Review and Measurements of Holo Audio May --- Probably the best discrete R2R DAC

[bunkbail]

OK, wow, those are spectacular. What are we looking at and what do you learn from it? <need popcorn >

Conclusion: MQA sounds the best

 

[WolfX-700]

There is much debate about D-S and R2R. For me-I remain strictly neutral-I don't think there is an essential difference between different architectures in sound-it's the sound that is played back, not the architecture.
As far as this product is concerned, I just respect the engineering implementation and the effort behind it.

 

[WolfX-700]

Similarly, in the case where discrete R2R can make such excellent measurement results. Let those who promote "discrete R2R not care about measurement" go to hell!

With enough effort, the measurement results of discrete R2R can be very good!

 

[DonH56]

I designed a 16-bit DAC based on the R2R idea many years ago (~1998) with near-perfect linearity, settling to 16-bits in one cycle and operating at 35 MS/s (it would actually clock to ~1 GS/s but linearity suffered; it did hit around 80 dB SFDR at 100 MS/s for Nyquist output). It was a single-chip (IC) since "that's what I do" (did). It was a huge effort but could be realized discretely as well (Accuphase did it, IIRC, long ago, though not at 35 MS/s). Needs laser trimming and all sorts of fancy tricks to ensure thermal stability and all that jazz. The major "secret sauce", and what I believe to be the case here, is to segment the MSBs into unary (unit) steps so the actual R-2R section is "only" the lower bits so you can achieve the desired matching (linearity) and settling. The idea is to split say the top 4-6 bits into ~16 to 64 (2^4 to 2^6) individual unit (unary) steps, and then the matching needs to be only 1/16 to 1/64 of the target resolution. So say the MSB, segmented into 64 (technically 63) unary steps, requires individual steps to match to "only" 1 part in 2^16/2^6 = 65536/64 = 1024 or about 0.1% instead of 0.0015%. Not that 0.1% is easy, but a fairly sophisticated (and thus time-intensive even for ATE) trim algorithm sufficed.

Knowing what it takes, I am very (make that extremely) impressed by the performance of this product.

This thread touches on multibit DACs a bit, including a brief handwaving look at segmented design: https://www.audiosciencereview.com/...ital-audio-converters-dacs-fundamentals.1927/

Similarly a brief look at delta-sigma DACs: https://www.audiosciencereview.com/...igma-delta-digital-audio-converters-dac.1928/

Please note those are old threads meant to (try to) provide an overview that does not go too deep. In other words, probably too demanding for someone starting from scratch and far too simplistic for someone actually designing them. We cover both extremes on ASR, natch.

FWIWFM - Don

 

[yummy]

unless this was posted by Amir, this will be some Chinese cooking something, even there're posts in Chinese above, vendors everywhere

 

[JohnYang1997]

unless this was posted by Amir, this will be some Chinese cooking something, even there're posts in Chinese above, vendors everywhere

what do you mean?

 

[Rja4000]

Hi
I like your THD+N vs Frequency plot:
The idea to combine both 20kHz and 90kHz Bandwidth on the same plot gives immediately more information. In this case, most noise is above 20kHz - as it's usually the case.

 

[Thomas savage]

unless this was posted by Amir, this will be some Chinese cooking something, even there're posts in Chinese above, vendors everywhere

Not sure what your aiming with this , I'm very keen on China as far as audio goes.

Plenty of companies there seem to invest in measuring equipment and value proper design while other western based audio outfits try and pin the tail on the donkey after 10 pints of kool aid.

We won't tolerate prejudiced towards our Chinese friends.

 

[Veri]

unless this was posted by Amir, this will be some Chinese cooking something, even there're posts in Chinese above, vendors everywhere

I really, really feel @WolfX-700 is a perfectly objective party here. Let's not bash China for no reason...

 

[Killingbeans]

I agree. This is clearly crossing the line of paranoia.

There's a huge difference between WolfX-700 and the people manufacturing Bung & Olafsen and the likes thereof.

Simply assuming that all of the nearly 1½ billion people living in China is up to no good, is just disrespectful to say the least.

EDIT: Okay, looking through yummy's posts I spot a distinct phobia aimed at anything from East Asia

 

[mansr]

I designed a 16-bit DAC based on the R2R idea many years ago (~1998) with near-perfect linearity, settling to 16-bits in one cycle and operating at 35 MS/s (it would actually clock to ~1 GS/s but linearity suffered; it did hit around 80 dB SFDR at 100 MS/s for Nyquist output). It was a single-chip (IC) since "that's what I do" (did). It was a huge effort but could be realized discretely as well (Accuphase did it, IIRC, long ago, though not at 35 MS/s). Needs laser trimming and all sorts of fancy tricks to ensure thermal stability and all that jazz. The major "secret sauce", and what I believe to be the case here, is to segment the MSBs into unary (unit) steps so the actual R-2R section is "only" the lower bits so you can achieve the desired matching (linearity) and settling. The idea is to split say the top 4-6 bits into ~16 to 64 (2^4 to 2^6) individual unit (unary) steps, and then the matching needs to be only 1/16 to 1/64 of the target resolution. So say the MSB, segmented into 64 (technically 63) unary steps, requires individual steps to match to "only" 1 part in 2^16/2^6 = 65536/64 = 1024 or about 0.1% instead of 0.0015%. Not that 0.1% is easy, but a fairly sophisticated (and thus time-intensive even for ATE) trim algorithm sufficed.

Moreover, with a unary array you can use scrambling techniques to decorrelate the effects of mismatches from the signal.

When it comes to building a product, I really don't see the point in going to all this trouble yourself when you can simply buy a DAC chip that does it all for a few dollars.

 

[scott wurcer]

I designed a 16-bit DAC based on the R2R idea many years ago (~1998)

We did a 16bit low frequency DAC circa 1988 based on a Kelvin-Varley divider, it had the same glitch energy problem as the AD5791 and got no traction with audio customers. There are lots of instrument and motion control devices that need direct stepped voltage out.

 

[DonH56]

Moreover, with a unary array you can use scrambling techniques to decorrelate the effects of mismatches from the signal.

When it comes to building a product, I really don't see the point in going to all this trouble yourself when you can simply buy a DAC chip that does it all for a few dollars.

I dug into the shuffling/scrambling approach outlined by Rudy van de Plassche but didn't have the headroom or power budget in the technology I was using (CBJT). It also introduced a spur at the scrambling frequency (I went to a random source to decorrelate that). In CMOS probably would've worked well.

Differentiation (technical and/or marketing) means there's always a place for a different, or at least vintage, design. These days I tend to agree that plenty of chips exhibit exemplary performance.

We did a 16bit low frequency DAC circa 1988 based on a Kelvin-Varley divider, it had the same glitch energy problem as the AD5791 and got no traction with audio customers. There are lots of instrument and motion control devices that need direct stepped voltage out.

I have said many times I don't think I've invented anything not done by somebody with tubes in the 1930's or whenever. Glitch suppression was a big focus an, even though it was pretty small, it was a significant problem in distortion and settling. And of course de-glitching circuits are about as hard as getting the DAC itself to settle. The work you and others did at ADI certainly set the standard (for a lot of things).

I have done a few K-V designs over the years. Getting the switches (in CMOS this time) to have high enough linearity and low enough switching glitches was a killer, and layout to match routes and maintain global and local matching and tolerance of gradients (process, thermal, voltage, etc.) was a nightmare. At high resolution everything matters and it is all a PITA.

Anyway, I was too excited when I wrote my post, but too late to edit now. The point was not that I had done something like it, but that it is HARD, and seeing that kind of performance is exciting.

 

[Alan S]

Probably not. R2R is a conceptually simple, brute-force way to do D2A conversion that turns out to be tricky to implement well. At one time I was enamored with R2R idea, and still own a recent Holo Spring L1, but can't say that it's any better than a decent D-S DAC at 1/4 the price.

R2R：steampunk
D-S：cyberpunk

 

[PierreV]

When it comes to building a product, I really don't see the point in going to all this trouble yourself when you can simply buy a DAC chip that does it all for a few dollars.

Definitely not rational. But when something like that is achieved with pure design/engineering skills and delivers on its promises, I can't help being somewhat in awe.

Plus, $4000 definitely falls in Stereophile's "budget components" these days

 

[scott wurcer]

Anyway, I was too excited when I wrote my post, but too late to edit now. The point was not that I had done something like it, but that it is HARD, and seeing that kind of performance is exciting.

All true, I was just adding info no commentary was intended. I was equally excited by someone de-glitching the AD5791 to that level of performance but the industry is so far down the D-S hole only the fringe cares.

 

[mansr]

Definitely not rational. But when something like that is achieved with pure design/engineering skills and delivers on its promises, I can't help being somewhat in awe.

It's impressive in the same way as someone paddling a kayak across the Atlantic instead of buying a plane ticket. There's no benefit to the end user from them choosing to do things the hard way.

 

[pkane]

It's impressive in the same way as someone paddling a kayak across the Atlantic instead of buying a plane ticket. There's no benefit to the end user from them choosing to do things the hard way.

It's the art of engineering. Sometimes seeing what's possible, no matter how impractical or unnecessary, is still useful.

 

[scott wurcer]

There's no benefit to the end user from them choosing to do things the hard way.

This is where the potential for snake oil enters such as "mine has no global feedback". You won't have to look too hard to find folks that claim there is something unknown or immeasurable going on in D-S DAC's.

 

[gvl]

... but the industry is so far down the D-S hole only the fringe cares.

So it's a hole? Care to elaborate?