感知無線電頻譜偵測與傳輸之聯合最佳化

Abstract

隨著無線網路技術的應用越來越普及，對無線頻寬的需求也越來越高。然而無線頻譜的資源是有限的，目前大部分可用的無線頻寬都已經分配給固有的無線網路技術所使用，造成新的技術逐漸面臨沒有足夠頻寬資源的問題。為了解決這樣的問題，感知無線電的概念被提出來以增進頻譜的使用效率；在感知無線電技術中，頻譜偵測與傳輸控制是很重要的兩個基本課題，然而過去的研究卻將這兩個重要的課題分開討論，限制了感知無線電技術整體可達到的效能。在本論文裡，我們針對不同的頻譜偵測模型(sensing model)研究頻譜偵測與傳輸功率控制的聯合最佳化問題。首先，我們針對hard sensing model提出一個聯合最佳化問題並且提出求解演算法，我們發現，與傳統上將頻譜偵測與傳輸功率分開最佳化的方法相較，透過聯合最佳化，頻譜使用效率會有一定的提升。接著，我們針對soft sensing model提出頻譜偵測與傳輸功率的聯合最佳化問題，我們發現在聯合最佳化的情況下，soft sensing model的表現比hard sensing model還好。但因為soft sensing model聯合最佳化問題的求解複雜度較高，為了兼具soft sensing model的效能與hard sensing model的易解特性，我們提出multi-level sensing model以及求解演算法。研究結果發現，multi-level sensing model確實改善了hard sensing model的效能，而且達到與soft sensing model相當接近的效能。

In this thesis, we investigate the problem of joint optimization between spectrum detection and transmission in cognitive radio networks. We first formulate the joint optimization problem over the detection threshold and transmission power in the conventional hard sensing model that needs to explicitly determine the state of the primary user. We then propose an algorithm to solve the joint optimization problem. Besides the hard sensing model, we also consider the soft sensing model in the joint optimization framework, where the secondary user does not need to explicitly determine the state of the primary user after sensing. Instead, in soft sensing the secondary user determines the transmission power based on the received spectrum sensing metric. We formulate the problem and propose an algorithm for solving the joint optimization problem in soft sensing. Evaluation results show that soft sensing can achieve better average throughput than hard sensing. It, however, suffers from a higher complexity compared to hard sensing. To strike a balance between the performance of hard sensing and complexity of soft sensing, we propose the concept of multi-level sensing in cognitive radio networks. In the unified sensing model, conventional hard sensing can be considered as a special case of two-level sensing and soft sensing can be considered as multi-level sensing with possibly infinite number of levels. We formulate an optimization problem and propose an algorithm for solving the problem. Evaluation results show that multi-level sensing can indeed improve the average throughput while reducing the system complexity. Finally, we leverage the framework of multi-level sensing for studying the impact of the estimation inaccuracy for different sensing models. We find that multi-level sensing can reduce performance difference between soft sensing and hard sensing when inaccurate estimation exists.