標題: Timing accuracy of meteor cameras
occult
管理員
Rank: 9Rank: 9Rank: 9



UID 1
精華 0
積分 0
帖子 3307
閱讀權限 200
註冊 2007-1-14
用戶註冊天數 6284
用戶失蹤天數 1936
狀態 離線
221.127.229.236
分享 
發表於 2009-10-24 00:15  資料 文集 私人訊息 
Timing accuracy of meteor cameras
Time accuracy is very complex problem especially on Windows system.
The major time value source for meteor observation on windows are
following 4 values.
1. Frame timing of asynchronous video camera
2. Multimedia timer of windows
3. NTP corrected clock time of windows
4. GPS time stamp which is super imposed to analog video stream

There are two aspects in evaluating the accuracy of time values.
A. Long term absolute time accuracy.
B. Short term time stability.

Each time source has following characteristics.

1. Frame timing of asynchronous video camera
It is driven from independent free running crystal oscillator.
Though it does not provide absolute time, but it is the only one time
source that has almost perfect short term time stability.
The accuracy of frame timing is less than 1 micro sec.
Therefore, the interval of video frame can be believed as most
accurate time span, unless there is no frame drops.
Therefore, UFOAnalyzer computes the frame interval time from assigned
FPS, and using the absolute time of the first frame, it re-decides the
time of all frames. (Also it checks the frame drop by media time stamp in
the vido stream).

2. Multimedia timer of windows
It has 10 nanosecond resolution. The accuracy depends on the windows
system clock. Therefore it should not be used for the measurement.
Any application program can refer this time, but the the reference
timing cannot be determined because it depends on the task dispatching
mechanism of Windows.
Usually it should be considered as having 0 to 0.3 second delay.
UFOCapture uses this clock for the time stamp super impose, because
it is the only time source that has sub second resolution in Windows.

3. NTP corrected clock time of windows
This is the most popular absolute time source on Windows.
If one uses NTP appropriately, the absolute time accuracy of system
time possibly be the order of 0.1sec.
I am using one stratum 1 server through internet with interval of 10
minutes on 5 computers (9 cameras). From its correction log, the accuracy
seems less that 10 msec usually, but sometime it records more than 300msec error.
The resulted time stamp of each camera differs up to 0.3 seconds.
This may be the limit of the overall absolute time accuracy
(including the task switching delay) on Windows.
As for short term time stability, NTP on Windows does not provide any
stability.
Because the time adjustment is done at a moment of correction. Time
jumps at the moment. (I believe that the xntp in linux may be far better than
windows because it uses phase lock loop and adjust time gradually)
The selection of NTP server and the adjustment interval is important.
One adjustment in a week with time.windows.com(Windows default) may
bring over 100 seconds of error.

4. GPS time stamp which is super imposed to analog video stream
By using some equipment that impose the time stamp to the video image,
one can know the GPS time of the events. It is believed as the most
reliable time.
I have tested this for a year, there I found two problems.
The biggest problem was the lost of satellites.
It happened several times in a week.(I used Garmin Mars15. There may
be better one.)
There is no record of the number of the satellites in each frame, I
could not know whether the stamped time was correct or not.
The second problem is the short term time stability. Most of the low
cost time stampers corrects its time at a moment of correction using 1sec pulse of GPS.
It means the time may jump.
After all, it gives almost equal reliability to NTP correction for
long term.
Therefore I am using NTP usually, and use GPS for short term special
observation of TLEs.

My current conclusion (what UFOseries does) is,
1. Believe the accuracy of the frame timing of video camera for the
measurement of meteor velocity. (almost 1 micro second accuracy).
Therefore, I believe that the major cause of the velocity error is
not the time but the measurement of the position of lights.
2. Use NTP to know the absolute time of the event(almost 0.3sec
accuracy).
頂部