• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Measurement of R2R DAC

Dialectic

Addicted to Fun and Learning
Patreon Donor
Joined
Sep 26, 2017
Messages
855
Likes
531
Location
New York
#1
In light of the considerable hype about R2R DACs and the shortcomings of historic R2R implementations, I'm curious whether any ASR member might be able to loan a recent R2R DAC to Amir for measurement.

I nevertheless doubt that any ASR member actually owns an R2R DAC, so I imagine that this request will go unanswered.
 

Dialectic

Addicted to Fun and Learning
Patreon Donor
Joined
Sep 26, 2017
Messages
855
Likes
531
Location
New York
#3
Discrete R2R DACs don't come cheap. Though I'm not certain, I think the Denafrips Ares, at S$888 (now approximately US$670), is the most inexpensive discrete R2R DAC now available.

IC-based DACs based on the PCM1704 or other chips are probably cheaper, but in the last year or two, such DACs have not been the focus of the R2R hype.
 

March Audio

Major Contributor
Manufacturer
Patreon Donor
Joined
Mar 1, 2016
Messages
3,507
Likes
3,017
Location
Perth Western Australia
#4
In light of the considerable hype about R2R DACs and the shortcomings of historic R2R implementations, I'm curious whether any ASR member might be able to loan a recent R2R DAC to Amir for measurement.

I nevertheless doubt that any ASR member actually owns an R2R DAC, so I imagine that this request will go unanswered.
I had a Soekris dac board dam1021. I will post measurements later but essentially very low noise floor, higher harmonic distortion than typical SD dacs ( but not excessive - nature of the beast with discrete r2r down to resistor tolerence) and sounded imo fwiw, very good.

They do diy boards, complete dacs and dac headphone amps.

http://www.soekris.dk
 
Last edited:

watchnerd

Major Contributor
Beer Hero
Joined
Dec 8, 2016
Messages
4,419
Likes
1,492
Location
Seattle Area
#5
Discrete R2R DACs don't come cheap. Though I'm not certain, I think the Denafrips Ares, at S$888 (now approximately US$670), is the most inexpensive discrete R2R DAC now available.

IC-based DACs based on the PCM1704 or other chips are probably cheaper, but in the last year or two, such DACs have not been the focus of the R2R hype.
The Schiit Modi Multibit seems to have had a lot of hype about being one of the cheapest R2R DACs one can buy.

$249 new, can probably get one cheaper used.
 

Dialectic

Addicted to Fun and Learning
Patreon Donor
Joined
Sep 26, 2017
Messages
855
Likes
531
Location
New York
#6
The Schiit Modi Multibit seems to have had a lot of hype about being one of the cheapest R2R DACs one can buy.

$249 new, can probably get one cheaper used.
Hype from the folks at Schiit doesn't count, and I think we can predict how the Schiit will measure. :)
 

March Audio

Major Contributor
Manufacturer
Patreon Donor
Joined
Mar 1, 2016
Messages
3,507
Likes
3,017
Location
Perth Western Australia
#8
Ok, I havent got all the measurements I was thinking of but still here are some. I cant remember my rationale for the 2kHz or high frequency jitter but hey :) . If anyone iscreally interested I do have access to the board (with friend) and can take more measurements with a Keysight u8903.

As mentioned the only cause for concern is the even harmonic distortion, but at 100 dB down I dont view it as a real issue. Also note I had the board with higher 0.05% tolerence resistors. A 0.01% board is available which will reduce the harmonics. Noise floor was very low and otherwise pretty much free from spuria (ignore the 50Hz and harmonic pick up - measurement issue). For ref its a 131072 point fft kaiser window I think.

-20dB tone
dam1021 -20dB FFT.png


-60dB
dam1021 -60dB FFT.png


-100dB
dam1021 -100dB FFT.png


Jitter

dam1021 jitter.png


-100dB waveform
dam1021 -100dB waveform.png


IMG_20170712_191040-1.jpg
 
Last edited:

dallasjustice

Major Contributor
Joined
Feb 28, 2016
Messages
1,269
Likes
714
Location
Dallas, Texas
#9
Don’t R2R DACs typically require slightly different measurements to find the pimples? What would be the best measurement to show glitch errors? Also, aren’t the NOS R2R DACs rolled off in HF?
 

DonH56

Major Contributor
Technical Expert
Patreon Donor
Joined
Mar 15, 2016
Messages
2,723
Likes
2,864
Location
Monument, CO
#10
1. Don’t R2R DACs typically require slightly different measurements to find the pimples?
2. What would be the best measurement to show glitch errors?
3. Also, aren’t the NOS R2R DACs rolled off in HF?
1. No. At least I think not; what do you define as the "pimples"? In any event I apply IEEE Standard measurements to any DAC (or ADC) irrespective of the architecture.

2. Glitches show up in time or frequency domain measurements, same as any other DAC.

3. Rolloff occurs due to both output loading and sampling rate. A Nyquist signal from a DAC that uses the full clock period for the signal is down ~3.9 dB due to the sinc function (it's in the math). Same thing happens with oversampled designs, but they have a much higher sampling rate so it does not cause issues. Note you can also oversample using an R2R DAC, and apply noise shaping to reduce the in-band quantization noise, if you wish. Most modern delta-sigma designs utilize multi-bit digitizers.

Note R2R DACs are rarely completely binary; most all have a unary (unit-bit) MSB section to reduce matching requirements. A 16-bit R2R DAC needs better than 1/2^16 (0.0015%) matching in the MSBs; move to a segmented architecture that makes the top four MSBs unary (so 15 unit cells then R2R for the rest of the bits) and the matching becomes 16x easier (~0.02%). Since Soekris claims 200+ resistors in a signed-magnitude design I'd guess they are segmenting six or seven bits. Really tempted to pick up one of those demo boards...

Mismatches among resistors (and switches or anything else) can introduce nonlinearities that cause harmonic distortion and/or raise the effective noise floor. So, even if the resistors are well-matched, a gradient across the array from e.g. thermals (change in temp from one end to the other, or from duty cycle variation in the resistors and switches) or voltage compliance or whatever can cause linearity issues. That means even if your resistors are 0.01% accurate in a segmented design, you may still not achieve full precision. Achieving 16+ bits in an R2R design, segmented or not, is hard.

HTH - Don
 
Last edited:

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
18,981
Likes
19,285
Location
Seattle Area
#11
Ok, I havent got all the measurements I was thinking of but still here are some. I cant remember my rationale for the 2kHz or high frequency jitter but hey :) . If anyone iscreally interested I do have access to the board (with friend) and can take more measurements with a Keysight u8903.
Thanks. I would really like to see nearly full scale 12 Khz tone/J-Test with FFT length of 32K so it matches my measurements. The ones shown use much longer FFTs which artificially lowers the noise floor a lot.

As it is, the distortion spikes are quite high compared to even cheap DACs I have test.
 

Cosmik

Major Contributor
Joined
Apr 24, 2016
Messages
2,774
Likes
1,514
Location
UK
#12

DonH56

Major Contributor
Technical Expert
Patreon Donor
Joined
Mar 15, 2016
Messages
2,723
Likes
2,864
Location
Monument, CO
#13
Thanks. I would really like to see nearly full scale 12 Khz tone/J-Test with FFT length of 32K so it matches my measurements. The ones shown use much longer FFTs which artificially lowers the noise floor a lot.

As it is, the distortion spikes are quite high compared to even cheap DACs I have test.
Hmmm... Long FFTs provide higher resolution so if the noise floor is limited by the FFT then it will drop on a per-bin basis (sometimes called "processing gain"). The net SNR should be the same for a given device no matter how long the FFT, and of course the ultimate noise floor will be limited by either the intrinsic noise of the device or the roughly 9N quantization noise floor for N-bit digitization.

Any little ripples in the resistor ladder cause spurs so it is really hard to get the same noise floor as a delta-sigma design. But, you don't have to deal with noise shaping and all the other issues of oversampling converters.
 

DonH56

Major Contributor
Technical Expert
Patreon Donor
Joined
Mar 15, 2016
Messages
2,723
Likes
2,864
Location
Monument, CO
#14
So for audio, what's the point of using them?
Marketing mystique? Delta-sigma designs have their foibles, e.g. they require high-order digital filters, and of course require higher clock rates and you have to deal with much more out-of-band noise. A lot of the early issues with delta-sigma designs have been solved for decades but if you get into the design at a low level there are still some unknown aspects and equations without closed-form solutions (so you rely on approximations and simulations, the latter being just bigger, better approximations). Delta-sigma designs tend to handle impulses more poorly than more conventional architectures so are not well-suited for some applications (or require special design techniques). For audio there are noise modulation and transient issues that delta-sigma designs are more susceptible to, but there are modulation-like and transient issues conventional designs also exhibit, so I am not sure there is a clear winner. In terms of cost it is way easier and cheaper to generate digital filters than trim analog circuits and keep them stable over time and PVT (process/voltage/temperature) and modern processing favors digital circuits over analogs so...

Some folk claim NOS converters sound better to them. I can think of reasons why that would and would not be true; too much depends upon other factors like the output buffer and so forth so I couldn't really say.

Hey, if turntables are undergoing a resurgence, why not early DAC technologies as well? :)
 

Jakob1863

Addicted to Fun and Learning
Joined
Jul 21, 2016
Messages
573
Likes
140
Location
Germany
#15
@BE718,

from your graph, the distortion isn´t 100 dB down as your reference level at 2 kHz is around - 20/22 dBr and due to the quite high level "higher" harmonics the total distortion sums up.
Those higher harmonics make one a bit suspicious of their sonic impact......
 

DonH56

Major Contributor
Technical Expert
Patreon Donor
Joined
Mar 15, 2016
Messages
2,723
Likes
2,864
Location
Monument, CO
#16
p.s. Before somebody better comes along and blasts me as incompetent, I should note that I have not been actively designing data converters for the past few years though I have followed them somewhat, but there may be breakthroughs I missed. The fundamentals have not changed AFAIK. I have in the past designed a number of data converters, ADCs and DACs, using a variety of architectures including segmented R2R and binary converters, slope and dual-slope, flash and folded-flash, successive approximation, delta and delta-sigma, phase ADCs, ADCs using Hadamard sequences and other interesting tricks, etc. Most of these were at much higher rates than audio, however, so my experience is limited to ~16-18 bits (and if you think 16 bits at 20 kHz is hard, try doing it at 100 MHz, or several GHz).

p.p.s. A signed-magnitude converter implemented the way they describe, using separate sections for (+) and (-) sides, means the two sides have to be very precisely aligned in level and time to keep glitches from occuring at the crossover. That may account for the higher distortion, although other factors like the output buffers and layout are likely also major players.
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
18,981
Likes
19,285
Location
Seattle Area
#17
Hmmm... Long FFTs provide higher resolution so if the noise floor is limited by the FFT then it will drop on a per-bin basis (sometimes called "processing gain"). The net SNR should be the same for a given device no matter how long the FFT, and of course the ultimate noise floor will be limited by either the intrinsic noise of the device or the roughly 9N quantization noise floor for N-bit digitization.

Any little ripples in the resistor ladder cause spurs so it is really hard to get the same noise floor as a delta-sigma design. But, you don't have to deal with noise shaping and all the other issues of oversampling converters.
I know it doesn't change what is being measured Don. I am asking for the shorter FFT with the same source signal so that his graphs can then be compared to my DAC measurements without hand compensation for the difference in FFT size. They larger FFT also visually makes things look better even more than mine than they are.
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
18,981
Likes
19,285
Location
Seattle Area
#18
p.s. Before somebody better comes along and blasts me as incompetent, I should note that I have not been actively designing data converters for the past few years though I have followed them somewhat, but there may be breakthroughs I missed. The fundamentals have not changed AFAIK. I have in the past designed a number of data converters, ADCs and DACs, using a variety of architectures including segmented R2R and binary converters, slope and dual-slope, flash and folded-flash, successive approximation, delta and delta-sigma, phase ADCs, ADCs using Hadamard sequences and other interesting tricks, etc. Most of these were at much higher rates than audio, however, so my experience is limited to ~16-18 bits (and if you think 16 bits at 20 kHz is hard, try doing it at 100 MHz, or several GHz).

p.p.s. A signed-magnitude converter implemented the way they describe, using separate sections for (+) and (-) sides, means the two sides have to be very precisely aligned in level and time to keep glitches from occuring at the crossover. That may account for the higher distortion, although other factors like the output buffers and layout are likely also major players.
Hi Don. One of the techniques used that I have seen is upsampling first. As you know, doing so then reduces the number of effective bits needed for the conversion. Some go as high as a few Megahertz here.
 

DonH56

Major Contributor
Technical Expert
Patreon Donor
Joined
Mar 15, 2016
Messages
2,723
Likes
2,864
Location
Monument, CO
#19
I know it doesn't change what is being measured Don. I am asking for the shorter FFT with the same source signal so that his graphs can then be compared to my DAC measurements without hand compensation for the difference in FFT size. They larger FFT also visually makes things look better even more than mine than they are.
Ah, got it, misunderstood your statement about the noise floor.

Sorry Amir - Don
 

DonH56

Major Contributor
Technical Expert
Patreon Donor
Joined
Mar 15, 2016
Messages
2,723
Likes
2,864
Location
Monument, CO
#20
Hi Don. One of the techniques used that I have seen is upsampling first. As you know, doing so then reduces the number of effective bits needed for the conversion. Some go as high as a few Megahertz here.
Sure, upsampling before applying the data to a DAC operating at a higher rate accomplishes essentially the same thing as oversampling, whilst allowing you to shape the signal on the way (for better or worse). I have read about audio converters running in the 5 MHz range or so IIRC but I have not really kept up so your knowledge is better than mine. As an aside my old Pioneer (now retired) used an SRC (sampling rate converter) to upsample everything before the output DACs. The RF analogue would be to do upconversion to spread the mixing spurs for filtering before down converting to baseband.

Each doubling of the sampling rate gains about 0.5 bits (3 dB) in SNR for a given (fixed and filtered) output bandwidth in a conventional design (i.e. without noise shaping like the R2R DACs discussed here). The gain is much higher as you move to noise shaping and multibit delta-sigma designs (you gain about L+0.5 bits for a modulator of order L, plus additional SNR gains from multibit converters in the loop(s)). Modern designs use pretty high-order loops and multibit quantizers in the loop to gain 10's of dB plus the benefits of oversampling.
 

Similar threads

Top Bottom