自主性裝置對裝置網路之效能：分析式模擬評估

Abstract

隨著全球行動數據的流量快速增長，裝置對裝置(Device-to-Device, D2D)通訊近來備受關注。 有別於利用大蜂巢網路(macro-cellular network, MCN)的大型基地台之通訊架構，D2D通訊提供了許多優勢也滿足各種應用需求。例如提供D2D鏈路較高的傳輸速率，低時間延遲，低發射功率和低干擾，更重要的是有助於分流基地台與核心網路(MCN)的負載，提升頻譜使用效率。然而，在現行通訊網路環境下，考慮不影響既有系統的通訊品質且支援大量D2D之通訊，仍有許多技術與應用上的挑戰，也缺少實體層上的實際方案。本篇論文所要探討與解決的問題為：在不藉由既有4G長期演進技術(LTE)網路的協助為前提下，如何建立一套自主性系統來偵測鄰近的裝置，並且利用數學分析來提升模擬效能以及評估整體系統設計的效應。 在我們的設計架構中，提出了裝置與裝置之間利用偵測頻譜資訊來達到自行聚集，進一步建立配對鏈路的實體層方法。基於鄰近裝置有較大機率會感測到相似頻譜，部分機率會因為環境的障礙物而導致感測到的頻譜差異較大。由於同時多個裝置之間的感測頻譜資訊交換，必須設計一套方法來解決彼此間干擾與碰撞的問題。我們利用一套頻譜資訊壓縮機制來充分利用可使用的資源，透過對子載波與量化後頻譜高度作壓縮得到一組向量，將向量視做受雜訊影響的里德-所羅門(RS)碼字(codeword)後來加以利用其特性解碼。每一組RS碼字對應到一跳頻探詢信號(probing signal)。鄰近裝置透過時頻區段上的能量偵測即可知道附近是否有裝置詢問。另外，除了找尋和本身相同聚落(cluster)的裝置外，我們使用RS碼錯誤和抹除錯誤(Error-and-Erasures)的方式來找到鄰近但其感測頻譜和本身不相似的額外裝置，大幅提升了在一個範圍內所能找到的裝置數與系統效能。此外，我們更進一步的提出系統效能評估的數學分析式，降低整體系統效能評估因為多重變數造成的複雜度，也經由電腦的分析式模擬來加以證實我們的方法在不透過MCN下所能大量支援D2D通訊以及其可靠度。

Direct short range device-to-device (D2D) communications offer many potential advantages against conventional base station based cellular technology, e.g., high transmission rates, lower latency, reduced transmit power and interference level. It also offloads traffics from existing macro-cellular networks (MCNs) and support emergency communications (public safety) when the core network breaks down.

To achieve the above-mentioned advantages in a D2D network, a device or user equipment (UE) must be able to discover other devices or UEs in its vicinity and establish the corresponding links. The purpose of this thesis is to present feasible and efficient distributed solution to these two critical issues for a D2D communication system underlaying an MCN.

In our scheme, device ID is based on the spectrum sensed by each D2D device. We presume that devices in close proximity would observe similar spectra thus are more likely to have same ID. These devices with the same ID are to be assigned to the same cluster. Since Reed-Solomon (RS) codes are non-binary maximum distance separable (MDS) codes, they are perfect candidate codes to model the sensed quantized spectra and generate frequency-hopped (FH) probing waveforms that have near-optimal dissimilarity.

To detect multiple probing signals, we develop an efficient scheme which takes advantage of the RS code's errors-and-erasure decoding capability. We are able to detect spectra not only those which are similar to that sensed by the detector but also those which are dissimilar. An enhanced probing signal design is proposed to enable a receiving device to find the most appropriate device among multiple probing devices which bear the same ID.

To estimate the discovery and link setup probabilities, we adopt an analytic simulation scheme. This scheme is very efficient in that it combines the analysis on the effects of both small and large scale fading and the simulated averaging on the geometry of devices and neighboring base stations.