為什么CST MWS設置不同AR filter的初始時間結果不同？

在下新手，
1.首先想請問各位大神AR filter的原理是什么。我看了一下幫助文件完又找了CST高級概念（截圖），但是還是不太理解。

是不是說從first time step開始就產生一個很短的時間間隙（skip time steps?）然后計算預測下一個時間段的時域波形，如果預測和實際吻合了就停止。如果時域信號掃了10ns，給它從2ns開始算，它算到5ns就停止了，也就是說在它所計算的3ns之內已經提取到了足夠的信息用來濾波（這是不是自回溯的概念？），5ns-10ns是不算了的，那以后我再計算此類結構的話是不是只要掃到5ns+就可以了？
2.數(shù)據(jù)處理的時候用不同的初始位置，濾波完了之后的結果很不一樣啊，這是為什么呢？雖然峰位基本一致，但是1.87THz處的峰位形狀變化實在是太怪異了，一會是峰，一會是谷（實際上也應該是谷），不知道怎么才能得到最好的濾波結果啊。

25ps的logfile，其它時間的結果也是類似的，analysis1的energy error還不錯，analysis2的就不行了（e-001左右）；Average deviation 就是在e-003量級，怎么也下不去……
===================================================================
Analysis 1 of 2 for stimulation port 1, mode 1
===================================================================
First time step : 2.499977e-011 s
Skip time steps : 10
Max. frequency : 4.935000e+000 THz
Max. order of filter : 40
Relative Window length: 3.000000e+000
Input signal samples : 7977
Input signal length : 4.902240e-011 s
-------------------------------------------------------------------------------
step | window range | rel. wnd. | filter | energy | pulses to
| s | length | order | error | calculate
------|---------------------------|-----------|--------|-----------|-----------
1 | 1.9543e-011 - 3.2448e-011 | 5.25 | 36 | 6.56e-005 | 1.32e+001
2 | 1.9604e-011 - 3.2510e-011 | 5.25 | 34 | 6.43e-005 | 1.33e+001
3 | 1.9665e-011 - 3.2571e-011 | 5.25 | 32 | 1.11e-004 | 1.33e+001
4 | 1.9727e-011 - 3.2632e-011 | 5.25 | 34 | 8.80e-005 | 1.33e+001
5 | 2.1632e-011 - 3.2694e-011 | 4.50 | 32 | 3.72e-005 | 1.33e+001
6 | 2.1694e-011 - 3.2755e-011 | 4.50 | 32 | 4.08e-005 | 1.34e+001
7 | 2.1755e-011 - 3.2817e-011 | 4.50 | 32 | 6.45e-005 | 1.34e+001
8 | 2.1816e-011 - 3.2878e-011 | 4.50 | 30 | 8.48e-005 | 1.34e+001
9 | 2.1878e-011 - 3.2940e-011 | 4.50 | 30 | 1.53e-004 | 1.34e+001
10 | 2.0096e-011 - 3.3001e-011 | 5.25 | 40 | 1.52e-004 | 1.35e+001
-------------------------------------------------------------------------------
filter step 5 was chosen for further calculations.
-------------------------------------------------------------------------------
===================================================================
Analysis 2 of 2 for stimulation port 1, mode 1
===================================================================
First time step : 2.499977e-011 s
Skip time steps : 10
Max. frequency : 4.935000e+000 THz
Max. order of filter : 40
Relative Window length: 3.000000e+000
Input signal samples : 7977
Input signal length : 4.902240e-011 s
-------------------------------------------------------------------------------
step | window range | rel. wnd. | filter | energy | pulses to
| s | length | order | error | calculate
------|---------------------------|-----------|--------|-----------|-----------
1 | 1.9543e-011 - 3.2448e-011 | 5.25 | 11 | 1.91e-001 | 1.32e+001
2 | 1.9604e-011 - 3.2510e-011 | 5.25 | 11 | 1.95e-001 | 1.33e+001
3 | 1.9665e-011 - 3.2571e-011 | 5.25 | 11 | 2.00e-001 | 1.33e+001
4 | 1.9727e-011 - 3.2632e-011 | 5.25 | 11 | 1.98e-001 | 1.33e+001
5 | 1.9788e-011 - 3.2694e-011 | 5.25 | 11 | 1.87e-001 | 1.33e+001
6 | 1.9850e-011 - 3.2755e-011 | 5.25 | 11 | 1.86e-001 | 1.34e+001
7 | 1.9911e-011 - 3.2817e-011 | 5.25 | 11 | 1.98e-001 | 1.34e+001
8 | 1.9973e-011 - 3.2878e-011 | 5.25 | 11 | 2.04e-001 | 1.34e+001
9 | 2.0034e-011 - 3.2940e-011 | 5.25 | 11 | 2.04e-001 | 1.34e+001
10 | 2.0096e-011 - 3.3001e-011 | 5.25 | 11 | 2.05e-001 | 1.35e+001
11 | 2.0157e-011 - 3.3063e-011 | 5.25 | 11 | 2.10e-001 | 1.35e+001
-------------------------------------------------------------------------------
filter step 6 was chosen for further calculations.
-------------------------------------------------------------------------------
===========================================================
Stimulation at port 1 (mode 1)
Average deviation in s-parameter energy balance: 1.142665e-002
Maximum deviation in s-parameter energy balance: 1.118762e+001
===========================================================
===================================================================
Maximum deviation in s-parameter energy balance = 11.1876
(regarding all successful filter runs)
===================================================================

補充：計算頻率0.1-3THz。