the mth radio frequency is transmitted to the mth subsurface through a feeder line, and the mth inverse on the mth subsurfaceThe radiating element being according to analogue precoding F a The mth diagonal matrix of (a)
Is +.q diagonal element->
To set a phase shift, where m=1..m, q=1..q, M is the number of radio frequency links and the number of sub-surfaces,/-for each sub-surface>
N is the total passive reflection element number on the intelligent reflection surface, which is the passive reflection element number on the sub-surface, +.>
An analog precoding matrix for the intelligent reflecting surface, < >>
diag { } represents transforming a vector into a diagonal matrix, ++>
θ q,m A phase shift of the q-th passive reflective element for the m-th subsurface; the signals transmitted to the intelligent reflecting surface are subjected to phase shift processing through the passive reflecting element, and the signals subjected to phase shift processing are transmitted to K users.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Claims (6)
1. The intelligent reflecting surface joint feedback and mixed precoding method based on deep learning is realized by an off-line training stage and an on-line prediction stage; the off-line training phase is characterized by comprising the following steps:
s11: the user terminal receives the channel state information, and calculates a digital precoding matrix of the base station and an analog precoding matrix of the intelligent reflecting surface according to the channel;
s12: constructing a combined feedback and mixed precoding deep learning model of the intelligent reflecting surface, preparing a digital precoding matrix of a channel and a base station and an analog precoding matrix of the intelligent reflecting surface into a data set, and training the combined feedback and mixed precoding deep learning model of the intelligent reflecting surface;
in the step S12, the combined feedback and hybrid precoding deep learning model of the intelligent reflecting surface is composed of K identical channel compression networks and hybrid precoding networks, where K is the number of users; the kth channel compression network comprises three fully connected layers, k=1, K, the first and second layers having dimensions 1024 and 512 respectively, the activation function is Relu, the dropoff layer is arranged after the activation function, the third layer dimension is Q, wherein Q is the dimension of channel compression, and the activation function and the dropoff layer are not arranged; the first layer in the mixed precoding network is a full-connection layer, the dimension is 2048, the second layer is a Reshape layer, vectors are reconstructed into a three-dimensional matrix of 32 x 2, the third layer and the fourth layer are convolution layers, the number of convolution kernels is 16, the size of the convolution kernels is 2 x 2, the convolution layers are subjected to batch normalization, an activation function is Relu, the fifth layer and the sixth layer are full-connection layers, the dimension is 1024, the activation function is Relu, the seventh layer is a full-connection layer, the dimension is T, T=N+2MK, N is the number of passive reflection elements of the intelligent reflection surface, and M is the number of radio frequency links;
s13: splitting a trained combined feedback and mixed precoding deep learning model of an intelligent reflecting surface into a mixed precoding network and K channel compression networks, wherein the mixed precoding network is 4-10 layers of the combined feedback and mixed precoding deep learning model of the intelligent reflecting surface, the kth channel compression network is the kth network in the K networks of 1-3 layers in the combined feedback and mixed precoding deep learning model of the intelligent reflecting surface, k=1, K is the number of users, the K channel compression networks and weights thereof are deployed in sequence on the K user terminals, and the mixed precoding network and weights thereof are deployed on a base station terminal;
the online prediction stage comprises the following steps:
s21: k user terminals acquire real-time channel state information, input real-time channels into a channel compression network to acquire compressed channel state information, and feed back the compressed channel state information to a base station terminal;
s22: the base station end inputs feedback information of K users into a mixed precoding network, obtains a digital precoding matrix and an analog precoding matrix according to the output of the mixed precoding network, and transmits the analog precoding matrix to the intelligent reflecting surface through a feed source;
s23: the base station end performs digital precoding on the transmission signal according to the digital precoding matrix to obtain a digital precoding signal;
s24: and transmitting the digital pre-coded signals to the sub-surface of the intelligent reflecting surface through the feed source, performing phase shift processing on the digital pre-coded signals by the passive reflecting elements on the sub-surface according to the analog pre-coding matrix, and transmitting the phase shifted signals to the user side.
2. The method for joint feedback and hybrid precoding of intelligent reflecting surface based on deep learning according to claim 1, wherein in step S11, the step of calculating the digital precoding matrix of the base station and the analog precoding matrix of the intelligent reflecting surface according to the channel is as follows:
t1, initializing probability vector
Where i=1,.. ite ，N ite For the number of iterations->
Vector element p in (b) n Represents the probability of +1 selected from { +1, -1}, 1 N The vector is an all-1 vector with dimension of N multiplied by 1, and N is the number of passive elements of the intelligent reflecting surface;
t2, using probability vectors
Generating C analog precoding vectors of dimension nx1 with set { +1, -1}>
Will->
Analog precoding matrix as intelligent reflector>
Diagonal element of>
Is 0, C digital precoding matrix +.>
Wherein->
ρ is the transmit power, +.>
For the channel between the smart reflector and K users, < >>
A link matrix from the radio frequency to the intelligent reflecting surface;
t3 according to
Calculating spectral efficiency R for the ith iteration (i) Wherein
For the signal-to-interference-and-noise ratio of the kth user, wherein +.>
Digital precoding matrix->
Is the kth column vector, h k For intelligent reflection surface to user k channel, sigma 2 Is the noise power; sequencing the C calculated frequency spectrum efficiencies from high to low, and selecting the first N elite Spectral efficiency, N elite The analog precoding vectors corresponding to the frequency spectrum efficiency are combined into N elite Matrix of XN->
T4, calculation
N-th column vector->
Number S of +1 in n Where n=1,..n, N, will +.>
Probability vector +.1 as the i+1 th iteration>
The value of the nth element;
t5, will
Repeating the steps T2-T5 as probability vector of the (i+1) th iteration, and repeating the iteration N ite Secondary times;
t6, N ite In +1 iteration, use the N ite Generated probability vectors for multiple iterations
Repeating step T2 according to
Calculating spectral efficiency, wherein->
And->
Respectively the N th ite A digital precoding matrix and an analog precoding generated by +1 iterations; sequencing the C calculated spectral efficiencies from high to low, selecting the highest spectral efficiency, converting the corresponding analog precoding vector into an analog precoding matrix, and taking the analog precoding matrix as an optimal analog precoding matrix F a The corresponding digital precoding matrix is used as the optimal digital precoding matrix F b 。
3. The method of deep learning based joint feedback and hybrid precoding of intelligent reflecting surface according to claim 1, wherein in the step S12, in the data set of the deep learning model of joint feedback and hybrid precoding of intelligent reflecting surface, the input of the kth channel compression network is the channel h from the intelligent reflecting surface to the user k k The output is
Where k=1,.. a ,F b Respectively an analog precoding matrix and a digital precoding matrix, vec T () Represents vectorizing and transpose the matrix, diag () represents taking the diagonal elements of the matrix,
the real part and the imaginary part of the complex number are respectively represented, and the angle represents the transformation of the complex number into an angle.
4. The method for joint feedback and hybrid precoding of intelligent reflecting surface based on deep learning according to claim 1, wherein in the step S12, when training the deep learning model of joint feedback and hybrid precoding, the loss function is set as the mean square error
Tag for data set->
For outputting data, where F a ,F b Analog precoding matrix, digital precoding matrix, respectively, of the tag +.>
Analog precoding matrix, digital precoding matrix, vec, respectively, of the estimates of the model output during training T () Representing vectorizing and transpose the matrix, diag () representing taking the diagonal elements of the matrix,/->
The real part and the imaginary part of the complex number are respectively represented, and the angle represents the transformation of the complex number into an angle.
5. The intelligent reflection surface joint feedback and mixed pre-coding method based on deep learning as claimed in claim 1, wherein in the step S22, the output of the network is
Wherein the method comprises the steps of
Analog precoding matrix and digital precoding matrix which are respectively output by the online prediction stage model, and vec T () Representing vectorizing and transpose the matrix, diag () representing taking the diagonal elements of the matrix,/->
Respectively representing the real part and the imaginary part of a complex number, and transforming the complex number into an angle and +.>
Is converted into complex form, which is multiplied by N x N identity matrix to obtain analog precoding matrix +.>
Will->
Transforming into matrix form of MxK, combining real part and imaginary part to obtain digital precoding matrix +.>
Wherein N is the number of passive reflecting elements of the intelligent reflecting surface, M is the number of radio frequency links, and K is the number of users.
6. The intelligent reflecting surface joint feedback and mixed pre-coding method based on deep learning as claimed in claim 1, wherein in the step S24, the digital pre-coded signal on the mth radio frequency is transmitted to the mth sub-surface through the feeder line, and the q-th reflecting element on the mth sub-surface is used for performing the analog pre-coding according to the analog pre-coding F a The mth diagonal matrix of (a)
Is +.q diagonal element->
To set a phase shift, where m=1,..m, q=1,., Q, M is the number of radio frequency links and the number of sub-surfaces,
for the number of passive reflective elements on the sub-surface, N is the total number of passive reflective elements on the intelligent reflective surface,
is an analog precoding matrix of the intelligent reflecting surface,
diag { } represents transforming vectors into pairsCorner matrix, < >>
θ q,m A phase shift of the q-th passive reflective element for the m-th subsurface; the signals transmitted to the intelligent reflecting surface are subjected to phase shift processing through the passive reflecting element, and the signals subjected to phase shift processing are transmitted to K users.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210681193.6A CN115085782B (en) | 2022-06-15 | 2022-06-15 | Intelligent reflecting surface joint feedback and mixed precoding method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210681193.6A CN115085782B (en) | 2022-06-15 | 2022-06-15 | Intelligent reflecting surface joint feedback and mixed precoding method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115085782A CN115085782A (en) | 2022-09-20 |
| CN115085782B true CN115085782B (en) | 2023-06-16 |
Family
ID=83254199
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210681193.6A Active CN115085782B (en) | 2022-06-15 | 2022-06-15 | Intelligent reflecting surface joint feedback and mixed precoding method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115085782B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113162666A (en) * | 2021-04-19 | 2021-07-23 | 北京科技大学 | Intelligent steel-oriented large-scale MIMO hybrid precoding method and device |
| CN113595606A (en) * | 2021-07-29 | 2021-11-02 | 东南大学 | Base station precoding and intelligent reflection surface phase shift joint optimization method based on deep reinforcement learning |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106031068B (en) * | 2014-04-02 | 2019-05-28 | 华为技术有限公司 | A kind of communication means and device based on beam forming |
| CN107733485B (en) * | 2016-08-12 | 2021-09-03 | 中兴通讯股份有限公司 | Feedback method and device of channel state information |
| CN106803958B (en) * | 2017-01-12 | 2019-12-27 | 同济大学 | Digital-analog hybrid video transmission method based on superposition modulation coding |
| CN110557177A (en) * | 2019-09-05 | 2019-12-10 | 重庆邮电大学 | DenseNet-based hybrid precoding method in millimeter wave large-scale MIMO system |
| CN113691288B (en) * | 2021-08-23 | 2022-04-19 | 北京理工大学 | Joint pilot frequency, feedback and multi-user hybrid coding method based on deep learning |
| CN113839697B (en) * | 2021-09-23 | 2022-06-28 | 南通大学 | Joint feedback and hybrid precoding design method based on deep learning |
-
2022
- 2022-06-15 CN CN202210681193.6A patent/CN115085782B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113162666A (en) * | 2021-04-19 | 2021-07-23 | 北京科技大学 | Intelligent steel-oriented large-scale MIMO hybrid precoding method and device |
| CN113595606A (en) * | 2021-07-29 | 2021-11-02 | 东南大学 | Base station precoding and intelligent reflection surface phase shift joint optimization method based on deep reinforcement learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115085782A (en) | 2022-09-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111294095B (en) | IRS (inter-range instrumentation Standard) assisted large-scale MIMO (multiple input multiple output) wireless transmission method based on statistical CSI (channel State information) | |
| CN112235026B (en) | Mixed beam design method of MIMO-OFDMA terahertz communication system | |
| CN113839697B (en) | Joint feedback and hybrid precoding design method based on deep learning | |
| CN110557177A (en) | DenseNet-based hybrid precoding method in millimeter wave large-scale MIMO system | |
| CN108736943B (en) | Hybrid precoding method suitable for large-scale MIMO system | |
| CN111835406B (en) | Robust precoding method suitable for energy efficiency and spectral efficiency balance of multi-beam satellite communication | |
| CN111147113B (en) | Multi-beam satellite communication robust precoding method for energy efficiency guarantee | |
| CN110943768B (en) | Mixed precoding codebook joint design method of millimeter wave large-scale MIMO system | |
| CN110365388B (en) | Low-complexity millimeter wave multicast beam forming method | |
| CN107809275B (en) | Finite feedback hybrid precoding method based on millimeter wave MIMO system | |
| CN106571858B (en) | Hybrid beam forming transmission system | |
| CN113556164B (en) | Energy efficiency priority-based beam forming optimization method in IRS-assisted SWIPT system | |
| CN110011712B (en) | Millimeter wave large-scale multi-input multi-output-oriented hybrid precoding method | |
| CN112260737A (en) | Multi-beam satellite communication robust precoding method with total energy efficiency and minimum energy efficiency balanced | |
| CN114567399B (en) | Multi-antenna multipath channel state information modeling and feedback method of air base station | |
| CN111726151A (en) | Resource allocation method and device based on wireless energy-carrying communication | |
| CN114465643B (en) | Mixed precoding method of millimeter wave large-scale MIMO antenna system based on gradient descent method | |
| CN108123741B (en) | Overlapping sub-array (OSA) based beamforming method and apparatus | |
| CN109067446B (en) | Mixed precoding method for multi-antenna multi-user large-scale antenna | |
| CN114844541B (en) | Hybrid precoding and feedback method based on deep learning | |
| CN113572503A (en) | Low-complexity improved mixed beam forming method based on GP | |
| CN113852402A (en) | IRS (intelligent communications system) -assisted NOMA-MIMO (non-multiple input multiple output) high-capacity access method | |
| CN105915272A (en) | Iterative beam forming method based on compressed sensing | |
| CN117527020A (en) | Combined active and passive beamforming optimization in intelligent reflector-assisted wireless energy-carrying communication system | |
| CN112398513A (en) | Beam forming method of massive MIMO system |
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 |