Minimising Latency

All other topics about Ventuz here.

Minimising Latency

Postby joysprod » 25 Jul 2013, 14:33

On a recent show using Ventuz, my client supplied me camera feeds from some Sony HD Camcorders with an SDI output. The cameras fed a Panasonic HD vision mixer, and I then took the feed into Ventuz. My system was top spec with AMD W8000, S400 sync card and Deltacast HDSDI card. I know in the past we have had about 2.5 to 3 frames of latency through our system depending on input.

The lipsync was very close to being unacceptable for a live awards show. I suspect that the cameras were probably a frame and a half to two frames delayed out of the SDI, as from experience I have found camcorders have quite a bit of internal processing going on, adding to the delay. We tried all sorts from running the system at standard def rates, 720p/25 720p/50 and 1080i with appropriate changes to the native vision mixer rate. Unfortunately I have found that camera suppliers are not interested in genlocking their cameras to a house sync anymore, preferring to just use the frame stores in the vision mixers. Of course, if this is just fed directly to a Plasma or LCD screen, you can get away with it! Unfortunately adding a Ventuz input/output and a projector just tips it over the edge. The brain seems quite happy to have about 2 frames of delay without noticing it!

Nothing seemed to improve the latency, and in the end we were helped by the audio engineer adding a couple of frames of delay to make the lip sync look acceptable (just) and the audio not too much like a train station announcement for the presenter.

When you add in the delay in the projectors (professional Christie HD units) we found that in an ideal world, we had to delay by 200ms to get everything absolutely spot on. This is just not achievable in a real world situation, as while the audience view at the rear of the room (about 60 metres from the stage) would look correct with live camera, the front row view of the live camera image would be very wrong as they could hear the un-amplified voice of the presenter.

I came across this Nvidia input card while searching for details on the new, and not yet released K6000 graphics output card...

http://www.nvidia.com/object/quadro-dig ... eline.html

I was wondering if this is (1) supported by Ventuz, and (2) an improvement on the standard Deltacast input and AMD/Nvidia output card scenario as regards reducing the amount of latency?

Regards

Peter
joysprod
 
Posts: 185
Joined: 20 Jan 2012, 12:24
Location: United Kingdom

Re: Minimising Latency

Postby ErikB » 06 Aug 2013, 23:05

Hi Peter,

well 3 frames latency is really the theoretical minimum (well okay, 2 is, but that leaves littlle room for processing) You need one frame to read, one frame to process and one frame to write. Getting below this is only possible with tricks like starting processing on the first scanlines, before the rest have arrived. Apart from that, regardless of whether you write directly to GPU or otherwise, you still need to wait until the entire picture is written before you can work on it, and you cannot write it until you are finished processing it. Because the frames come in at exactly the frame rate they are played out, it requires one frame to get it, one frame to write it.

Even a Spyder or Encore have 2- 3 frames delay for the same reason.

What the Nvidia option allows is to get around bus limitations - so you can get more SDI in. It does not necessarily make it faster. It means, for example, one could in theory support more simultaneous live SDI feeds before the bus falls apart.
User avatar
ErikB
 
Posts: 202
Joined: 10 Jan 2012, 12:52

Re: Minimising Latency

Postby chriss0212 » 07 Aug 2013, 06:48

hi peter

if you realy have just 2-3 frames of delay in the total chain: you can be realy happy! on bigger events you have mostly a chain like this: camera (some delay depending on manufacturer), video mixer (1-2frames) presentation switcher (~2frames), projector(hopefully <1 frame).
yes, lipsync is always a problem but its a fact what we have to accept ;( delay on the speaker: mostly not accepted by the speaker, because he is getting confused by himself ;)

next; many (or maybe all) hardware has MORE delay in genlocked mode!

greetz

christian
User avatar
chriss0212
 
Posts: 586
Joined: 18 Jan 2012, 20:56
Location: wuppertal


Return to General Discussion