改善偵測延遲之次級使用者頻譜分享系統

Abstract

本論文考慮一個可實現的基於偵測頻譜分享系統，系統中包含不同 層級的次要使用者，在不同的次要使用者之間有偵測延遲，在訊息大 小有限的情形之下，偵測延遲會導致對較高層級使用者的干擾。此研 究目標為找到此系統中，不同層級的次要使用者傳輸率的極限。我們 提出一些傳輸方法，目的為解決偵測延遲所造成的干擾，並且比較採 用不同方法的表現。傳統上解決干擾問題，低階使用者利用避免干擾 或降低傳輸能量的方法，來減少對高階使用者的影響。採用這種方法 的收發端較容易實現，然而，若是偵測延遲的長度相較於訊息長度已 不可忽略，對高層級使用者就會造成不良的使用體驗。而另一種方法 為干擾消除，透過重傳干擾的機制，可以使較高層級使用者接收干擾 到自己的訊息，利用此訊息回復先前遭到干擾的訊息。若使用此方法， 收發端的複雜度勢必較高，且會造成解碼時間的延遲。理論上，減掉 干擾可以使系統達到較高傳送速率，且次級使用者如同無偵測延遲。 此研究採用有限長度的消息理論分析(Finite blocklength regime)，如此較貼近實際上的傳輸情況。與其他避免干擾的傳輸方法作比較，使用 干擾消除的方法，在偵測延遲影響較大的情形之下，重傳干擾可以獲 得較好的效果。

This thesis considers a sensing-based spectrum sharing among dynamic secondary users with different priorities. In the system, transmission of secondary users with lower priorities should not degrade performance of those with higher priorities. Our goal is to characterize undamental performance limits of the system which has sensing delay. Under our consideration, the delay period of low-priority is comparable to message size and it causes significant interference to high-priority users, since the message size of high-priority is limited. This work is focused on finite blocklength results of different spectrum sharing techniques. Traditionally, low-priority user stops transmission or lower transmission power to avoid causing interference to high-priority users. It turns out that the performance of such interferen-ceavoidance schemes degrades as the sensing delay becomes comparable to the blocklength of the high-priority users. Instead, we propose novel methods that harness simple retransmission from low-priority users to realize interference cancellation at the receivers of high-priority users. The finite blocklength performances of the proposed schemes outperform traditional ones when the sensing delay is comparable to the blocklength of high-priority users. The additional cost lies in the complexity of the receivers and/or the induced decoding delay which is acceptable. To obtain these theoretical results, we first derive the equivalent channels under these spectrum sharing schemes, which are no longer i.i.d. or memoryless, and then extend the finite blocklength analysis for point-to-point channels to these equivalent channels.