Video ADC's

[ashleysmithd]

Hi,

I'm researching video ADC's for a presentation, and has a few questions if anyone could help me out.

From the research I've done, it looks like video ADC's use 'Flash ADC's' (aka Direct conversion ADC's), however due to the number of comparators required to convert analogue video (ie PAL) they are very impractical for any application above 8-bits in resolution.

This however contradicts the specifications of say the AJA D5D (http://www.aja.com/pdf/support/manuals_conv/AJA_manual_D5D.pdf) and other converters including from Blackmagic as they state 12-bit resolution. Obviously I understand SDI runs at 10-bits so it looks like it's just over-quantising but down-rezzing to 10-bits but at 16x the resolution of 8-bit, do these ADC's really consist of a Flash ADC?

Also, some converters now claim that they do not clip anything below 0v or above .7v, but then what values are assigned to anything below 0v? I've never heard of negative RGB values. I can understand that peak white in PAL atleast is 0.7v, which relates to approximately 235 on the RGB scale, so there's some headroom there.

If anyone could bring some clarity to this that would be great.

Thanks.

[ashleysmithd]

I had another quick question on flash ADC's, the diagram below shows that the output from the comparators will equal 0 when the input voltage is lower than the reference voltage. However, would that not mean there is always a clear divide between the output, meaning it's impossible to get 256 values? As per the diagram, there can only be 00000000, 11111111, or a divided mixture 00001111, 00111111 or 00000011 etc.



Apologies I'm fairly new to this, just trying to get my head around the basics.

Thanks.

[PID_Stop]

Goodness... you're taking me back more than thirty years to equipment like the DPS-1 and the original Ampex ACR digital time base corrector, which used discrete components for analog to digital conversion!

As you observe, the upside to flash ADC is that it is exceedingly fast; the downside is the amount of hardware required, and special care needed to avoid invalid data as the converter transitions between values (using a Gray code, where successive values only yield a one-bit difference in output data, is one approach... using a sample and hold circuit to avoid input changes while the data is being read is another). On the opposite end of the scale is successive approximation, which involves comparing the input signal against a local DAC and adjusting the DAC data until it matches the input. SA is slow, but uses minimal hardware. I have seen a hybrid of the two, but the problem is making the converter linear between different flash regions.

Remember, the analog input to a box like this AJA only needs to be sampled at a relatively low rate (by modern standards): typically, four times color subcarrier, or 14.18MHz for NTSC, so successive approximation is not an unreasonable approach.

A key question that comes up is where the decoding from a composite format to components is to be performed: it was pretty common to separate luminance from chrominance and derive the color components with analog circuitry before feeding them to ADCs; in this scenario, there is no particular benefit to using an ADC with greater word size than the output format demands (which is the point I think you are making). The other approach is to take the composite analog signal as it stands and convert that to a serial composite data stream, and do the transcoding to component digitally. If you go this route, it makes sense to do the conversion with greater precision so that the transcoding may be done with greater accuracy (rounding errors wind up getting truncated out of the final result, rather than corrupting the least significant digits). If I were designing a new piece of equipment, I would use the latter method: it is far easier to make composite to component decoding multi-standard in software, and digital signal processing permits filters with far greater performance and stability than one can easily achieve with analog circuitry.

You ask about clipping and negative values: again, remember that in analog plants, it is quite common to find equipment with AC-coupled inputs (this is even true with most waveform monitors)... you need an ordinary oscilloscope set for DC coupling on the input to really appreciate how much the video can float over time (and usually as a function of average picture level). The fix for this is to precede the ADC with some sort of DC restoration circuit: the most typical method is to use a sync detector to clamp the signal to a particular voltage reference during the back porch. In that way, you can pretty well guarantee that your video will not exceed the limits of the converter -- at least, as long as your input video stays within valid limits.

Again, remember to differentiate between composite versus component video: as you suggest, components should always be positive values. Composite video, on the other hand, contains color subcarrier that can go as low as -20IRE... so if you have clamped the back porch to ground, you will indeed have perfectly valid content that falls below ground.

I'm hustling to write this between phone calls, so it might be a bit disjointed... please let me know if this makes sense, and is responsive to what you are looking for!


Regards,

-- Jeff

[PID_Stop]

I had another quick question on flash ADC's, the diagram below shows that the output from the comparators will equal 0 when the input voltage is lower than the reference voltage. However, would that not mean there is always a clear divide between the output, meaning it's impossible to get 256 values? As per the diagram, there can only be 00000000, 11111111, or a divided mixture 00001111, 00111111 or 00000011 etc.

Apologies I'm fairly new to this, just trying to get my head around the basics.

Thanks.

For a conversion scheme like this, the number of comparators you need is always one less than the number of possible digital values. It becomes a bit more intuitive if you start with a small model and scale it up. Consider a two-bit converter, which has three comparators that can produce four distinct states. If each comparator has its own threshold level Vn:

Output 00: Input < V1
Output 01: Input >= V1, but < V2
Output 10: Input >= V2, but < V3
Output 11: Input >= V3

Similarly, an eight-bit converter will require only 255 comparators to yield 256 discrete output states.

Again, referring back to your original post's question about negative values... if you're directly converting composite video, you can clamp the back porch to ground and make the reference voltages on the resistor ladder –0.25 and +0.75 volts -- the digital output will then cover the input range from -40 to +120IRE.

Never apologize for trying to learn!

-- Jeff

[ashleysmithd]

Wow thanks for the informative reply Jeff, that certainly clears a lot of questions up.

You mentioned sampling at 4x the colour sub-carrier frequency, is this rule typical in most ADC's? From what I've learnt PAL is normally sampled at 13.5MHz in accordance with the Nyquist-Shannon theorem (double and then add a bit), 4x the sub-carrier would make it 17.73MHz which I guess would produce an even more accurate sample and reduce noise.

Would you say that more composite PAL/NTSC to SDI ADC's are Flash ADC based or Successive Approximation? It actually make's more sense to use Successive Approximation ADC's as they sample in the range of hundreds of MHz, minus the 4095 comparators you would need for a 12-bit Flash ADC. I'm just trying to work out what area to focus more on in my presentation.

Again, many thanks for the reply it was extremely helpful.

Daniel

[PID_Stop]

Wow thanks for the informative reply Jeff, that certainly clears a lot of questions up.

You mentioned sampling at 4x the colour sub-carrier frequency, is this rule typical in most ADC's? From what I've learnt PAL is normally sampled at 13.5MHz in accordance with the Nyquist-Shannon theorem (double and then add a bit), 4x the sub-carrier would make it 17.73MHz which I guess would produce an even more accurate sample and reduce noise.

Would you say that more composite PAL/NTSC to SDI ADC's are Flash ADC based or Successive Approximation? It actually make's more sense to use Successive Approximation ADC's as they sample in the range of hundreds of MHz, minus the 4095 comparators you would need for a 12-bit Flash ADC. I'm just trying to work what area to focus more on in my presentation.

Again, many thanks for the reply it was extremely helpful.

Daniel

About sample rates: again, it's important to distinguish between sampling a composite signal versus sampling the components. For composite video, sampling 3fsc is marginally adequate to overcome Nyquist issues, and general practice (I think there's even a SMPTE recommendation on this) is 4fsc. As you point out, that does indeed yield a rate of 17.72MHz.

The 13.5MHz rate you allude to is the sample rate for the luminance channel only, in a component (Y / R-Y / B-Y) system; each of the chrominance components are sampled at half that, or 6.75MHz. (Incidentally, the same rates apply to either 525 or 625 line component formats.) If you assume ten bits per sample, times (13.5 + 6.75 + 6.75) million samples per second, you wind up with a bit rate of 270Mb/s -- the most common SMPTE 259M rate, for component serial digital video. (If you take those composite samples we mentioned in the past paragraph and transmit those serially at ten bits per sample, you get another familiar bitrate: 177Mb/s, also defined in the 259M standard, but generally regarded as obsolete.)

One important point to remember is that in a composite converter, the sample rate clock is locked to incoming color burst, and the samples are taken at specific phase angles: 57°, 147°, 237°, and 327° (I had to look that up, it isn't something I memorized back in the school for Sony BVH machines!). Remember, color subcarrier is both amplitude modulated (a function of chroma saturation) and phase modulated (a function of hue); by having the sample points occur at specifically defined points with respect to the color burst reference, it becomes practical to discern luminance, chroma level and chroma phase directly from the samples themselves, without having to do additional computation; moreover, it allows the recovered chroma phase to be quite accurate.

Nowadays, I rather suspect that successive approximation is most likely how ADC is done, but I can't really prove that beyond inferring from the fact that it's technically feasible, and from the relatively low component count in contemporary equipment. Unfortunately, the days where equipment came with several manuals of schematics and detailed theory of operations tutorials are gone -- it's not unusual for equipment to ship with no documentation whatsoever, forcing the user to wade through web sites for the most rudimentary information. That's why I hang on to ancient manuals for long-gone equipment: they often explain things that apply to far newer devices.

-- Jeff

[ashleysmithd]

Brilliant, thanks Jeff. I've gone with successive approximation, but have explained why within the presentation.

One component I have to try and squeeze into my (8-minute..) presentation is the the testing of the hardware's efficiency. I've done some looking around and found a few bits and pieces including an FFT spectral analysis and histogram statistical analysis, but is there any particular hardware dedicated to testing the efficiency of video ADC's that you know of? Or any particular methods you would recommend talking about? It would be good to do a small case study and explain some of the theory behind it.

Many thanks

Daniel

[PID_Stop]

Brilliant, thanks Jeff. I've gone with successive approximation, but have explained why within the presentation.

One component I have to try and squeeze into my (8-minute..) presentation is the the testing of the hardware's efficiency. I've done some looking around and found a few bits and pieces including an FFT spectral analysis and histogram statistical analysis, but is there any particular hardware dedicated to testing the efficiency of video ADC's that you know of? Or any particular methods you would recommend talking about? It would be good to do a small case study and explain some of the theory behind it.

Many thanks

Daniel

I have never seen any sort of analysis on efficiency as you describe, per se -- I suspect that's because historically, the design of this sort of equipment is driven by the capability of ICs that are available at the time. Case in point: we used to have an Ampex ACR-25 videotape recorder, a rather large machine that played brick-sized cassettes containing up to six minutes of two-inch quadruplex format tape. It was a product of the early 1970s, and used a mixture of RTL and TTL logic families, with most of the ICs being fairly small scale -- quad gates, dual flip flops, and such. The first revision had analog time-base correction using variable delay lines; the second revision ACR-25B featured a digital time base corrector, and our station had one of these. As I recall, the TBC's ADC occupied four or five plug-in circuit boards, each something like 8x10 inches, each crammed full of 14- and 16-pin DIP ICs and a handful of 8-pin metal can ICs. One of our engineers had an unfortunate slip with a scope probe one day, and managed to short the +12 volt supply to the +5 bus -- wiping out dozens of ICs. We spent at least a week trying to repair the damage before finally giving up and ordering replacement boards from Ampex. When we opened the box, we found one board with a huge (80-pin or thereabouts) TRW single-IC ADC, surrounded by a small handful of resistors, caps, and a couple ICs; the other boards were totally blank, except for traces to get signals through the card frame. I would not be at all surprised if that monster TRW chip contained a big ladder/comparator system; in the late 1970s, the logic families available didn't run fast enough to permit successive approximation at the speeds necessary for video; it was impressive simply to find memory chips that could handle the clock rates.

Incidentally, you don't know fun until you try to troubleshoot a late '70s vintage frame synchronizer that has streaks of stuck pixels... you're talking about going through several large circuit boards with rows and rows of 1K or 2K static RAM chips, trying to find cases where the data coming out is invalid. If I didn't have short hair, it would have been gone from pulling it out!

Anyway, logic is so compact and runs so fast nowadays that it's quite practical to do tasks like ADC through successive approximation; given sufficient processing power, it's much cheaper and more reliable to use a small amount of nominally inefficient hardware augmented with firmware, than to use very efficient hardware alone. And from the perspective of a designer, it's easier to create a product using fairly generic blocks rather than doing more intricate and task-specific design. I guess the point I'm trying to make is that efficiency can be assessed in a number of ways: so long as the device meets its basic design objectives, you can look at factors like how much time the processor is idle, or how much of the power the device consumes is wasted through heat. On the other hand, efficiency could also be a reflection of how little design time (and cost) was required, how expensive the parts are to implement comparative designs, and how expensive the finished design is to build. And if you can design a circuit that can be employed in multiple devices, that has efficiency implications as well.

Time to get back in the trenches... have a great day!

-- Jeff

[w9wi]

...the second revision ACR-25B featured a digital time base corrector, and our station had one of these. As I recall, the TBC's ADC occupied four or five plug-in circuit boards, each something like 8x10 inches, each crammed full of 14- and 16-pin DIP ICs and a handful of 8-pin metal can ICs. One of our engineers had an unfortunate slip with a scope probe one day, and managed to short the +12 volt supply to the +5 bus -- wiping out dozens of ICs. We spent at least a week trying to repair the damage before finally giving up and ordering replacement boards from Ampex.

Hang on, you worked at WISC-TV back in the mid-1980s?

Exactly the same thing happened to us. We had two ACR-25s; one's TBC wasn't behaving properly, so one of our techs went probing in both of them comparing waveforms.

His probe slipped, doing exactly what you describe above.

Unfortunately, he was probing the good TBC at the time. Obviously it only remained "good" for a few more milliseconds...

[PID_Stop]

Hang on, you worked at WISC-TV back in the mid-1980s?

Exactly the same thing happened to us. We had two ACR-25s; one's TBC wasn't behaving properly, so one of our techs went probing in both of them comparing waveforms.

His probe slipped, doing exactly what you describe above.

Unfortunately, he was probing the good TBC at the time. Obviously it only remained "good" for a few more milliseconds...

Nope, this was WIXT-TV (now WSYR-TV) circa 1982. We only had one ACR, so until we got the replacement boards we ran with two film chains and three quad machines (two VR2000s and one VR1200). The ACR died with sufficient regularity that we were actually pretty good at running breaks from spot reels and hot-changing shows. It also helped weed out new hires who couldn't handle pressure...

Nowadays the equipment is so reliable that when it does fail, the operators have no idea what to do to keep on the air.

-- Jeff

[eadler]

That's why I hang on to ancient manuals for long-gone equipment: they often explain things that apply to far newer devices.

Tek (Tektronix) has a number of manuals from old scopes that explain a wide variety of analog video signaling issues (including some how-to and how-it-works documentation that is independent of equipment). This is all available from their website. I would expect that they may have some similar documentation on digital signals, comparing analog to digital, and how to convert between the two.

[rockmanac]

Nowadays the equipment is so reliable that when it does fail, the operators have no idea what to do to keep on the air.

You call tech support!

[ashleysmithd]

Just wanted to say a quick thanks to Jeff for your help, the presentation went very well.

Kind Regards

[PID_Stop]

Just wanted to say a quick thanks to Jeff for your help, the presentation went very well.

Kind Regards

Hooray! Glad to help.

-- Jeff