Research On Key Technologies Of Dynamic Detection Of Idle Frequency Band In Wireless Communication

With the rapid development of technologies such as autonomous driving,the Internet of Things,and artificial intelligence,the demand for spectrum resources has grown dramatically,leading to an increasingly strained wireless spectrum.Wireless communication is facing a new round of technical challenges.Traditional static spectrum allocation techniques are no longer able to meet the demands of high capacity and high data rate wireless communications.There is an urgent need to explore efficient wireless spectrum allocation strategies in order to improve spectrum utilization and keep up with the rapid development of wireless communication technologies.Spectrum sensing technology,through the detection of spectrum opportunities,enables dynamic spectrum access.Model-driven spectrum sensing algorithms,such as energy detection and covariance matrix detection,employ threshold adaptation strategies.However,the sampling points in these algorithms are often fixed and have high values,resulting in high sensing complexity.On the other hand,with the deep integration of artificial intelligence and wireless communication technology,data-driven spectrum sensing techniques have gained significant attention.Data-driven spectrum sensing techniques iterate on linear and nonlinear fusion of sample signals,effectively extracting prominent features that differentiate between authorized and unauthorized signals,thus enhancing sensing accuracy.However,data-driven spectrum sensing techniques face challenges such as poor robustness in sensing performance,high sample collection costs,data privacy concerns,and data silos,significantly affecting the stability and accuracy of spectrum sensing.To address these challenges,this paper investigates the technology of idle frequency band spectrum detection in wireless communication.In terms of model-driven spectrum sensing techniques,a spectrum sensing scheme based on adaptive sampling points is proposed.By adapting and coordinating the sampling points with the sensing thresholds,the performance of spectrum sensing is further improved.In the context of data-driven spectrum sensing techniques,a distributed spectrum sensing strategy based on transfer learning and a collaborative spectrum sensing method based on federated learning are proposed.These approaches involve model transfer between a central server and local nodes,as well as joint model training of local sensing nodes and the central server,aiming to overcome issues such as poor robustness in spectrum sensing,data silos,and data privacy concerns.The main contributions and innovations of this thesis are summarized as follows:1)Adaptive Sampling Point Spectrum Sensing Algorithm:To reduce the complexity of spectrum sensing,this paper proposes a spectrum sensing scheme based on adaptive sampling points at the sensing terminal.It adaptively adjusts the sensing sampling points based on the environment’s signal-to-noise ratio,enabling efficient detection of idle spectra in dynamic electromagnetic environments.Firstly,a convolutional neural network-long shortterm memory network is employed for real-time assessment of the signal-to-noise ratio in the environment.Building upon this,a spectrum sensing scheme is proposed where the sampling points at the sensing terminal adaptively change according to the environment’s signal-to-noise ratio,with the objective function being the sum of false alarm probability and missed detection probability.When the signal-to-noise ratio fluctuations are small,the adaptive sampling spectrum sensing method is used for detecting idle spectra.Conversely,when the signal-to-noise ratio fluctuations are significant,an improved energy detection algorithm is employed for spectrum sensing.2)Distributed Spectrum Sensing Scheme based on Transfer Learning:To address the issue of low robustness in spectrum sensing,transfer learning is applied to spectrum sensing technology,resulting in a proposed distributed spectrum sensing scheme based on transfer learning.Initially,partial observed signals from each sensing node are collected,and a macro-level dataset is constructed on the central server for macro-level sensing model training.The trained macro-level model is then distributed to the sensing nodes,where local nodes perform model fine-tuning to mitigate the impact of electromagnetic environment fluctuations on intelligent spectrum sensing algorithms and improve the robustness of the sensing model.Experimental results demonstrate that this scheme can achieve good sensing performance with limited samples.Furthermore,compared to non-transfer learning models,it exhibits higher detection accuracy and stronger sensing stability.3)Collaborative Spectrum Sensing Method based on Federated Learning:To address issues such as data privacy and data silos among sensing users,a collaborative spectrum sensing algorithm based on federated learning is proposed.Firstly,a local node spectrum data feature extraction network is designed and trained.The converged model weights are then sent to the fusion center for federated averaging aggregation,resulting in a global model that is distributed to all sensing nodes.The proposed approach enriches the training sample quantity effectively through the collaborative training of local model weights.It ensures the security of data in practical application scenarios and improves the sensing accuracy of the system.