My Question For Amir

[Steve Williams]

This comes from a thread in a different forum but I feel the answer is best served by putting it in this forum

Why do DACs lose their accuracy/have more distortion as you increase their sampling rate!

 

[Bruce B]

I guess my 768kHz and DSD5.6 DAC's are really out of wack!!

 

[amirm]

I think I am being setup for a beating!!!

 

[Ron Party]

Come on Amir. Admit it. You love a good beating every now and then. It's okay. You're amongst friends.

 

[amirm]

OK Ron... Please watch my back especially with this Bruce character....

Let me put my toe in the water first. Let's take a random high performance DAC from TI. In this case, let's use the Burr-Brown DSD-1792A (TI owns Burr-Brown). And then drill down into the spec to see its THD+N:

THD+N at VOUT = 0 dB:

Sampling Rate     THD+N Level
44.1 KHz            0.0004%
96 KHz              0.0008%
192 KHz             0.0015%

Let's look at a visual plot of this measured differently and note that for THD, the smaller the number, the better.

For grins, let's look at what happens to channel separation. This time, bigger numbers are better:

So at least objectively, all else being equal, the DAC struggles to produce the same performance at higher sampling rates.

 

[Steve Williams]

You said Bur-Brown DAC.

Is this problem for all DACs

 

[amirm]

Sadly, most DAC vendors do not document THD performance down to the level that TI does. They do sheepishly imply the same thing with a footnote that their THD rating is at 48Khz and such.

Here is another company, AKM, which provides some documentation. This is their highest performing DAC: http://www.akm.com/datasheets/ak4399_f03e.pdf

48Khz THD = -105 db
96Khz THD = -102 db

So the combination of noise+distortion is twice as high (3 db difference) at 96 Khz compared to 48 Khz. In this DAC, performance does not get worse (nominally) at 192 Khz but there is just not enough detailed testing to really trust that assertion.

 

[RBFC]

Is there any difference when you look at the newer 32 bit DACs, such as the Sabers or Crystals? Could 24 bit performance suffer less, even at the higher sampling rates? If the DAC is designed to perform on a higher bit sample, could working at a lower "exertion" keep its distortion in line at higher rates?

Lee

 

[Bruce B]

So the combination of noise+distortion is twice as high (3 db difference) at 96 Khz compared to 48 Khz. In this DAC, performance does not get worse (nominally) at 192 Khz but there is just not enough detailed testing to really trust that assertion.

So is the noise linear from say 0 - 22k as opposed to 0 - 352.8k?


Maybe we should at least listen to these DAC's instead of the looking at graphs. I know the PD MPS-5 measured pretty poorly.

 

[DonH50]

There are many noise sources in the system and in the data converters. Some increase with frequency (bandwidth, actually), some do not, and very few increase linearly. Thermal noise, for example, increases with the square root of bandwidth (or temperature). Flicker noise can increase linearly, and some phase noise terms (related to clock jitter) are supra-linear. Ideal quantization noise is only a function of the resolution (number of bits) for perfect data conversion. Something rather difficult to achieve in the real world, alas...

Nonlinearity (distortion) also increases as frequency increases, due to all sorts of reasons. Slew-rate limitations, more modulation from nonlinear capacitances, varying impedance effects, etc.

Realizing I am not Amir, I shall defer to him. - Don

 

[amirm]

Hey, I can use all the help I can get . Welcome to the forum by the way Don.

 

[DonH50]

Thanks Amir, but I think you can hold your own. Glad to help when I can!

I have friends working at the old BB (now TI) design center, some of whom I have known since the early 1980's. Sharp, sharp people...

 

[amirm]

Burr-Brown people were ahead of the game to be sure with stellar innovation in integrated DACs. Hope acquisition by TI did not change that spirit. I don't think the TI brand carries anywhere the same weight as BB.

 

[DonH50]

I actually did not hear that much about it either way so hopefully not. As for TI's rep helping them, certainly not in the audio world. I was a bit surprised they didn't carry the BB name further, sort of like Maxim did with Dallas.

My experience is with much higher-speed converters, but the basics are the same. I just can't implement a high-order delta-sigma loop, let alone the post-processing, at several GHz, though have worked on several low-order modulators up there.

 

[Old Listener]

Why do DACs lose their accuracy/have more distortion as you increase their sampling rate!

Two possible practical reasons:

1. Most if not all DACs capable of 24/96 and 24/192 operation are low bit (delta-sigma) types. The clock for the delta-sigma steps runs at a much higher speed than the sample clock or the bit clock. There is often a upper limit on the speed of that clock than requires compromise at high sample rates. So conversion at 192 KHz may involve coarser steps than conversion at a lower rate. This is spelled out in the data sheets of some DACs.

2. The capability of the DSP section of the DAC that does oversampling and digital filtering may also require compromises at higher sample rates. These DSP sections can perform some maximum number of adds and multiplys in a time interval. Better algorithms and longer polynomials often require lots of arithmetic. Up the sample rate and the DAC may have to use a lower quality algorithm or a polynomial with fewer terms. I don't have any evidence about the extent of this limitation in practice.

Bill

 

[muralman1]

Hey Amir, you are on the right path. I just argue 1x oversampling produce the purity of the music better than any other DAC type I have listened to.

 

[DonH50]

Quick comments:
1. 1x oversampling is not oversampling.
2. Oversampling can be done with normal, Nyquist DACs (it only adds 1/2-bit for every doubling of sampling rate, vs. roughly L+1/2 bits for an L-order modulator).
3. PCM (pulse-code modulated) or delta-sigma (a subset of PCM) DACs usually utilize oversampling and noise shaping.
4. Noise shaping PCM/DS DACs do not have to oversample (though the vast majority do).
5. In common usage, the debate is "oversampling DACs" vs. "non-oversampling" or "Nyquist" DACs.

FWIWFM - Don

 

[muralman1]

Don H50 you are right about 1x oversampling. I use that term from time to time because Audio Note uses it.

 

[DonH50]

LOL I meant to say "1x sampling"... Audio Note is probably trying to leverage off all the oversampling marketing and just stuck with the term.

One thing I left off is averaging multiple DACs to minimize linearity errors -- that can be done with any type of DAC as well.

No worries! - Don