[HK-Steve]

As I am looking to purchase a scaler this year, I found through another thread(thanks Oferlaor)that the CS-1 has about a 67ms (NTSC) 80ms (PAL) delay.


Do All scalers have delays like the CS-1?


Are any scalers better than others for the length in the delay?


What would be an acceptable maximum amount of delay before you would really start to notice it?


I was looking at the NRS, CS-1, ROCK or ROCK Lite.


Many Thanks


Steve

 

[Namlemez]

The think the Faroujda NRS has about the same delay. Not sure about the Rock, but I would believe so.


People usually use their receiver to delay the audio (not all can do this) or get a digital device to put in line to do this. Some jitter correctors pull double duty as audio delay devices as well.

 

[Allan Jayne]

I am led to believe that all scalers that have a de-interlacer that recognizes and optimizes for 3-2 pulldown will have a delay of 50 to 65 milliseconds. It takes this many fields being sized up simultaneously in order to detect the 3-=2 pulldown. And if they include a 3d comb filter, an additional 30 milliseconds can be expected on composite input.


Video hints:
http://members.aol.com/ajaynejr/viddoubl.htm

 

[oferlaor]

You guys are way off base.


The Sil504 (which is what Center Stage uses) has the 67ms/80ms delay (NTSC/PAL) because it uses 4 fields to do its motion compensated video deinterlacing.


The NRS has 33ms/40ms (NTSC/PAL again) delay because DCDi requires just 2 fields for deinterlacing. Together with the typical living room delay for a center speaker (I think I calculated it once at around 15ms), this delay is negligable and will not be noticed.


The ClearMatrix algorithm (which the Rock+ and LEEZA uses) also has the same delay as the NRS.


The Vigatec Dune has the same delay as Center Stage, but the new Dune units (which change their deinterlacing chip from Silicon Image to Fardouja) will cut down the delay by half (i.e., the same as NRS, Rock and LEEZA).


The Terranex uses a 7 field delay for motion compensation (the theory goes is that the more information you can look at in the past, the more informaiton you can bring into the current frame). That translates to 117ms in NTSC and 140ms in PAL. However, the Terranex has a built-in audio delay box, so I doubt anyone would even care...

 

[Rogerdon]

oferlaor,

I thought the NRS was 64 ms. This was posted a while back and they quoted Jeff Davis as the source.

I want to make sure I set it correctly in my Denon 5800U.


Roger.

 

[oferlaor]

AFAIK, it is a 2 field delay.


I couldn't find a detailed spec sheet on the net. Does anyone have a IC spec sheet on the FLI2200?

 

[Dizzman]

How could you detect 3:2 pulldown with less than 5 fields? I am not questioning, just wondering. i would think that the pattern would not be apparent in less than that.


Could somebody post the math on figuring out delay? I divided one second into 59.94 parts, then multiplied by 5 to get an 83 millisecond delay.

 

[trbarry]

Quote:

How could you detect 3:2 pulldown with less than 5 fields? I am not questioning, just wondering. i would think that the pattern would not be apparent in less than that.

It may take a history of as much as 5 or more fields to establish a 3:2 pattern but if you already have that pattern you can confirm it with maybe only a 1 field delay. In any 3:2 pattern any even field should be matched with either the previous or next odd field (or should just be dropped). But this part only requires a one field lookahead. You could do it on no additional lookahead if you were willing to be wrong on pattern changes.


I've never been certain how to properly calculate the delay since it involves knowing how everyone else is talking about it. But it is my opinion that if a digital scaler only waited to receive a single complete field (not frame) and then displayed it that would already be at least a one field delay, greater than 1000/60 = 16.67 milliseconds. But I'm not sure that is what others mean or if there is an officially proper definition.


- Tom

 

[JeffDM]

The FLI2200 product brief doesn't say anything about delays or the number of fields it needs. I could have sworn that Faroudja ads claim that they track four fields to do everything they need. The product brief does mention an "intra-field mode" which it can manage to make do without extra external memory in cost sensitive applications, but no claim is made as to the performance sacrafices in doing this.

 

[oferlaor]

JefDM,


I looked at the same data sheet. We really should try and find the spec sheet for the actual chip (the info should be there). If not there, then it should be in the application notes.


They can track as many as they want, but again, just like Tom said - the VIDEO field delay is the problem, not the number of fields tracked. The two numbers can be. unrelated.


Without memory, the sacrifice would be clearly in how good the PQ is. Without memory, the unit cannot do motion compensation very well (it would have to use internal memory), so the PQ would clearly suffer.

 

[trbarry]

Also remember that design economics have changed drastically the last few years. It used to be that saving a few copies of a screen buffer was a considerable memory expense but now adds an insignificant percentage cost to an average scaler.


This is one of the main items that DScaler has leveraged to get away with what it does. Each 720x240 display field maybe takes an extra .4 MB of memory in maybe a 256 MB computer. In a Rock or Leeza the economics are probably similar. But this has led to casually using more memory instead of dedicated hardware, at least in more recently designed models.


So I'd assume the memory was there.


- Tom