本論文提出了在無線網路裡，利用接收訊號強度量測來計算次梯度最佳化與訊號特徵法的室內定位系統。而我們也採用的目標函式是加權最小平方估計，之所以會採用這個目標函式是因為他擁有好的函式凸型性以及能夠對抗對遮蔽效應。在分散式方法裡，無線感測器利用遞迴的方式使得目標函數近似次梯度法。本論文也提出可調變的步長，我們主要是藉由次梯度值與最小的移動距離來加速收斂速度。在追蹤定位上除了次梯度最佳化也搭配訊號特徵法，不但可以藉此得到更快速的收斂速度，也讓初始位置得到改善。一個是大步走的訊號特徵法，一個是小步走的次梯度法，倆倆互相搭配運用，稱之為高爾夫定位演算法，合併了訊號特徵法與次梯度法的優點，搭配使用。此外，收斂分析也證明了我們所設計的可調整式步長是會收斂的。在模擬方面，我們提出的演算法比傳統的分配式定位演算法都來的精確，不論是固定位置分析或是追蹤分析，並且收斂速度也比較快。最後，在硬體實作方面，我們利用了硬體描述語言ISE來實現高爾夫演算法，並且比較了定點數與浮點數的誤差值，以及觀察實現的電路行為是否有追蹤的動作。In this paper, we propose a subgradient optimization method for localization based on received signal strength in wireless sensor network. The objective function of weighted least-squares estimation is adopted, which shows good convexity and has immunity to shadowing effect. We also approximate the subgradient of the objective function by a recursive form so that it can be implemented in a decentralized manner within each sensing node. A variable step size is proposed to take into consideration both the subgradient and minimum adjustment to accelerate convergence. To improve the accuracy of positioning in a large-scale sensor network, a fingerprint method is incorporated. The localization method allows a large jump in movement. So, our proposed golf localization combines the fingerprint method with subgradient optimization.Furthermore, the convergence analysis is also given to show the feasibility of our design for the step size. From simulation results, we can see the proposed algorithm has better accuracy and convergence rate than the conventional decentralized algorithms to localize a stationary or moving target in wireless sensor network. Finally, we are going to introduce the hardware, and we use ISE simulation to show our proposed golf algorithm, and then compare the error value between the fixed point and floated point. So, we can look out true whether the implement circuit do tracking or not.
