Why run an AD1852 DAC at 128X instead of 512X?

[RocketScientist]

Does anyone know why a manufacture would choose to run the Analog Devices AD1852 DAC at 128 times the sampling rate instead of the recommended 512X? This is a commercial DAC that was measuring oddly, and thanks to help from Abraxalito and Jakob2 here, I discovered the 128X issue. But it's still a mystery why the designer didn't follow the datasheet requirements?

The DAC seems to be working properly below 20 Khz, but has strong image problems above 20 Khz. It's driven from a TI DIR9001. Is there some valid reason to "misuse" the AD1852 in this way?

[SoNic_real_one]

What is that "512x sampling"? In the data sheet I see only 2x, 4x, 8x. That means internal Sigma-Delta rate of 128x, 64x or 32x depending of the incoming SR.
If you mean "master clock at 512fs", that is 24.576MHz clock - someting different from "samplig rate". And it can run either 256 fS, 384 fS, 512 fS, 768, or 1024 fs.
Also, keep in mind, that usually, higher you go with overclocking in this simple DAC, higher the distortions go...

Now... you said "commercial" and "manufacture". Is that China made?

[RocketScientist]

Thanks for the reply. We're talking about the same thing. I said "128 times the sampling rate"--i.e. Fs.

According to Analog Devices, in the table you show above, the chip requires a clock of at least 256xFs for operation with 48 Khz audio (if I'm understanding the table correctly). At higher input rates, you can use the 4X or 2X modes and then 128xFs is "allowable".

In this DAC, pin 10 is pulled high which is forcing it into 4X mode. And, as I understand it, that mode is only supposed to be used with an input rate of 96 Khz. But it's also running in that mode for 44/48 Khz inputs. There's no SRC. And the DIR9001 is configured for a 128xFs master clock.

I'm trying to figure out if there's some benefit to "underclocking" the DAC at 44/48 Khz, if it's a mistake, or they just cut corners? The AD1852 is a relatively expensive DAC. If someone went to the trouble to use it over chips that are 1/3 the cost it doesn't make sense they would cut such an obvious corner unless there was a good reason? And yes, AFAIK, it's a China design.

[abraxalito]

There are two possibilities that I can see, and I think you covered them in your earlier thread. Its either incompetence or its because it sounded better to the designer.

To me its entirely plausible that it might sound better - lower frequency operation probably means lower levels of noise. Here I don't mean quantisation noise from the modulator - that probably would be higher in the audio band running at 128fs. I mean crosstalk from digital to analog. In my experience (not with this particular DAC but in playing around with various digital circuits) lower noise levels means less 'digital grunge' or sibilance. This I think is the reason there's a significant minority that likes NOS. Running 128fs rather than 256fs is a step nearer to NOS

[RocketScientist]

Thanks abraxalito. I know HiFiMan is a fan of NOS old-school DAC chips and also uses "alternative" filtering with less stop band attenuation in some of their products claiming better sound. But some of their products have very measurable distortions (like a rolled off high end) within the audio band that may account for the differences people hear. And, obviously, some people and companies do things just to differentiate their products from others--sometimes even when that involves making them perform worse.

In this case, I was mainly curious if there are technical (vs subjective) advantages behind the odd design. The sort of "digital crosstalk" you suggest would show up in a variety of measurements. It's likely more of a problem with cheap highly integrated DAC/CODEC chips where everything is jammed onto one die.

There are some good reasons for steeply suppressing the images and other junk above 20 Khz coming out of a DAC. As feedback falls beyond 20 Khz, high frequency distortion rises--sometimes dramatically. And you can get IMD distortion products down in the audio range-much like those from the twin tone 19/20 Khz CCIF measurement.

So, intuitively, I'd say all that "mirrored energy" reflected around/above 22 Khz is more likely to make things sound worse rather than better. But I also know certain kinds of filters can have audible effects as well. I'm typically a "follow the datasheet" kinda guy but I'm open to other options when they make sense.

[SoNic_real_one]

NOS for some people sounds better exactly because of the extra image frequencies generated by poor filtration that they percieve as a "fuller" sound. Personally I think that is garbage that wans't present in originally sound. In that specific DAC, 128x MCLK would mean 4X OS at 44.1kHz incoming samplerate... if it is working (the datasheet doesn't show it).
BTW, the DAC curves show that the "sweet spot" for that DAC is at 96kHz. 192kHz increase the distortion figures.

So.. who knows what was the real purpouse?

[abraxalito]

Originally Posted by RocketScientist

.. some of their products have very measurable distortions (like a rolled off high end) within the audio band that may account for the differences people hear.

The sound quality differences I hear aren't down to distortions like that. I admit it would be better to fix those things - get the specs and the sound good Not least because then the market size is so much bigger.

In this case, I was mainly curious if there are technical (vs subjective) advantages behind the odd design. The sort of "digital crosstalk" you suggest would show up in a variety of measurements.

I've been hunting myself for such measurements for a couple of years now. If you've got suggestions, I'm all ears

It's likely more of a problem with cheap highly integrated DAC/CODEC chips where everything is jammed onto one die.

I think chip designers are smart guys, smarter than the average implementer of DACs. So I doubt they build-in the noise problems. But there's plenty of anecdotal evidence that off-chip I/V sounds better than on-chip opamps so perhaps they've not got it totally licked.

There are some good reasons for steeply suppressing the images and other junk above 20 Khz coming out of a DAC. As feedback falls beyond 20 Khz, high frequency distortion rises--sometimes dramatically.

Quite. But everything in engineering is a trade-off, there are no free lunches that I've yet found. So steeper filtering leads to poorer time-domain responses. Steeper filtering has a cost in silicon area and in power consumption. Our job as engineers is to optimize all the constraints.

So, intuitively, I'd say all that "mirrored energy" reflected around/above 22 Khz is more likely to make things sound worse rather than better.

Yes, but when I did a quick listen recently between a DAC with an SAA7220 in circuit and then again with it out of circuit, there was no comparison - without sounded so much better. I trust my ears, not my thinking these days.

[abraxalito]

Originally Posted by SoNic_real_one

NOS for some people sounds better exactly because of the extra image frequencies generated by poor filtration that they percieve as a "fuller" sound.

I'm betting you have precisely zero evidence for that assertion right?

Personally I think that is garbage that wans't present in originally sound.

I agree, better off not to have it there. But if removing it makes things sound worse, then the medicine is worse than the disease.

Also, the DAC curves show that the "sweet spot" for that DAC is at 96kHz. 192kHz increase the distortion figures.

Which curves are you looking at?

[SoNic_real_one]

Originally Posted by abraxalito

Yes, but when I did a quick listen recently between a DAC with an SAA7220 in circuit and then again with it out of circuit, there was no comparison - without sounded so much better. I trust my ears, not my thinking these days.

That is mainly because the SAA7200 was a poor design, done in rush. High jitter specially. Not because inherently the 4X OS is flawed.

Originally Posted by abraxalito

Which curves are you looking at?

Page 9, 11 of 20. 48 and 96kHz are sensible equal. 192kHz is some 10dB higher. You need to read "between lines" - they don't have the full measurements there.

[abraxalito]

Originally Posted by SoNic_real_one

That is mainly because the SAA7200 was a poor design, done in rush.

Have a reference for this? I'm reverse engineering the SAA7220 at the moment, so all information is gratefully received I've already found one potential aspect of its design which I wouldn't replicate, too early to say yet whether that's 'poor design' or not.

High jitter specially. Not because inherently the 4X OS is flawed.

I also take the view that 4X OS isn't inherently flawed - that's why I'm designing my own digital filter.