4
\$\begingroup\$

I will be using the OPA694 for some video processing. I would like to know how much delay it introduces, so I can account for it. Here are some specifications of the op-amp:

Rise Time and Fall Time    G = +2, VO = 0.2 V  Step 0.8 ns
Settling Time to 0.01%     G = +2, VO = 2 V    Step 20 ns
Settling Time to 0.1%      G = +2, VO = 2 V    Step 13 ns

I won't be using a 2+ gain, only a +1. My signal will be an RGB line and range from -0.3 V to 0.7 V. A rough estimate will do, and as I understand it can vary with the application.

I'm asking this because I'll be taking an input which will be roughly 35 nanoseconds behind, delaying it to sum it to another signal. E.g. Sync is 35 ns late and RGB are all on time 0 ns. To add sync to G, I will delay G 35 ns and use an op-amp to add G and sync together. The op-amp now means line G is 35 + x ns behind R and B at 0 ns.

Also, how much does 1 ns matter in video? If it is less than a pixel it's fine.

\$\endgroup\$
12
  • \$\begingroup\$ Also, how would any delay matter ? It's a video signal, human eyes are slow so even if the picture on a screen is delayed by 1 ms, no one would notice. \$\endgroup\$ Sep 6, 2016 at 15:18
  • 1
    \$\begingroup\$ @FakeMoustache: "How would any delay matter?" That's a stupid question. If you're trying to sample the data, a half-pixel delay relative to whatever you're using as a timing reference (usually HSYNC) will completely mess up the results. \$\endgroup\$
    – Dave Tweed
    Sep 6, 2016 at 15:24
  • 6
    \$\begingroup\$ Can we try taking the question seriously as written please? It's an important property that's not straightforward to read off a datasheet. \$\endgroup\$
    – pjc50
    Sep 6, 2016 at 15:38
  • 2
    \$\begingroup\$ At 1080p60 you have 7.8ns per pixel (1/(1080*1960*60)). You have a signal range of 1V but real world pixels don't generally go from black to white instantly, there will be a couple of pixel transition region so a realistic max signal change will be 0.5 to 0.3V per pixel. From the data you've given the settling time for a 0.2V step is 0.8ns or 1/10th of a pixel. 10% of a pixel seems a small enough delay to consider insignificant. In other words at 1080p60 the delay doesn't look significant for all but the sharpest edges. For low resolutions/frame rates you have a massive margin. \$\endgroup\$
    – Andrew
    Sep 6, 2016 at 15:52
  • 1
    \$\begingroup\$ @DaveTweed "How would any delay matter?" No that's not a stupid question, you missed the point. I wanted to provoke OP into thinking about that delay. When you say "of course it matters" you're already thinking of some delay somewhere in the chain where it matters. Later the question was edited with some info about a Sync signal that was not there previously. So I made OP think about it and improve his question. Still think my question was stupid ? \$\endgroup\$ Sep 6, 2016 at 19:01

2 Answers 2

3
\$\begingroup\$

80ps differential RGB may not be visible in 1080p but 800ps delay may be noticable. 1080p60Hz uses an 8ns pixel and RGB skew or diff. delay causes colour shift of black vertical edges from white to the delayed colour for a single pixel only.

Differential Delay is the rate of change of phase with frequency.

The LCD monitor first syncs up to VS and HS and then syncs pixel clock f to the video signal , so it may be best to passive combine Sync + G so that RGB pixels stay in sync. But it may not be noticeable.

Since pixel clock f(P) depends on f(VS)* N pixels per screen resolution HS delay is not critical.

If LCD pixel clock is out of sync with image information, aliasing patterns will be visible. If pixel clock is skewed in phase then the alignment of pixel information will be skewed differential delay of any signal should be <<1% of pixel clock as the pixel clock phase resolution in LCD monitors is at least 1/256 of a pixel.

There are a couple test patterns which check the sharpness of a monitor and the quality of the video.

One is alternate pixel 0-100% IRE with 8 bit resolution in white or any colour, which 1/2 pixel clock rate. If black white pixels are sharp , then both video and LCD have adequate bandwidth. If grayish then it does not. If the tint changes , it means one colour has visible delay relative to the others at 1/2 f(P).

The next test is incremental RGB from 0 to 100% IRE looking for missing codes in DAC monotonicity.

So what differential delay is negligible? It depends on the quality of your monitor or interface specs.

A 1080p display @60Hz uses a pixel clock of 1080x1920x60 = 124.416 MHz Thus <1% or 1/256 of 8ns pixel is the best differential delay quality that can be achieved.

But what is noticable? That depends on the test pattern and your expectations?

Can you notice 1 pixel colour bleeding on the edge of a high contrast image?

Yes, using 8 point font without gamma pixel blending enabled on your PC and without sharpening.

note LCD sharpening may enhance low video quality but degrades hires fonts with colour smearing by differential phase in HF gain, and also overshoot in luminance.

Can you notice 1% of a pixel delayed? Not likely unless you have expensive studio monitors. (as @Dave Tweed said, 10% may be realistic or 800 ps)

If you wish to check if your Monitor is "in sync" get the free DPT.exe and use the 1st pattern 1pix alternate black/white then 2pix alt. then linear fade RGB to black then Linear fade RGB to black and compare saturation thresholds and adjust/cal. gamma, black,contrast in your video card software.

\$\endgroup\$
4
  • \$\begingroup\$ Nice explanation of how timing can affect video signals. How would I passively add sync to G? \$\endgroup\$ Sep 6, 2016 at 17:32
  • \$\begingroup\$ depends. R divider from 5V sync pattern but still 75 Ohm source. \$\endgroup\$ Sep 6, 2016 at 17:36
  • \$\begingroup\$ So if both the lines are NOT terminated (75 ohms) Using a Resistive Divider with both resistors being 75 ohms would lead to a terminated line as well as adding them? \$\endgroup\$ Sep 6, 2016 at 17:47
  • \$\begingroup\$ Each RGB signal source must be 75 ohms and terminated with same in monitor. \$\endgroup\$ Sep 6, 2016 at 19:07
7
\$\begingroup\$

Using an LTspice simulation with 100 MHz 0.5 V peak sine wave input and the Figure 36 unity gain buffer (+/-5 V supplies), and a 20 pF load, I get about 130 ps.

Enter image description here

Enter image description here

Enter image description here

You can see from Figure 5 (in the datasheet) that it is much less than 1 ns.

\$\endgroup\$
6
  • \$\begingroup\$ It seems the images are missing \$\endgroup\$ Sep 6, 2016 at 17:26
  • 1
    \$\begingroup\$ @UmarMahmood Oh, all right, but I suggest you do this yourself too. \$\endgroup\$ Sep 6, 2016 at 17:34
  • \$\begingroup\$ I see from Figure 5 that 10-90% slew rate is almost 2 dots or 2ns but spec is 1700V/us typ or 1.7V/ns or <1ns for 1V step \$\endgroup\$ Sep 6, 2016 at 17:40
  • 3
    \$\begingroup\$ @TonyStewart I don't think we should be looking at slew rate unless we think it's slew-rate limited. \$\endgroup\$ Sep 6, 2016 at 17:48
  • \$\begingroup\$ I'm not too worried about slew rate as the max voltage step is theoretically 0v to 0.7v for a 100% white image \$\endgroup\$ Sep 6, 2016 at 17:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.