The Art of Listenng - Comparing DACs

[Bruce B]

In many years of conducting measurements, I've measured only one DAC that had any trouble with 0 dBFS and it has numerous other blatant design flaws as whoever designed it apparently slept through their engineering classes, or never took any.

OdBFS shouldn't be a problem for ANY converter. It's when you have inter-sample peaks that the distortion becomes a problem. Most "good" converters can handle these peaks really well. It's the poor designs and the ones where the manufacturer has cut corners to meet a certain price point where hot signals become a problem.

 

[NwAvGuy]

OdBFS shouldn't be a problem for ANY converter. It's when you have inter-sample peaks that the distortion becomes a problem. Most "good" converters can handle these peaks really well. It's the poor designs and the ones where the manufacturer has cut corners to meet a certain price point where hot signals become a problem.

Thanks. That's a good point. Trust me, a simple 0 dBFS sine wave is a problem for at least one converter from an audiophile company.

I read an AES paper on intersample clipping a while back and it's something (as I recall) that's easy enough to test for. But it's also my understanding (please correct me if I'm wrong) most reasonably well mastered music avoids intersample peak problems by design. I believe it only requires leaving a dB or two of headroom to avoid the problem completely?

Obviously, there are some hot mixes that emphasize "loud" above all else. My experience with those, however, is they're so horribly distorted on the CD, even a really lousy DAC's flaws are inaudible.

It depends on the DAC architecture (and especially how integrated the design is) but it might be as simple as setting a single resistor value to prevent intersample clipping on playback. In that case, there are no corners to be cut or extra cost involved. The DAC designer need only choose the right gain values relative to 0 dBFS and the power supply rails. Some designers might want to squeeze out every last dB of S/N and risk intersample clipping which to me is a poor design trade off. I guess it depends on how much of an audible problem real world recordings are.

I'll look into incorporating a standard intersample clipping test for the dScope so I can hopefully include it in my suite of DAC measurements for future tests. It might be one of those less obvious things that helps define the "best" from the "nice try" products. I might have missed it, but it's not something I've seen measured in typical reviews.

 

[DonH50]

OdBFS shouldn't be a problem for ANY converter. It's when you have inter-sample peaks that the distortion becomes a problem. Most "good" converters can handle these peaks really well. It's the poor designs and the ones where the manufacturer has cut corners to meet a certain price point where hot signals become a problem.

I really wish we could meet with a white board and maybe a beer; I suspect you know all I am about to say and we could skip the board. I started to try to explain but it got too long and deep. There's no such thing as "inter sample peaks" over time; some peak will exceed the dynamic range at the sample point eventually. If properly filtered, you can't get a peak over about 1 lsb over FS without clipping, and then 0 dBFS no longer applies -- it's +1 dBFS or whatever. Since we are talking DACs, not ADCs, there's no way to generate peaks, inter-sample or not, greater than 0 dBFS at the DAC's output. Well, for a perfect DAC, and in the long term...

Here's what I believe, and have measured, and you can undoubtedly go to ADI and other web sites for more information: DACs glitch. Real-world designs have overshoot, ringing, and output glitches that exceed 0 dBFS, sometimes significantly. Those signals can overdrive and saturate the output buffer amplifier, causing all sorts of nasty distortion. I suspect that is what you are hearing with "hot" signals; the DAC itself may be part of the problem by generating the glitches, but the amplifier stage(s) after it muck up the signal by saturating and not recovering quickly.

Unless you are speaking of "DAC" as most audiophiles use the term, including the preprocessing, actual digital-to-analog converter, and output filters and buffer amps, all as "the DAC".

Most converters (ADCs and DACs) are tested at -1 dBFS. For various reasons performance tends to fall a little right at full-scale.

Your last line is spot-on, and of course applies to many things in many fields... - Don

 

[NwAvGuy]

Unless you are speaking of "DAC" as most audiophiles use the term, including the preprocessing, actual digital-to-analog converter, and output filters and buffer amps, all as "the DAC".

I can't speak for Bruce, but the above is the definition I'm using--i.e. the whole DAC "box" from digital input to analog output. It's been proven intersample clipping is real under some circumstances, I just don't know how much of a problem it is in the real world.

 

[DonH50]

You both probably are. As a DAC designer (among other things) I keep getting tripped up by audiophile terminology. I'd plead ignorance except that I've been an audiophile longer than I've been an engineer...

 

[NwAvGuy]

You both probably are. As a DAC designer (among other things) I keep getting tripped up by audiophile terminology. I'd plead ignorance except that I've been an audiophile longer than I've been an engineer...

So, as a DAC designer, do you agree intersample clipping is easy enough to avoid? It seems to me even a cheapo DAC (end-to-end) can avoid the problem without any significant compromises or extra expense?

 

[DonH50]

My work has been a world away from audio -- mostly lower resolution (8 - 16 bits) and much higher frequencies (1 - 10 GS/s and above). I do not know what "intersample clipping" means to you -- could you define it, please? I should probably read up on it before babbling further.

For an ADC, you need to sure input signal peaks never exceed (ideally) 1/2 lsb above or below the ADC's full-scale range to avoid clipping. For a DAC, presumably the system knows the limits and will never apply more than a FS input to the DAC (it is all digital at that point). At the DAC's output, glitches and other things that exceed FS happen all the time. What needs to be done to avoid problems after the DAC can range from simple to very complicated, so I would not agree in general that it is easy or cheap.

The impact of non-ideal DAC output on the rest of the "DAC" may be an interesting discussion, but I am wary of taking this thread too far off-topic so perhaps Bruce can chime in before I say any more -- I have gotten in trouble for deviating in the past.

 

[NwAvGuy]

My work has been a world away from audio -- mostly lower resolution (8 - 16 bits) and much higher frequencies (1 - 10 GS/s and above). I do not know what "intersample clipping" means to you -- could you define it, please? I should probably read up on it before babbling further.

This might help:

http://www.tcelectronic.com/media/Level_paper_AES109.pdf

 

[Bruce B]

For an ADC, you need to sure input signal peaks never exceed (ideally) 1/2 lsb above or below the ADC's full-scale range to avoid clipping. For a DAC, presumably the system knows the limits and will never apply more than a FS input to the DAC (it is all digital at that point). At the DAC's output, glitches and other things that exceed FS happen all the time. What needs to be done to avoid problems after the DAC can range from simple to very complicated, so I would not agree in general that it is easy or cheap..

This is one of the last resort tools that a mastering engineer has. When all compression and gain staging has been done and the client still wants it "louder", on a very well made converter, you can clip the inputs depending on how the converter will handle it. Some converters can handle clipping really well where you really can't hear it. On the other side, some converters will distort like crazy any time you reach even near 0dBfs.
Like I said, it all depends on how well the manufacturer has paid attention to detail.

 

[DonH50]

Are we talking ADCs or DACs? Clipping the inputs of an ADC is different than clipping the output of a DAC, though there are some similarities on the analog side.

I'll read the paper, thanks.

 

[NwAvGuy]

Are we talking ADCs or DACs? Clipping the inputs of an ADC is different than clipping the output of a DAC, though there are some similarities on the analog side.

I'll read the paper, thanks.

I think we're all mostly talking about DACs. At least I am. But the "cure" to this problem is supposedly on the mastering side of things.

The authors of the paper used artificial test tracks to exercise DACs in various CD players. They all had trouble to some degree. What they didn't do was correlate the problem to real world music.

The theory says you can have 2 successive samples that are at less than full scale that represent a peak greater than full scale. And I would think DAC designers would account for this? How a DAC deals with heavily clipped data (3 or more samples at full scale) is perhaps a slightly different question.

 

[DonH50]

I understand the issue, and I have done mastering in the past and am generally familiar with the process today. If I get time maybe I'll draw pictures. The problem is the DAC has no way of knowing, barring a predictive decoder before it, that samples are going to be for a signal that's more than full scale. For example, maybe it's a flat-topped wave (square wave) and you don't mean for it to have a rounded top? That said, such decoders exist, and provide the digital equivalent of a limiter circuit. I can think of several ways the DAC itself might deal with the issue, including a few extra lsbs at the top and bottom (over range bits), and filter circuits that allow overshoot in the wave form at the output so the signal is allowed to exceed full-scale after the DAC. I believe the majority of DACs these days are flavors of delta-sigma modulators, and with higher-order loops the signal has to be limited for stability. In those cases, a few samples "over the limit" can be handled. The design will dictate how well they are handled. You can build in a few extra bits in the DAC's modulator to provide headroom (remember the modulator in a DAC is all digital; it is usually sampled-analog in the ADC for a delta-sigma design). I suspect the better ones do, and the corresponding headroom in the output buffers, while the lesser ones do not.

Music gets complex enough to approximate noise signals, and the NPR test is one of the most stringent tests of data converter linearity. Fortunately, we tolerate quite a bit of distortion before it gets noticeable. In fact, I think most of us are more likely to notice its lack (when taken away) than its presence.

Interesting subject! - Don

 

[NwAvGuy]

Music gets complex enough to approximate noise signals, and the NPR test is one of the most stringent tests of data converter linearity. Fortunately, we tolerate quite a bit of distortion before it gets noticeable. In fact, I think most of us are more likely to notice its lack (when taken away) than its presence.

Interesting subject! - Don

Yeah it is. I'm looking forward to doing some of my own tests. I think Noise Power Ratio is more used in telecom than in audio. I'm not sure if that's due to the higher resolutions (bits) in audio or some other reason. But yeah, it's considered the "acid test" in some camps.

I met with an Analog Devices applications engineer the other day on a non-audio project. He said he hardly gets any questions about high fidelity DACs, Codecs, etc. He said all that stuff is seen as a commodity and companies usually design in whatever is cheapest (which generally rules them out), lowest power, in the right form factor, etc. He said their DSPs get a lot more attention for audio. It sounds like most of their ADC/DAC business is medical, aerospace, instrumentation, etc.

Next time I see him, I'll ask about intersample clipping just for fun and see if I get a blank stare?

 

[DonH50]

Yes, NPR was developed for, and still used mostly by, the telecom crowd. It's not only a hard test to perform well upon, it's also a hard test to set up. Most folk use ENOB based upon the IEEE Standard, at least that I have seen.

He'll probably know the term; I tripped up on its application to a DAC. It's a problem on the ADC input, especially in pulse systems (e.g. RADAR).