Converter chips ?

[roon]

I´ve been trying to get the idea of the difference in quality of A/D converters.
I know that listening is the way but still I´d like to know the cold facts.

Fx I´ve been able to locate these facts:

ProtoolsHD use an AKM AK5394A chipset

Behringer ULTRAMATCH PRO SRC2496 use an AKM AK5381 chipset (it seems)

AKM chipset catalog


Now, is AKM generally a good chipset and does that imply that this new behringer stereo-converter is decent ?


It has been imposible for me to find detailed information about the ADC´s in M-audio FW410 and FW1814 hardware - does anybody know which chipset is inside, or how to find out ?


The general question of interest for everybody is:
Which chipsets are great and which are not so great ?

[maz]

Good question, I wanted to ask the same exact thing. For example, who does the convert chip on say RME multiface or Apogee rosetta 200 or lynx 2 is different vs say m.audio audiophile 192 or digi002?

Beside the chip, which one have a better API / drivers? I know that chip does make a huge difference, but also how the sound card communicates with your computer software/hardware also makes a difference. This communication is done by the API and the drivers, so that also has to be taken into consideration ( I don't know which ones are the best....)


Lastly, the sample rate also makes a difference. For example a 192 samples sound might sound better that a 96khz using the same card. (I don;'t know how much better, but it should sound closer in quality to the source.

My question is this though, would a RME 192 sound better that a Lynx 96??? RME although excellent is not as good as a lynx, but I wonder higher sample rme at 192 would be better vs Lynx recording at 96

Anyways, great topic

[Bolt Thrower]

I'm very curious about this as well, but it seems like this is going to be one of those very closely guarded secrets. It's not in the best interests of the converter brands (low end or high end) to discuss the details of the chips they select for any particular product. Kind of like showing the Wizard behind the curtain; or, perhaps the Emperor's New Clothes would be a better metaphor?

For example, Crystal Audio makes many A/D/A converter chipsets; some are nicer than others but the brand is the same. So when I observe that at least some MOTU units use Crystal Audio chips (or was it Cirrus?), as does Hoontech (a Korean manufacturer of low-cost gear), and also a large number of on-board soundcards from various PC motherboard manufacturers -- not to mention the legions of cheap soundcards out there -- it really begs the question of what makes better converters better. I believe that ADATs and MOTU units use the same brand of converter chips, and with all the shit I hear about the poor converters in an ADAT I really have to wonder.

It's certainly obvious enough that something like an Apogee Mini-Me completely destroys the A/D capability of MOTU, Hoontech, PC mobos, and Soundblaster type soundcards. I think Apogee use a programmable chip (something like a Xilinx type product), rather than an off-the-shelf converter.

It could be something like better error correction, interpolation, clocking, or application of psychoacoustics rules that make these higher end products actually sound better... but the question of separating that from the off-the-shelf converter chipsets (and it seems there really aren't that many manufacturers of these) seems crucial to better understanding the difference in technical value versus the marketing BS to which so many of us fall victim.

[Confusionator]

The specs are easily available at http://www.akm.com/ and http://www.Cirrus.com.

However, the "magic sauce" is always in the analog filtering, clocking, and power supply designs for converter boxes. The chips' specs themselves tell you next to nothing.

I mean, come on: do you really think that an E-MU sounds better than a Lavry, even though the E-MU uses a better-spec'd AKM part?

I hate to say it, but you have to use your ears to answer these questions. (And don't forget the advice of the forum members here--very helpful!)

[ddageek]

There is so much more than just the chip! Yes the chip maters but rember there is the clocking, analog circuit ect! Makeing judgement soly based on the chip is like sayin he used an Elam 251 therefore it must be an excellent vocal!
A designers choice is based on what he feels is the best for hus appliction (and in many cases there are other factors ie price point availability consistancy reliability ect)
as hard as it is ne really must listen before making choices not just look at the specks!

[Bolt Thrower]

OK great, so we've established that there's more to a good converter than the chipset. Does this mean that professional converters sound much better because of everything other than the chip?

If so, then why does the selection of chipset matter at all? It sounds like we may be saying it, ultimately, makes little-to-no difference?

Likewise, is there essentially zero benefit in considering the choice of chipset when comparing converters?

Many folks aren't able to buy all the possible products on credit, and then return the ones they didn't like -- or somehow obtain them all on loan from a store for detailed listening tests (which can involve the store placing an auth on your card for the purchase price of each item, which in turn locks up that dollar amount on your card, which in practical terms is much like going ahead and just buying them all).

I think what the OP is getting at is how to cut through the hype surrounding the various converter products, relative to the actual converter chips involved. Since it's obvious that the rest of the product is vital to the sound, I think it would be helpful if some of the Knowledgables around here could relate the concept of Marketing to the OP's question...

[roon]

As I said from the beginning, I know that sound is about listening
No need to point that out again and again, everybody knows that.

This is just about the specs and the overall factual difference of converters.

I don´t know anything about this as I always use my ear, but now I´m just curious if it´s possible to shine some light of understanding on the subject.

Thanks for all the useful information about what makes a good converter

So, if it´s not the chipset that makes Lavry better than E-MU, what is it then ?

[Confusionator]

As I said earlier, analog filtering, clocking, and power supply are the major factors.

[drosophila]

Originally Posted by roon

As I said from the beginning, I know that sound is about listening
No need to point that out again and again, everybody knows that.

This is just about the specs and the overall factual difference of converters.

I don´t know anything about this as I always use my ear, but now I´m just curious if it´s possible to shine some light of understanding on the subject.

Thanks for all the useful information about what makes a good converter

So, if it´s not the chipset that makes Lavry better than E-MU, what is it then ?

So, read the specs on the various chips and put together a spreadsheet comparing the various signal to noise/dynamic range/etc specs. At the end of the day, you'll probably at least have narrowed that list down to a few particular units - from there, all you can do is listen to them, weigh features and their importance to you, and ultimately pick one based on that criteria that suits your budget. I did this when I was shopping around and wound up picking the E-Mu 1820m.

Apogee may use Xilinx.. as I recall, so does RME. Given threads in this forum, however, many people will argue for the respective merits and flaws between them. The parts utilized provides half the picture, the rest is in the actual implementation.

Think of it this way: pick two people, give them each the same type of hammer, the same bucket of nails, a stack of wood, and tell them to build a bookshelf. I'd be willing to bet that even if they build virtually identical bookshelves, there will be minute differences that will make one guy say Person A did a better job and yet another say that Person B did things more to his liking.

[maz]

So does the conversion rate matter also?

FOr example does an audiophile 192 at 192khz sound better than rme recording at 96???

[Bolt Thrower]

Originally Posted by Confusionator

As I said earlier, analog filtering, clocking, and power supply are the major factors.

And Marketing is not a major factor?

You seem to imply that the chips are all so over-engineered that they can't hurt (or help) the design, no matter which ones are used.

[roon]

Originally Posted by Confusionator

As I said earlier, analog filtering, clocking, and power supply are the major factors.

I thought clocking was done by the chipset. Nice to know - how is clocking done ?

In what way is the powersupply important ? Something with stability ? Are firewire-supplied units a bad thing then ?

[roon]

I´m definately not surprised that a lot of other factors play a role in making a good converter. But still, when regarding the chipset:

Is it possible to make a short list of some bad chipsets and some good chipsets ?

Fx I´ve read a lot of negative stuff about the m-audio units, is this due to chipset or something else ?

[Bolt Thrower]

Originally Posted by roon

I´ve read a lot of negative stuff about the m-audio units, is this due to chipset or something else ?

I'd change up your mode of questioning.... otherwise you're likely to get more responses repeating the mantra that analog filtering, clocking, and power supply are the major factors.

Unless we're all being completely duped by product marketeers, M-Audio is clearly willing and able to cut a lot of corners when it comes to the rest of the design. Otherwise, how could their prices be so low?

I for one am still hoping that some brave soul with knowledge will clarify if in fact the converter chips themselves don't really matter, since those components are almost never discussed.

That said, Confusinator may be 100% correct. In which case it seems to follow that converter advertisements don't mention the chipset used because they no longer make a huge difference.

[Duardo]

I don't know if anyone would say that the chipsets don't matter, but they're certainly not all that there is to it and probably not even the most important part of the equation, as long as they're decent...E-Mu and MOTU both like to advertise the fact that they use the same chips (I believe they've actually said 'the same converters', which seems a bit deceptive or at least misleading to me) as Digi's 192, but the three do not sound the same.

Apogee got their start making filters for the big expensive multitrack machines...and they certainly made a difference, even though the converter chips themselves didn't change. Certainly the analog circuitry, power supply, clock, etc make a huge difference in the quality of a converter...that's certainly why the better ones are as expensive to make as they are, considering the relatively low cost of the chips themselves.

-Duardo

[ddageek]

Guys what I was trying to say is chip selection is only part of the equation and to make generalizations based on one component is stupid! Each designer has his own biases for his choice of chip or analog circut some sonic some not! The problem with compareing some designs using a specific chip and then makeing Generalizations about said chip is that you would have to listen to all designs using that chip and sort out which used other common design elements (as well as peak and make sure the manufacture hadn't made changes from their spec! even then you would need to listen to multiplle samples of each design!

[roon]

OK, I get it.

What is it with the powersupply then ? And the clocking ? And the analog filtering ?

How do I check it out ? (besides listening)

Do these aspects show in specs ?

[Bolt Thrower]

Originally Posted by roon

OK, I get it.

What is it with the powersupply then ? And the clocking ? And the analog filtering ?

How do I check it out ? (besides listening)

Do these aspects show in specs ?

The answer is most likely to be, "get an electrical engineering degree if you really want to understand"...

[Duardo]

Unfortunately, specs won't tell you how a converter sounds...most of them come pretty close to being perfect, spec-wise, and the ones that do spec out better don't always spec out better anyhow. So pretty much it comes down to listening.

-Duardo

[tINY]

http://www.semiconfareast.com/adc.htm
http://www.hitequest.com/Kiss/adc_terms.htm
http://www.maxim-ic.com/appnotes.cfm/appnote_number/748
http://www.maxim-ic.com/appnotes.cfm/appnote_number/641



Here is some light reading for you....

There is a lot of talk about jitter, but I'm not sure I buy it. As long as your peak jitter is less than 2 sample periods (or one at lower conversion rates) and your analog signal is well filtered, you won't hear any difference (and you'll have a hard time measuring it with traditional audio measurements).

The drivers in the computer that you are using to store the captured audio don't matter until they are so bad that they loose samples.




-tINY

[David Rick]

Originally Posted by tINY

There is a lot of talk about jitter, but I'm not sure I buy it. As long as your peak jitter is less than 2 sample periods (or one at lower conversion rates) and your analog signal is well filtered, you won't hear any difference (and you'll have a hard time measuring it with traditional audio measurements).

The drivers in the computer that you are using to store the captured audio don't matter until they are so bad that they loose samples.

You've got a basic misconception here. What you say is true if you're just talking about tranfering pre-existing data from the hard disk on one computer, to the hard disk on another via a digital link. But if you actually intend to listen to the audio during the transfer, then the jitter will certainly affect the performance of the DAC you're listening through. (Better designed DAC's will be less affected.) This degradation is not permanent. Usually data will reach the second hard disk just fine, but it will sound like crap on the way.

The worst case is recording though an ADC when the ADC is not the clock master. Jitter on the ADC's clock reference will translate to noise and distortion in the resulting digital audio. Think about it: if you sampling a changing waveform in the wrong place, then you get the wrong value! You can see this in the spectrum of a pure sine as side-bands around the main tone. Again, some ADC's are more sensitive to this than others -- it depends on the design. And some ADC's actually sound better running off a (very good) external clock than they do on their own. (Don't buy those. It indicates poor design.)

Getting back to the original subject of this thread, AKM, Cirrus (Crystal), and TI (Burr-Brown) make both good converter chips and fair-to-poor ones. The latter are cheaper, and intended for consumer equipment. The performance of the good ones depends a lot on the care used in designing the associated external circuitry. And it's not just the parts that matter. It's easy to screw up a perfectly good converter chip simply by having a poor circuit board layout.

David L. Rick

[maz]

I know I'm never gonna hear the end of this but:


Is there an audible difference in quality of AD, and DA conversion of DIGI 002 VS, Apogee mini me, VS multiface RME, vs Audiophile vs Lynx two????

Please please please let me know.

Also is there a reason to record at 192 vs 96? Is there any improvements? Eitherway I'm gonna down sample to 44.1..

Thanks,

[tINY]

It depends on the charater of the jitter. You can measure peak jitter and you can measure RMS jitter. Both are important.

But, unless you get jitter on the order of 10uS, you wont be able to hear it. You Probably wouldn't heat a 100uS peak jitter, as this is slight glitch in the 10kHz range. Decent clocks have jitter on the order of 50-100pS RMS and maybe 1nS peak. The frequencies associated with this are well beyond the range the LPF is passing to the filter.

They are also well above the passband of the DAC output filter.

So, if you are using a decent internal clock, you are fine. A little buffering goes a long way. If you rely on the AES word clock for your DAC reference, all bets are off.





-tINY

[dasbin]

I have heard designers say that the single most important sonic element in a converter is the layout and design. In other words, how clean can you keep the signal as you route it to the next element, how good is your ground, and how few caps can you pass it through, how clean is your power?
Look at the Benchmark DAC-1. It uses mostly standard parts found in way less-expensive converters - and in some cases far cheaper parts. The sound of that converter is almost purely in the design.

[Sam Lord]

Originally Posted by tINY

It depends on the charater of the jitter. You can measure peak jitter and you can measure RMS jitter. Both are important.

But, unless you get jitter on the order of 10uS, you wont be able to hear it. You Probably wouldn't heat a 100uS peak jitter, as this is slight glitch in the 10kHz range. Decent clocks have jitter on the order of 50-100pS RMS and maybe 1nS peak. The frequencies associated with this are well beyond the range the LPF is passing to the filter.

They are also well above the passband of the DAC output filter.
-tINY

First 3 sentences right. But 10uS jitter is way out of bounds unless it is totally uncorrelated with the signal, in which case it's just a noise source. But that is never, ever the case. Correlated jitter generates harrmonic distortion even at insanely small levels, like the 100ps range you mention. For a long time this wasn't measured and thus was not seen as a fundamental flaw with early AD and DA conversion.
Regards, Sam

[tINY]

Alright then, expain this to me....

If you have an analog signal filtered with a 20kHz, 2nd order filter, like a 96/16 ADC. That means that rise time is limited to about 18uS. So, you go from 10-90% of full scale in 18uS...

Alright - engineering fart....

Should be "you can't hear jitter on the order of 10-100nS". Still, it shouldn't be an issue with reasonable clocks.

If you can hear 100pS peak jitter, you are hearing something that's not encoded.

Back to the analysis - that's 52,000 steps (LSB) in 18uS for a 16 bit system. Which is one step in about 350pS. So a 100pS peak jitter corresponds to 1/3 LSB with a 20kHz band-limited signal in a 16 bit system.

It's still more than -100dB in a 24 bit system.

A 10nS jitter corresponds to -86dB.....




-tINY

[Bolt Thrower]

Originally Posted by tINY

Alright - engineering fart....

Thanks for taking the time to explain about jitter vs. audible frequency and noise level... I think I got something out of it.

[Sam Lord]

Originally Posted by tINY

...If you can hear 100pS peak jitter, you are hearing something that's not encoded.

Back to the analysis - that's 52,000 steps (LSB) in 18uS for a 16 bit system. Which is one step in about 350pS. So a 100pS peak jitter corresponds to 1/3 LSB with a 20kHz band-limited signal in a 16 bit system.

It's still more than -100dB in a 24 bit system.

A 10nS jitter corresponds to -86dB.....
-tINY

Thanks tINY, good numbers. (They look right to me...) So does anything important happen at 100ps jitter, -100dB? I wouldn't have thought so. Although, since with 24 bit ADCs folks often record at a lower VU setting, sensibly taking advantage of say 120dB DR in the converter. So, that jitter level sits just a little higher than the ADC's THD figure, and maybe more than 20 dB below the mic's THD + noise (when driven) measure. I *sure* don't know why we notice harmonic hash 20 or more dB below harmonic hash from even a great mic, but that is where the low-range ADCs measure, and the top models are way below that . The same situation occurs with DACs. Thanks for responding, Sam