CN114710187B - Power distribution method for multi-cell large-scale MIMO intelligent communication under user number dynamic transformation scene - Google Patents
Power distribution method for multi-cell large-scale MIMO intelligent communication under user number dynamic transformation scene Download PDFInfo
- Publication number
- CN114710187B CN114710187B CN202210491259.5A CN202210491259A CN114710187B CN 114710187 B CN114710187 B CN 114710187B CN 202210491259 A CN202210491259 A CN 202210491259A CN 114710187 B CN114710187 B CN 114710187B
- Authority
- CN
- China
- Prior art keywords
- user
- users
- cell
- power
- channel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000009826 distribution Methods 0.000 title claims abstract description 32
- 238000004891 communication Methods 0.000 title claims abstract description 25
- 230000009466 transformation Effects 0.000 title claims abstract description 12
- 239000013598 vector Substances 0.000 claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 25
- 230000008859 change Effects 0.000 claims abstract description 5
- 238000004088 simulation Methods 0.000 claims abstract description 4
- 230000003595 spectral effect Effects 0.000 claims abstract description 4
- 238000005457 optimization Methods 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 16
- 230000005540 biological transmission Effects 0.000 claims description 10
- 238000005562 fading Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000013135 deep learning Methods 0.000 claims description 6
- 238000013439 planning Methods 0.000 claims description 6
- 230000007613 environmental effect Effects 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000012856 packing Methods 0.000 claims description 4
- 239000000654 additive Substances 0.000 claims description 3
- 230000000996 additive effect Effects 0.000 claims description 3
- 230000001427 coherent effect Effects 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims description 3
- 230000007812 deficiency Effects 0.000 claims description 2
- 230000008054 signal transmission Effects 0.000 claims description 2
- 230000001360 synchronised effect Effects 0.000 claims description 2
- 238000004422 calculation algorithm Methods 0.000 abstract description 8
- 238000013459 approach Methods 0.000 abstract description 2
- 238000013528 artificial neural network Methods 0.000 abstract 2
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000013468 resource allocation Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 238000012795 verification 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/391—Modelling the propagation channel
-
- 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
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Electromagnetism (AREA)
- Power Engineering (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a power distribution method for multi-cell large-scale MIMO intelligent communication under a user number dynamic transformation scene, which comprises the following steps: constructing a large-scale MIMO network communication environment model under a scene of dynamic change of the number of users, generating user data with different numbers and different positions in a simulation environment, calculating channel state information of the users, obtaining self characteristics of each user and surrounding environment characteristics of each user, combining the self characteristics with the surrounding environment characteristics as characteristic vectors, and providing the characteristic vectors as input to a transducer network; and obtaining optimal power distribution by using a traditional geometric programming method as an output label of a transducer network, and training and optimizing parameters of the neural network to enable the neural network to be converged to a stable state. The user spectral efficiency based on different power allocation algorithms is calculated for evaluating its performance. The invention can approach complex traditional algorithms with less computing resources and cost, and at the same time, different networks do not need to be trained for different numbers of users.
Description
Technical Field
The invention relates to a power distribution method under a multi-cell large-scale MIMO scene based on a transducer model, and belongs to the fields of wireless communication and deep learning.
Background
Wireless communication technology has become a widely used communication scheme in connection with our production and life. The number of wireless voice and data communications has grown at an exponential rate for decades. An important issue with this is how to develop existing wireless communication technologies to meet the ever-increasing demands and ever-increasing quality of service expectations. With the formal commercial use of 5G, the 5G era has come, and a large-scale Multiple Input Multiple Output (MIMO) technology is a valuable solution, where a large number of large-scale antenna arrays are configured at a base station end to serve multiple users simultaneously, so that the system can obtain better space diversity and space multiplexing effects, can avoid interference more effectively, and becomes one of key technologies for solving the communication data service requirements. A cellular network consists of a set of Base Stations (BSs) and a set of User Equipments (UEs). Each UE is connected to one BS, which serves it. The Downlink (DL) is a signal transmitted from a BS to a respective terminal, and the Uplink (UL) is a transmission transmitted from a terminal to a respective BS.
In massive MIMO networks, resource allocation is important for handling inter-user interference. Many of the resource allocation problems in Massive MIMO are easier to solve than in conventional systems, since channel hardening makes the utility function dependent only on large scale fading coefficients that are stable over time. In resource management, firstly, considering the power distribution of a base station, the base station distributes different powers to all users in a service cell so as to achieve the purpose of reducing interference and increasing network throughput, the optimization problem is usually non-convex and difficult to directly solve, and the traditional solution is mostly based on iterative algorithm successive approximation optimal solution, and has high complexity. On the other hand, based on the rapid development of artificial intelligence, power allocation can be solved by training a deep neural network model, and the problem is solved by using the strong predictive power of the neural network model. Some methods utilize traditional iterative methods to obtain optimal labels, use deep learning models to perform supervised training, and map user position or channel information with optimal power allocation. There are also scholars that use reinforcement learning strategy selection capabilities to transform the power distribution problem into a markov problem for solution. The algorithm based on deep learning avoids iteration and time cost required by the traditional optimal algorithm, the trained model is based on linear operation, the complexity is low, but the model is mostly only suitable for the situation that the number of users is determined, different sample sets are required to be used for training to obtain neural network models with different parameters and different structures aiming at different numbers of users, and the number of users often changes in actual situations.
The invention introduces a current popular transform model, modifies the model according to the problem of power distribution, and the model obtained after certain data training can simultaneously process the problem of power distribution when different users are in a certain scene, without reconstructing and training the model according to different numbers of users, and the final effect can approximate to a complex traditional iterative optimization algorithm.
Disclosure of Invention
The invention aims to: the invention provides a power distribution method for multi-cell large-scale MIMO intelligent communication under a user dynamic transformation scene, which aims to solve the problems that the complexity of solving the power distribution problem is high by the traditional method and the deep learning method always needs to train different network models according to different numbers of users.
The invention adopts the technical scheme that:
a power distribution method for multi-cell large-scale MIMO intelligent communication under a user number dynamic transformation scene comprises the following steps:
(1) And constructing a multi-cell large-scale MIMO network communication model under a scene of dynamic change of the number of users, calculating to obtain channel state information of the users, and constructing own characteristics of each user and surrounding environment characteristics of each user as characteristic vectors of each user.
(2) Setting different user numbers and user positions, establishing an optimization objective function, obtaining optimal power distribution by using a traditional geometric planning method as a label, and calculating the user spectrum efficiency under the power distribution.
(3) And constructing a transducer network model aiming at the scene.
(4) And constructing a data set required by deep learning for training the model, wherein the input is the self characteristics of the user and the surrounding environment characteristics of the user, and the output is the power distributed by the user by the cell base station of the user, and training is carried out until the model is converged.
Preferably, the specific steps of step (1) comprise:
(1.1) setting parameter information of a multi-cell multi-user scene to be simulated, wherein considering Downlink (DL) transmission of a massive MIMO network with L cells, each cell comprises a BS with M antennas and K UEs, setting uplink transmission power P ul of a user, maximum signal transmission power P max of a base station end and bandwidth W; determining a channel modeling mode; establishing a two-dimensional coordinate system of an area to be simulated, determining the position of each BS, and generating position information of K UE (user equipment) at random in each cell, wherein the size and the position (x, y) of the number K are not fixed;
(1.2) channel modeling assuming that it obeys Rayleigh distribution, using Representing the channel between BS j to UE k in cell l.
Wherein,Is the spatial correlation matrix at the base station end.
Obtaining a precoding vector by using pilot signals and MMSE channel estimationSatisfy ||w li||2 = 1.
Preferably, the specific steps of step (1.2) comprise:
(1.2.1) the distance of cell l from BS j to UE k can be calculated after the position determination Calculating the large-scale fading coefficient of the signals transmitted by the AP i to the position of the UE j by utilizing a wireless signal free space path loss formula, calculating the large-scale fading coefficient matrix of each pair of BS to the UE, wherein the large-scale fading coefficient of one antenna from the BS j to the UE k in the cell l is/>
Y determines the intermediate channel gain at a reference distance of 1 km. In theoretical studies, the parameters y and α can be calculated from one of a number of established propagation models;
(1.2.2) estimating a channel vector using pilot-based channel training. It is assumed that the BS and UEs are fully synchronized and operate according to a Time Division Duplex (TDD) protocol, where the DL data transmission phase is preceded by a training phase of channel estimation. The pilot reuse factor is 1 and the UEs in each cell use the same pilot. BS j obtains the total UL pilot power for each UE and standard MMSE estimation technique Is estimated as:
Wherein the noise P tr denotes the total power of the uplink pilot.
(1.2.3) Design of precoding UL and DL channels are interacted within one coherent block in consideration of UL-DL duality, which allows the BS to also calculate/select precoding vectors using UL channel estimation. A common heuristic is chosen, using M-MMSE precoding:
(1.3) calculating the signal and interference of the AP to the UE according to the AP connected with each UE by using the channel information matrix and the precoding information;
the BS in cell l transmits a DL signal, Is the signal power.
The first term is effective signal, the second term is interference of other users in the cell to the target user, the third term is interference of other users outside the cell to the target user, the fourth term is noise,Is independent additive receiver noise.
Preferably, the specific step of step (2) comprises:
(2.1) calculating an estimate SE of the spectral efficiency of all users;
the DL traversal channel capacity estimate for UE k in cell j is:
τ c represents the product of the coherence time and the coherence bandwidth, τ d represents the duration of the downstream data transmission in the coherence time.
The SINR of the user terminal is calculated as follows;
Representing the expected value.
(2.2) The objective function is selected to maximize the effective SINR product:
This objective function was chosen because of consideration:
The lower bound to maximize the sum of SE is sought, where the "1+" term is ignored in each logarithm, which has little effect on high SINR users, but underestimates the spectral efficiency of users with weaker signals. Thus, maximizing the effective SINR product enables UEs with weakest signals to have higher SE than directly maximizing SE sum. This objective function also ensures that each UE gets a non-zero SE, so this utility function is more fair than maximizing the sum of SE.
(2.3) Final optimization objectives are:
the optimization problem can be effectively solved by converting the optimization problem into a geometric planning problem, converting the geometric planning problem into a convex planning problem through variable transformation, and obtaining the final optimal output power P= (ρ 1,...,ρN) by using a traditional optimization method, wherein the sum of the number of users is calculated
Preferably, the specific step of step (3) comprises:
(3.1) a transducer model is built for the problem to be solved herein, unlike the classical transducer model for processing text translation, in the communication system herein, the number of users is changed, but each user always corresponds to one allocated power, so the output network sequence length is always equal to the input sequence length and corresponds to one. The dimension of the input characteristic of each user is higher, the dimension of each power value output is 1, the converter model decoder part is considered to be modified, the modified model is processed for each user, serialization input is not needed, the self characteristic of the user is directly connected in series with the surrounding environment characteristic influencing the user, the self characteristic is used as an input, the optimal power distribution value of the user is used as an output, and the modified model is easier to train.
(3.2) Only a part of the decoder part input of the modified converter model is the output of the decoder, and the calculation mode of Q, K, V attention is unchanged, namely:
Wherein, the dimension of the feature vector of d k is represented.
Preferably, the specific step of step (4) comprises:
The sample feature of the data set (4.1) comprises two parts, namely a user feature and an environment feature, wherein the user feature comprises information S received by a user, a user position (x, y), a channel gain (g 1,...,gL) between the user and L base stations, the surrounding environment feature comprises interference (I 1,...,IN) of other users on the user, the interference of the user on the user is set to be 0, and the sum of the number of all users in the N-finger environment.
(4.2) Because the number of users N in the environment is dynamically changed, the length of the interference vector varies with the number of users. Consider a similar approach to processing a text transducer model, using Padding to patch all environmental feature vectors to the same length. If the number of users is less than the maximum number of users Nmax that the model can solve, zero is filled in for the environmental feature vector length deficiency, so the finally constructed feature vector is (S, x, y, g 1,...,gL,I1,...,IN, packing), packing= (0,..0), so that the input feature dimensions are consistent. The data set label is the optimal output power of each user obtained by using the traditional optimization method;
And (4.3) generating user characteristic data and labels by using a simulator, wherein the modified converter model is processed once for each user, so that the number of the users in different numbers of scenes is basically consistent in the finally obtained user samples because the samples obtained by simulating one communication are larger than the situations of a small number of users for scenes with relatively fewer users.
(4.4) Selecting an appropriate batchsize and number of iterations epoch during training, designing the loss function to be the predicted optimal powerMSE from true optimum power P *, i.e./>The training process is repeated until the network converges and the loss tends to be stable.
The beneficial effects are that:
The invention can solve the power distribution under the scene of the number of the large-scale MIMO dynamic users in multiple cells, the effect approximates to the result of the traditional optimization algorithm, and the same model can be used for solving different scenes of the change of the positions and the number of the users. After training the model parameters, the complexity is low when the model is used for prediction.
Drawings
Fig. 1 is a flow chart of a power distribution method for implementing multi-cell large-scale MIMO intelligent communication under a user dynamic transformation scenario in accordance with the present invention;
FIG. 2 is a diagram of a transducer model used in the practice of the present invention;
FIG. 3 is a flow chart of the use of the model of the present invention to obtain its power allocation scheme for new user data;
FIG. 4 is a graph showing the performance effect of the embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings and examples.
Fig. 1 shows a flow chart of a power distribution method for implementing multi-cell large-scale MIMO intelligent communication under a user number dynamic transformation scenario according to the present invention, including:
step S1: and constructing a large-scale MIMO network communication model under the multi-cell dynamic user number scene, calculating to obtain channel state information of the users, and constructing self characteristics of each user and surrounding environment characteristics of each user as characteristic vectors.
Step S11: setting a multi-cell multi-user scene to be simulated, setting the number of cells to be 4, randomly generating position information of K UE (users) in each cell, wherein K is less than or equal to 6, namely the total user number N is less than or equal to 24, the area of each cell is 250 multiplied by 250M, forming a 1000 multiplied by 1000M multi-cell scene, each cell comprises a base station BS positioned in the middle of the cell, the number of antennas of each BS is M=100, the uplink transmission power of the user is=100 mw, the angle standard difference in a local scattering model in a network is 10 degrees, the bandwidth is 5MHz, the hardware noise intensity of a receiver is 7dB, and the channel modeling mode refers to Rayleigh channel modeling;
Step S12: channel modeling assumes that it follows Rayleigh distribution, using Representing the channel between BS j to UE k in cell l.
Wherein,Is the spatial correlation matrix at the base station end.
Obtaining a precoding vector by using pilot signals and MMSE channel estimationSatisfy ||w li||2 = 1.
Step S121: after the position is determined, a large-scale fading coefficient matrix from each pair of BS to UE can be calculated, and the large-scale fading coefficient of one antenna from BS j to UE k in cell l is as follows
Gamma determines the gain of the intermediate channel at a reference distance of 1 km. Gamma is-148 dB and alpha is 3.76;
step S122: the precoding design allows for UL-DL duality, where UL and DL channels interact within one coherent block, which allows the BS to also use UL channel estimation to calculate/select precoding vectors. A common heuristic method and M-MMSE precoding are selected:
step S13: calculating the signal and interference of the BS to the UE according to the BS connected with each UE by using the channel information matrix and the precoding information;
BS in cell l transmits DL signals to users Is the signal power.
The first term is effective signal, the second term is interference of other users in the cell to the target user, the third term is interference of other users outside the cell to the target user, the fourth term is noise,Is independent additive receiver noise.
Step S2: setting different user numbers and user positions, establishing an optimization objective function, calculating optimal power distribution by using a traditional method, and calculating the user spectrum efficiency under the power distribution.
Step S21: the objective function is selected to maximize the effective S1NR product:
Step S22: the final optimization objective is:
the optimization problem can be converted into a geometric planning problem to be effectively solved, and the final optimal output power P= (ρ 1,...,ρN) is obtained. Base station herein The base station may allocate more power per user on average when the number of users is small.
Step S3: and constructing a transducer network model aiming at the scene.
Step S31: FIG. 2 shows a diagram of a modified transform model used in the present invention, wherein the activation function comprises GELUs. GELUs can be seen as a combination of dropout, zoneout, relus, GELUs multiplies the input by a mask of 0,1, and the mask is generated randomly depending on the input according to probability.
GELU(x)=xP(X<=x)=xΦ(x)
Where Φ (x) is a probability function of a normal distribution, the mathematical formula for the gel (x) approximation calculation assuming a standard normal too distribution is as follows:
Only a part of the input of the decoder part of the modified converter model is the output of the decoder, and the calculation mode of Q, K, V attention is unchanged, namely:
Wherein, the dimension of the feature vector of d k is represented. The feature vector length herein is 31.
Step S32: for the problem to be solved herein, the modified transducer model processes for each user, and no serialization input is needed, but the self characteristics of the user and the surrounding environment characteristics influencing the self characteristics are directly connected in series to be used as a characteristic vector as input, and the optimal power distribution value ρ * of the user is a label.
Step S4: and constructing a data set required by deep learning for training the model, wherein the input is the self characteristics of the user and the surrounding environment characteristics of the user, and the output is the power distributed by the user by the cell base station of the user, and training is carried out until the model is converged. And generating a new user data input model to obtain an optimal power distribution result.
Step S41: the user characteristics of the data set comprise two parts, namely self characteristics and environmental characteristics, wherein the self characteristics of the user comprise information S received by the user, user positions (x, y), channel gains (g 1,...,gL) between the user and each base station, the global characteristics comprise interference (I 1,...,IN) of all other users on the user, and the label is to obtain final optimal output power rho * by using a traditional optimization method; n refers to the sum of all the number of users in the environment. N is dynamically changing, all environmental feature vectors are padded to the same length using Padding, the final constructed feature vector is (S, x, y, g 1,...,gL,I1,...,IN-1, padding) = (0,..0), so that the input feature length is consistent at 31.
Step S42: the simulator is utilized to generate user characteristic data and labels, and as the modified transducer model is processed once for each user, the simulation of a scene with relatively fewer users is performed, the sample obtained by simulating one communication is larger than that of a small number of users, and the simulation times of the scene with fewer users are properly increased. Finally 160000 training samples are obtained, and the number of samples of the verification set and the test set is 10% of that of the training samples.
Step S43: at training time batchsize is 512, the loss function is designed to be the predicted optimal powerMSE from true optimum power P *, i.e./>The training process is repeated until the network converges and the loss tends to be stable, and the total iteration is 150 times.
Step S44: new user data can be processed by using the trained transducer model, as shown in fig. 3, and the characteristics of each user are built into the form input model of (S, x, y, g 1,...,gL,I1,...,IN-1, padding), so that the optimal power distribution output can be obtained. Fig. 4 shows that the performance of the power allocation of the model herein at different numbers of users is significantly higher than the average allocation and is sufficient to approximate a complex conventional iterative algorithm.
The foregoing examples of the present invention are only for illustrating the calculation model and the calculation flow of the present invention in detail, and are not limiting to the embodiments of the present invention, but are not intended to be exhaustive of all the embodiments, and all obvious variations or modifications that come within the scope of the invention are still within the scope of the invention.
Claims (4)
1. The power distribution method for multi-cell large-scale MIMO intelligent communication under the user number dynamic transformation scene is characterized by comprising the following steps:
(1) Constructing a multi-cell large-scale MIMO network communication model under a scene of dynamic change of the number of users, calculating to obtain channel state information of the users, and constructing self characteristics of each user and surrounding environment characteristics of each user as characteristic vectors of each user;
(2) Setting different user numbers and user positions, establishing an optimization objective function, obtaining optimal power distribution by using a traditional geometric planning method as a label, and calculating the user spectrum efficiency under the power distribution;
(3) Constructing a transducer network model aiming at the scene;
(4) Constructing a data set required by deep learning and used for training a model, wherein the input is a characteristic vector of a user, the output is power distributed by a cell base station of the user, and the training is carried out until the model is converged;
The specific steps of the step (1) comprise:
setting parameter information of a multi-cell multi-user scene to be simulated, wherein for DL transmission of a large-scale MIMO network with L cells, each cell comprises a BS with M antennas and K UEs, and setting uplink transmission power P ul of a user, maximum signal transmission power P max of a base station end and bandwidth W; establishing a two-dimensional coordinate system of an area to be simulated, determining the position of each BS, and generating position information of K UE (user equipment) at random in each cell, wherein the size and the position (x, y) of the number K are not fixed;
(1.2) channel modeling assuming that it obeys Rayleigh distribution, using Representing the channel between BS j to UE k in cell i;
Wherein, Is a space correlation matrix of a base station end; obtaining precoding vector/>, by utilizing pilot signals and MMSE channel estimationSatisfy @ w li∥2 = 1;
(1.3) calculating signal and interference of the BS to the UE according to the BS connected with each UE by using the channel information matrix and the precoding information;
The BS in cell l transmits a DL signal, Ρ lk is the signal power;
The first term is effective signal, the second term is interference of other users in the cell to the target user, the third term is interference of other users outside the cell to the target user, the fourth term is noise, Is independent additive receiver noise;
the specific steps of the step (2) comprise:
(2.1) calculating an estimate SE of the spectral efficiency of all users;
the DL traversal channel capacity estimate for UE k in cell j is:
τ c represents the product of the coherence time and the coherence bandwidth, τ d represents the duration of the downlink data transmission in the coherence time;
the SINR of the user terminal is calculated as follows;
Representing the expected value;
(2.2) the objective function is selected to maximize the effective SINR product:
This objective function was chosen because of consideration:
(2.3) final optimization objectives are:
obtaining final optimal output power P= (ρ 1,…,ρN) by using a traditional optimization method, and summing the number of users
2. The power allocation method for multi-cell large-scale MIMO intelligent communication in the user dynamic transformation scenario of claim 1, wherein: the specific steps of the step (1.2) comprise:
(1.2.1) position determination and calculation to obtain the distance of cell l from BS j to UE k Calculating the large-scale fading coefficient of the signals transmitted by the AP i to the position of the UE j by utilizing a wireless signal free space path loss formula, calculating the large-scale fading coefficient matrix of each pair of BS to the UE, wherein the large-scale fading coefficient of one antenna from the BS j to the UE k in the cell l is/>
Gamma determines the gain of the intermediate channel at a reference distance of 1km, the parameters gamma and alpha being calculated according to one of a number of established propagation models;
(1.2.2) estimating a channel vector using pilot-based channel training; assuming that BS and UEs are fully synchronized and operate according to TDD protocol, the DL data transmission phase is preceded by a training phase of channel estimation; the pilot reuse factor is 1, and the same pilot is used by the UE in each unit; BS j obtains the total UL pilot power for each UE and standard MMSE estimation technique Is estimated as:
Wherein the noise Ρ tr refers to the total power of the uplink pilot;
(1.2.3) design of precoding considering UL-DL duality, UL and DL channels are interacted within one coherent block, which enables the BS to calculate/select precoding vectors also using UL channel estimation; a common heuristic is chosen, using M-MMSE precoding:
3. The power allocation method for multi-cell large-scale MIMO intelligent communication in the user dynamic transformation scenario of claim 2, wherein: the specific steps of the step (3) comprise:
(3.1) although the number of users can be changed, each user always corresponds to one allocated power, so that the length of the output network sequence is always equal to that of the input sequence and corresponds to one of the input network sequence; the dimension of the input characteristic of each user is higher, the dimension of each power value output is 1, a converter model decoder part is modified, the modified model is processed for each user, serialization input is not needed, the self characteristic of the user is directly connected in series with the surrounding environment characteristic influencing the user, the self characteristic of the user is used as an input, the characteristic vector is used as an input, and the optimal power distribution value of the user is output;
(3.2) only a part of the decoder part input of the modified converter model is the output of the decoder, and the calculation mode of Q, K, V attention is unchanged, namely:
。
4. the power allocation method for multi-cell massive MIMO intelligent communication in a user dynamic transformation scenario according to claim 3, wherein: the specific steps of the step (4) comprise:
(4.1) the sample feature of the data set comprises two parts, namely a user feature and an environment feature, wherein the user feature comprises information S received by a user, a user position (x, y), a channel gain (g 1,…,gL) between the user and L base stations, the surrounding environment feature comprises interference (I 1,…,IN) of other users on the user, the interference of the user on the user is set to be 0, and N refers to the sum of the number of all users in the environment;
(4.2) because the number of users N in the environment is dynamically changed, the length of the interference vector is changed with the change of the number of users; padding all the environmental feature vectors to the same length by using Padding; if the number of users is smaller than the maximum number of users Nmax which can be solved by the model, zero is filled in for the length deficiency of the environment feature vector, so the finally constructed feature vector is (S, x, y, g 1,…,gL,I1,…,IN, packing), packing= (0, …, 0), and the input feature dimensions are consistent; the data set label is the optimal output power of each user obtained by using the traditional optimization method;
(4.3) generating user characteristic data and labels by using a simulator, wherein the modified Transformer model is processed once for each user, so that for scenes with relatively fewer users, the samples obtained by simulating one communication are larger than the situations of a small number of users, and the fewer scene simulation times of the users are properly increased, so that the number of the user samples in different number of scenes in the finally obtained user samples is basically consistent;
(4.4) selecting an appropriate batchsize and number of iterations epoch during training, designing the loss function to be the predicted optimal power MSE from true optimum power P *, i.e./>The training process is repeated until the network converges and the loss tends to be stable. /(I)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210491259.5A CN114710187B (en) | 2022-05-07 | 2022-05-07 | Power distribution method for multi-cell large-scale MIMO intelligent communication under user number dynamic transformation scene |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210491259.5A CN114710187B (en) | 2022-05-07 | 2022-05-07 | Power distribution method for multi-cell large-scale MIMO intelligent communication under user number dynamic transformation scene |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114710187A CN114710187A (en) | 2022-07-05 |
CN114710187B true CN114710187B (en) | 2024-06-04 |
Family
ID=82176468
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210491259.5A Active CN114710187B (en) | 2022-05-07 | 2022-05-07 | Power distribution method for multi-cell large-scale MIMO intelligent communication under user number dynamic transformation scene |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114710187B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116484180B (en) * | 2023-06-21 | 2023-09-22 | 中国人民解放军国防科技大学 | System and method for extracting communication signal gene |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101483874A (en) * | 2009-01-06 | 2009-07-15 | 东南大学 | Uplink resource allocation method for distributed antenna MIMO-OFDM/SDMA system |
CN109474317A (en) * | 2019-01-07 | 2019-03-15 | 河南理工大学 | MR pre-processes the lower extensive MIMO bidirectional relay system power distribution method of hardware damage |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106899993B (en) * | 2017-04-02 | 2020-10-09 | 上海无线通信研究中心 | Network optimization method facing large-scale MIMO network and base station thereof |
-
2022
- 2022-05-07 CN CN202210491259.5A patent/CN114710187B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101483874A (en) * | 2009-01-06 | 2009-07-15 | 东南大学 | Uplink resource allocation method for distributed antenna MIMO-OFDM/SDMA system |
CN109474317A (en) * | 2019-01-07 | 2019-03-15 | 河南理工大学 | MR pre-processes the lower extensive MIMO bidirectional relay system power distribution method of hardware damage |
Also Published As
Publication number | Publication date |
---|---|
CN114710187A (en) | 2022-07-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Fernandes et al. | Inter-cell interference in noncooperative TDD large scale antenna systems | |
EP2154925B1 (en) | Cooperative multi-cell transmission method | |
JP5173573B2 (en) | Wireless communication apparatus and wireless communication method | |
CN112468196B (en) | Power distribution method in de-cellular large-scale MIMO system based on PZF precoding | |
CN111132333B (en) | Time distribution device and method based on time reversal in wireless power supply network | |
JP2009278205A (en) | Radio communication apparatus and radio communication method | |
CN114710187B (en) | Power distribution method for multi-cell large-scale MIMO intelligent communication under user number dynamic transformation scene | |
Chu et al. | Deep reinforcement learning based end-to-end multiuser channel prediction and beamforming | |
CN108667498B (en) | Effective capacity optimization method for multi-antenna transmission under feedback limitation | |
JP2008236066A (en) | Transmission method and apparatus for spatial multiplex transmission | |
US20230412430A1 (en) | Inforamtion reporting method and apparatus, first device, and second device | |
CN113038583A (en) | Inter-cell downlink interference control method, device and system suitable for ultra-dense network | |
CN115622595B (en) | High-energy-efficiency networking method for realizing self-adaptive large-scale URLLC | |
Zia et al. | Outage analysis of multi-user massive MIMO systems subject to composite fading | |
US10194404B2 (en) | Transmission control apparatus and transmission control method | |
CN105429687A (en) | Interference alignment method for minimizing interference power and dimension | |
CN108282201B (en) | User terminal scheduling method and device and communication system | |
Chung et al. | Semidynamic cell-clustering algorithm based on reinforcement learning in cooperative transmission system | |
CN111541472B (en) | Low-complexity machine learning assisted robust precoding method and device | |
CN110086591B (en) | Pilot pollution suppression method in large-scale antenna system | |
KR101323378B1 (en) | Transmission rate related information transmission apparatus and method for receiver selection in multi cell mimo downlink network, and receiver selection apparatus, method and system in multi cell mimo downlink network | |
Wang et al. | Resource Allocation in Multi-Cell Integrated Sensing and Communication Systems: A DRL Approach | |
CN110912590A (en) | Interference suppression precoding method of large-scale fading MIMO system based on channel inversion technology | |
JP2008211349A (en) | Radio communication device, and radio communication method | |
CN114915322B (en) | Multi-cell large-scale MIMO communication power distribution method based on semi-supervised learning and feature augmentation |
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 |