Research On Spectrum Sensing Task Allocation And Spectrum Sharing Technology In Real-time Mobile Scenarios

The development of wireless communication has exacerbated spectrum resource shortage,and dynamic spectrum access is considered an effective way to improve spectrum utilization and meet user communication needs.However,the cognitive radio used to detect spectrum holes is deployed based on a fixed sensor network.Its non-mobile sensors make it difficult to expand the sensing area,and the deployment and maintenance costs are high.It makes the service not geographically continuous and limits the deployment area of dynamic spectrum access in practical applications.Meanwhile,there are no mechanisms that support the secondary users' mobility and protect their privacy in existing dynamic spectrum access research.Existing works either assumes that secondary users cannot move in real-time or abuses their location to allocate channels,which leads to the disclosure of the secondary users' privacy.On the contrary,the excessive protection of location information makes it hard for the channel manager to avoid interference between secondary users by calculating the interference relationship between users.Both of the above two problems hinder the application of dynamic spectrum access in practice.Aiming at the fixed sensing area and high cost of the traditional sensor network,this paper applies the dynamic spectrum sensing task to replace the traditional sensor network.Instead of deploying sensors in the target area,the platform recruits workers with intelligent devices to obtain the channel sensing state.The workers are mobile,and there are no costs such as deployment and maintenance.The platform only needs to pay the workers a specific fee as a reward to obtain the target area's sensing data.However,the current dynamic spectrum sensing task that considers sensing errors is established in specific scenarios.These mechanisms are no longer applicable when the conditions that affect the sensing quality change.This paper proposes to use an error function to abstract the sensing error.When a new scene is encountered,only the calculation method of the error function needs to be changed,and the mechanism proposed in this paper is still applicable.Furthermore,this paper applies auction in game theory to allocate the sensing tasks.It uses mathematical derivation to obtain the task allocation method and payment determination function that maximizes the utility of the platform.Theoretical and experimental results show that the task allocation mechanism proposed in this paper satisfies the individual rationality and truthfulness in the auction.When the workers' error ranges from 5% to 20%,compared to the existing mainstream dynamic spectrum sensing task mechanisms,the proposed mechanism in this paper increases the utility of the platform by 7% on average in the case of the indivisible tasks,and 13% on average in the case of the divisible tasks.In response to the problem that existing mechanisms cannot meet secondary users' privacy and mobility requirements,this paper adopts reinforcement learning to transform the centralized channel allocation performed by the service provider into channel selection performed by secondary users.The difficulty is that secondary users consume too many resources and need to upload sensitive sample data in reinforcement learning.Thus,this paper applies federated learning to implement reinforcement learning and reduce secondary users' resource consumption and keep sensitive data in their local devices.In this way,the secondary user can directly use the global parameters to update the reinforcement learning model in the local device and access the channel in real-time to transmit data.However,the secondary user cannot make a comprehensive choice when selecting the channels to access since they cannot obtain other secondary users' information.This paper proposes a multi-priority strategy to solve the above problem and reduce secondary users' interference while improving channel utilization.At the same time,this paper also proposes two algorithms to achieve global aggregation,respectively supporting the synchronous update and asynchronous update of the global parameters for secondary users.The experimental results show that the mechanism proposed in this paper effectively protects the privacy of the secondary users when they transmit data.The proposed mechanism averagely reduces the interference between secondary users and the data transmission time by 42% and 19%,respectively,and improves the satisfaction and channel utilization by 8% and 20%,respectively.Compared with the existing dynamic spectrum access mechanisms based on reinforcement learning,the mechanism proposed in this paper effectively reduces the resource and time consumption of the secondary user on model training.Furthermore,the secondary users can instantaneously use global parameters to access the network and transmit data.