What causes 60 ms shutter lag?

[mribble]

I've designed the Camera Axe trigger which generally is used for very high speed events such as photographing a bullet in motion.  This requires a flash (often a very specialized airgap flash) due to shutter lag and the general characteristics of cmos sensors.  However, sometimes I want to photograph something slower such as lightning or insects in flight where shutter speeds matter.  Today the accepted approach is an external shutter which can reduce shutter lag to around 5ms from 60+ms seen on most dslr cameras.  That's some background so people hopefully don't say 60 ms is faster than human reflexes because that's not relevant here.

Anyways I'd like to understand where that 60 ms of shutter lag comes from on modern dslr cameras.  I have found that some of the last Canon film cameras had very low shutter lags (10ms), but when we came to digital it increased to 40-80ms.  I know on my Canon cameras (30d and 60d) mirror lockup doesn't reduce shutter lag.  The main areas that seem to take time are mirror lockup (5-10ms), opening shutter (3 ms), stopping down lens aperture (5 ms but can happen in parallel to other events).  This leaves around 40 ms of unexplained shutter lag.  Some searching on the web indicates that cmos sensors might require around 40 ms of warm up time to reduce noise in images, but that is not confirmed and if it is correct I'd like to know what's actually happening during that time.

Now my question is does anyone here know more where this extra shutter lag is coming from in dslr cameras?  I'm asking here because I think people who work on firmware adjustments might have run across more info about hardware than your average photography forums. 

If anyone knows of a way to modify the firmware to skip some of this shutter lag or do it preemptively even at the cost of increased noise that would be a very useful modification to people using these external triggers.  I'd be willing to sponsor a cash or hardware reward for work in this area if people here think that is possible.

[Walter Schulz]

Some searching on the web indicates that cmos sensors might require around 40 ms of warm up time to reduce noise in images

Do you have a link where I can find such statements?
It sounds like utter nonsense to me. Noise goes up in semiconductors (and other stuff) with rising temperature, not down.

[dmilligan]

How does the camera take a picture?

There is a dummy readout step, IDK how much of the time this accounts for. It is probably used for correlated double sampling, though I don't really know this. But if that is true then that would be considered "noise" reduction I guess (the noise here being unwanted electrical offsets, aka FPN)

Have you considered using ML's own various triggers?

I have made a lightning trigger for the full res silent mode: http://www.magiclantern.fm/forum/index.php?topic=12891.0

I works by taking silent pictures as fast as it can and analyzing them afterwards for a change in luminance and saving them if one is detected. In this sense it is "preemptive", and you don't have to worry about lag, it could perhaps even be made to work with an external trigger. There are some limitations though.

There's also this: http://www.magiclantern.fm/forum/index.php?topic=6303.0

The silent burst end trigger could also be useful for you, it fills up a buffer with silent pictures and then saves the last however many frames fit in the buffer when you release the trigger. So it is also sort of "preemptive". This mode is a lower resolution though.

[mribble]

Walter, here is a link to someone saying that Chuck Westfall from Canon said the following.  Chuck would know this, but if he really said it is up to debate.  I would love some time with a Canon engineer to learn more about sensors...  http://www.dpreview.com/forums/thread/1482463?page=3

"If the EOS-1D Mark II were a 35mm SLR, your requests would be reasonable. In fact, we did get release time lag down to 6 milliseconds with the EOS-1N RS, and release time lag for the EOS-1v with the mirror locked was also quite short. However, the EOS-1D Mark II is a digital SLR, and it therefore has a different set of requirements to prepare for exposure. The most significant requirement in this context is the initialization of the image sensor for noise processing. Initialization takes place during the 40 millisecond interval after the shutter button is fully depressed, and this step can neither be reduced nor eliminated. "

Dmilligan, thanks for the links!  The first one about how a camera takes a picture is very interesting.  Having not worked with ML before it will take me some time to digest it, but I will probably have some questions about it later.   The lightning modes you mentioned initially don't seem to directly solve my problem, but I'll think about it more...

[mribble]

I'm pretty ignorant about ML.  This has convinced me to try it someday because it does have some cool features, but for now I'm staying focused on my work with external triggers.  I like that you guys are solving some of the same problems, but there are certain cases such as photographing explosions with microsecond accuracy that won't be handled in a camera for the foreseeable future.  I love that people are solving both sides of the problem.

For me the two lightning modes seem to slow to be interesting for my work.  Of the two the second ring buffer mode seems most interesting.  When I start playing around with ML I might install that module and have a play with it.  It might be good enough for some insect photography, but in that case you really want higher resolutions and external shutters are the norm there.

The SCSState stuff is quite interesting, but I'll admit I don't understand it all even after reviewing the state diagrams for awhile.  I think my lack experience with ML is probably the cause.  Is it possible to skip certain steps such as scsDummyCapEnd and see what happens, or is everything started on scsReleaseStart and the log is just reporting back as certain stages complete?  Is it reasonable to get timing information added to those logs?

[a1ex]

Most cameras do a dummy capture (probably for FPN corrections), with two exceptions: 500D and 1100D.

I think it's worth checking the shutter lag on these two.

[mribble]

I don't have either of those cameras to test, but I found this review online which says the prefocused lag is 113 ms.  This is a high lag time, but it has been my experience that less expensive cameras often have longer prefocus lags.  I don't understand why this is.

http://www.imaging-resource.com/PRODS/T3/T3A6.HTM

Here's another site that has shutter lags for a bunch of other cameras, but they don't have the cameras you mentioned.  http://www.impulseadventure.com/photo/shutter-lag.html