Research On Location-Assisted Intelligent Resource Allocation Technology

With the development of communication technology,the application of millimeter wave communication is becoming more and more extensive.The propagation characteristics of millimeter wave make it face large losses under reflection,scattering,transmission,and bad weather conditions.Therefore,beamforming technology is often used to improve coverage.For users located at the edge of adjacent cells,when the users in different cells move to a relatively close location,the base station serving them may allocate beams pointing to the same area.At this time,if these users are allocated the same frequency resources,large inter-user interference will occur.Considering actual scenarios,there are not only communication devices such as base stations,but also some devices such as cameras that are not directly used for communication.The information obtained by the camera is helpful to improve the performance of the communication system.The vision-aided wireless communication system refers to such a system.The visual information can be obtained to judge and avoid the mentioned interference and help to improve the performance of the communication system.Therefore,the research work in this thesis studies the relevant schemes as follows:(1)User Location Extraction Scheme based on Computer Vision and FingerprintIn order to allocate frequency resources based on user location and avoid inter-user interference at the edge of adjacent cells,it is necessary to extract user location from the system.We have proposed a user location extraction scheme based on computer vision and fingerprint,which could make full use of the information in the vision-aided wireless communication system.The proposed scheme attempts to extract the user’s location using visual images captured in the system.The captured image is forwarded to an object detector to find the user’s location in the image.The obtained user location in the image is then transformed into the location in the world coordinate.If the target user cannot be detected,the proposed scheme utilizes the fingerprint-based method to obtain the user’s location.Simulation results show that the proposed scheme can achieve high accuracy when the probability of viewing users through cameras is high;When the probability is relatively low,the proposed scheme can still work to extract the user location,thus conducive to applying the scheme in scenarios at night or with a small number of cameras.(2)Resource Allocation Scheme based on User Location and Reinforcement LearningIn order to avoid inter-user interference at the edge of adjacent cells,it is necessary to judge the occurrence of the interference according to the user location and change the frequency resources allocated to users.We have proposed a resource allocation scheme based on user location and reinforcement learning.Specifically,We have designed the reward for the agent based on an optimal action sequence and proposed a method to judge the interference according to user location and then affect the system’s resource allocation.Simulation results show that the proposed scheme can effectively avoid inter-user interference at the edge of adjacent cells.Compared with a default resource allocation scheme based on channel gain not considering user location,the proposed scheme can avoid the mentioned interference and obtain larger system capacity.(3)Design and Implementation of a Data Simulation System for Vision-Aided Wireless CommunicationConsidering that collecting and organizing required data from a real environment costs a lot of human resources,and that simulation data can also meet the research needs to a certain extent,a data simulation system can be designed to support research in this thesis.We have designed and implemented a simulation system for vision-aided wireless communication.The simulation system transforms the moving objects in the scenario at each sampling time and then adopts the ray tracing algorithm to obtain rays from base stations to users.After that,the simulation system constructs the channel,generates beams,and finally obtains the data needed for the research.The process of wireless data generation using the proposed simulation system corresponds to the process of vision data generation using an open source software called Blender.Compared with the scheme of using scripts to repeatedly read and write external files and call external simulation processes to obtain required data,using the proposed system can avoid these frequent data exchanges and thus improve efficiency.In addition,as the generated wireless data correspond to the vision data,the proposed simulation system can support not only the research in this thesis but also other related research for vision-aided wireless communication.