應用於物聯網之利用訊號特性的預測式輪詢演算法

Abstract

在本篇論文當中，我們提出了應用在物聯網之下，利用信號特性的預測式輪詢演算法。在一段時間之中，由於龐大的通訊設備(MTC設備)數量及有限的媒介存取資源，基地台無法有效率地將媒介存取資源分配給每個MTC設備。基於考慮MTC設備延遲容忍性以及大部分的MTC設備都是以上傳資料為主，我們希望能從訊號的觀點，從眾多的MTC設備中，找出擁有重要的資訊的MTC設備，並將媒介存取資源優先分配給這些MTC設備。我們認為若是一個MTC設備的訊號跟他之前的訊號有很大差異，其訊號對基地台視為比較重要的。因此，我們藉由隨機程序中的自我迴歸模型建立每個MTC裝置的訊號。藉由此模型，基地台可以根據之前收到的訊號及時間差來預測一個MTC設備現在的訊號，並將媒介存取資源優先分配給預測誤差較大的MTC裝置，如此一來我們便可降低系統的總體預測誤差。經由數學推導降低系統的計算複雜度，我們以一階及二階的自我迴歸模型來實現我們的此預測式輪詢演算法，並模擬結果來證明我們所提出的演算法相較於循環輪詢演算法是較好的。

In this thesis, we propose a novel signal-centric predictive polling scheme for medium access control in the machine-to-machine(M2M) communication networks. Due to the large number of machines and the limited resource for medium access control, it is infeasible for all machines to successfully send data in a time slot. Conventional medium access control schemes do not take into consideration of the values of signals inside packets. The proposed signal-centric predictive polling schemes always poll the machines having the most valuable signals. In particular, we evaluate the value of a signal at a machine based on the statistical correlation between the signal and the ones from the machine to the base station in the past. We adopt the widely used autoregressive (AR) model to calculate the mean squared error (MSE) of prediction based on the polling decision. To make the optimal polling decision in a time slot, we formulate a discrete optimization problem. The total MSE will be minimized by polling machines with the largest errors. To improve the fairness, we propose using a polling metric based on both MSE of prediction and polling delay. We justify the proposed algorithm by simulation results and show that both signal-centric predictive polling schemes could significantly outperform the round-robin scheme.