Higher Sample Rate or possibly Better Converters?

[mpgbari]

I'm still quite new to this game and am looking to "upscale" my current set up. I primarily record acapella music and am in need of recording 6-8 simultaneous audio tracks. I'm currently running on Digi 002 which as you all know has only 4 mic in's. I plan on adding (soon) a PreSonus M80 8 channel mic pre. This will give me mic ins I need. Now for the big question:

I can run the M80 directly into my ProTools session and let the 002 handle the AD conversion. This would allow me to record sessions up to 96 kHz. I realize that the ADC's on the 002 aren't the greatest but at least decent.

I was toying with the idea of getting a separate AD/DA converter, perhaps the RME AD8-DS. I'd run the PreSonus M80 pre's into the RME and then lightpipe optical out into the 002. From what I understand, the best sample rate I can get through lightpipe is 48 kHz.

So my question is, which might serve me better, running directly from the PreSonus M80 into 002 (which can support up to 96 kHz) OR, run the M80 into the RME converter which will ultimately deliver only 48 kHz.

I know higher sample rate doesn't necessarily mean better, I'm just wondering if the 48 sample rate I'd get out of the RME converters would be better than a higher sample rate converted by the 002 (all other things being equal)?

Thanks so much for your input!!!!

[Smithcok]

Maybe consider getting the black lion audio upgrade?

I dont think just changing the sample rate up will give you what you want.

[Tarktones]

If you're going to CD, I'd say it's probably simpler to keep your recording at 44.1k. I'd say if you can afford some nice converters, going in lightpipe at 44.1k would give you the best results. You might want to consider investing in the Black Lion Audio 002 Mod. Just a thought.

*edit* I took a long time typing and Smithcok's post snuck in. hahaha Oh well, there's two votes for the BLA upgrade.

[mpgbari]

I had briefly looked into that option as well ... I think that's what I might do as it sounds like everyone has been quite pleased with the results, and, it's really at a pretty reasonable $. The only concern is parting with the 002 for that long :)

Thanks so much guys!

[DAS]

I've heard some users claim that higher sample rates sound significantly better on their 002's. I always question how rigorous testing methods are for these sorts of thing so I'd take it with a grain of salt....but since you have one you can certainly try that. If it makes a big difference for you then maybe you don't need to buy an external converter.

[Tarktones]

I had briefly looked into that option as well ... I think that's what I might do as it sounds like everyone has been quite pleased with the results, and, it's really at a pretty reasonable $. The only concern is parting with the 002 for that long :)

Thanks so much guys!

I was only away from mine for about a week and a half. It came back and I was quite pleased with the results.

[Rad]

Yes, the theory is pretty clear that higher sample rates do not make any difference whatsoever in terms of accuracy of capturing the waveform. 44.1 is the best to go for CD recording, so I vote for that too.
That being said, some shoddier converters _can_ sound different at different sample rates. Not because the theory is wrong, but simply because the convertors are crappy. Typically where things go wrong most often is in the anti-aliasing filter, which has to be audibly transparent and not to roloff in the audible range, or cause group delay in the trebles, or in any other way de-phase the signal. And this is hard to do, hence the difference.

[wfhscoyotes]

What exactly does the Black Lion upgrade do?

[Tarktones]

Replaces all of the analog stages (op amps, mic pres, etc...) and improves the word clock.

[Smithcok]

If you're going to CD, I'd say it's probably simpler to keep your recording at 44.1k. I'd say if you can afford some nice converters, going in lightpipe at 44.1k would give you the best results. You might want to consider investing in the Black Lion Audio 002 Mod. Just a thought.

*edit* I took a long time typing and Smithcok's post snuck in. hahaha Oh well, there's two votes for the BLA upgrade.

I'm pretty sure I've done that to you before so at halftime its:
Smithcok - 1
Tarktones - 1

The Black Lion audio makes a fairly substantial difference imo. Get that.

[Audioholic]

I am curious as to how the New 003 will compare to the Mods on the 002. If there is still a substantial difference.

mpgbari,
If I were you, and thinking of dropping close to 2 grand on some mic pre's, I would seriously consider adding something with a digital card, so you can go digital into pro tools, and still maintain your 8 analog ins.

http://www.sweetwater.com/store/detail/ISA428/

this has an option of adding a digital card, though in the long run it will be more expensive, you will get 4 nice pre's and an additional 4 in to use with anything else going digital into pro tools. Or, take a look at the Digimax line from presonus. Paying close to 1800 on a pre that will run through the 002's converters isn't my first choice. I bypass the 002 analog stage as much as I can, not that its horrible, but with a better digital converted signal, it does sound better.

Take advantage of the adat in on that sucker, maximize your input capability. thats my 2 cents.

[dpd]

Yes, the theory is pretty clear that higher sample rates do not make any difference whatsoever in terms of accuracy of capturing the waveform.

nit: Above about 20 Khz, higher sample rates are required for accuracy


44.1 is the best to go for CD recording, so I vote for that too.
That being said, some shoddier converters _can_ sound different at different sample rates. Not because the theory is wrong, but simply because the convertors are crappy.

Agreed - Depending on the design, converters can do some very odd things. For example, they may have have artifacts at one set of sample rates (e.g. 44.1 / 88.2) and have zero artifacts at other sample rates (e.g. 48 / 96). I have confirmed this via direct spectral comparisons of some converters. There are other non-linearities, there are different noise-shaping filters in different converters, overall sampling timing accuracies can be different.


Typically where things go wrong most often is in the anti-aliasing filter, which has to be audibly transparent and not to roloff in the audible range, or cause group delay in the trebles, or in any other way de-phase the signal. And this is hard to do, hence the difference.

I don't think any anti-aliasing filter (AAF), when protecting a native sampling 44.1 Khz A/D can be totally transparent (in terms of lack of amplitude rolloff and constant group delay) up to the Nyquist rate. The AAF design challenges and the need to obtain accurate conversion to 16 bits and beyond forced the industry into the sigma-delta designs where the AAF has essentially zero impact in the audible range. But, I don't necessarily agree that the AAF is the big problem.

[Rad]

nit: Above about 20 Khz, higher sample rates are required for accuracy


22.05 kHz, to be exact. But since human hearing doesn't go that high, for all practical purposes 44.1 should suffice. (60% of people over 25 don't hear over 17kHz, 0% hear above 20k - ref. Ortofon hearing tables; See also Nika Aldrich, Digital Audio explained for the Audio Engineer.)



I don't think any anti-aliasing filter (AAF), when protecting a native sampling 44.1 Khz A/D can be totally transparent (in terms of lack of amplitude rolloff and constant group delay) up to the Nyquist rate.

Agreed. With oversampling the filter can be made to roloff way higher than the audible range, which is what you essentially say about linear-phase digital anti-aliasing filters used on converters with sigma-delta modulators.



But, I don't necessarily agree that the AAF is the big problem.

For modern convertors with sigma-delta modulators, it remains critical because the non-linearities you mention above are essentially absent in them - sigma delta's are either one-bit or multi-bit with randomized amplitude ranges, so they are either deterministically linear by construction (as is the case with 1-bit sigma-delta modulators) or stochastically linear (for the multi-bit). However, despite this when the signal passes through the reconstruction filter at the end, it is always an _analog_ filter as opposed to the digital AAF at the front which can be made to be linear-phase. To put it simply, the step-function type wave coming out of the DAC needs to be "smoothed" by filtering the odd harmonics comprising the "kinks" between samples in order to reconstruct the original waveform. So, the quality of the reconstruction filter (steep enough to attenuate above-Nyquist frequencies, yet audibly transparent w/o much group delay) remains very critical as to what we hear, in terms of phase shift and frequency response.

Regarding noise shaping, agreed that different manufacturers do different things and that can create artifacts. The noise shaping filter implements what is called an ARIMA process to the quantization error, by essentially making it a moving average with lags. The number of lags and their coefficients are critical to how "smooth" the result, so it takes a lot of calibration to get a converter that sounds "natural".

Along this line, it is almost scary to think how many steps digital audio passes through during processing. And the bad thing is that, given this, it is essentially possible to make it sound any way you want while still formally being accurate to the original wave. Say is digital audio with 8-lag noise shaping more "accurate" than one with 6-lag noise shaping? Tough question, damn it! But anyway, my point was that convertors make a difference.

[brianbfw]

Without a doubt, converters are a HUGE HUGE factor in sound quality, especially with acoustic instruments, voices, etc.

I play classical guitar. I personally use Apogee Rosetta 200. I believe the Apogee coverters sound better then the Digitech coverters on Pro Tools rigs. Nothing against Pro Tools, but that's just my tainted ears. I've recorded many sessions on both, and to me, the Apogee coverters allways produce wider stereo image, tighter bass, very transparent, and an overall clarity in the mix i can't see to hear when using of other coverters.

[DAS]

... better then the Digitech coverters on Pro Tools...

I think you meant "Digidesign" converters. No worries. Just wanted to clarify so nobody watching from the sidelines gets confused. Good discussion. Feel free to resume.

[mpgbari]

mpgbari coming back at you ... first, thanks to everyone for their input (although, being a newbie, some of it was a little over my head :)

I spoke to a guy at Black Lion Audio today ... he claims the "converters" in the 002 aren't really the problem; it's a combination of the word clock (he described it as being a little off is like taking a fuzzy picture), and the analog stages; they "muddy up the waters before the signal makes to it the converters."

So, their claim is to revamp the analog stages before the signal hits the converters and put in a new clock. Additionally, the analog outs are modded so that when you listen back and monitor your mix you're in better shape. So far from what I've read (and heard on their site), people are pleased with the results. That said, I could now run say 6-8 mics into the modded analog stages from a good mic pre (I liked the features on the PreSonus M80, especially the "saturation adjustment" and hopefully have a nice clean result. I'm wondering if anyone thinks it is better (quality-wise) to convert to digital outside the 002 (like with the PreSonus DigiMax 96( and run lightpipe in knowing that that data would be limited to 48k.

Any additional thoughts ... once again, thanks for the input.

[Audioholic]

I think if you want to maximize your input count, then going with something like the digimax is the way to go. You could get the digimax and the mod for cheaper then the mp80, and have more inputs at once. I haven't used the mp80 so I don't know if its worth it, but I know the digimax is very decent and you will now how 12 mic inputs and 4 line's to spare.

I just know that even with the mods, if you go in all analog, you will be limited to 8 inputs period (apart from a spdif of course)

[dpd]

22.05 kHz, to be exact. But since human hearing doesn't go that high, for all practical purposes 44.1 should suffice. (60% of people over 25 don't hear over 17kHz, 0% hear above 20k - ref. Ortofon hearing tables; See also Nika Aldrich, Digital Audio explained for the Audio Engineer.)

yup - assuming you have a perfect filter. We tend to use about 2.25:1 or 2.56:1 ratios to enable us to design cost-effective filters (sonar processing).


So, the quality of the reconstruction filter (steep enough to attenuate above-Nyquist frequencies, yet audibly transparent w/o much group delay) remains very critical as to what we hear, in terms of phase shift and frequency response.

Or, oversample on the D/A end to make a wide transition band and use either 1st order or gaussian filters.


Regarding noise shaping, agreed that different manufacturers do different things and that can create artifacts. The noise shaping filter implements what is called an ARIMA process to the quantization error, by essentially making it a moving average with lags. The number of lags and their coefficients are critical to how "smooth" the result, so it takes a lot of calibration to get a converter that sounds "natural".

Along this line, it is almost scary to think how many steps digital audio passes through during processing. And the bad thing is that, given this, it is essentially possible to make it sound any way you want while still formally being accurate to the original wave. Say is digital audio with 8-lag noise shaping more "accurate" than one with 6-lag noise shaping? Tough question, damn it! But anyway, my point was that convertors make a difference.

Any data on how that effects a true random waveform in terms of effective bits or other error-measurements? One of our system analysts tried to do a Matlab model of the internal guts of the sigma-delta A/Ds (the noise shaping, primarily) we use on a system in order to model a huge transient step would smear in the converter. We never did get the internal design details to do a good model, but the signal really got smeared for a short time until everything converged. (I don't remember the number of affected samples)

Thanks for the info

[Rad]

Dpd,

What you say about the modeling of the process by folks in your company is interesting. Do you work in an audio engineering company, as it sounds? Because to me what you describe sounds like impulse response analysis, for which Matlab is generally well suited. (By the way, Stata corp now has a competing optimized matrix language called Mata - I'd be curious who wins at the end). But getting the right coefficients for IR is crucial, so no wonder that without knowledge of the design of the convertor, the model was hard to do. What would be interesting is to look at the recovery pattern and speed of convergence, because I bet that's where different designs make a huge difference. (By a twist of fate, now I'm spending 90% of my time dealing with exactly this kind of analysis, however, not for electrical but for economic systems which is considerably less fun - in practice we work with data that almost never give "good" results.)

Rad

[dpd]

Dpd,

What you say about the modeling of the process by folks in your company is interesting. Do you work in an audio engineering company, as it sounds? Because to me what you describe sounds like impulse response analysis, for which Matlab is generally well suited. (By the way, Stata corp now has a competing optimized matrix language called Mata - I'd be curious who wins at the end). But getting the right coefficients for IR is crucial, so no wonder that without knowledge of the design of the convertor, the model was hard to do. What would be interesting is to look at the recovery pattern and speed of convergence, because I bet that's where different designs make a huge difference. (By a twist of fate, now I'm spending 90% of my time dealing with exactly this kind of analysis, however, not for electrical but for economic systems which is considerably less fun - in practice we work with data that almost never give "good" results.)

Rad

We had to model the impulse response of an entire sonar system, starting with the A/D and some interesting signal processing involving some AGC, etc. The time smear of the system, starting with the sigma-delta A/Ds impacted the performance during the settling time caused by the impulse.

Matlab owns our shop - it would cost me huge to change horses at this point.