CN112187387A - Novel reinforcement learning method based on rasterization user position automatic antenna parameter adjustment - Google Patents

Novel reinforcement learning method based on rasterization user position automatic antenna parameter adjustment Download PDF

Info

Publication number
CN112187387A
CN112187387A CN202011002214.4A CN202011002214A CN112187387A CN 112187387 A CN112187387 A CN 112187387A CN 202011002214 A CN202011002214 A CN 202011002214A CN 112187387 A CN112187387 A CN 112187387A
Authority
CN
China
Prior art keywords
user
macro
base station
reinforcement learning
sinr
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011002214.4A
Other languages
Chinese (zh)
Inventor
高晖
林元杰
许文俊
曹若菡
陆月明
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.)
Beijing University of Posts and Telecommunications
Original Assignee
Beijing University of Posts and Telecommunications
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 Beijing University of Posts and Telecommunications filed Critical Beijing University of Posts and Telecommunications
Priority to CN202011002214.4A priority Critical patent/CN112187387A/en
Publication of CN112187387A publication Critical patent/CN112187387A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • H04B17/3912—Simulation models, e.g. distribution of spectral power density or received signal strength indicator [RSSI] for a given geographic region
    • 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
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/06—Testing, supervising or monitoring using simulated traffic

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Electromagnetism (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

A novel reinforcement learning method based on automatic antenna parameter adjustment of a rasterized user position is used in the following scenes: a dual-layer heterogeneous cellular network including a plurality of macro base stations and micro base stations with a plurality of macro users. In order to enable a plurality of users moving at high speed in a complex network environment to always keep high weighting and high speed, a novel reinforcement learning method based on automatic antenna parameter adjustment of a rasterized user position is provided. The method comprises two steps: (1) in the off-line modeling stage, the greatest advantage is that the time overhead and the computational complexity in the on-line learning process can be reduced; (2) and (3) an online learning stage: based on the real-time SINR value fed back by the user, the antenna parameter configuration which can lead the user weighting and the speed R to reach the maximum is provided by utilizing the proposed novel reinforcement learning method. Compared with the traditional method, the method has the advantages that the applicable scene is closer to the current network situation, the reinforcement learning has a good effect on time sequence prediction, and meanwhile, the method based on the grid user position has expansibility.

Description

Novel reinforcement learning method based on rasterization user position automatic antenna parameter adjustment
Technical Field
The invention relates to a method for automatically and jointly adjusting multiple parameters of an antenna, in particular to a novel deep reinforcement learning method for automatically adjusting parameters of the antenna based on rasterized user position information in a double-layer heterogeneous cellular network with strong user mobility, and belongs to the technical field of wireless communication.
Background
With the popularization of 5G, the current network structure gradually develops to a dense and complex topology structure. With the continuous improvement of the user's perception of network quality, it is difficult for the conventional network coverage optimization method to meet the development process of high-speed and low-delay networks, and the adjustment of antenna parameters has a great influence on the network coverage. The current antenna tuning is still based on the traditional manual tuning, but the disadvantages of the current antenna tuning are obvious, for example, with the deployment of a new base station or the change of a surrounding communication environment, the labor cost and the time cost of the traditional manual tuning in a dense network are greatly increased, and even more, the trial and error cost in the manual tuning process is high. Therefore, it is very urgent to realize the automatic adjustment of the antenna parameters, and only then, the method can deal with the intensive and complex network conditions which are continuously developed in new period.
The existing schemes for jointly adjusting a plurality of self-optimized antenna parameters only basically adjust one parameter of the downward inclination angle of the antenna, but the invention jointly adjusts three parameters of the downward inclination angle of the antenna, the horizontal wave width of half power and the vertical wave width of half power. Even if the existing scheme of jointly adjusting multiple parameters aims at the application scene which is basically a static environment and a single-layer cellular network, the slightly complicated heterogeneous network environment with multiple cells and simple samples is far from the current network situation under the 5G network, the studied scene is the double-layer heterogeneous cellular network under the user high dynamic situation with more complicated and close to the real network topology structure, and therefore, the scheme design of the invention has more practical application value.
In practical application, situations that a plurality of traditional methods are difficult to deal with are encountered, for example, interference of macro base stations of other cells to macro users, interference of micro base stations of the cell to macro users and the like, and user density and user mobility in a coverage area of the macro base stations are high. Therefore, a deep reinforcement learning method is needed to simplify the optimization target, predict the user state change situation at each moment after the current moment in real time, and ensure that all user weights and rates can be kept at a high level to adjust the antenna parameters at the current moment. Meanwhile, a method combining offline modeling and online learning is utilized, a model with good expansibility is abstracted out to perform offline modeling training and stored, and the burden of online training is greatly reduced.
Aiming at the problems that the conventional method is poor in expansibility, the complex existing network condition is not fully considered, and the like, the invention provides a novel reinforcement learning method based on automatic antenna parameter adjustment of a rasterized user position.
Disclosure of Invention
In view of the above, the present invention provides a novel reinforcement learning method for antenna tuning based on rasterized user position automation, so that a plurality of users moving at high speed in a complex network environment can always maintain high weighting and high speed. The invention relates to a scheme design combining antenna parameter self-optimization and deep reinforcement learning methods in a network coverage area with strong user mobility.
In order to achieve the above object, the present invention provides a novel reinforcement learning method based on rasterized user position automation antenna parameter adjustment, which is used in the following scenarios: the dual-layer heterogeneous cellular network comprises a plurality of macro base stations, a micro base station and a plurality of macro users, and the user mobility is strong. Because the scene is complex and the dynamic change of the user is very frequent, the traditional method is not suitable for antenna parameter adjustment. Therefore, we propose the following method, characterized in that: the method comprises the following two operation steps:
(1) an off-line modeling stage:
the method comprises the steps of performing rasterization processing on a cellular network by analyzing user position information of a network coverage area with strong user mobility and a signal-to-interference-plus-noise ratio (SINR) fed back by a user, performing cluster analysis on user data information in each grid by repeatedly performing snapshot, and finally abstracting a user position information model of a typical macro base station sector.
The specific steps of rasterization are as follows:
repeatedly snapshotting a macro cell and the surrounding environment thereof to acquire data; setting the size of a grid, and rasterizing a macro cell, wherein all the grids are positioned on the same horizontal plane and have no gradient; when a user moves to an area contained in a certain grid, the user is considered to be positioned in the center of the grid, and the number of users contained in each grid in the same time has no upper limit; calculating the distance between a user positioned at the center point of each grid of the macro cell and the macro base station, and the pitch angle and the azimuth angle of the user positioned in the grid relative to the macro base station; carrying out mean value processing on all user SINR values, path loss values and interference between users and a base station obtained in each grid respectively; and combining the rasterized macro cell to obtain a base station-user interference model, a path loss change model, a user position and a corresponding mean value SINR model.
(2) And (3) an online learning stage:
the novel deep reinforcement learning method predicts the state change of a user in the moving process by using a Markov model, matches the optimal antenna parameter configuration corresponding to the user state at different moments through repeated training until the neural network hyper-parameter convergence involved in the deep reinforcement learning, and finally provides the suggestion of the optimal configuration of the antenna parameters of the base station through statistical analysis.
The novel deep reinforcement learning method comprises the following specific steps:
the method comprises the steps of designing action, state, reward and hyper-parameter updating rules required by a reinforcement learning model, and also comprising an optimization target and constraint conditions, wherein the state is used as input, and the action is used as output; initializing each parameter, and inputting a new state (namely a set of SINR values of each macro user); updating the hyper-parameters according to a gradient descent principle from the current moment, continuously giving the reward and action at each moment, and continuously iterating the weighting and the rate R until the hyper-parameters are converged; after the hyperparameter converges, outputting the action with the largest occurrence times (namely a set of corresponding downward inclination angle, vertical half-power wave width and horizontal half-power wave width) in the n iteration processes of maximizing R under the state; repeating the method until all the states are used as input to obtain corresponding action as output; and outputting all the state-action pairs and outputting and selecting the number of the states of each action, thereby finishing all the processes of the automatic antenna parameter adjustment.
The invention relates to a novel depth reinforcement learning method for automatic antenna parameter adjustment based on rasterized user position information. The advantages are that: on the premise of simplifying the existing network conditions and reducing time overhead and calculation complexity, the state and position information fed back by users under most conditions of the existing network is completed, and the optimal configuration of the antenna closest to the actual condition is given so as to meet the condition that all users under the coverage of the antenna always keep weighting and the rate value maximized; meanwhile, the novel deep reinforcement learning method is used for processing actual problems encountered in the antenna parameter adjusting process in more detail, and the best matching mode meeting the data transmission development trend of the existing network user is strived to be achieved.
The key points of the innovation of the method are as follows: compared with the traditional mode of manually adjusting parameters of the antenna, the invention introduces a deep reinforcement learning method, designs various parameters required by a grid user position information model to extract an antenna gain model by utilizing a method combining off-line modeling and on-line learning, provides a set of complete antenna parameter joint adjustment scheme for a large number of user states and position information in a complex network through a novel deep reinforcement learning algorithm, greatly reduces time expenditure and calculation complexity, has strong expansibility, and enables the result to be more fit with the actual situation of the existing network.
Drawings
Fig. 1 is an application scenario of the present invention: a dual-layer heterogeneous cellular network.
Fig. 2 is a flow chart of the novel reinforcement learning method based on the rasterized user location automated antenna tuning of the present invention.
Fig. 3 is a simulation statistical chart of the number of states for selecting different antenna configurations in the case where there are 4 macro users in a typical sector and the SINR granularity of the divided users is 3dB in the embodiment of the present invention.
Fig. 4 is a simulation diagram of SINR granularity variation of divided users, where the percentage of the number of states configured by the selected optimal antenna parameters to all the states is different when the number of users in a sector is different in the embodiment of the present invention; and selecting a simulation diagram that the percentage of the state quantity of the optimal antenna parameter configuration to all the states changes along with the user quantity in the sector when the SINR granularity of the divided users is different.
(remark: optimal antenna parameter configuration means action maximizing the forward in the state case, and satisfying the high percentage of the number of states selecting the action in all the states)
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the accompanying drawings.
Referring to fig. 1, an application scenario of the method of the present invention is described first: the double-layer heterogeneous cellular network comprises K macro base stations and 2K micro base stations, wherein the K macro base stations which are configured with 3 directional antennas are all located in the centers of K macro cells and respectively cover a 120-degree sector, the 2K micro base stations are uniformly distributed in the K macro cells, each macro cell is averagely provided with U macro users, and the macro users are configured with a single omnidirectional antenna. And the macro user selects and connects with a nearby macro base station generating the maximum SINR value according to the signal-to-interference-plus-noise ratio (SINR), the connected macro base station is marked as K, and signals generated by the remaining K-1 macro base stations are regarded as interference for the macro user. The micro base station serves only a specific user and the signal it generates is also seen as interference to the macro user.
With reference to fig. 2, the following two operating steps of the method of the invention are described:
(1) an off-line modeling stage: the method comprises the steps of performing rasterization processing on a cellular network by analyzing user position information of a network coverage area with strong user mobility and a signal-to-interference-plus-noise ratio (SINR) fed back by a user, performing cluster analysis on user data information in each grid by repeatedly performing snapshot, and finally abstracting a user position information model of a typical macro base station sector.
(2) And (3) an online learning stage:
designing reinforced learning model
a. Design mechanism for state: the SINR value for user u is defined as follows:
Figure BDA0002694726000000051
for U users in the macro cell k, the SINR value of the user U is converted in dB level,
the conversion formula is: gamma rayu[n]＝10log10ρu[n]，
Then state is ═ gamma1[n],γ2[n],...,γU-1[n],γU[n]]；
b. Design mechanism for action:
action＝[θt,θ3dB,φ3dB]，
wherein the initial action is actioninit＝[15°,10°,75°]Possible values of the respective parameters are shown below as the down tilt angle thetat＝[0°,3°,6°,9°,12°,15°]
Vertical half power wave width theta3dB＝[4.4°,6.8°,9.4°,10°,13.5°]
Horizontal half power wave width phi3dB＝[45°,55°,65°,70°,75°,85°]；
c. Design mechanism for reward:
Figure BDA0002694726000000061
wherein
Figure BDA0002694726000000062
α is defined as a learning rate in reinforcement learning;
d. so that the optimization objective is
Figure BDA0002694726000000063
s.t.C1:
Figure BDA0002694726000000064
C2:
Figure BDA0002694726000000065
C3:
Figure BDA0002694726000000066
Wherein the content of the first and second substances,
Figure BDA0002694726000000067
Figure BDA0002694726000000068
for all the sets of states, the state is,
Figure BDA0002694726000000069
all sets of action;
② designing updating rule of hyper-parameter
The loss function is:
Figure BDA00026947260000000610
wherein
Figure BDA00026947260000000611
The updating rule for gradient decreasing to obtain the hyperparameter w is as follows
Figure BDA00026947260000000612
When n/5 is an integer, the structure,
Figure BDA00026947260000000613
otherwise, it is not updated
Figure BDA00026947260000000614
Design mechanism of characteristic value
x＝[x1,x2,x3,...,xU-1,xU],
xu＝γu[0]+ΔAu-Δβu(s,a)+ΔLu,
Wherein Δ AuFor the value of the gain change, Δ β, during the movement of the user uu(s, a) is an interference variation value,. DELTA.LuIs a path loss variation value;
design mechanism for dividing SINR value granularity
The SINR value of the user in the current network data is basically in the range of 0-30dB, and as the SINR value of the user is an infinite number of discrete values, a data initialization process is performed for the convenience of analysis:
for example, a rounding mode is adopted, the SINR granularity at this time is 1dB, and 0-30dB is divided into 31 discrete values; the SINR granularity may also be designed to be 2dB, which is divided into 16 discrete values, and so on. In the simulation implementation test, 2dB, 3dB, 4dB, 5dB and 6dB are adopted to divide the granularity of the SINR value, and the influence on the simulation implementation test result is researched.
Initializing each parameter, inputting a new state, updating the hyper-parameter according to a gradient descending principle from the current moment, continuously giving a reward and an action at each moment, and continuously iterating R until the hyper-parameter is converged; after the hyper-parameter is converged, outputting the state to enable the R to reach the most frequently occurring action in the n iteration processes with the maximum R; repeating the method until all the states are used as input to obtain corresponding action as output; and outputting all the state-action pairs and outputting and selecting the number of the states of each action, thereby finishing all the processes of the automatic antenna parameter adjustment.
In order to show the practical performance of the method of the present invention, the applicant performs multiple simulation implementation tests, a network configuration model in a test system is an application scenario shown in fig. 1, and the results of the simulation tests are shown in fig. 3 and fig. 4, and the simulation is performed under the conditions of different numbers of users in a typical sector and different user SINR partition with different granularity.
As can be seen from fig. 3, in the case that there are 4 macro users in a typical sector and the partition SINR granularity is 3dB, the statistics of the number of states of the optimal antenna configuration is selected, and the antenna configuration with too small number of selected states is not displayed. 94.78% of the states select 20 antenna parameter configurations from 180 antenna parameter configurations, in other words, only switching among 20 antenna parameter sets is needed, so that 94.78% of the user states can be always kept at a high weighting sum rate.
As can be seen from fig. 4, with the method of the present invention, for the same partition SINR granularity, with the change of the number of UEs, the overall selected optimal antenna parameter configuration combination does not change, which indicates that with the increase of the number of users, the selection of the optimal antenna parameter configuration combination is not affected (on the premise of not seriously affecting the quality of the communication environment), and the ratio of the number of states of selecting the optimal antenna parameter combination is in an increasing trend; on the premise that each sector has the same number of UEs, the ratio of the number of states of the optimal antenna parameter configuration combination is basically increased along with the process of gradually reducing the division SINR granularity from 6dB to 2 dB.
The above description is only exemplary of the present invention and should not be taken as limiting the invention, as any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (3)

1. A novel reinforcement learning method based on automatic antenna parameter adjustment of a rasterized user position is used in the following scenes:
the dual-layer heterogeneous cellular network comprises a plurality of macro base stations, a micro base station and a plurality of macro users, wherein the base stations are provided with a plurality of directional antennas, and the macro users are provided with a single omnidirectional antenna; the macro user selects and connects with a macro base station which generates the maximum SINR value nearby according to the signal plus interference noise ratio (SINR); the micro base station only serves a specific user, and the signal generated by the micro base station is regarded as interference to the macro user, and besides, the macro user can be subjected to interference signals from a macro base station of a neighboring cell; in order to enable a plurality of users moving at high speed in a complex network environment to be always under high weighting and high speed, a novel reinforcement learning method based on automatic antenna tuning parameters of a rasterized user position is provided, three parameters of an antenna downward inclination angle, a vertical half-power wave width and a horizontal half-power wave width are jointly tuned, and the method comprises the following two operation steps:
(1) an off-line modeling stage: by repeatedly snapshotting a macro cell and the surrounding environment thereof, 3 models, namely a base station-user interference model, a path loss change model, a user position and a corresponding mean value SINR model, are obtained based on a rasterization method, and the offline modeling has the biggest advantage of reducing the time overhead and the calculation complexity in online learning on the premise of abstracting an expandable model; (2) and (3) an online learning stage: and predicting the moving state of the user at each moment by using the provided novel reinforcement learning method based on the real-time SINR value fed back by the user, and continuously iterating until the hyper-parameter is converged, and then providing antenna parameter configuration which can enable the weighting and the speed of the user to reach the maximum.
2. The method of claim 1, wherein:
the concrete steps of the modeling stage of the lifted-off line are as follows:
(1) repeatedly snapshotting a macro cell and the surrounding environment thereof to acquire data;
(2) and setting the size of the grids, and rasterizing the macro cell, wherein all the grids are positioned on the same horizontal plane and have no gradient. When a user moves to an area contained in a certain grid, the user is considered to be positioned in the center of the grid, and the number of users contained in each grid in the same time has no upper limit requirement;
(3) calculating the distance between a user positioned at the center point of each grid of the macro cell and the macro base station, and the pitch angle and the azimuth angle of the user positioned in the grid relative to the macro base station;
(4) carrying out mean value processing on all user SINR values, path loss values and interference between users and a base station obtained in each grid respectively;
(5) and (4) combining the data processed in the steps (3) and (4) with the rasterized macro cell to obtain a base station-user interference model, a path loss change model, a user position and a corresponding mean SINR model.
3. The method of claim 1, wherein:
the specific steps of the proposed online learning phase are as follows
(1) The method comprises the steps of designing action, state, reward, an optimization target and a constraint condition required by a reinforcement learning model, and also comprising an updating rule of a hyper-parameter, wherein the state is used as input, and the action is used as output;
(2) initializing each parameter, and inputting a new state (namely a set of SINR values of each macro user);
(3) from the current moment, updating the hyper-parameters according to a gradient descent principle, continuously giving the reward and action at each moment, and continuously iterating R;
(4) repeating (2) - (3) until the hyperparameter converges;
(5) after the hyperparameter converges, outputting the action (namely a set of corresponding downward inclination angle, vertical half-power wave width and horizontal half-power wave width) which is most frequently appeared in the n iteration processes when R reaches the maximum under the state;
(6) repeating (2) - (5) until all states are used as input to obtain corresponding action as output;
(7) and outputting all the state-action pairs and outputting and selecting the number of the states of each action, thereby finishing all the processes of the automatic antenna parameter adjustment.
CN202011002214.4A 2020-09-22 2020-09-22 Novel reinforcement learning method based on rasterization user position automatic antenna parameter adjustment Pending CN112187387A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011002214.4A CN112187387A (en) 2020-09-22 2020-09-22 Novel reinforcement learning method based on rasterization user position automatic antenna parameter adjustment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011002214.4A CN112187387A (en) 2020-09-22 2020-09-22 Novel reinforcement learning method based on rasterization user position automatic antenna parameter adjustment

Publications (1)

Publication Number Publication Date
CN112187387A true CN112187387A (en) 2021-01-05

Family

ID=73957095

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011002214.4A Pending CN112187387A (en) 2020-09-22 2020-09-22 Novel reinforcement learning method based on rasterization user position automatic antenna parameter adjustment

Country Status (1)

Country Link
CN (1) CN112187387A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113300749A (en) * 2021-03-30 2021-08-24 北京邮电大学 Intelligent transmission beam optimization method based on machine learning enabling

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108375363A (en) * 2017-12-05 2018-08-07 中国移动通信集团福建有限公司 Antenna bearingt angular deflection check method, device, equipment and medium
CN109379752A (en) * 2018-09-10 2019-02-22 中国移动通信集团江苏有限公司 Optimization method, device, equipment and the medium of Massive MIMO
CN109714093A (en) * 2018-11-05 2019-05-03 南京邮电大学 A kind of joint antenna selection method towards isomery cellular network
CN109963287A (en) * 2017-12-26 2019-07-02 中国移动通信集团湖北有限公司 Antenna directional angle optimization method, device, equipment and medium
CN110572835A (en) * 2019-09-06 2019-12-13 中兴通讯股份有限公司 method and device for adjusting antenna parameters, electronic equipment and computer readable medium
CN111062466A (en) * 2019-12-11 2020-04-24 南京华苏科技有限公司 Method for predicting field intensity distribution of cell after antenna adjustment based on parameters and neural network
WO2020093631A1 (en) * 2018-11-06 2020-05-14 五邑大学 Antenna downtilt angle measurement method based on depth instance segmentation network
CN111246497A (en) * 2020-04-10 2020-06-05 卓望信息技术（北京）有限公司 Antenna adjustment method based on reinforcement learning
CN111309779A (en) * 2020-01-19 2020-06-19 北京互联无界科技有限公司 Method and system for realizing beam parameter sample collection based on clustering technology

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108375363A (en) * 2017-12-05 2018-08-07 中国移动通信集团福建有限公司 Antenna bearingt angular deflection check method, device, equipment and medium
CN109963287A (en) * 2017-12-26 2019-07-02 中国移动通信集团湖北有限公司 Antenna directional angle optimization method, device, equipment and medium
CN109379752A (en) * 2018-09-10 2019-02-22 中国移动通信集团江苏有限公司 Optimization method, device, equipment and the medium of Massive MIMO
CN109714093A (en) * 2018-11-05 2019-05-03 南京邮电大学 A kind of joint antenna selection method towards isomery cellular network
WO2020093631A1 (en) * 2018-11-06 2020-05-14 五邑大学 Antenna downtilt angle measurement method based on depth instance segmentation network
CN110572835A (en) * 2019-09-06 2019-12-13 中兴通讯股份有限公司 method and device for adjusting antenna parameters, electronic equipment and computer readable medium
CN111062466A (en) * 2019-12-11 2020-04-24 南京华苏科技有限公司 Method for predicting field intensity distribution of cell after antenna adjustment based on parameters and neural network
CN111309779A (en) * 2020-01-19 2020-06-19 北京互联无界科技有限公司 Method and system for realizing beam parameter sample collection based on clustering technology
CN111246497A (en) * 2020-04-10 2020-06-05 卓望信息技术（北京）有限公司 Antenna adjustment method based on reinforcement learning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
EREN BALEVI等: "A Novel Deep Reinforcement Learning Algorithm for Online Antenna Tuning", 《2019 IEEE GLOBAL COMMUNICATIONS CONFERENCE (GLOBECOM)》 *
EREN BALEVI等: "Online Antenna Tuning in Heterogeneous Cellular Networks With Deep Reinforcement Learning", 《IEEE TRANSACTIONS ON COGNITIVE COMMUNICATIONS AND NETWORKING》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113300749A (en) * 2021-03-30 2021-08-24 北京邮电大学 Intelligent transmission beam optimization method based on machine learning enabling

Similar Documents

Publication Publication Date Title
Meng et al. Power allocation in multi-user cellular networks: Deep reinforcement learning approaches
US20230171008A1 (en) Systems and methods for wireless signal configuration by a neural network
CN111683375B (en) Unmanned aerial vehicle deployment optimization method for unmanned aerial vehicle-assisted wireless cellular network
Xia et al. Generative neural network channel modeling for millimeter-wave UAV communication
CN113438002B (en) LSTM-based analog beam switching method, device, equipment and medium
CN113114343B (en) High-energy-efficiency intelligent dynamic beam forming method for multi-beam satellite
CN114422363B (en) Capacity optimization method and device for unmanned aerial vehicle-mounted RIS auxiliary communication system
CN112866904B (en) Channel-training-free large-dimension communication beam alignment method based on beam index map
CN113473480A (en) Improved reinforcement learning network coverage optimization method facing cellular network
Hashimoto et al. SICNN: Spatial interpolation with convolutional neural networks for radio environment mapping
CN113300749A (en) Intelligent transmission beam optimization method based on machine learning enabling
CN114980169A (en) Unmanned aerial vehicle auxiliary ground communication method based on combined optimization of track and phase
CN103945518A (en) Beam-forming-based power distribution method for cognitive radio system
CN114826462B (en) Beam domain channel augmentation method for large-scale MIMO statistical port selection
CN113239632A (en) Wireless performance prediction method and device, electronic equipment and storage medium
CN112187387A (en) Novel reinforcement learning method based on rasterization user position automatic antenna parameter adjustment
Sun et al. Long-term transmit point association for coordinated multipoint transmission by stochastic optimization
Fonseca et al. Adaptive height optimization for cellular-connected UAVs: A deep reinforcement learning approach
Wen et al. Machine learning based mm-wave 60 GHz channel modeling for 5G wireless communication systems
CN110505604B (en) Method for accessing frequency spectrum of D2D communication system
Reininger et al. Multi-stage optimization for mobile radio network planning
CN114980156B (en) AP switch switching method of honeycomb millimeter wave-free large-scale MIMO system
CN113242068B (en) Intelligent communication beam collision avoidance method based on deep reinforcement learning
Tekgul et al. Sample-efficient learning of cellular antenna parameter settings
CN112351449B (en) Massive MIMO single-cell weight optimization method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20210105