A detailed look at KIWI OSD video timestamps

Page last updated: 31st Dec 2007

For Japanese readers , CLICK HERE , for a Japanese translation of this page - thanks to Kazuhisa Miyashita (author of LiMovie software).

The purpose of this page is to present recent findings from precision video timing experiments. It is for people who have a video time inserter and wish to have a deeper understanding of what is actually being measured, and how their camera works from a timing perspective.

[KIWI OSD Timing] Consider this example (by Dave Gault) of KIWI OSD timing a star reappearing from behind a waning moon.

KIWI OSD timestamps the vertical synchronising (Vsync) pulses generated by the camera. Reading the time directly from the screen, the leading Vsync was at 17:54:02.528 and the trailing Vsync was at 17:54:02.545 - a delta of 17ms, implying an EIA camera. This was field number 15,078 since the GPS sync'd timer began timestamping the video stream.

The question is - How do these times relate to the actual "optical exposure"?


I must admit, until recently I understood the optical exposure to be perfectly aligned with the above two time stamps. However with the help of an exciting new device designed by Gerhard Dangl (Austria) called VEXA ("Video EXposure Analyser") we have found that the optical exposure is not quite aligned to the Vsync pulses. However people should not panic that their previous timings with KIWI OSD are wrong. The difference between the Vsync markers and the "exposure window" is less than one millisecond, and for most practical purposes (especially with a "field" uncertainty of 16 milliseconds) I feel people can still treat the two millisecond timestamps as the 'beginning' and 'end' of the optical exposure.

Although it is a very exciting and well designed device - not everyone needs to have VEXA to understand the fine timing details of their camera. Soon Gerhard hopes to have an egroup for VEXA, where we hope to have a summary for the different cameras commonly used, so people can know (without requiring their own VEXA) how their video cameras behave from a timing perspective.

Although it is early days yet, so far we have found three important aspects with VEXA. 
 - delta between Vsync and end of optical exposure.
 - which end of the field does the camera start excluding photons, as light level increases.
 - the hysteresis effect of exposure time with light level change - for cameras like PC164, 
   that have automatic exposure only modes.
I would like to discuss the first two aspects with regards the PC164 camera.

Delta between Vsync and end of optical exposure

Recently Walt Morgan (California USA) conducted some very interesting experiments with video timing and WWV audio delays using Digital Video (DV). In one of his experiments, Walt exploited the fact that EIA video runs slightly "slow" to integer 60 fields per second.

Walt used his PC164 EIA camera ( click here for an explanation of why 59.94 fields per second) to video the KIWI OSD 1pps led flash.

The start of the LED flash appears to 'drift' through the video field at a rate of 1 millisecond per second (60/59.94). So over 17 seconds the flash drifts from the END of a video field to the START of a video field. By choosing the right field, you can check to see the video time inserter is correctly time stamping to a one millisecond accuracy. Sadly this is something those of us using the PAL system cannot do. Walt sent me the following LiMovie image (a negative to enhance contrast). Walt was curious why field number 142,635 showed the LED just turning on (the gray smudge) yet the END time was 999 milliseconds. He wondered if perhaps KIWI OSD was time stamping 1 millisecond LATE. I would have been puzzled by Walt's image if it weren't for Gerhard's experiments with VEXA - showing that indeed the camera keeps exposing AFTER the Vsync pulse.

Dave Gault (Australia) then measured the PC164 with VEXA, and found the camera keeps exposing for about 550 microseconds AFTER the leading edge of the Vsync pulse. I was then able to give an explanation (directly follows Walt's image) for why the LED flash can appear in a video field with a timestamp of 999 milliseconds. [Walt Morgan's Early Flash image]
[Explanation of Walt's Early Flash]

So for the first time, we can see the relationship between the Vsync and optical exposure. This sub-millisecond offset (between Vsync and end of optical exposure) does not impact greatly on most video timing - however I think it is important to appreciate the link between what is time stamped and the optical exposure window. With PAL cameras, Gerhard and I have measured a delta of about 780 microseconds from the leading edge of the Vsync to the END of the optical exposure.

Which end of a field, is a short exposure taken at?

The following is one of my favourite images taken using KIWI OSD - it shows the International Space Station (ISS) 'flying' in front of Mare Nectaris on the Moon.
[ISS at edge of Mare Nectaris]

We can see from the image that the exposure time must have been way under 1 millisecond (because the ISS is not blurred) but what is the EXACT time of the exposure?

Ed Morana used VEXA to test his Watec 902H CCD Video Camera, and found that at high illumination, the camera (like the PC164) records at the END of the video field. So now we can give a "precise" time for the above event as: 
07:47:04.633 UTC +/- 1 millisecond
I have a PAL CMOS color camera that operates quite differently to the PC164 camera. Using VEXA I found as the light level increases, the camera (OV7910 sensor) starts cutting exposure from "both" ends of the video field, whereas the PC164 starts cutting exposure from the START of the field. So different manufacturers use different methods to reduce exposure time within cameras.

There are more experiments and testing to be done with Gerhard's VEXA to find out details of how automatic exposure works, and what part of the video field is exposed when the camera is viewing a high intensity scene.

I hope this page has helped those with video time inserters understand video timing issues a little better. With the help of devices like VEXA, and people doing innovative experiments - we can all learn a little more of the Art and Science of video timing. 

My thanks to:

Gerhard Dangl (Austria) for his development of VEXA.
Dave Gault (Australia)
Ed Morana (USA)
Walt Morgan (USA)
KIWI OSD Homepage:

Regards, Kiwi Geoff