以統計距離分布為基礎之感知無線電合作式頻譜偵測技術

Abstract

在此論文中，我們討論了在感知無線電的環境中，使用了距離量測 (Distance measure)在機率分佈下之頻譜感測。使用距離量測當作效能之度量的原因是基於我們並不能輕易的從log likelihood ratio中得到近似解。所以在每一個次要使用者 (Secondary users)中，我們採用距離量測作為度量。 這篇論文主要分為兩部分。在第一部分中，我們考慮集中式偵測的方式：每一個次要使用者傳送沒有量化過的訊號到共同接收器 (Fusion center)中來偵測是否有頻譜洞 (Spectrum hole)存在。在此，我們使用兩種距離量測方法，J-divergence及L2 distance來設計決策方法(Decision rule)。實際上，我們嘗試使用最佳功率分配(Optimal power allocation)及最佳線性組合(Optimal linear combination)兩種方法使偵測頻譜洞之偵測機率最大同時維持主要使用者收到的干擾在一定的程度之內。從模擬結果可以看出，使用距離量測下所得到的偵測機率的確比相同功率分布(Equal power allocation)及相同加權組合(Equal weighting combination)來的好。 在第二部分中，我們考慮了使用設限(Censoring)之非集中式偵測。設限代表了次要使用者只傳有資訊的資料到共同接收器，否則便不傳任何資料。因為高斯混合模型(Gaussian mixture model)的關係，依然很難最大化偵測機率。所以我們一樣使用距離量測的方式當作效能之度量。從模擬結果中我們可以看到，在設限方法下的偵測機率，的確比原本沒有任何限制的方法還好。

In this thesis, we discuss the problem of collaborative spectrum sensing in cognitive radio networks from the perspective of distance measures between probability distributions. The rationale behind using the distance measures as the performance metric lies on the difficulty of having a closed-form expression for the log likelihood ratio. We adopt the distance measure as the metric to design the decision criterion in each of the secondary users in the cooperative environment.

The thesis is mainly consisted of two parts. In the first part, we consider the case of centralized detection in which every secondary user sends un-quantized signal to the fusion center for the ultimate detection of the spectrum hole. We use two distance measures, J-divergence and L2 distance, to design the local decision rule. In particular, we attempt to devise a power allocation scheme among secondary users, as well as a combination scheme to gather received signals in the fusion center, for maximizing the probability of detection of a spectrum hole while keeping the interference observed by the primary user within a predetermined level. The analytical and simulated results show that we can improve the detection probability by optimizing the distance measures as compared to the equal power allocation and equal weight combination. In the second part, we consider the case of decentralized detection with censoring. The censoring means the secondary users only transmit informative observations to the fusion center or keep silent. In this case, it’s also hard to maximize detection probability because of the underlying Gaussian mixture model (GMM). We again use the distance measures as the performance metric. Simulation results show that the detection probability of the censoring scheme is better than that of the non-censoring approach.