在新世代的無線網路中，高精準度、高穩定性的頻率及時間將是不可或缺的關鍵要素。本論文採用調適性類神經模糊推論系統（ANFIS）為基礎，來建置時間及頻率備援系統。本系統有2種運作模式：一般模式(Normal Mode)及繼任模式(Holdover Mode)。在一般模式下，以高精準度銫原子鐘(Cesium Clock)之頻率為校準基礎，透過模糊推論控制器（Fuzzy Controller）來校正恆溫石英晶體震盪器（OCXO），並記錄毎秒的控制量。當銫原子鐘無法正常運作時，繼任模式將會使用ANFIS的預測來接管系統，使用ANFIS分析資料庫之數據來預測下一秒控制量，並經由數位類比轉換器提供電壓加以控制，以延長OCXO之精準度。實驗結果顯示，一般模式下的OCXO精準度由10-9改善到10-13；繼任模式中之效能則可維持到10-12。本研究之系統可適用於電信、量測、儀器、控制、航運、電力系統及各種各樣的應用。 Accuracy and stability are two essential issues of frequency and time in the new generation wireless networking. A backup system for time and frequency calibration based on adaptive neural-fuzzy inference system (ANFIS) is presented in this paper. The System has two kinds of operation way: normal mode and holdover mode. In normal mode, an oven-controlled crystal oscillator (OCXO) is steered by using fuzzy controller to synchronize with the primary cesium atomic clock, and recorded the control amount of per second. While the primary cesium atomic clock cannot operate correctly, a holdover mode operation which utilizes an ANFIS predictor will takes over the system. In particular, the ANFIS is applied to analyze datum of the database for predicting the controlling amount of the next second when the system enters holdover mode. Experimental results show that the frequency stability of the OCXO can be improved from a few parts in 10-9 to 10-13 for averaging times, as well as the performance could be maintained within a few parts in 10-12 in the holdover mode. The system of this research is applicable to telecommunication, measurement, instrumentation, control systems, navigations, power systems, and in numerous other applications.
