CN114844538A - Millimeter wave MIMO user increment cooperative beam selection method based on wide learning - Google Patents

Millimeter wave MIMO user increment cooperative beam selection method based on wide learning Download PDF

Info

Publication number
CN114844538A
CN114844538A CN202210474863.7A CN202210474863A CN114844538A CN 114844538 A CN114844538 A CN 114844538A CN 202210474863 A CN202210474863 A CN 202210474863A CN 114844538 A CN114844538 A CN 114844538A
Authority
CN
China
Prior art keywords
user
wide
training
learning
network
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.)
Granted
Application number
CN202210474863.7A
Other languages
Chinese (zh)
Other versions
CN114844538B (en
Inventor
张铖
黄永明
俞菲
张璐佳
陈乐明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southeast University
Original Assignee
Southeast University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southeast University filed Critical Southeast University
Priority to CN202210474863.7A priority Critical patent/CN114844538B/en
Publication of CN114844538A publication Critical patent/CN114844538A/en
Application granted granted Critical
Publication of CN114844538B publication Critical patent/CN114844538B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • 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
    • H04B7/0426—Power distribution
    • H04B7/043—Power distribution using best eigenmode, e.g. beam forming or beam steering
    • 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)
  • Power Engineering (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

The invention discloses a millimeter wave MIMO user increment cooperative beam selection method based on wide learning, which comprises the following steps: aiming at the problem of downlink beam selection in a multi-point cooperation millimeter wave large-scale MIMO scene, each user collects downlink wide beam responses and transmission narrow beam responses, trains a local wide learning network, and selects beams based on predicted narrow beam responses. Further, by modeling the training problem for each user's local network as a distributed optimization problem with consistency constraints, efficient sharing of training data can be achieved with D2D communication between neighboring users. Furthermore, an incremental updating mode of the user local network in a cooperation mode is designed, and the training complexity of the network can be effectively reduced. The method fully utilizes the capability of the distributed wide learning to mine the response relation between the multi-base-station wide beam response and the transmission narrow beam response under the condition of a small sample, and can realize the low-complexity and low-overhead beam selection of the fast time-varying scene multipoint cooperative millimeter wave large-scale system.

Description

Millimeter wave MIMO user increment cooperative beam selection method based on wide learning
Technical Field
The invention relates to the field of wireless communication network optimization and intelligent communication, in particular to a millimeter wave MIMO user increment cooperative beam selection method based on wide learning.
Background
Due to the advantage of large bandwidth of millimeter wave frequency band, meanwhile, massive MIMO beam forming can effectively compensate millimeter wave propagation fading by using array gain, and millimeter wave massive MIMO technology is considered as one of key technologies for realizing higher spectrum efficiency of 5.5G mobile communication system. Then, since millimeter wave massive MIMO requires data transmission using narrow beams, it is very sensitive to shadowing effects existing in a channel. To solve the above problem, coordinated multipoint transmission is considered as a main mode of the millimeter wave massive MIMO system to achieve its performance gain. In a multi-point cooperative millimeter wave large-scale MIMO system, the required beam training overhead is significantly increased along with the increase of the number of cooperative base stations, and in order to reduce the training overhead, a related low-overhead beam training scheme is proposed by research based on the idea of combining traditional model driving and data driving with Machine Learning (ML).
In the existing typical ML-based solution, user uplink omnidirectional wide beam response and uplink narrow beam response samples collected at the base station side in an offline stage are utilized. Since the wide beam response from the user to multiple base stations can approximately reflect the user's location and propagation environment characteristics, the factors that determine the narrow beam response include the channel response and the pattern in the narrow beam determined by the user's location and propagation environment. By training the mapping relation between the multi-base station uplink wide beam response and the multi-base station uplink narrow beam response, the base station side can be helped to predict the uplink narrow beam response only by using the uplink wide beam response at the online stage, and corresponding beam selection is made.
The above typical ML scheme cannot be directly applied to beam selection for downlink transmission in a scenario where reciprocity between uplink and downlink of a channel is poor, such as a Frequency Division Duplex (FDD) mode. Therefore, the user side needs to complete mapping training based on the collected downlink wide beam response and downlink narrow beam response of the multiple base stations, so as to support prediction of beam selection, and feed back the prediction to the base station side.
Different from model training of a base station side, a user side needs more elaborate design due to the fact that computing power, storage capacity and interaction capacity of the user side are relatively weak, training data sharing of multiple users and complexity control of model updating and the like, and especially under a fast time-varying scene, the requirements on collection cost control of samples and real-time guarantee of model updating are higher.
Disclosure of Invention
In view of this, the present invention aims to provide a millimeter wave MIMO user increment cooperative beam selection method based on wide learning, so as to solve the problem that the current low-overhead beam training method based on wide beam response cannot be directly applied to a millimeter wave multipoint cooperative downlink transmission system without reciprocity of uplink and downlink channels, and optimize ML model design on a user side, so as to adapt to weak calculation power, storage capability and multi-user interaction capability on the user side. The method realizes the lightweight of the ML model at the user side, and simultaneously ensures the performance under the scene that a small training sample and the model need to be frequently updated.
In order to achieve the purpose, the invention adopts the following technical scheme:
a millimeter wave MIMO user incremental cooperative beam selection method based on wide learning, the method comprising:
step S1, constructing a downlink received signal receiving expression and a corresponding reachable rate expression of the downlink received signal receiving expression on the user side of the multi-point cooperative millimeter wave large-scale MIMO system; then, constructing an actual effective rate expression of the user side according to the reachable rate expression and the tracking period of the channel time variation of the system; constructing a global optimization problem model which takes the maximization of the system effectiveness and the speed as the target based on the actual effective speed expression;
step S2, constructing and initializing a local wide learning mapping network on each user side based on the wide learning architecture; taking multi-base station wide beam responses collected by each user and beam equivalent channel strength corresponding to transmission narrow beams as a local training set, wherein the multi-base station wide beam responses are taken as input features, equivalent rate index sample beams corresponding to the transmission narrow beams are taken as output, affine transformation is carried out on the input features through setting feature nodes and enhancement nodes based on a width learning framework, and the optimization problem of a network model is modeled as a weight problem for solving the affine transformation;
step S3, equivalently converting the multi-user beam selection problem into a distributed optimization problem with consistency constraint according to a distributed optimization theory, designing an iterative interactive method to train a local wide learning mapping network of each multi-user, and realizing effective sharing of a multi-user local training set, thereby reducing the requirement on the number of training samples;
step S4, introducing a collaborative incremental learning method for rapidly updating collaborative model parameters when data are newly added, and realizing a network model which is updated while being collected; performing iterative training on the data set generated in the step S2 by using the network established in the step S3;
and step S5, each user inputs the multi-base-station wide beam response collected in real time into the local mapping network to predict the beam selection of the user.
Further, in the multi-point cooperative millimeter wave massive MIMO system, it includes:
b base stations and U users, wherein the number of antennas of a single base station is M, the users configure omnidirectional antennas for signal reception, and the number of system subcarriers is K;
the user receives the cooperation downlink transmission service provided by the B base stations through an orthogonal frequency division multiple access mode;
considering the situation that a single base station uses a single beam to serve one user, and multiple users move randomly in a certain area, the base stations are set as
Figure BDA0003624997700000021
Set of users as
Figure BDA0003624997700000022
Set of subcarriers as
Figure BDA0003624997700000023
The set of subcarriers allocated to user u is
Figure BDA0003624997700000024
The size is recorded as
Figure BDA0003624997700000025
Satisfy the requirement of
Figure BDA0003624997700000026
Recording the response of a downlink antenna domain channel from a base station b to a user u on a subcarrier k as h b,u,k Then, the modeling is as follows:
Figure BDA0003624997700000031
in this formula, L represents the number of scattering paths, α, of the propagation channel b,u,l Denotes the complex gain of the l-th path, f k Denotes the center frequency, τ, of the k-th subcarrier l Represents the path delay of the ith path; a (-) denotes a pilot vector of the base station multi-antenna array.
Further, for the multipoint cooperation millimeter wave massive MIMO system, the downlink received signal receiving expression and the reachable rate expression corresponding to the downlink received signal receiving expression at the user side are obtained by the following method:
considering that downlink transmission is based on typical mixed precoding, the analog beam sets of all base stations are
Figure BDA0003624997700000032
And satisfy F H F＝FF H ＝I M The base station b selects the second from F
Figure BDA0003624997700000033
Column(s) of
Figure BDA0003624997700000034
As analogue beams
Figure BDA0003624997700000035
Serving user u, defining an analog beamforming matrix for a plurality of base stations
Figure BDA0003624997700000036
The cooperative center control node collects the beam equivalent channel state information of each base station, and designs the information on the sub-carriers according to the maximum ratio transmission criterion
Figure BDA0003624997700000037
The numeric precoding vector of the upper service user u is:
Figure BDA0003624997700000038
in the formula, the first and second sets of data are represented,
Figure BDA0003624997700000039
wherein
Figure BDA00036249977000000310
Representing the response of the downlink antenna domain channel from B base stations to user u on subcarrier k.
Then the downlink received signal of the user u on the kth subcarrier is represented as:
Figure BDA00036249977000000311
in this formula, s u,k ～CN(0,P u ) Is a data symbol, n u,k ～CN(0,σ 2 ) For the receiver noise, σ, of user u on the k-th subcarrier 2 Is the noise power; when in use
Figure BDA00036249977000000312
When the system adopts equal power distribution between users and subcarriers, the method comprises the following steps
Figure BDA00036249977000000313
Wherein P is the total transmitting power of the base station;
correspondingly, the received signal-to-interference-and-noise ratio of user u on subcarrier k is expressed as:
Figure BDA00036249977000000314
the corresponding achievable rates are:
Figure BDA00036249977000000315
in the formula, B w Representing the system bandwidth.
Further, for the multi-point cooperative millimeter wave massive MIMO system, a global optimization problem model with the goal of maximizing system effectiveness and rate is obtained by the following method, including:
defining the tracking period of the system for channel time variation as T, namely, the system needs to perform beam training again every T to update the precoding design for data transmission; first T of each cycle r Time, the base station carries out beam training, and the rest time carries out data transmission;
thus, the actual effective rate achieved by user u is represented as:
Figure BDA0003624997700000041
the beam selection problem is expressed as an optimization problem that maximizes system efficiency and rate through beam selection:
Figure BDA0003624997700000042
since there is no interference between users accessing through OFDMA, the above problem translates into:
Figure BDA0003624997700000043
optimization due to the above problem requires selection of an optimal beam from a cascaded selectable analog beam space of B base stationsIn combination, the search space size is M B The conversion is thus made as follows:
Figure BDA0003624997700000044
a low-complexity sub-optimal solution of the original problem is obtained.
Further, the step S2 specifically includes:
to solve the problem constructed in step S1, each base station
Figure BDA0003624997700000045
Using pilots in the training phase
Figure BDA0003624997700000046
Making wide beams
Figure BDA0003624997700000047
And each analog beam to be selected
Figure BDA0003624997700000048
The training of (2),
Figure BDA0003624997700000049
represents the training power;
then each user
Figure BDA00036249977000000410
On the sub-carrier
Figure BDA00036249977000000411
The above obtained pilot received signal is:
Figure BDA00036249977000000412
after pilot frequency matching, the user obtains each wave beam response
Figure BDA00036249977000000413
Is estimated as：
Figure BDA00036249977000000414
Each user
Figure BDA00036249977000000415
By collecting N multi-base-station wide-beam response samples
Figure BDA00036249977000000416
Corresponding analog wave beam equivalent rate index sample to be selected
Figure BDA00036249977000000417
Figure BDA00036249977000000418
Training a local wide learning network;
the specific process is as follows:
memory users
Figure BDA00036249977000000419
The wide beam response sample N collected from base station b is 1, …, with N arranged as
Figure BDA00036249977000000420
Figure BDA00036249977000000421
The response samples of the beam to be selected are
Figure BDA0003624997700000051
All the wide beam response samples collected from the B base stations are arranged into
Figure BDA0003624997700000052
Corresponding to-be-selected beam response sample arrangement
Figure BDA0003624997700000053
After data arrangement, the input of the user u local wide learning network model corresponding to the nth sample is as follows:
Figure BDA0003624997700000054
wherein
Figure BDA0003624997700000055
The symbol represents taking a complex phase operator;
output is as
Figure BDA0003624997700000056
Then the N samples correspond to the input and output matrices as:
Figure BDA0003624997700000057
further, in the step S2, an algorithm based on a width learning framework is adopted to perform nonlinear transformation and affine transformation on the original feature of the wide beam response signal received by the user; the algorithm defines characteristic nodes and enhancement nodes, calculates the characteristic nodes from input data, further generates the enhancement nodes from the characteristic nodes, and then splices the two nodes together as affine transformation; based on the structure, the optimization of the model only needs to solve the weight of affine transformation;
based on a wide learning framework, the specific processing process of the input sample in the feature transformation network is as follows:
first, using input data X u Mapping to I-group feature nodes
Figure BDA0003624997700000058
And J group enhanced nodes
Figure BDA0003624997700000059
Tables of F and EShowing the number of the characteristic nodes and the enhanced nodes of each group, and specifically showing that:
Figure BDA00036249977000000510
Figure BDA00036249977000000511
wherein Z u ＝[Z u,1 ,Z u,2 ,…,Z u,I ]A cascade matrix of characteristic nodes of group I, H u ＝[H u,1 ,...,H u,J ]A cascaded matrix of enhanced nodes for the J set;
Figure BDA00036249977000000512
and
Figure BDA00036249977000000513
respectively representing the connection weight and the bias of the feature generation network and the feature enhancement network, and generally adopting random generation;
Figure BDA00036249977000000514
a column vector representing elements all 1; phi (-) and xi (-) both represent linear or non-linear activation functions;
then, the user local wide learning network pair joint feature enhancement node
Figure BDA00036249977000000515
Performing affine transformations
Figure BDA00036249977000000516
Output of
Figure BDA00036249977000000517
Based on the minimum mean square error criterion and utilizing L2 norm regularization to improve the generalization performance of the network, the network weight matrix W is learned widely u The optimization problem of (a) is modeled as:
Figure BDA0003624997700000061
in the formula, | · luminance | | F Expressing a Frobenius norm, and lambda expresses a regularization factor;
the optimal solution is as follows:
Figure BDA0003624997700000062
in the formula, I IF+JE An identity matrix of dimension IF + JE is represented.
Further, the step S3 includes:
designing a mechanism for a user to perform cooperative training according to a downlink wide beam response signal based on an ADMM algorithm so as to reduce the calculation burden of each user node, and regarding the converted distributed optimization problem as a problem that the user shares information through cooperation, specifically comprising the following steps:
if the propagation environments between the U users and the base station have similarity in space, the models of local wide learning should tend to be consistent, that is, the models are consistent
Figure BDA0003624997700000063
I.e. joint representation of network training optimization problem for multiple users as
Figure BDA0003624997700000064
In the formula, the first step is that,
Figure BDA0003624997700000065
the shared training based on all user local training samples can be realized by solving the problems, and the performance of the wide learning method under the condition of small samples is improved;
based on the ADMM algorithm, the above problem is converted into:
Figure BDA0003624997700000066
Figure BDA0003624997700000067
in the formula, the first step is that,
Figure BDA0003624997700000068
is an introduced auxiliary variable matrix;
the solution to this problem requires only that each user utilize locally collected training data { X } u ,Y u Performing information interaction with adjacent users through an inter-user D2D communication protocol;
the iterative solution process is as follows:
Figure BDA0003624997700000069
Figure BDA00036249977000000610
O u (t)＝O u (t-1)+W u (t)-W 0 (t
in the formula, t represents iteration times, a parameter rho is a punishment coefficient for controlling and estimating consistency constraint, and rho>0；
Figure BDA0003624997700000071
Figure BDA0003624997700000072
Iteratively repeating the above process t max Secondly, the distributed solution of the optimization problem is realized.
Further, the step S4 includes:
when the wireless environment changes due to the movement of users or the change of the environment, the wide learning network of each user needs to be updated in a self-adaptive manner;
defining individual users
Figure BDA0003624997700000073
The newly collected training samples for adapting to the environmental changes are
Figure BDA0003624997700000074
Figure BDA0003624997700000075
Obtaining new features and enhanced nodes through feature processing of the local wide learning network
Figure BDA0003624997700000076
To avoid data characteristics obtained directly for the current accumulation
Figure BDA0003624997700000077
And output tag
Figure BDA0003624997700000078
Figure BDA0003624997700000079
Recalculating new weight width learning weight
Figure BDA00036249977000000710
The incremental update is performed as follows:
definition of
Figure BDA00036249977000000711
Woodbury identity equation D-UBV inversed by matrix -1 ＝D -1 +D -1 U(B -1 -VD -1 U) -1 VD -1 ，
Figure BDA00036249977000000712
Restated as:
Figure BDA00036249977000000713
in the formula, the content of the active carbon is shown in the specification,
Figure BDA00036249977000000714
no recalculation is required for the known symmetric matrix at the time of the current update.
Further, the step S5 includes:
in the on-line execution phase, each user
Figure BDA00036249977000000715
Wide beam response of multiple base stations only using current time slot
Figure BDA00036249977000000716
Obtaining joint feature enhanced node through nonlinear feature transformation of local wide learning network
Figure BDA00036249977000000717
Obtaining the predicted value of the analog beam index to be selected after affine transformation processing
Figure BDA00036249977000000718
Determining each base station according to the predicted value
Figure BDA00036249977000000719
Analog beam sequence number of
Figure BDA00036249977000000720
And feeds back to the base station to enable the base station to perform subsequent downlink data transmission;
the average effective rate of each user on each subcarrier is given by:
Figure BDA00036249977000000721
in the formula, 2BT d Training B wide beam responses for a base station
Figure BDA0003624997700000081
And B selected analog beams
Figure BDA0003624997700000082
The time spent, wherein T d Representing the time it takes for a single beam training.
Further, in the step S3, a penalty coefficient ρ of the algorithm is set to 0.001, and a regularization constraint coefficient λ is set to 2 -5 Number of interactive iterations t of the training process max ＝5；
In the aspect of network structure, the number of feature nodes of the wide learning network is set to 300, and the specific setting of the enhanced nodes is as follows:
when the number of local training samples is not more than 100, the number of the enhanced nodes is set to be 100, when the number of the local training samples is more than 100 and not more than 300, the number of the enhanced nodes is set to be 500, when the number of the local training samples is more than 300 and not more than 500, the number of the enhanced nodes is set to be 1000, and when the number of the local training samples is more than 500, the number of the enhanced nodes is set to be 2000;
the enhancement layer activation function is designed to Tansig in the following specific form:
Figure BDA0003624997700000083
the invention has the beneficial effects that:
compared with the prior art, the technical scheme of the invention has the following beneficial technical effects: the millimeter wave MIMO user increment cooperative beam selection method based on wide learning provided by the invention utilizes distributed increment updating design aiming at the wide learning model, can realize effective sharing of training data between adjacent users and low complexity updating of the local wide learning model of each user, can obviously reduce multi-user interaction overhead under the condition of no obvious performance loss compared with centralized training beam selection, and simultaneously meets the requirements on the number of local training samples of the users and the real-time property of model updating under a fast time-varying scene.
Drawings
Fig. 1 is a schematic flowchart of a millimeter wave MIMO user increment cooperative beam selection method based on wide learning in embodiment 1;
fig. 2 is a schematic view of a communication scenario of the multi-point cooperative millimeter wave massive MIMO downlink communication system provided in embodiment 1;
fig. 3 is a schematic structural diagram of a coordinated multiple point millimeter wave massive MIMO downlink communication system provided in embodiment 1;
FIG. 4 is a model diagram of the wide learning network provided in example 1;
fig. 5 is a schematic diagram illustrating a principle of a millimeter wave MIMO user increment cooperative beam selection method based on wide learning in embodiment 1;
fig. 6 and 7 are graphs comparing the performance of the millimeter wave MIMO user incremental cooperative beam selection method based on wide learning in embodiment 1 with that of the conventional method.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
Referring to fig. 1 to fig. 7, the present embodiment provides a millimeter wave MIMO user incremental cooperative beam selection method based on wide learning, where the method flow is shown in fig. 1, and the method includes the following steps:
s1, constructing a cooperative transmission millimeter wave large-scale MIMO system model under the scene of dynamic change of user positions, and establishing a multi-user beam selection optimization problem model taking the maximum system effectiveness and speed as targets;
specifically, in this embodiment, the step S1 specifically includes:
a scene is constructed and specific channel data is generated in this embodiment by employing an open source Deep MIMO data set. The data set channel data is obtained through simulation of ray tracing simulation software Wireless InState according to scene environment parameters, and the simulator can simulate hundreds of rays emitted from a transmitting end and generate channel parameters such as an arrival angle, a departure angle, time delay, path loss and the like for each propagation path reaching a receiving end.
The data set can simulate a scenarized real channel environment, rather than a statistical channel model commonly used in past theoretical analysis. The constructed DeepMIMO channel can capture the dependency relationship of environment, materials and the positions of a transmitter and a receiver, and is widely applied to millimeter wave and large-scale MIMO research in industry and academia. The DeepMIMO data set is based on an outdoor street scene of 18 base stations and 100 multi-user location points, and generates a DeepMIMO channel data set allowing customization according to a group of parameters which can be adjusted by researchers, such as the number of antennas, the number of OFDM subcarriers, the number of channel paths and the like. The 'O1' scene in the Deep MIMO dataset is horizontally a main street 600m long (along the Y-axis) and 40m wide (along the X-axis), with a total of 12 base stations, 6 on each side. The user position grid is positioned along the main street, and is 550m long and 35m wide. Starting from the top right, 15m from the street start, and ending at the bottom left, 35m from the street end, the grid includes 2751 lines, from R1 to R2751. Wherein, each row has 181 user positions, the distance between every two adjacent user positions is 20cm, and the receiving height of each position user antenna is constant to be 2 m.
In this embodiment, for a multi-point cooperative millimeter wave large-scale MIMO downlink system, it is assumed that there are B base stations and U users, the number of antennas of a single base station is M, the users configure omnidirectional antennas for signal reception, and the number of subcarriers of the system is K.
Multiple users receive the coordinated downlink transmission service provided by the B base stations through an Orthogonal Frequency Division Multiple Access (OFDMA) mode.
Consider again the typical case where a single base station serves one user with a single beam, and multiple users move randomly within a certain area. Recording a set of base stations as
Figure BDA0003624997700000091
Set of users as
Figure BDA0003624997700000092
Set of subcarriers as
Figure BDA0003624997700000093
The set of subcarriers allocated to user u is
Figure BDA0003624997700000094
The size is recorded as
Figure BDA0003624997700000095
Satisfy the requirement of
Figure BDA0003624997700000096
The response of a downlink antenna domain channel from a base station b to a user u on a subcarrier k is recorded as h b,u,k The modeling can be as follows:
Figure BDA0003624997700000101
in this formula, L represents the number of scattering paths, α, of the propagation channel b,u,l Denotes the complex gain of the l-th path, f k Denotes the center frequency, τ, of the k-th subcarrier l Represents the path delay of the ith path; a (-) denotes a pilot vector of the base station multi-antenna array.
If the base station uses a Uniform rectangular planar antenna array (UPA) with half-wavelength array element spacing, the number of antennas in the y and z directions in the yz plane is W and H, i.e., M is WH. The steering vector for the UPA array can be expressed as:
Figure BDA0003624997700000102
in this formula, θ b,u,l Phi and phi b,u,l Respectively, the azimuth angle and the downtilt angle of the incident array signal, an
Figure BDA0003624997700000103
Figure BDA0003624997700000104
Wherein
Figure BDA0003624997700000105
d is the spacing of adjacent array elements.
Considering that downlink transmission is based on typical mixed precoding, the analog beam sets of all base stations are
Figure BDA0003624997700000106
Such as a standard Discrete Fourier Transform (DFT) matrix, satisfying F H F＝FF H ＝I M The base station b selects the second from F
Figure BDA0003624997700000107
Column(s) of
Figure BDA0003624997700000108
As analogue beams
Figure BDA0003624997700000109
Serving user u, defining an analog beamforming matrix for a plurality of base stations
Figure BDA00036249977000001010
The cooperative center control node can collect the Channel State Information (CSI) of each base station, and design the CSI on a subcarrier according to the Maximum Ratio Transmission (MRT) criterion
Figure BDA00036249977000001011
The numeric precoding vector of the upper service user u is
Figure BDA00036249977000001012
Wherein
Figure BDA00036249977000001013
Wherein
Figure BDA00036249977000001014
Representing the response of the downlink antenna domain channel of B base stations to user u on subcarrier k.
The downlink received signal of the user u on the k-th subcarrier can be represented as:
Figure BDA00036249977000001015
in this formula, s u,k ～CN(0,P u ) Is a data symbol, n u,k ～CN(0,σ 2 ) For the receiver noise, σ, of user u on the k-th subcarrier 2 Is the noise power; when in use
Figure BDA00036249977000001016
When the system adopts equal power distribution between users and subcarriers, the method comprises the following steps
Figure BDA00036249977000001017
Where P is the total transmit power of the base station.
Accordingly, the received Signal-to-Noise-and-Interference-Ratio (SINR) of user u on subcarrier k can be expressed as:
Figure BDA0003624997700000111
the corresponding achievable rates are:
Figure BDA0003624997700000112
in the formula, B w Representing the system bandwidth.
The base station needs to acquire a certain CSI for design
Figure BDA0003624997700000113
In a massive MIMO system, a beam training mode is usually adopted, and a large amount of pilot/time overhead is consumed for complete beam training. Defining the tracking period of the system for channel time variation as T, i.e. the system needs to perform beam training again every T to update the precoding design for data transmission. First T of each cycle r And in time, the base station performs beam training, and the rest time is used for data transmission.
Thus, the actual effective rate available to user u can be expressed as:
Figure BDA0003624997700000114
the beam selection problem can be expressed as an optimization problem that maximizes the system efficiency and rate through beam selection:
Figure BDA0003624997700000115
since there is no interference between users accessing through OFDMA, the above problem can be translated into
Figure BDA0003624997700000116
Since optimization of the above problem requires selection of the optimal beam combination from the cascaded alternative simulated beam space of the individual B base stations, the search space size is M B Practical systems tend to be difficult to withstand; the conversion may be made as follows,
Figure BDA0003624997700000117
a low-complexity sub-optimal solution of the original problem is obtained.
For the problem after conversion, if the base station side can obtain the equivalent CSI information of the complete beam domain
Figure BDA0003624997700000118
It is possible to maximize
Figure BDA0003624997700000119
c b,u Representing a user
Figure BDA00036249977000001110
Equivalent rate index available under base station b. But due to the large overhead of beam training time required, i.e.
Figure BDA00036249977000001111
Smaller, may result in poor overall performance. How to ensure c b,u On the premise of no obvious loss, the beam training time T is reduced as much as possible r And becomes the key to solve the problem.
More specifically, in the present embodiment, as shown in fig. 2, a 36m × 40m area covered by rows R1166 to R1366 in a Deep MIMO main street is selected for a communication scene, sample data at positions 201 × 181 ═ 36281 are collected in total, and the distance between adjacent position points is 20 cm. The system works in a millimeter wave frequency band, the carrier frequency is 60GHz, the working bandwidth is 300MHz, the number of base stations is 3, an 8 multiplied by 4 uniform planar array is configured, the number of cooperative users is 2, and all the antennas are configured with omnidirectional single antennas. Both mobile users access the base station through the OFDMA mode and can exchange information through the D2D protocol. Consider two mobile users no more than 15m apart during the collaboration period.
Step S2, each user utilizes the collected multi-base-station wide beam responses and the beam equivalent channel intensity corresponding to the transmission narrow beam to construct a local training set; and taking the multi-base-station wide beam response as an input characteristic, taking an equivalent rate index sample beam corresponding to each transmission narrow beam as an output, and initializing the user local wide learning mapping network.
Specifically, in the present embodiment, the step S2 includes:
research has proved that the wide beam transmitted by the base station can help the user to sense the environmental characteristic information between the user and the transmitting base station, and an implicit user position representation can be obtained by using the wide beam responses of a plurality of base stations.
Since the beam selection is determined by the beam domain equivalent CSI, and the CSI information is directly related to the user position, the mapping relation between the wide beam response of the multiple base stations and the beam selection can be judged.
On the other hand, although different users are served by different subcarriers, the narrow beam response strength of the channel can be judged to be independent of the subcarrier frequency according to the channel model. Therefore, as for the problem constructed in step S1, a feasible solution is that each base station
Figure BDA0003624997700000121
Using pilots in the training phase
Figure BDA0003624997700000122
(for simplicity, this may be a pilot sequence in practice) for broad beam formation
Figure BDA0003624997700000123
And each analog beam to be selected
Figure BDA0003624997700000124
The training of (2),
Figure BDA0003624997700000125
representing the training power.
Then each user
Figure BDA0003624997700000126
On the sub-carrier
Figure BDA0003624997700000127
To obtain a pilot received signal of
Figure BDA0003624997700000128
After pilot frequency matching, the user can obtain each wave beam response
Figure BDA0003624997700000129
The estimation of (d) is:
Figure BDA00036249977000001210
each user
Figure BDA00036249977000001211
By collecting N multi-base-station wide-beam response samples
Figure BDA00036249977000001212
Corresponding analog wave beam equivalent rate index sample to be selected
Figure BDA00036249977000001213
Figure BDA00036249977000001214
A local wide learning network is trained.
The specific process is as follows:
memory users
Figure BDA00036249977000001215
The wide beam response sample N collected from base station b is 1, …, with N arranged as
Figure BDA00036249977000001216
Figure BDA0003624997700000131
The response samples of the beam to be selected are
Figure BDA0003624997700000132
All the wide beam response samples collected from the B base stations are arranged into
Figure BDA0003624997700000133
Corresponding to-be-selected beam response sample arrangement
Figure BDA0003624997700000134
After data arrangement, the input of the user u local wide learning network model corresponding to the nth sample is as follows:
Figure BDA0003624997700000135
wherein
Figure BDA0003624997700000136
The symbol represents taking a complex phase operator.
Output is as
Figure BDA0003624997700000137
Then the N samples correspond to the input and output matrices as:
Figure BDA0003624997700000138
more specifically, in the embodiment of the present invention, in order to better fit the network, the training data set is subjected to a maximum normalization process, and all inputs of the local wide learning network model of user u are divided by a constant scaling factor Δ of amplitude norm Defined as:
Figure BDA0003624997700000139
for the output of the local wide learning network model, response samples of each beam to be selected of each base station b are obtained by a user u
Figure BDA00036249977000001310
Respectively carrying out normalization processing, and operating as follows:
Figure BDA00036249977000001311
specifically, in the embodiment of the invention, each user local wide learning network has 2BK u Each input corresponds to K u Amplitude and phase of wide beam response to B base stations on the sub-carrier; each user local wide learning network has BM outputs corresponding to the equivalent rate indexes of each analog beam to be selected of B base stations. The simulation divides the 36281 location points into a training set and a testing set, the ratio is 8:2, and the sizes are 29024 and 7256 respectively, and the training set and the testing set are used by two cooperative users.
Specifically, in the embodiment of the invention, the original feature of the wide beam response signal received by the user is subjected to nonlinear transformation and affine transformation by adopting an algorithm based on a width learning framework. The algorithm defines feature nodes and enhancement nodes, calculates the feature nodes from input data, further generates the enhancement nodes from the feature nodes, and then splices the two nodes together as affine transformation. Based on the structure, the optimization of the model only needs to solve the weight of affine transformation.
More specifically, based on the wide learning framework, the specific processing procedure of the input sample in the feature transformation network is as follows:
first, input data X is utilized u Mapping to I-group feature nodes
Figure BDA0003624997700000141
And J group enhanced nodes
Figure BDA0003624997700000142
F and E respectively represent the number of the characteristic nodes and the enhanced nodes of each group, and are specifically represented as follows:
Figure BDA0003624997700000143
Figure BDA0003624997700000144
wherein Z u ＝[Z u,1 ,Z u,2 ,…,Z u,I ]Is a group I characteristicCascade matrix of symbolic nodes, H u ＝[H u,1 ,...,H u,J ]A cascaded matrix of enhanced nodes for the J set;
Figure BDA0003624997700000145
and
Figure BDA0003624997700000146
respectively representing the connection weight and the bias of the feature generation network and the feature enhancement network, and generally adopting random generation;
Figure BDA0003624997700000147
a column vector representing elements all of 1; both φ (-) and ξ (-) represent linear or nonlinear activation functions.
Then, the user local wide learning network pair joint feature enhancement node
Figure BDA0003624997700000148
Performing affine transformations
Figure BDA0003624997700000149
Output of
Figure BDA00036249977000001410
Based on the minimum mean square error criterion and utilizing L2 norm regularization to improve the generalization performance of the network, the network weight matrix W is learned widely u The optimization problem of (a) can be modeled as:
Figure BDA00036249977000001411
in the formula, | · luminance | | F Expressing a Frobenius norm, and lambda expresses a regularization factor; the optimal solution is as follows:
Figure BDA00036249977000001412
in the formula, I IF+JE Representing an identity matrix of dimension IF + JE.
Step S3, equivalently converting the multi-user beam selection problem into a distributed optimization problem with consistency constraint according to a distributed optimization theory, designing an iterative interactive method to train a local wide learning mapping network of each multi-user, and realizing effective sharing of a multi-user local training set, thereby reducing the requirement on the number of training samples;
specifically, in this implementation, a mechanism for a user to perform cooperative training according to a downlink wide beam response signal is designed based on an ADMM algorithm to reduce the computational burden of each user node, and a converted distributed optimization problem is regarded as a problem that the user shares information through cooperation, which specifically includes the following steps:
if the propagation environments between the U users and the base station have similarity in space, the models of the local wide learning should tend to be consistent, that is, the models of the local wide learning should tend to be consistent
Figure BDA00036249977000001413
That is, the network training optimization problem of multiple users can be jointly expressed as
Figure BDA00036249977000001414
In the formula, the content of the active carbon is shown in the specification,
Figure BDA00036249977000001415
by solving the problems, shared training based on all user local training samples can be realized, and the performance of the wide learning method under the condition of small samples is improved.
However, the problem needs to be solved based on the wide beam characteristic transformation a of all users and the corresponding analog beam response Y, and the corresponding actual system for information interaction between users is difficult to bear. The following converts the above problem into, based on the ADMM algorithm:
Figure BDA0003624997700000151
Figure BDA0003624997700000152
in the formula, the first step is that,
Figure BDA0003624997700000153
is the introduced auxiliary variable matrix.
The solution to this problem requires only that each user utilize locally collected training data { X } u ,Y u And a small amount of information interaction with neighboring users is performed through an inter-user D2D communication protocol.
The iterative solution process is as follows:
Figure BDA0003624997700000154
Figure BDA0003624997700000155
O u (t)＝O u (t-1)+W u (t)-W 0 (t)
in the formula, t represents iteration times, a parameter rho is a punishment coefficient for controlling and estimating consistency constraint, and rho>0。
Figure BDA0003624997700000156
Figure BDA0003624997700000157
Iteratively repeating the above process t max Second, distributed solving of the above optimization problem can be achieved.
More specifically, in the embodiment of the present invention, as shown in fig. 4, an ADMM algorithm based on a wide learning framework is adopted to perform weight update in user cooperation. As shown in FIG. 5, the user cooperation interaction process continuously loops until the set interaction iteration number t is reached max . In the embodiment of the invention, the penalty coefficient rho of the algorithm is set to be 0.001, and the regularization constraint coefficient lambda is set to be 2 -5 Number of interactive iterations t of the training process max 5. In the aspect of a network structure, the number of characteristic nodes of the wide learning network is set to 300, the number of enhanced nodes is flexibly set according to the change of the number of local training samples, the taken parameters are all configurations capable of obtaining better performance, and the specific settings are as follows: when the number of local training samples is not more than 100, the number of the enhanced nodes is set to 100, when the number of the local training samples is more than 100 and not more than 300, the number of the enhanced nodes is set to 500, when the number of the local training samples is more than 300 and not more than 500, the number of the enhanced nodes is set to 1000, and when the number of the local training samples is more than 500, the number of the enhanced nodes is set to 2000. Regardless of the nonlinear activation function of the feature layer, the enhancement layer activation function is designed as Tansig in the following specific form:
Figure BDA0003624997700000158
and step S4, introducing a cooperative incremental learning method, wherein each user can quickly update the local learning network by using a small amount of collected new samples when the environment changes, and effective tracking of the multi-user beam selection on the fast time-varying environment is realized.
Specifically, in this embodiment, when the wireless environment changes due to the movement of the user or the environment itself, the wide learning network of each user needs to be updated adaptively. Defining individual users
Figure BDA0003624997700000161
The newly collected training samples for adapting to the environmental changes are
Figure BDA0003624997700000162
Figure BDA0003624997700000163
Obtaining new features and enhanced nodes through feature processing of the local wide learning network
Figure BDA0003624997700000164
Figure BDA0003624997700000165
To avoid data characteristics obtained directly for the current accumulation
Figure BDA0003624997700000166
And output tag
Figure BDA0003624997700000167
Recalculating new weight width learning weight
Figure BDA0003624997700000168
(the main calculated amount is
Figure BDA0003624997700000169
The inversion operation of (d) in the following manner.
Definition of
Figure BDA00036249977000001610
Woodbury identity equation D-UBV inversed by matrix -1 ＝D -1 +D -1 U(B -1 -VD -1 U) -1 VD -1 In a clear view of the above, it is known that,
Figure BDA00036249977000001611
restated as:
Figure BDA00036249977000001612
in the formula, the first step is that,
Figure BDA00036249977000001613
no recalculation is required for the known symmetric matrix at the time of the current update.
Figure BDA00036249977000001614
The calculation only involving
Figure BDA00036249977000001615
Inversion of the dimensional symmetric matrix if
Figure BDA00036249977000001616
The computational overhead of the incremental update method described above is much less than the direct inversion method.
In the embodiment of the present invention, it should be noted that, because real and effective samples in an actual communication scenario are difficult to obtain, incremental learning mainly considers the condition that the number of samples collected by a user is limited, and makes full use of the advantage of an incremental user cooperation network on small data volume. The number of local sample data sets used for training is sequentially increased from 50 to 1000.
Step S5: each user inputs the multi-base-station wide beam response collected in real time into a local mapping network to predict the self beam selection.
As shown in FIG. 3, the phases are executed online, with each user executing
Figure BDA00036249977000001617
Wide beam response of multiple base stations only using current time slot
Figure BDA00036249977000001618
Figure BDA00036249977000001619
Obtaining joint feature enhanced node through nonlinear feature transformation of local wide learning network
Figure BDA00036249977000001620
Obtaining the predicted value of the analog beam index to be selected after affine transformation processing
Figure BDA00036249977000001621
Based on the predicted values, each base station can be determined
Figure BDA00036249977000001622
Analog beam sequence number of
Figure BDA00036249977000001623
And feeds back to the base station to enable the base station to perform subsequent downlink data transmission.
The average effective rate of each user on each subcarrier is given by:
Figure BDA00036249977000001624
in the formula, 2BT d Training B wide beam responses for a base station
Figure BDA0003624997700000171
And B selected analog beams
Figure BDA0003624997700000172
The time spent, wherein T d Representing the time it takes for a single beam training. In the embodiment of the invention, T is set to be 1ms, and T d Set to 0.01 ms.
The embodiment of the invention utilizes fig. 6 to compare the user effective rate curves of the designed user incremental cooperative beam selection method based on the wide learning with the user side centralized beam selection method based on the wide learning, the user side fully distributed beam selection method based on the wide learning, the user side centralized beam selection method based on the deep learning and the traditional exhaustive beam scanning search method. The Genie-aid curve in the figure represents an ideal situation, and the base station side does not need any training overhead, namely T r The optimal beam selection can be made in the case of 0, representing an upper rate performance limit. The Basline curve is set to the average effective rate obtained by using the traditional exhaustive beam scanning search method for beam training. In the designed user increment cooperative beam selection method based on wide learning, t is set max 5. As can be seen from fig. 6, the wide learning-based user-side fully distributed beam selection method has poor performance, but is superior to the deep learning-based user-side centralized beam selection method, and particularly in the case of a small number of samples, the advantage is more obvious. This is because the wide learning network in this case is easier to get a reasonable tradeoff of training sample number versus model complexity. Compared with the traditional exhaustive beam scanning search method, when the total training sample number is less than 100, the wide learning-based user-side fully distributed waveThe beam selection method can also achieve performance advantages with an average effective rate improvement of about 7.5%. The above results indicate that the wide learning network is suitable for the beam selection problem concerned by this patent as a typical lightweight learning network. Further, when the number of local training samples is increased to 400, the designed wide learning-based user incremental cooperative beam selection method has the beneficial average effective rate of 6.529bps/Hz, which is improved by about 6.9% compared with the wide learning-based user-side fully-distributed beam selection method and is improved by about 29% compared with the traditional exhaustive beam scanning search method, and has obvious performance improvement; and with the increase of the number of training samples, the performance approaches to the user side centralized beam selection method based on the wide learning. Further, the incremental updating mechanism designed by the invention achieves almost the same performance as the method based on wide learning direct retraining, and the effectiveness is proved.
Fig. 7 shows a variation relationship between the beam selection success probability of the user incremental cooperative beam selection method based on the wide learning and the number of user cooperative iterations, and compared with the centralized method based on the wide learning and the fully distributed method based on the wide learning, it can be seen that when the number of samples of the local input data set is 500, the beam selection success rate of the user incremental cooperative beam selection method based on the wide learning increases with the increase of the number of interactive iterations among the cooperative users, and compared with the fully distributed method, the method has an obvious advantage that the beam selection accuracy is improved by about 6%, and when the number of iterations is 5, an effect close to the centralized method has been obtained.
The invention is not described in detail, but is well known to those skilled in the art.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.

Claims (10)

1. A millimeter wave MIMO user increment cooperative beam selection method based on wide learning is characterized by comprising the following steps:
step S1, constructing a downlink received signal receiving expression and a corresponding reachable rate expression of the downlink received signal receiving expression on the user side of the multi-point cooperative millimeter wave large-scale MIMO system; then, constructing an actual effective rate expression of the user side according to the reachable rate expression and the tracking period of the channel time variation of the system; constructing a global optimization problem model which takes the maximization of the system effectiveness and the speed as the target based on the actual effective speed expression;
step S2, constructing and initializing a local wide learning mapping network on each user side based on the wide learning architecture; taking multi-base station wide beam responses collected by each user and beam equivalent channel strength corresponding to transmission narrow beams as a local training set, wherein the multi-base station wide beam responses are taken as input features, equivalent rate index sample beams corresponding to the transmission narrow beams are taken as output, affine transformation is carried out on the input features through setting feature nodes and enhancement nodes based on a width learning framework, and the optimization problem of a network model is modeled as a weight problem for solving the affine transformation;
step S3, equivalently converting the multi-user beam selection problem into a distributed optimization problem with consistency constraint according to a distributed optimization theory, designing an iterative interactive method to train a local wide learning mapping network of each multi-user, and realizing effective sharing of a multi-user local training set, thereby reducing the requirement on the number of training samples;
step S4, introducing a collaborative incremental learning method for rapidly updating collaborative model parameters when data are newly added, and realizing a network model which is updated while being collected; performing iterative training on the data set generated in the step S2 by using the network established in the step S3;
and step S5, each user inputs the multi-base-station wide beam response collected in real time into the local mapping network to predict the beam selection of the user.
2. The wide learning based millimeter wave MIMO user incremental cooperative beam selection method according to claim 1, wherein in the multi-point cooperative millimeter wave massive MIMO system, it comprises:
b base stations and U users, wherein the number of antennas of a single base station is M, the users configure omnidirectional antennas for signal reception, and the number of system subcarriers is K;
the user receives the cooperation downlink transmission service provided by the B base stations through an orthogonal frequency division multiple access mode;
considering the situation that a single base station uses a single beam to serve one user, and multiple users move randomly in a certain area, the base stations are set as
Figure FDA0003624997690000011
Set of users as
Figure FDA0003624997690000012
Set of subcarriers as
Figure FDA0003624997690000013
The set of subcarriers allocated to user u is
Figure FDA0003624997690000014
The size is recorded as
Figure FDA0003624997690000015
Satisfy the requirement of
Figure FDA0003624997690000016
The response of a downlink antenna domain channel from a base station b to a user u on a subcarrier k is recorded as h b,u,k Then, the modeling is as follows:
Figure FDA0003624997690000021
in this formula, L represents the scattering of the propagation channelNumber of paths, α b,u,l Denotes the complex gain of the l-th path, f k Denotes the center frequency, τ, of the k-th subcarrier l Represents the path delay of the ith path; α (-) denotes a steering vector of the base station multi-antenna array.
3. The method for selecting the millimeter wave MIMO user increment cooperative beam based on the wide learning as claimed in claim 2, wherein for the multi-point cooperative millimeter wave massive MIMO system, the downlink received signal receiving expression and the reachable rate expression corresponding thereto at the user side are obtained by:
considering that downlink transmission is based on typical mixed precoding, the analog beam sets of all base stations are
Figure FDA0003624997690000022
And satisfy F H F＝FF H ＝I M The base station b selects the second from F
Figure FDA0003624997690000023
Column(s) of
Figure FDA0003624997690000024
As analogue beams
Figure FDA0003624997690000025
Serving user u, defining an analog beamforming matrix for a plurality of base stations
Figure FDA0003624997690000026
The cooperative center control node collects the beam equivalent channel state information of each base station, and designs the information on the sub-carriers according to the maximum ratio transmission criterion
Figure FDA0003624997690000027
The numeric precoding vector of the upper service user u is:
Figure FDA0003624997690000028
in the formula, the first and second sets of data are represented,
Figure FDA0003624997690000029
wherein
Figure FDA00036249976900000210
Representing the response of the downlink antenna domain channel from the B base stations to the user u on the subcarrier k;
then the downlink received signal of the user u on the kth subcarrier is represented as:
Figure FDA00036249976900000211
in this formula, s u,k ～CN(0,P u ) Is a data symbol, n u,k ～CN(0,σ 2 ) For the receiver noise, σ, of user u on the k-th subcarrier 2 Is the noise power; when in use
Figure FDA00036249976900000212
When the system adopts equal power distribution between users and subcarriers, the method comprises the following steps
Figure FDA00036249976900000213
Wherein P is the total transmitting power of the base station;
correspondingly, the received signal-to-interference-and-noise ratio of user u on subcarrier k is expressed as:
Figure FDA00036249976900000214
the corresponding achievable rates are:
Figure FDA00036249976900000215
in the formula, B w Representing the system bandwidth.
4. The wide learning-based millimeter wave MIMO user incremental cooperative beam selection method according to claim 3, wherein for the multi-point cooperative millimeter wave massive MIMO system, a global optimization problem model with the goal of maximizing system efficiency and rate is obtained by the following method, comprising:
defining the tracking period of the system for channel time variation as T, namely, the system needs to perform beam training again every T to update the precoding design for data transmission; first T of each cycle r Time, the base station carries out beam training, and the rest time carries out data transmission;
thus, the actual effective rate achieved by user u is represented as:
Figure FDA0003624997690000031
the beam selection problem is expressed as an optimization problem that maximizes the system efficiency and rate through beam selection:
Figure FDA0003624997690000032
since there is no interference between users accessing through OFDMA, the above problem translates into:
Figure FDA0003624997690000033
since optimization of the above problem requires selection of the optimal beam combination from the cascaded alternative simulated beam space of the individual B base stations, the search space size is M B The conversion is thus made as follows:
Figure FDA0003624997690000034
a low-complexity sub-optimal solution of the original problem is obtained.
5. The method for selecting millimeter wave MIMO user increment cooperative beam based on wide learning according to claim 4, wherein the step S2 specifically comprises:
to solve the problem constructed in step S1, each base station
Figure FDA0003624997690000035
Using pilots in the training phase
Figure FDA0003624997690000036
For making wide beam
Figure FDA0003624997690000037
And each analog beam to be selected
Figure FDA0003624997690000038
The training of (2),
Figure FDA0003624997690000039
represents the training power;
then each user
Figure FDA00036249976900000310
On the sub-carrier
Figure FDA00036249976900000311
The above obtained pilot received signal is:
Figure FDA00036249976900000312
after pilot frequency matching, the user obtains each wave beam response
Figure FDA00036249976900000313
The estimation of (d) is:
Figure FDA00036249976900000314
each user
Figure FDA00036249976900000315
By collecting N multiple base station wide beam response samples
Figure FDA00036249976900000316
Corresponding analog wave beam equivalent rate index sample to be selected
Figure FDA00036249976900000317
Figure FDA00036249976900000318
Training a local wide learning network;
the specific process is as follows:
memory users
Figure FDA0003624997690000041
The wide beam response sample N collected from base station b is 1, …, with N arranged as
Figure FDA0003624997690000042
Figure FDA0003624997690000043
The response samples of the beam to be selected are
Figure FDA0003624997690000044
All the wide beam response samples collected from the B base stations are arranged into
Figure FDA0003624997690000045
Corresponding to-be-selected beam response sample arrangement
Figure FDA0003624997690000046
After data arrangement, the input of the user u local wide learning network model corresponding to the nth sample is as follows:
Figure FDA0003624997690000047
wherein
Figure FDA0003624997690000048
The symbol represents taking a complex phase operator;
output is as
Figure FDA0003624997690000049
Then the N samples correspond to the input and output matrices as:
Figure FDA00036249976900000410
6. the millimeter wave MIMO user increment cooperative beam selection method based on wide learning according to claim 5, wherein in the step S2, an algorithm based on a wide learning framework is adopted to perform nonlinear transformation and affine transformation on an original feature of a wide beam response signal received by a user; the algorithm defines characteristic nodes and enhancement nodes, calculates the characteristic nodes from input data, further generates the enhancement nodes from the characteristic nodes, and then splices the two nodes together as affine transformation; based on the structure, the optimization of the model only needs to solve the weight of affine transformation;
based on a wide learning framework, the specific processing process of the input sample in the feature transformation network is as follows:
first, using input data X u Mapping to I-group feature nodes
Figure FDA00036249976900000411
And J group enhanced nodes
Figure FDA00036249976900000412
F and E respectively represent the number of the characteristic nodes and the enhanced nodes of each group, and are specifically represented as follows:
Figure FDA00036249976900000413
Figure FDA00036249976900000414
wherein Z u ＝[Z u,1 ,Z u,2 ,…,Z u,I ]A cascade matrix of characteristic nodes of group I, H u ＝[H u,1 ,...,H u,J ]A cascaded matrix of enhanced nodes for the J set;
Figure FDA00036249976900000415
and
Figure FDA00036249976900000416
respectively representing the connection weight and the bias of the feature generation network and the feature enhancement network, and generally adopting random generation;
Figure FDA0003624997690000051
a column vector representing elements all 1; phi (-) and xi (-) both represent linear or non-linear activation functions;
then, the user local wide learning network pair joint feature enhancement node
Figure FDA0003624997690000052
Performing affine transformations
Figure FDA0003624997690000053
Output of
Figure FDA0003624997690000054
Based on the minimum mean square error criterion, the network generalization performance is improved by utilizing L2 norm regularization, and the network weight matrix W is learned widely u The optimization problem of (a) is modeled as:
Figure FDA0003624997690000055
in the formula, | · luminance | | F Expressing a Frobenius norm, and lambda expresses a regularization factor;
the optimal solution is as follows:
Figure FDA0003624997690000056
in the formula, I IF+JE Representing an identity matrix of dimension IF + JE.
7. The millimeter wave MIMO user incremental cooperative beam selection method based on wide learning according to claim 6, wherein the step S3 comprises:
designing a mechanism for a user to perform cooperative training according to a downlink wide beam response signal based on an ADMM algorithm so as to reduce the calculation burden of each user node, and regarding the converted distributed optimization problem as a problem that the user shares information through cooperation, specifically comprising the following steps:
if the propagation environments between U users and the base station have similarity in space, the models of local wide learning should tend to be consistent, i.e. W u →W,
Figure FDA0003624997690000057
I.e. network training for multiple usersThe optimization problem is jointly expressed as:
Figure FDA0003624997690000058
in the formula, the first step is that,
Figure FDA0003624997690000059
the shared training based on all user local training samples can be realized by solving the problems, and the performance of the wide learning method under the condition of small samples is improved;
based on the ADMM algorithm, the above problem is converted into:
Figure FDA00036249976900000510
Figure FDA00036249976900000511
in the formula, the first step is that,
Figure FDA00036249976900000512
is an introduced auxiliary variable matrix;
the solution to this problem requires only that each user utilize locally collected training data { X } u ,Y u Performing information interaction with adjacent users through an inter-user D2D communication protocol;
the iterative solution process is as follows:
Figure FDA0003624997690000061
Figure FDA0003624997690000062
O u (t)＝O u (t-1)+W u (t)-W 0 (t)
in the formula, t represents iteration times, a parameter rho is a punishment coefficient for controlling and estimating consistency constraint, and rho>0；
Figure FDA0003624997690000063
Figure FDA0003624997690000064
Iteratively repeating the above process t max Secondly, the distributed solution of the optimization problem is realized.
8. The millimeter wave MIMO user increment cooperative beam selection method based on wide learning of claim 7, wherein the step S4 includes:
when the wireless environment changes due to the movement of users or the change of the environment, the wide learning network of each user needs to be updated in a self-adaptive manner;
defining individual users
Figure FDA0003624997690000065
The newly collected training samples for adapting to the environmental changes are
Figure FDA0003624997690000066
Figure FDA0003624997690000067
Obtaining new features and enhanced nodes through feature processing of the local wide learning network
Figure FDA0003624997690000068
To avoid data characteristics obtained directly for the current accumulation
Figure FDA0003624997690000069
And output tag
Figure FDA00036249976900000610
Figure FDA00036249976900000611
Recalculating new weight width learning weight
Figure FDA00036249976900000612
The incremental update is performed as follows:
definition of
Figure FDA00036249976900000613
Woodbury identity equation D-UBV inversed by matrix -1 ＝D -1 +D -1 U(B -1 -VD -1 U) -1 VD -1 ，
Figure FDA00036249976900000614
Restated as:
Figure FDA00036249976900000615
in the formula, the first step is that,
Figure FDA00036249976900000616
no recalculation is required for the known symmetric matrix at the time of the current update.
9. The millimeter wave MIMO user increment cooperative beam selection method based on wide learning of claim 8, wherein the step S5 includes:
in the on-line execution phase, each user
Figure FDA00036249976900000617
Wide beam response of multiple base stations only using current time slot
Figure FDA00036249976900000618
Go through the bookJoint feature enhancement node obtained by nonlinear feature transformation of terrain learning network
Figure FDA0003624997690000071
Obtaining the predicted value of the analog beam index to be selected after affine transformation
Figure FDA0003624997690000072
Determining each base station according to the predicted value
Figure FDA0003624997690000073
Analog beam sequence number of
Figure FDA0003624997690000074
And feeds back to the base station to enable the base station to perform subsequent downlink data transmission;
the average effective rate of each user on each subcarrier is given by:
Figure FDA0003624997690000075
in the formula, 2BT d Training B wide beam responses for a base station
Figure FDA0003624997690000076
And B selected analog beams i b,u ,
Figure FDA0003624997690000077
Time spent wherein T d Representing the time it takes for a single beam training.
10. The wide-learning-based millimeter wave MIMO user incremental cooperative beam selection method according to claim 9, wherein in the step S3, a penalty coefficient p of the algorithm is set to 0.001, and a regularization constraint coefficient λ is set to 2 -5 Number of interactive iterations t of the training process max ＝5；
In the aspect of network structure, the number of feature nodes of the wide learning network is set to 300, and the specific setting of the enhanced nodes is as follows:
when the number of local training samples is not more than 100, the number of the enhanced nodes is set to be 100, when the number of the local training samples is more than 100 and not more than 300, the number of the enhanced nodes is set to be 500, when the number of the local training samples is more than 300 and not more than 500, the number of the enhanced nodes is set to be 1000, and when the number of the local training samples is more than 500, the number of the enhanced nodes is set to be 2000;
the enhancement layer activation function is designed to Tansig in the following specific form:
Figure FDA0003624997690000078
CN202210474863.7A 2022-04-29 2022-04-29 Millimeter wave MIMO user increment cooperative beam selection method based on wide learning Active CN114844538B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210474863.7A CN114844538B (en) 2022-04-29 2022-04-29 Millimeter wave MIMO user increment cooperative beam selection method based on wide learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210474863.7A CN114844538B (en) 2022-04-29 2022-04-29 Millimeter wave MIMO user increment cooperative beam selection method based on wide learning

Publications (2)

Publication Number Publication Date
CN114844538A true CN114844538A (en) 2022-08-02
CN114844538B CN114844538B (en) 2023-05-05

Family

ID=82568761

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210474863.7A Active CN114844538B (en) 2022-04-29 2022-04-29 Millimeter wave MIMO user increment cooperative beam selection method based on wide learning

Country Status (1)

Country Link
CN (1) CN114844538B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117377062A (en) * 2023-11-08 2024-01-09 中科晶锐(苏州)科技有限公司 Machine learning sample enhancement method for positioning by using channel state information

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017000752A1 (en) * 2015-07-01 2017-01-05 东南大学 Downlink training sequence design method of fdd large-scale mimo system
CN110971279A (en) * 2019-12-30 2020-04-07 东南大学 Intelligent beam training method and precoding system in millimeter wave communication system
CN111917447A (en) * 2020-08-12 2020-11-10 电子科技大学 Low-frequency auxiliary hybrid precoding design method based on beam selection
WO2021103325A1 (en) * 2019-11-28 2021-06-03 东南大学 Millimeter wave/terahertz network large-scale mimo radio transmission method
CN113300746A (en) * 2021-05-24 2021-08-24 内蒙古大学 Millimeter wave MIMO antenna and hybrid beam forming optimization method and system
CN113411110A (en) * 2021-06-04 2021-09-17 东南大学 Millimeter wave communication beam training method based on deep reinforcement learning
CN113965233A (en) * 2021-10-19 2022-01-21 东南大学 Multi-user broadband millimeter wave communication resource allocation method and system based on deep learning

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017000752A1 (en) * 2015-07-01 2017-01-05 东南大学 Downlink training sequence design method of fdd large-scale mimo system
WO2021103325A1 (en) * 2019-11-28 2021-06-03 东南大学 Millimeter wave/terahertz network large-scale mimo radio transmission method
CN110971279A (en) * 2019-12-30 2020-04-07 东南大学 Intelligent beam training method and precoding system in millimeter wave communication system
CN111917447A (en) * 2020-08-12 2020-11-10 电子科技大学 Low-frequency auxiliary hybrid precoding design method based on beam selection
CN113300746A (en) * 2021-05-24 2021-08-24 内蒙古大学 Millimeter wave MIMO antenna and hybrid beam forming optimization method and system
CN113411110A (en) * 2021-06-04 2021-09-17 东南大学 Millimeter wave communication beam training method based on deep reinforcement learning
CN113965233A (en) * 2021-10-19 2022-01-21 东南大学 Multi-user broadband millimeter wave communication resource allocation method and system based on deep learning

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117377062A (en) * 2023-11-08 2024-01-09 中科晶锐(苏州)科技有限公司 Machine learning sample enhancement method for positioning by using channel state information
CN117377062B (en) * 2023-11-08 2024-04-09 中科晶锐(苏州)科技有限公司 Machine learning sample enhancement method for positioning by using channel state information

Also Published As

Publication number Publication date
CN114844538B (en) 2023-05-05

Similar Documents

Publication Publication Date Title
CN111010219B (en) Reconfigurable intelligent surface-assisted multi-user MIMO uplink transmission method
Qi et al. Integrating sensing, computing, and communication in 6G wireless networks: Design and optimization
CN113411110B (en) Millimeter wave communication beam training method based on deep reinforcement learning
Zhang et al. IRS empowered UAV wireless communication with resource allocation, reflecting design and trajectory optimization
CN110289895B (en) Large-scale MIMO downlink power distribution method based on energy efficiency and spectrum efficiency joint optimization
Liu et al. Intelligent reflecting surface aided NOMA for millimeter-wave massive MIMO with lens antenna array
CN109104225A (en) A kind of optimal extensive MIMO Beam Domain multicast transmission method of efficiency
CN114726687B (en) Channel estimation method of intelligent reflection surface auxiliary millimeter wave large-scale MIMO system
CN114268350A (en) Robust beam forming design method under non-ideal channel in intelligent reflector auxiliary communication
Sun et al. A robust secure hybrid analog and digital receive beamforming scheme for efficient interference reduction
KR20220013906A (en) Deep learning based beamforming method and apparatus for the same
CN114844538B (en) Millimeter wave MIMO user increment cooperative beam selection method based on wide learning
CN114640379A (en) Beam optimization method and system based on intelligent reflecting area array element grouping
Chafaa et al. Federated channel-beam mapping: from sub-6ghz to mmwave
Lavdas et al. A machine learning adaptive beamforming framework for 5G millimeter wave massive MIMO multicellular networks
Mahmood et al. Deep Learning Meets Swarm Intelligence for UAV-Assisted IoT Coverage in Massive MIMO
CN115664471B (en) Millimeter wave MIMO base station cooperative beam selection method based on wide learning
Amiriara et al. IRS-user association in IRS-aided MISO wireless networks: Convex optimization and machine learning approaches
Xu et al. Intelligent reflecting surface enabled integrated sensing, communication and computation
Zhou et al. Joint trajectory and hybrid beamforming design for multi antenna UAV enabled network
CN113595609B (en) Collaborative signal transmission method of cellular mobile communication system based on reinforcement learning
Huang et al. Self-attention reinforcement learning for multi-beam combining in mmWave 3D-MIMO systems
CN114745032A (en) Non-cellular large-scale MIMO intelligent distributed beam selection method
CN109347528B (en) 3D-MIMO downlink multi-user scheduling and self-adaptive transmission method
Wang et al. Deep Learning-Based Radio Map for MIMO-OFDM Downlink Precoding

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