Study Of Adaptive Cross-layer Optimization Mechanism For Cognitive Radio Systems

The scarcity of spectrum and operation of complex networks become inevitable crucial problems in future wireless communication systems, therefore, a kind of intelligent communication system based on cognitive science, namely, cognitive radio system is proposed by academia. For the rapid development of digital signal processing, computer networks and artificial intelligence, it is possible to realize cognitive radio systems by high integration of these technologies. How to make cognitive radio systems sense environment, learn from environment by artificial intelligence and adapt to evolution of environment by reconfiguration of cross-layer operational parameters so as to achieve flexible and reliable communication meeting the requirements of subscribers in complex communication systems and efficient utilization of spectrum has ever been one important research topic of modern wireless communication systems, and also been the objective of this paper.This dissertation mainly focuses on the adaptive cross-layer radio resource management of cognitive radio systems, such as sensing and access of spectrum, transmission and scheduling. More specifically, the main research contents and innovations are as follows:1. Firstly, a cross-layer optimization framework of dynamic spectrum sensing and access, transmission and scheduling in cognitive radio systems is proposed. According to the evolution of spectrum availability and sensing results, spectrum occupancy of licensed users is model as an alternative renewal process in the framework. Accordingly, statistical characteristic of spectrum availability is induced and spectrum sensing results is formulated as a discrete time Markov chain. In the proposed constrained Markov decision process (CMDP) for optimization framework, the evolution of state does not affected by actions. Consequently, traditional LP is modified to simplify solution of the CMDP. To cope with the curse of dimensionality incurred by the overmany number of variables belonging to the CMDP, policy separation approach is employed to work out optimal access police and optimal transmission police respectively. Moreover, heuristic algorithms by which it is no use for solving LP are proposed. 2. Based on the previous study, the cognitive radio systems are considered to be deployed in the scenario that environment parameters are undiscovered, that is, the systems do not have a priori knowledge about state transition probabilities of CMDP. Therefore, a kind of reinforcement learning, namely, R learning is employed by the systems to learning nearly optimal policy from the environment. The tradeoff between the QoS depending on buffer occupancy and the long-term average power efficiency is involved in the nearly optimal policy. Since R learning only adapts to unconstrained MDP (UMDP), Lagrangian multiplier approach is utilized to convert the CMDP to a corresponding UMDP. The nice monotone property of Lagrangian multiplier facilitates the search of the proper multiplier by proposed golden section search method. In addition, state space compaction and action set reduction are proposed respectively to reduce the storage cost and accelerate the convergence of R Learning. Meanwhile, the proposition that R learning policy employing the state space compaction and action set reduction converges to optimal policy under some reasonable assumption is proved.3. The work described above focuses on single user and single link, and is extended to the cognitive radio networks in which multiple users compete for spectrum. By auction theory of microeconomic, a spectrum allocation mechanism based on repeated multi-bid auction is proposed. In this mechanism, users of the system are the bidders and the access point or base station acts as an auctioneer in an auction round. Each bidder gives a multi-bid for spectrum to satisfy the utility of it. The winner determination made by the auctioneer can work out the allocation result and improve system efficiency by maximizing the revenue of the network. To satisfy different QoS criteria, three different optimization objectives of spectrum allocation and corresponding value functions of bidders are defined respectively. Compared with other allocation mechanisms, repeated multi-bid auction mechanism has some advantages:it is decentralized in nature and requires little signaling exchange and computational expense.