Analyzing A/D D/A Converters on specifications- not personal bias

[Sacalait]

Is there a way of looking at specifications of converters to determine that one will "sound better" than another? For example, the Prism high-end converters must have some specs that set it apart from the sub $1K stuff. It would be great to be able to read specs and know what quality you're getting as apposed to wading through boatloads of posts with personal bias.

I understand that specs aren't "everything". (My signature is "results beat logic- always".) I have a Fender American Strat I just almost gave away because it sucked so bad. But even in that case had I read a spec of the "pickups are wound like a 1999 Stratocaster" I may have opted for something else.

[Audiop]

I'm afraid the standard numbers you can see is not enough. Sure you want a certain level of performance as judged by standard measurements but it seems like it's not enough, not IME at least.

That said there is no mystique going on, it's just about finding all the little things that have influence on the sound and measure that in a proper way. I have done some listening tests and measurements and I am not sure how to measure to get all necessary info.

You can be sure on one thing though, if you can hear it you can measure it. It's "just" about using the right stimuli and right analysing tools.


/Peter

[audiovisceral]

dynamic range is a good one. top converters pull 115-125 db of dynamic range these days. average ones 100-115 db.

this just relates to how cutting edge their chips are though. the newer 120+ chips do sound more detailed imo, but there are too many other bigger, unquantifiable characteristics to consider as well, so whatever. gotta use your ears.

[Audiop]

If dynamic range would be a limiting factor then there would be audible noise, which normaly is not the case, or?


/Peter

[audiovisceral]

Originally Posted by Audiop

If dynamic range would be a limiting factor then there would be audible noise, which normaly is not the case, or?


/Peter

ya i think higher dynamic range = higher sound to noise. more detailed transients, etc. dunno exactly.

anyway, 24 bit has a theoretical maximum dynamic range of 144 dB, compared to 96 dB for 16 bit. we haven't maxed out on 24 bit yet. 20 db to go. theoretically.

Audio bit depth - Wikipedia, the free encyclopedia

again, it's really kind of just a statistic though.

[mxeryus]

Until all manufacturers agree on a certain standard, they will all use their own specific (no pun inteded) figures, measurements and facts. So if a certain converter has an excellent S/N ratio but bandwith limited to 12kHz and just a picosecond delay at 44.1 kHz, does it make the choice easier ?
I personally think that there are no really bad converters (except from the two I made in 1986). The total concept of the converter (including avoiding jitter, for example) is more important. Does it interface or interfere with the rest of your setup? Does it have it's own character? Any good dealer will let you try a certain convertor.
Look at Dan Lavry's site for some excellent background http://www.lavryengineering.com/foru...ital_Audio.pdf

[Audiop]

Yes, AFAIU dynamic range is the distance between max output, 0dBFS, and the intrinsic noisefloor of the converter. This means that for program material that peaks out close to 0dBFS and are not very dynamic you will never hear the noisefloor even in redbook CD format.

A 24bit system gives us a useable 20-25dB extra. If you have program material that peaks out at 0dBFS you will need to play 120-125dB peak SPL to hear the noise in the weaker parts of the music. Most people don't have amps and speakers that handle that.

A typical monitor may have 88dBSPL/2.83V/1m sensitivity, A 200W amp will give aprox. 111dB depending on the room. You'll need an amp that put out more than 1000W to peak at about 118dB.

The situation may change if relatively dynamic material is recorded at -20dBFS peak or so, since then you'll have to stepmup the gain which brings the noisefloor up to where it may start to be audible.

Mic noise is way much severe than converter noise and the major enemy for high dynamic range.

Well, that's my take on it at least! :-)


/Peter

[Audiop]

Originally Posted by mxeryus

Until all manufacturers agree on a certain standard, they will all use their own specific (no pun inteded) figures, measurements and facts. So if a certain converter has an excellent S/N ratio but bandwith limited to 12kHz and just a picosecond delay at 44.1 kHz, does it make the choice easier ?

Well there are standards! :-)

Only we need more of them I guess one could say.

I have tried to get help to measure things that is not really explained by typical standard measurements but none seem to be interested.


/Peter

[iomegaman]

You can use an outstanding converter in a very unimpressive way...screw up the isolation barrier with psu noise and what-not...the specs on Mackie stuff can be very impressive...if you get my drift.

Not everything labeled as excellent is.

YMMV

[Jim Williams]

Take 2 converters using the same converter chip and they sound different.

Mytek and Apogee are examples.

The analog makes the sound difference, clocking less so.

The problem with "Pro" audio is you often don't even find the best converter chips used, it's mostly mid-level stuff. How many use the PCM1792 DAC?

Only a couple.

Jim Williams
Audio Upgrades

[myrtlebacker]

Here's how I test A/D D/A - without personal bias (and without ears).

I use Audacity and load the attached test file. I play it back and record it at the same time. I try to keep the record level at 0.8 but it's not important to get it perfect. Close is enough.

Then I just visually inspect the differences, first from a zoomed out view. You can find coarse problems in converters very quickly with that. Converters just hate sawtooths Sometimes funny things happen, like the waveform sort of twisting or dropping down to the floor.

Otherwise the deformation of the zoomed in waveform gives a clue as to the quality of the filter implementation. Also look a the end and the start of the waveforms.

[Audiop]

myrtlebacker,

about this:

Sometimes funny things happen, like the waveform sort of twisting or dropping down to the floor.

Don't you think that what you have seen is drop outs?

I've had clicks and pops with firewire interfaces on my laptop and without ASIO I get lots of clicks and pops.


/Peter

[u b k]

Originally Posted by Audiop

You can be sure on one thing though, if you can hear it you can measure it.

funnily enough, i don't share your sureness on that notion.

for starters, you have to deal with the fact that hearing (like all the senses) is a subjective process that involves reassembling, recompiling, and rejecting a ton of data, and there is a tremendous amount of 'filling in the blanks,' more than most people would care to admit. we often hear things that do not actually exist in the source.

then there's the reality that two people often hear and describe the exact same sounds in different, often incompatible ways. how do you compensate for that?

reality may well be reducible by mathematics, but our experience of it is not.


gregoire
del
ubk
.

[Jim Williams]

Originally Posted by u b k

funnily enough, i don't share your sureness on that notion.
for starters, you have to deal with the fact that hearing (like all the senses) is a subjective process that involves reassembling, recompiling, and rejecting a ton of data, and there is a tremendous amount of 'filling in the blanks,' more than most people would care to admit. we often hear things that do not actually exist in the source.
then there's the reality that two people often hear and describe the exact same sounds in different, often incompatible ways. how do you compensate for that?
reality may well be reducible by mathematics, but our experience of it is not.
gregoire
del
ubk
.

I tend to agree with that. I own and have used some of the best test gear, Audio Precision, HP network analyzers, etc. These are very powerful tools. They will tell you when something is wrong. They won't tell you when something is right, only in their limited ability to measure audio are they useful. Once the errors are removed from an audio design the real work begins, the listening tests.

Jim Williams
Audio Upgrades

[David Spearritt]

Originally Posted by Jim Williams

The analog makes the sound difference, clocking less so.

Not as it should be. A converter should impart no analog colouration, that's not its job.

Dynamic range is of little consequence, who cares if there is a 10db difference at the bottom of 110dB range when all your signal is confined to the top 10dB. For classical music it is of little importance as well because the concert hall noise floor is still up at minus 80dB.

Clocking accuracy and jitter removal are by far the most important specs that influence sound quality. I am amused by a lot of "high end" converters that don't publish such data. Its hard to take the manufacturers seriously in such cases.

[Jim Williams]

We wish it was so simple, it's not. A dac design using onboard filters and output opamps in a voltage output design sounds different than a high quality current output dac followed by excellent current to voltage conversion. Even a BB PCM 1704 current output dac sounds different than a PCM1792 current output dac, even when using the same data input, digital input reciever and analog stages. I just did some listening comparisons between them last night, it's not hard to detect. Well, maybe with some of the overcompressed, clipped modern rock stuff. I don't go there though. MP-3 is usually good enough for that.

Jim Williams
Audio Upgrades

[Gearhero]

Jim, maybe you post some level matched un-labled examples so we can hear the difference? The best thing to do would be to post at least four examples made with two converters (A and B). See if people can match which examples came from which converters.

[living sounds]

Originally Posted by Gearhero

Jim, maybe you post some level matched un-labled examples so we can hear the difference? The best thing to do would be to post at least four examples made with two converters (A and B). See if people can match which examples came from which converters.

You need decent converters yourself to hear these things with unedited recordings. There are some very good examples on the Mytek website where they used tape as a source, with Mytek, Prism and Lavry converters. They all sound quite different.

[softwareguy]

I think that Jim's point is really a very simple and straightforward one.

When you buy a AD or DA you are not only buying the chip that does the actual conversion, you are buying the whole circuit around that chip that supports the receipt of the audio/bits and the output of audio/bits. You run into all of the same issues that we have with all of our gear, because there are decisions to be made about impedances, etc., all of which affect the actual sound of the circuit, and which cannot be avoided in the real world. Two manufacturers using exactly the same chip can end up with different sounding gear because it's living in the same electronic world as all of our other gear and is necessarily not 100% transparent.

Different topologies/parts/specs will, as always, yield different results.

[Audiop]

Originally Posted by u b k

for starters, you have to deal with the fact that hearing (like all the senses) is a subjective process that involves reassembling, recompiling, and rejecting a ton of data, and there is a tremendous amount of 'filling in the blanks,' more than most people would care to admit. we often hear things that do not actually exist in the source.

What's your definition of source here?

If a converter or preamp or whatever changes the sound audibly I can assure you it's possible to measure this in the voltage presented to the power amp or the speaker. A speaker do not fill in any blanks, a speaker reacts to the current that is pushed thru the voice coil.. period!

then there's the reality that two people often hear and describe the exact same sounds in different, often incompatible ways. how do you compensate for that?

That is not the way it is. In blind tests and studies when people doesn't know at forehand what to think what you describe does not exists. Subjective preference may differ but that's another thing and even there people tend to have more or less similar preferences.

reality may well be reducible by mathematics, but our experience of it is not.

At this level, without a doubt.There's no mystique even though some people seem to like that idea! :-)



/Peter

[u b k]

Originally Posted by David Spearritt

Not as it should be. A converter should impart no analog colouration, that's not its job.

my converter's job is whatever i say it is. yours may be different.

'should' is a bizarre concept in the realm of the arts. hell, it's a bizarre concept in general.


gregoire
del
ubk
.

[Audiop]

Originally Posted by Jim Williams

I tend to agree with that. I own and have used some of the best test gear, Audio Precision, HP network analyzers, etc. These are very powerful tools. They will tell you when something is wrong. They won't tell you when something is right, only in their limited ability to measure audio are they useful. Once the errors are removed from an audio design the real work begins, the listening tests.

Jim Williams
Audio Upgrades

With all respect Jim, tools are not enough you must measure the right thing also and knowing what to look for.

A modern soundcard, a scope, a PC and some software lets you do most of what you can do with "old time" expensive specialised tools.

About listening tests, I assume you mean blind level matched tests?

Those use to be very telling.


/Peter

[Jim Williams]

Originally Posted by Audiop

A modern soundcard, a scope, a PC and some software lets you do most of what you can do with "old time" expensive specialised tools.
/Peter

These are not old time tools. This is not using a Sound Technologies 1700 or older HP audio sets.

Modern soundcards are limited by their converters, they do not use the best ones. Measurements are therefore limited by the resolution of those converters.

With AP, the converter's errors and limits are mathmatically cancelled out so readings go far below the resolution of modern A/D chipsets. In the analog domain, IMD and noise readings are seen down into the -130~140 db range, far below any soundcard's resolution.

With soundcards as the testing platform, you are basically measuring the sound card or it's residuals added to the DIT.

Jim Williams
Audio Upgrades

[Audiop]

Originally Posted by Jim Williams

Modern soundcards are limited by their converters, they do not use the best ones. Measurements are therefore limited by the resolution of those converters.

Yes but I hope you don't believe that you have a human being can hear spurious signals/distortion products at levels of -150dBFS? That's about the range we have with modern converters.

With AP, the converter's errors and limits are mathmatically cancelled out so readings go far below the resolution of modern A/D chipsets. In the analog domain, IMD and noise readings are seen down into the -130~140 db range, far below any soundcard's resolution.

Sure but no one hear anything that far down with typicall program material. Also AFAIK even a soundcard can be used with external notch circuits to increase practicall resolution.

With soundcards as the testing platform, you are basically measuring the sound card or it's residuals added to the DIT.

Jim Williams
Audio Upgrades

Yes but if you are measuring something that have less distortion (let's say HD) than your modern converter you are not likely going to hear it anyway.

I do not know how to measure everything but I'm confident that if something cause a change in waveform big enough to be heard, than there will be tools to hunt it down.

There is nothing I have ever heard or seen that indicates anything else.


/Peter

[myrtlebacker]

Originally Posted by Audiop

Don't you think that what you have seen is drop outs?

No, I am sure it's not drop outs.

I've had clicks and pops with firewire interfaces on my laptop and without ASIO I get lots of clicks and pops.

I am not even using windows (for audio).

[Duardo]

my converter's job is whatever i say it is. yours may be different.

I don't disagree, but wouldn't you agree that, in general, the design goal of most converters is to be as accurate as possible? There are certainly a few exceptions (the Burl and CraneSong units come to mind) but in most cases converters are designed with accuracy in mind, no?

[space2012]

Originally Posted by myrtlebacker

Here's how I test A/D D/A - without personal bias (and without ears).

I use Audacity and load the attached test file. I play it back and record it at the same time. I try to keep the record level at 0.8 but it's not important to get it perfect. Close is enough.

Then I just visually inspect the differences, first from a zoomed out view. You can find coarse problems in converters very quickly with that. Converters just hate sawtooths Sometimes funny things happen, like the waveform sort of twisting or dropping down to the floor.

Otherwise the deformation of the zoomed in waveform gives a clue as to the quality of the filter implementation. Also look a the end and the start of the waveforms.

nice test,...
test A/D D/A - without personal bias & without ears!

[Audiop]

Originally Posted by myrtlebacker

No, I am sure it's not drop outs.

I am not even using windows (for audio).

I assume you're runing a Mac? Drop outs occure in those as well. My point being these things are problematic with so many pieces of hardware and software that has to play nice with eachother.

Then I just visually inspect the differences, first from a zoomed out view. You can find coarse problems in converters very quickly with that. Converters just hate sawtooths Sometimes funny things happen, like the waveform sort of twisting or dropping down to the floor.

If you're running a continous signal and the waveform all of a sudden take a dive down to zero I can't see any other explanation than a drop out in the signal flow. And this is often how clicks/pops looks like.


/Peter

[u b k]

Originally Posted by Duardo

I don't disagree, but wouldn't you agree that, in general, the design goal of most converters is to be as accurate as possible?

i completely agree with that, and believe it's a laudable goal. i only take exception when the idea shifts from 'the typical design goal is' to 'all converters should'.

i would also note that the vast majority of converters fall well short of the goal of transparency, at least to my ears... and i'm including most of the expensive ones here. they may sound great, but they change the sound.


gregoire
del
ubk
.