感知無線電網路中基於核心的動態頻譜存取

Kernel-Based Dynamic Spectrum Access in Cognitive Radio Networks

林柏江(Po-Chiang Lin)

指導教授 ： 林宗男

國立臺灣大學/電機資訊學院/電信工程學研究所/博士(2010年)

https://doi.org/10.6342/NTU.2010.00626

摘要

無線頻譜對於通訊而言，是有限且珍貴的資源。根據現今大部分無線頻譜管理單位採用的固定式頻譜分配策略，無線頻譜在空間、時間、以及頻域上，並沒有被充分利用。在感知無線電網路中，動態頻譜存取提供了改善頻譜使用效率的功能，其允許次要使用者在不干擾主要使用者的情況下，有機會使用無線頻譜。主要的挑戰是，在保護主要使用者的同時，讓次要使用者的整體效用最大化。
動態頻譜存取包括兩個主要問題：適當的分配可用的無線通道給次要使用者，以及在這些無線通道上控制傳輸功率。由於通道分配與功率控制會影響到主要使用者的干擾程度，以及感知無線電網路的效能，因此設計最佳化的通道分配與功率控制方法相當重要。
為了達到最佳的通道分配與功率控制，必須要得到通道增益的資訊。傳統的通道增益估測方法，要求傳送端與接收端切換到同一個通道，並根據引導符號來估測通道增益。因此這些方法在感知無線電網路中，變得花時間且沒有效率。另外，無線通道會被小規模的衰減所影響，因此每一次的通道增益量測，都會由於小規模衰減而產生誤差。在這分論文中，我們提出一個基於核心的通道增益估測方法。在這個方法中，我們採用支援向量回歸，來建立傳送端、接收端的位置資訊與相對應的通道增益之間的知識。這樣的機器學習方法可以有效的對抗雜訊，提供一個有效且快速的通道增益估測方法。我們並在真實環境中量測GSM訊號，用以評量所提出的通道增益估測方法的效能。實驗結果顯示，當有足夠的訓練數據時，我們的方法可以有效率的估測通道增益，均方根錯誤能夠小至2 dB。
既有的關於通道分配與功率控制問題的研究，通常表示成混合整數規劃問題。然而，這樣的問題型式通常是NP-hard。在這分論文中，我們分析通道分配與功率控制問題之間的關係，並轉換為一個非線性規畫問題。結合通道增益估測，我們使用內點動態頻譜存取最佳化演算法，解決動態頻譜存取的問題，此方法能在多項式時間內得到最佳解。模擬結果顯示，我們所使用的內點動態頻譜存取最佳化演算法，表現優於其他現有的演算法。

關鍵字

感知無線電網路 ； 動態頻譜存取 ； 最佳化 ； 機器學習

並列摘要

Wireless spectrum is a limited and valuable resource for communications. In accordance with the fixed spectrum allocation strategies adopted by most regulators nowadays, wireless spectrum is known to be underutilized in spacial, temporal, and spectral domains. The dynamic spectrum access (DSA) of cognitive radio networks (CRNs) provides the capability to improve the spectrum efficiency by allowing secondary users to access the spectrum opportunistically without interfering primary users. The challenge is to maximize the utilities of the secondary users while protecting the primary users.
In dynamic spectrum access, there exist two main problems. The first one is to allocate the available channels to the secondary users appropriately. The second one is to assign the transmission power to the assigned channels of secondary users. Due to the fact that the channel allocation and the power control problems would affect the aggregated interference of the primary users and the performance of the cognitive radio networks, it is important to design an optimal channel allocation and power control method.
In order to achieve optimal channel allocation and power control, the knowledge of all channel gains are necessary. Conventional channel estimation methods require a transmitter and a receiver to tune to the same channel and estimate the channel gain by sending a pilot signal. These methods are thus time-consuming and inefficient for the dynamic spectrum access in cognitive radio networks. Moreover, wireless channels are known to be affected by the small-scale fading. A one-time sample of a channel gain is thus noisy, and the small-scale fading would lead to channel gain estimation errors. In this dissertation we propose a kernel-based channel gain estimation method. In this method we adopt the support vector regression (SVR) to build the knowledge between the location information of each transmitter-receiver pair and the corresponding channel gain. Such a machine-learning method is noise-resistant. It provides an effective and efficient method to estimate the channel gain. We perform a real-world experiment to measure the GSM signals, and use the measurement to evaluate the performance of the proposed kernel-based channel gain estimation method. Experiment results show that with sufficient training data, the proposed method could efficiently estimate channel gains and achieve the root mean square error as low as 2 dB.
Previous works about the channel allocation and power control problem usually model the problem as a mixed integer programming problem. However, such problem formulation is NP-hard in general. In this dissertation we analyze the relationship between the channel allocation and the power control, and thus re-formulate the problem as a nonlinear programming problem. With the estimated channel gains, such problem formulation would be solved much more efficiently. We solve the dynamic spectrum access problem by an interior point DSA optimization algorithm. This algorithm could obtain the optimal solution in polynomial time. Simulation results show that the interior point DSA optimization algorithm outperforms other existing algorithms.

並列關鍵字

cognitive radio networks ； dynamic spectrum access ； optimization ； machine learning

參考文獻

[1] M. A. McHenry, “NSF spectrum occupancy measurements, project summary,”

[2] J. Mitola III, “Cognitive radio: An integrated agent architecture for software

[3] T. Shu and M. Krunz, “Coordinated channel access in cognitive radio networks:

A multi-level spectrum opportunity perspective,” in INFOCOM

video transmission over CDMA networks with multiuser detection,” IEEE