CN110557177A - DenseNet-based hybrid precoding method in millimeter wave large-scale MIMO system - Google Patents
DenseNet-based hybrid precoding method in millimeter wave large-scale MIMO system Download PDFInfo
- Publication number
- CN110557177A CN110557177A CN201910838213.4A CN201910838213A CN110557177A CN 110557177 A CN110557177 A CN 110557177A CN 201910838213 A CN201910838213 A CN 201910838213A CN 110557177 A CN110557177 A CN 110557177A
- Authority
- CN
- China
- Prior art keywords
- layer
- densenet
- neural network
- precoding
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 16
- 239000011159 matrix material Substances 0.000 claims abstract description 59
- 238000013528 artificial neural network Methods 0.000 claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 12
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 19
- 230000004913 activation Effects 0.000 claims description 12
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 210000005036 nerve Anatomy 0.000 claims description 2
- 238000004088 simulation Methods 0.000 claims description 2
- 238000004891 communication Methods 0.000 abstract description 5
- 238000013461 design Methods 0.000 description 6
- 239000013598 vector Substances 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000001228 spectrum Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000001965 increasing effect Effects 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 230000003595 spectral effect Effects 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 241000764238 Isis Species 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction 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
- H04B7/0456—Selection of precoding matrices or codebooks, e.g. using matrices antenna weighting
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Radio Transmission System (AREA)
Abstract
the invention relates to a Denset-based hybrid precoding method in a millimeter wave large-scale MIMO system, which belongs to the technical field of wireless communication and comprises an offline training stage and an online training stage, wherein the offline training stage comprises the steps of firstly, performing singular value decomposition on a channel matrix H to obtain an optimal unconstrained hybrid precoding F opt corresponding to the H, secondly, constructing a Denset neural network to aim at minimizing the F norm of the difference between the optimal unconstrained hybrid precoding F opt and the product F RF F BB of an analog precoding F RF and a digital precoding F BB, and optimizing the parameters of the Denset neural network by using SGD to obtain a trained Denset neural network, and the online stage of correspondingly outputting the optimal analog precoding matrix F RF and the digital precoding matrix F BB according to different channel conditions by using the trained Denset neural network.
Description
Technical Field
The invention belongs to the technical field of wireless communication, and particularly relates to a DenseNet-based hybrid precoding method in a millimeter wave large-scale MIMO system.
background
With the development of mobile internet and internet of things, wireless data traffic is increased dramatically; meanwhile, in a Massive large-connection scenario (Massive Machine Type of Communication, mtc), the number of devices connected to a wireless network will increase explosively due to a large-scale Internet of things (IOT) service, and the problem of shortage of spectrum resources is increasingly prominent. To meet the increasing data traffic of users, the development of fifth generation mobile communication systems has been promoted. The ultra-high frequency of the millimeter wave signal can provide a large amount of underutilized spectrum resources, so that millimeter wave communication is one of the key technologies of the 5G physical layer. Compared with most wireless systems at present, the carrier frequency is increased by 10 times, which causes the millimeter wave signal to increase by orders of magnitude in the path loss of the free space, and thus the millimeter wave communication outdoors is greatly hindered. The reduction of millimeter wave wavelength concentrates a large number of antenna elements on a small aperture, so that it is possible to implement a large-scale multiple-input multiple-output (MIMO) technology by configuring hundreds or thousands of antennas at the base station. In addition, large arrays may provide the beamforming gain needed to overcome path loss to offset the severe path loss of millimeter wave signals, while multiple data streams may be precoded to improve spectral efficiency. The combination of the millimeter wave and massive MIMO technology becomes inevitable.
The millimeter wave large-scale MIMO technology combined with precoding can multiplex a large number of data streams, more accurate beam forming is realized, and the frequency spectrum efficiency of a millimeter wave system is further improved. Conventional Multiple Input Multiple Output (MIMO) processing is typically digital at baseband, so that the phase and amplitude of the signal can be controlled simultaneously. However, digital processing requires dedicated baseband and rf hardware for each antenna element, resulting in high hardware cost and high power consumption, and such a transceiving structure cannot be realized at present; the hybrid precoding architecture separates the precoding process into digital baseband precoding and analog precoding composed of Analog Phase Shifters (APSs). The scheme only needs a small number of RF chains, thereby solving the problems of high cost and high power consumption of the traditional all-digital baseband precoding scheme.
Compared to digital precoding, analog precoding is typically achieved using phase shifters that impose constant modulus constraints on the elements of the radio frequency precoder, making the design of analog precoding more challenging. In the context of millimeter wave massive MIMO systems, although a lot of research has been devoted to improving hybrid precoding performance, there still exist many problems, two of the main challenges are extremely high computational complexity and poor system performance, and the existing millimeter wave massive MIMO systems do not fully utilize the structure existing in the channel, so as to achieve the traditional low complexity scheme at the cost of reducing the system hybrid precoding. Therefore, conventional studies have not fundamentally solved these problems. Deep learning is a non-trivial technique to deal with explosive data and to deal with complex non-linear problems. It is proved that deep learning is an excellent tool for dealing with the complex non-convex problem and the high computation problem, and the deep learning has super strong recognition and representation capabilities, and can capture the structure information of the hybrid precoding scheme through training, which is helpful for reducing the computation complexity and improving the spectrum efficiency of the system.
Disclosure of Invention
In view of this, the present invention provides a hybrid precoding method based on DenseNet in a millimeter wave massive MIMO system, which is used to solve the problems of high computational complexity and poor system performance of the existing hybrid precoding scheme.
in order to achieve the purpose, the invention provides the following technical scheme:
A DenseNet-based hybrid precoding method in a millimeter wave large-scale MIMO system comprises an offline training stage and an online stage, and specifically comprises the following steps:
(1) An offline training stage:
s1: singular value decomposition is carried out on the channel matrix H to obtain the optimal unconstrained hybrid precoding F corresponding to Hopt；
S2: constructing a DenseNet neural network to minimize optimal unconstrained hybrid precoding Foptand analog precoding FRFAnd digital precoding FBBProduct of FRFFBBThe F norm of the difference is taken as a target, and the parameters of the DenseNet neural network are optimized by using a random gradient descent algorithm (SGD) to obtain a trained DenseNet neural network;
(2) An online stage:
S3: correspondingly outputting an optimal analog pre-coding matrix F according to different channel conditions by utilizing the trained DenseNet neural networkRFAnd a digital precoding matrix FBB。
Further, the step S1 specifically includes: for channel matrixsingular Value (SVD) decomposition:
Wherein,
complex set is shown, rank (H) shows the rank of channel matrix H; n is a radical ofsRepresenting the number of data streams, NrAnd Ntthe number of receiving antennas and the number of transmitting antennas are respectively; optimal unconstrained precoder F corresponding to Hopt＝V1。
further, the step S2 specifically includes:
S21: constructing a DenseNet neural network, and optimizing a multilayer structure of the network by using an activation function; during initialization, establishing a mapping relation:
Wherein,respectively representing initial values of an analog precoding matrix and a digital precoding matrix, wherein omega is an input data set; training a DenseNet neural network with a loss function, loss function FlossComprises the following steps:
wherein | · | purple sweetFDenotes the F norm operation, FRF，FBBRespectively analog precoding and digital precoding;
S22: setting an error threshold tau and an iteration number j, initializing a weight w, and iteratively updating a weight matrix w through a random gradient descent algorithm to train to obtain the optimal DenseNet neural network.
Further, the DenseNet neural network is composed of 4 sequentially Connected neural network modules, each of which is composed of an Input (Input) layer, a convolution (convolution) layer, a Batch Normalization (Batch Normalization) layer, an Activation function (Activation) layer, a flat (Flatten) layer, a Fully Connected (full Connected) layer and an Output (Output) layer; the activation function uses relu (a) ═ max (0, a), where max (0, a) denotes taking the maximum value between 0 and a; the convolutional layer uses 32 convolutional kernels with the size of 3 × 3; the full connection layer comprises 1024 nerve units; the input to DenseNet is implemented for N times channel matrix H and is padded with a suitably sized Zero Padding (Zero Padding) layer.
Further, the connection mode among 4 neural network modules of the DenseNet neural network specifically is as follows:
network input is X0Through a neural network comprising L layers, if g is usedi(. h) represents the i-th layer nonlinear transformation, gi() is accumulated from a plurality of functions; layer 1 output X1＝g1(X0) (ii) a In order to optimize the transmission of the data stream, the output of each layer of the DenseNet is related not only to the input of the previous layer, but also to the output of all the previous layers, thus enhancing the reuse of the features. For example: for layer 2, the input is the input X of the network0and layer 1 output X1Is then the secondThe output of the layer is X2＝g2([X1,X0]) By analogy, the output of layer 4 is X4＝g4([X3,X2,X1,X0])，X3Is the output of the third layer.
Further, the step S22 specifically includes: in the data generation, generating N times of realization of a channel matrix H according to different channel conditions; dividing a channel matrix H into real parts Re { [ H ]]i,jAnd imaginary part Im { [ H ]]i,jExpressing Re {. cndot. } and Im {. cndot. } respectively taking a real part and an imaginary part of the channel matrix H; setting an initialization error threshold value tau, iteration times j and a weight w, iteratively updating a weight matrix w through a random gradient descent algorithm, and outputting the dimension of the weight matrix w at the output end of the neural networkwith a digital precoding and dimensionality ofPerforming real number to complex number conversion operation on the analog pre-coding matrix to obtain a digital and analog pre-coding matrix, whereinIndicating the number of RF chains.
The invention has the beneficial effects that: the method is characterized in that a neural network is built based on deep learning, aiming at the problems of high calculation complexity and poor system performance of the existing hybrid precoding scheme, the design of the hybrid precoder is regarded as a black box, millimeter wave channel characteristics are extracted by using the deep neural network, nonlinear operation among neurons is mapped into the hybrid precoding design, the learning network is trained in a large amount offline, and the network is deployed on line to reduce the calculation complexity of the system and obtain the optimal performance.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the invention may be realized and attained by the means of the instrumentalities and combinations particularly pointed out hereinafter.
drawings
for purposes of promoting a better understanding of the objects, aspects and advantages of the invention, reference will now be made to the following detailed description taken in conjunction with the accompanying drawings in which:
FIG. 1 is a schematic diagram of a millimeter wave massive MIMO system model;
FIG. 2 is a detailed structural diagram of the DenseNet-based optical fiber according to the present invention;
Fig. 3 is a general flow chart of the hybrid precoding based on DenseNet according to the present invention;
Fig. 4 is a schematic diagram of the training phase and the application phase of the DenseNet neural network.
Detailed Description
the embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention.
referring to fig. 1-4, fig. 1 is a single-user millimeter wave massive MIMO system model, assuming that the number of base station transmitting antennas is NtandOne RF chain, N in basebandsThe data stream is first of all sent by a digital precoder FBBIs precoded, and satisfies Ns≤Nt RF≤Ntafter passing through the respective RF chains, the digital domain signal from each RF chain is passed to an analog phase shifter to perform analog precoding. To fully achieve spatial multiplexing gain, the base station will typically be NsSending an independent data stream to Nrthe number of data streams for a user having a receiver antenna equal to the number of RF chains NsK, then the received signal y ═ y1,y2,...,yk]Can be expressed as:
Wherein,is a base station and NrDepending on the downlink channel matrix between the receiving antennas, p is the average received power,An analog pre-coding matrix is represented,A digital pre-coding matrix is represented,For transmitting symbol vectors and satisfying power constraintsSince F is implemented using an analog phase shifterRFThus, therefore, it isIs additive white Gaussian noise, whereinP＝FRFFBBIs a hybrid precoding matrix and satisfiesTo ensure the total transmission power limitation, | ·| non-calculationFRepresenting the F-norm operation.
Considering the millimeter wave propagation characteristic of high free space path loss, which results in limited spatial selectivity or scattering, while the large and dense antenna array deployed at the base station end results in a high level of correlation between the antennas, the combination of such dense arrangement is such that in a millimeter wave sparse scattering environmentThe traditional fading statistical channel model is in line with the inaccuracy of millimeter wave channel modeling. Therefore, the extended Saleh-Vallenzuela-based model is adopted to model the millimeter wave massive MIMO channel, and the mathematical structure of the millimeter wave channel can be accurately described. According to this model, assume a matrix channel H of Nclintegration of the distribution of scattering clusters, each cluster containing NrayA strip propagation path. Thus, the narrowband channel H can be represented as:
wherein alpha isi,lRepresents the complex gain of the ith propagation path in the ith scattering cluster, obeys a mean of 0 and a variance ofThe complex gaussian distribution of (a) is,Represents the average power of the ith cluster and satisfies Is a normalization factor, which is a function of, AndAngle of Arrival (AOA) and Angle of Departure (AOD) respectively representing azimuth angles, vectorsAndrespectively expressed in azimuthandReceive and transmit array response vectors. When the transmitting antenna and the receiving antenna array both adopt Uniform Linear Arrays (ULA), the array response vectorandThe following can be written:
where k is 2 pi/lambda, lambda is the wavelength,Is the inter-antenna spacing. The transmit antenna array response vector is represented in a matrix form:The corresponding vector of the receive antenna array may also be represented in a similar fashion.
if the transmission signal is transmitted through a channel, the spectral efficiency R of the whole system can be expressed as:
P＝FRF FBBFor the hybrid precoding matrix, the invention designs FRF，FBBMaximization of mutual information achieved by a Gaussian channel on a maximized mmWave channel;
From FRF,FBBBased on the design of (a), the precoding optimization problem can be expressed as:
wherein,A feasible set of analog precoding matrices is represented whose elements satisfy a constant modulus constraint. While maximizing spectral efficiency is approximately equivalent to minimizing the optimal hybrid precoding matrix FRFFBBAnd a full digital precoding matrix Foptthe euclidean distance of (a) thus further simplifies the design problem of the hybrid precoding matrix, and the hybrid precoding optimization problem can be rewritten as:
Wherein, FoptRepresented as an optimal all-digital precoding matrix.
as shown in fig. 3 to 4, based on the above system, the hybrid precoding algorithm based on the DenseNet network in this embodiment includes the following steps:
The method comprises the following steps: offline training phase
(1) For channel matrixSingular Value (SVD) decomposition:
Wherein
Complex set is shown, rank (H) shows the rank of channel matrix H; n is a radical ofsRepresenting the number of data streams, NrAnd NtThe number of receiving antennas and the number of transmitting antennas are respectively; optimal unconstrained precoder F corresponding to Hopt＝V1。
(2) And constructing a DenseNet neural network, and optimizing the multilayer structure of the network by using an activation function. During initialization, establishing a mapping relation:
Wherein,Respectively representing the initial values of the analog precoding matrix and the digital precoding matrix; to train the DenseNet neural network, the loss function is set to:
Wherein | · | purple sweetFrepresenting an F norm operation, omega being the input data set, FRF，FBBRespectively analog precoding and digital precoding.
In order to extract better features from the input channel matrix, the proposed DenseNet network consists of 4 sequentially connected neural network modules, as shown in fig. 2. The neural network module is composed of an Input (Input) layer, a convolution (Convolutional) layer, a Batch Normalization (Batch Normalization) layer, an Activation (Activation) layer, a flat (Flatten) layer, a Fully Connected (full Connected) layer and an Output (Output) layer, and in addition, a function relu (a) ═ max (0, a) is introduced as an Activation function of the neural network, and max (0, a) represents a nonlinear Output of the neural network. The input to DenseNet is implemented N times as the channel matrix H and is padded with a suitably sized Zero Padding (Zero Padding) layer. The convolutional layer of each neural network of DenseNet uses 32 convolutional kernels of size 3 × 3, one ZeroPadding layer, one ReLU activation function, and one fully-connected layer, where the fully-connected layer contains 1024 neural units.
The connection mode among the 4 DenseNet neural network modules is as follows: network input is X0Through a neural network comprising L layers, if g is usedi(. h) represents the i-th layer nonlinear transformation, gi(. h) is accumulated from a plurality of functions. Layer 1 output X1＝g1(X0). In order to optimize the transmission of the data stream, the output of each layer of the DenseNet is related not only to the input of the previous layer, but also to the outputs of all the previous layers, thus enhancing the reuse of the features. For example: for layer 2, the input is the input X of the network0And layer 1 output X1then the output of the second layer is X2＝g2([X1,X0]) The output of layer 4, in turn, is X4＝g4([X3,X2,X1,X0])。
In data generation, N realizations of the channel matrix H are generated according to different channel conditions. Splitting a channel matrix H into real parts Re { [ H ]]i,jand imaginary part Im { [ H ]]i,j}，Re{·}、Im{. denotes the real part and the imaginary part of the channel matrix H, respectively; setting an initialization error threshold value tau, iteration times j and a weight w, iteratively updating the weight matrix w through a random gradient descent algorithm, and obtaining an output dimension ofWith a digital precoding and dimensionality ofand performing real number conversion complex operation on the analog pre-coding to obtain a digital and analog pre-coding matrix.
step two: on-line stage
(3) Correspondingly outputting an optimal simulation precoding matrix F according to different channel conditions by using the trained DenseNet neural networkRFAnd a digital precoding matrix FBB。
Finally, the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions, and all that should be covered by the claims of the present invention.
Claims (6)
1. A DenseNet-based hybrid precoding method in a millimeter wave large-scale MIMO system is characterized by comprising an offline training stage and an online stage, and specifically comprising the following steps:
(1) An offline training stage:
S1: singular value decomposition is carried out on the channel matrix H to obtain the optimal unconstrained hybrid precoding F corresponding to Hopt；
S2: constructing a DenseNet neural network to minimize optimal unconstrained hybrid precoding Foptand analog precoding FRFAnd digital precoding FBBProduct of FRFFBBtargeting the F norm of the difference, using a stochastic gradient descent algorithm to DenseNetOptimizing parameters of the network to obtain a trained DenseNet neural network;
(2) an online stage:
S3: correspondingly outputting an optimal simulation precoding matrix F according to different channel conditions by using the trained DenseNet neural networkRFAnd a digital precoding matrix FBB。
2. The hybrid precoding method based on DenseNet in mmwave massive MIMO system according to claim 1, wherein the step S1 specifically includes: for channel matrixSingular value decomposition is carried out:
Wherein, complex set is shown, rank (H) shows the rank of channel matrix H; n is a radical ofsRepresenting the number of data streams, NrAnd NtThe number of receiving antennas and the number of transmitting antennas are respectively; optimal unconstrained precoder F corresponding to Hopt＝V1。
3. The hybrid precoding method based on DenseNet in mmwave massive MIMO system according to claim 2, wherein the step S2 specifically includes:
S21: constructing a DenseNet neural network, and optimizing a multilayer structure of the network by using an activation function; during initialization, establishing a mapping relation:
Wherein,Respectively representing initial values of an analog precoding matrix and a digital precoding matrix, wherein omega is an input data set; training a DenseNet neural network with a loss function, loss function Flosscomprises the following steps:
Wherein | · | purple sweetFDenotes the F norm operation, FRF，FBBRespectively analog precoding and digital precoding;
S22: setting an error threshold tau and an iteration number j, initializing a weight w, and iteratively updating a weight matrix w through a random gradient descent algorithm to train to obtain the optimal DenseNet neural network.
4. The DenseNet-based hybrid precoding method in MMW massive MIMO system according to claim 3, wherein the DenseNet neural network is composed of 4 sequentially connected neural network modules, each of which is composed of an input layer, a convolutional layer, a batch normalization layer, an activation function layer, a flat layer, a fully connected layer and an output layer; the activation function uses relu (a) ═ max (0, a), where max (0, a) denotes taking the maximum value between 0 and a; the convolutional layer uses 32 convolutional kernels with the size of 3 × 3; the full connection layer comprises 1024 nerve units; the input to DenseNet is implemented for the channel matrix H N times and is padded with a zero padding layer of appropriate size.
5. the DenseNet-based hybrid precoding method in the MMW massive MIMO system according to claim 4, wherein the connection mode among 4 neural network modules of the DenseNet neural network is specifically as follows:
Network input is X0through a neural network comprising L layers, if g is usedi(. h) represents the i-th layer nonlinear transformation, gi() is accumulated from a plurality of functions; layer 1 output X1＝g1(X0) (ii) a For layer 2, the input is the input X of the network0and layer 1 output X1then the output of the second layer is X2＝g2([X1,X0]) By analogy, the output of layer 4 is X4＝g4([X3,X2,X1,X0])，X3Is the output of the third layer.
6. The hybrid precoding method based on DenseNet in mmwave massive MIMO system according to claim 3, wherein the step S22 specifically includes: in the data generation, generating N times of realization of a channel matrix H according to different channel conditions; dividing a channel matrix H into real parts Re { [ H ]]i,jAnd imaginary part Im { [ H ]]i,jexpressing Re {. cndot. } and Im {. cndot. } respectively taking a real part and an imaginary part of the channel matrix H; setting an initialization error threshold value tau, iteration times j and a weight w, iteratively updating a weight matrix w through a random gradient descent algorithm, and outputting the dimension of the weight matrix w at the output end of the neural networkWith a digital precoding and dimensionality ofPerforming real number to complex number conversion operation on the analog pre-coding matrix to obtain a digital and analog pre-coding matrix, whereinIndicating the number of RF chains.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910838213.4A CN110557177A (en) | 2019-09-05 | 2019-09-05 | DenseNet-based hybrid precoding method in millimeter wave large-scale MIMO system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910838213.4A CN110557177A (en) | 2019-09-05 | 2019-09-05 | DenseNet-based hybrid precoding method in millimeter wave large-scale MIMO system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110557177A true CN110557177A (en) | 2019-12-10 |
Family
ID=68739234
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910838213.4A Pending CN110557177A (en) | 2019-09-05 | 2019-09-05 | DenseNet-based hybrid precoding method in millimeter wave large-scale MIMO system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110557177A (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110995327A (en) * | 2019-12-17 | 2020-04-10 | 电子科技大学 | Mixed beam forming optimization method and system for multi-carrier MIMO system |
| CN111092641A (en) * | 2019-12-18 | 2020-05-01 | 重庆邮电大学 | Hybrid precoding design method based on millimeter wave MIMO system deep learning |
| CN111181619A (en) * | 2020-01-03 | 2020-05-19 | 东南大学 | Millimeter wave hybrid beam forming design method based on deep reinforcement learning |
| CN111277312A (en) * | 2020-02-26 | 2020-06-12 | 电子科技大学 | Fixed subarray space-based millimeter wave beam forming method based on deep complex network |
| CN111726144A (en) * | 2020-06-24 | 2020-09-29 | 中南大学 | Hybrid precoding design method, device, medium and equipment based on initial value optimization |
| CN112165344A (en) * | 2020-09-16 | 2021-01-01 | 杭州电子科技大学 | Mixed precoding algorithm based on gradient descent method |
| CN112398513A (en) * | 2020-11-09 | 2021-02-23 | 杭州电子科技大学 | Beam forming method of massive MIMO system |
| CN112448742A (en) * | 2020-09-28 | 2021-03-05 | 上海电机学院 | Hybrid precoding method based on convolutional neural network under non-uniform quantization |
| CN112910520A (en) * | 2021-02-03 | 2021-06-04 | 广州市埃特斯通讯设备有限公司 | Convolutional neural network-based MIMO system beam training method |
| CN112910521A (en) * | 2021-02-27 | 2021-06-04 | 中电万维信息技术有限责任公司 | Deep learning-based MIMO mixed beam forming method |
| CN113193893A (en) * | 2021-04-30 | 2021-07-30 | 东南大学 | Millimeter wave large-scale MIMO intelligent hybrid beam forming design method |
| CN113242072A (en) * | 2021-05-10 | 2021-08-10 | 东南大学 | Hybrid beam forming method based on convolutional neural network |
| CN113839697A (en) * | 2021-09-23 | 2021-12-24 | 南通大学 | Joint feedback and hybrid precoding design method based on deep learning |
| CN114499605A (en) * | 2022-02-25 | 2022-05-13 | 北京京东方传感技术有限公司 | Signal transmission method, signal transmission device, electronic equipment and storage medium |
| CN114696878A (en) * | 2022-02-23 | 2022-07-01 | 浙江大学 | Combined design method and device based on hybrid precoding and channel estimation |
| CN115085782A (en) * | 2022-06-15 | 2022-09-20 | 南通大学 | Intelligent reflecting surface joint feedback and hybrid precoding method based on deep learning |
| CN115426013A (en) * | 2022-08-31 | 2022-12-02 | 东南大学 | RIS auxiliary communication system precoding and phase shifting method based on unsupervised learning |
| CN115473553A (en) * | 2022-08-03 | 2022-12-13 | 华中科技大学 | Method and device for intelligently inhibiting beam squint and electronic equipment |
| CN115549732A (en) * | 2022-09-13 | 2022-12-30 | 哈尔滨工业大学 | Large-scale MIMO mixed pre-coding method based on artificial intelligence |
| CN116112044A (en) * | 2022-08-09 | 2023-05-12 | 西安电子科技大学 | High-energy-efficiency hybrid precoding method based on neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109617585A (en) * | 2019-01-18 | 2019-04-12 | 杭州电子科技大学 | Mixing method for precoding based on part connection in the extensive MIMO of millimeter wave |
| CN109660297A (en) * | 2018-12-19 | 2019-04-19 | 中国矿业大学 | A kind of physical layer visible light communication method based on machine learning |
| US20190190578A1 (en) * | 2017-12-15 | 2019-06-20 | Motorola Mobility Llc | Coding main beam information in csi codebook |
-
2019
- 2019-09-05 CN CN201910838213.4A patent/CN110557177A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190190578A1 (en) * | 2017-12-15 | 2019-06-20 | Motorola Mobility Llc | Coding main beam information in csi codebook |
| CN109660297A (en) * | 2018-12-19 | 2019-04-19 | 中国矿业大学 | A kind of physical layer visible light communication method based on machine learning |
| CN109617585A (en) * | 2019-01-18 | 2019-04-12 | 杭州电子科技大学 | Mixing method for precoding based on part connection in the extensive MIMO of millimeter wave |
Non-Patent Citations (3)
| Title |
|---|
| HONGJI HUANG 等: "Deep-Learning-Based Millimeter-Wave Massive MIMO for Hybrid Precoding", 《IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY》 * |
| KAREN SIMONYAN 等: "Very Deep Convolutional Networks fo Large-scale Image Recognition", 《ICLR 2015》 * |
| OMAR EL AYACH 等: "Spatially Sparse Precoding in Millimeter Wave MIMO Systems", 《IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS》 * |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110995327A (en) * | 2019-12-17 | 2020-04-10 | 电子科技大学 | Mixed beam forming optimization method and system for multi-carrier MIMO system |
| CN110995327B (en) * | 2019-12-17 | 2021-05-04 | 电子科技大学 | Mixed beam forming optimization method and system for multi-carrier MIMO system |
| CN111092641A (en) * | 2019-12-18 | 2020-05-01 | 重庆邮电大学 | Hybrid precoding design method based on millimeter wave MIMO system deep learning |
| CN111092641B (en) * | 2019-12-18 | 2022-02-22 | 重庆邮电大学 | Hybrid precoding design method based on millimeter wave MIMO system deep learning |
| CN111181619A (en) * | 2020-01-03 | 2020-05-19 | 东南大学 | Millimeter wave hybrid beam forming design method based on deep reinforcement learning |
| CN111277312A (en) * | 2020-02-26 | 2020-06-12 | 电子科技大学 | Fixed subarray space-based millimeter wave beam forming method based on deep complex network |
| CN111726144A (en) * | 2020-06-24 | 2020-09-29 | 中南大学 | Hybrid precoding design method, device, medium and equipment based on initial value optimization |
| CN112165344A (en) * | 2020-09-16 | 2021-01-01 | 杭州电子科技大学 | Mixed precoding algorithm based on gradient descent method |
| CN112165344B (en) * | 2020-09-16 | 2021-09-21 | 杭州电子科技大学 | Mixed precoding method based on gradient descent method |
| CN112448742A (en) * | 2020-09-28 | 2021-03-05 | 上海电机学院 | Hybrid precoding method based on convolutional neural network under non-uniform quantization |
| CN112398513A (en) * | 2020-11-09 | 2021-02-23 | 杭州电子科技大学 | Beam forming method of massive MIMO system |
| CN112910520A (en) * | 2021-02-03 | 2021-06-04 | 广州市埃特斯通讯设备有限公司 | Convolutional neural network-based MIMO system beam training method |
| CN112910521A (en) * | 2021-02-27 | 2021-06-04 | 中电万维信息技术有限责任公司 | Deep learning-based MIMO mixed beam forming method |
| CN112910521B (en) * | 2021-02-27 | 2022-04-05 | 中电万维信息技术有限责任公司 | Deep learning-based MIMO mixed beam forming method |
| CN113193893A (en) * | 2021-04-30 | 2021-07-30 | 东南大学 | Millimeter wave large-scale MIMO intelligent hybrid beam forming design method |
| CN113242072A (en) * | 2021-05-10 | 2021-08-10 | 东南大学 | Hybrid beam forming method based on convolutional neural network |
| CN113839697A (en) * | 2021-09-23 | 2021-12-24 | 南通大学 | Joint feedback and hybrid precoding design method based on deep learning |
| CN114696878A (en) * | 2022-02-23 | 2022-07-01 | 浙江大学 | Combined design method and device based on hybrid precoding and channel estimation |
| CN114696878B (en) * | 2022-02-23 | 2023-10-17 | 浙江大学 | Combined design method and device based on mixed precoding and channel estimation |
| CN114499605A (en) * | 2022-02-25 | 2022-05-13 | 北京京东方传感技术有限公司 | Signal transmission method, signal transmission device, electronic equipment and storage medium |
| CN115085782A (en) * | 2022-06-15 | 2022-09-20 | 南通大学 | Intelligent reflecting surface joint feedback and hybrid precoding method based on deep learning |
| CN115473553A (en) * | 2022-08-03 | 2022-12-13 | 华中科技大学 | Method and device for intelligently inhibiting beam squint and electronic equipment |
| CN115473553B (en) * | 2022-08-03 | 2024-04-23 | 华中科技大学 | Method and device for intelligently inhibiting beam strabismus and electronic equipment |
| CN116112044A (en) * | 2022-08-09 | 2023-05-12 | 西安电子科技大学 | High-energy-efficiency hybrid precoding method based on neural network |
| CN115426013A (en) * | 2022-08-31 | 2022-12-02 | 东南大学 | RIS auxiliary communication system precoding and phase shifting method based on unsupervised learning |
| CN115549732A (en) * | 2022-09-13 | 2022-12-30 | 哈尔滨工业大学 | Large-scale MIMO mixed pre-coding method based on artificial intelligence |
| CN115549732B (en) * | 2022-09-13 | 2024-05-28 | 哈尔滨工业大学 | Large-scale MIMO hybrid precoding method, device and storage medium based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110557177A (en) | DenseNet-based hybrid precoding method in millimeter wave large-scale MIMO system | |
| CN111092641B (en) | Hybrid precoding design method based on millimeter wave MIMO system deep learning | |
| Bao et al. | Deep CNN and equivalent channel based hybrid precoding for mmWave massive MIMO systems | |
| CN113691288B (en) | Joint pilot frequency, feedback and multi-user hybrid coding method based on deep learning | |
| CN108880635A (en) | A kind of transmit-receive combination mixing method for precoding of the millimeter wave mimo system based on orthogonal code book | |
| CN113300746A (en) | Millimeter wave MIMO antenna and hybrid beam forming optimization method and system | |
| CN110943768B (en) | Mixed precoding codebook joint design method of millimeter wave large-scale MIMO system | |
| CN109302215B (en) | Hybrid precoding method based on row vector optimization | |
| CN110138427B (en) | Large-scale multi-input multi-output hybrid beam forming algorithm based on partial connection | |
| CN110661555B (en) | Hybrid precoding algorithm for partially connected phase shifter networks for massive MIMO | |
| CN108199753A (en) | A kind of method for precoding based on iteration minimum in millimetre-wave attenuator | |
| CN112468202B (en) | Low-complexity millimeter wave large-scale MIMO hybrid precoding method | |
| CN112653496B (en) | Mixed precoding method of millimeter wave large-scale MIMO system | |
| Tao et al. | Autoencoder neural network based intelligent hybrid beamforming design for mmWave massive MIMO systems | |
| CN112564754B (en) | Wave beam selection method based on self-adaptive cross entropy under millimeter wave Massive MIMO system | |
| CN112636794A (en) | Wave beam shaping method based on greedy algorithm | |
| Balevi et al. | Unfolded hybrid beamforming with GAN compressed ultra-low feedback overhead | |
| Li et al. | Multi-task deep learning based hybrid precoding for mmWave massive MIMO system | |
| CN113572503B (en) | Low-complexity improved mixed beam forming method based on GP | |
| Thuc et al. | A metaheuristics-based hyperparameter optimization approach to beamforming design | |
| CN110233649A (en) | A kind of dynamic subarray design method based on millimeter wave mimo system | |
| CN112398513A (en) | Beam forming method of massive MIMO system | |
| Hu et al. | Hybrid beamforming with deep learning for large-scale antenna arrays | |
| Bo et al. | Deep learning based low-resolution hybrid precoding design for mmWave MISO systems | |
| CN108683441B (en) | Multi-user beam forming method in hybrid pre-coding |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191210 |