New 28-bit DAC coming out.

[KSTR]

I feel that I have to comment on the review from Bob Katz on this new ‘multi-path’ DAC design, because people in this thread seem to think the measurements of the unnamed 'standard' DAC were from an ADI-2 DAC FS. As you will see this is definitely not the case. Furthermore the review measurements show very unusual, bad results, that need both clarification and correction. [...]

The best way for a comparison of the technology vs. current single-path would have been a switch option in the D1 where one could defeat the low-level DAC and the main DAC then would do all the work as usual, with no other changes to anything else.

I would assume that Imersiv have implemented this "legacy" option for development and testing anyway but decided to not give it to the customer...

 

[antcollinet]

Hello everyone,

I feel that I have to comment on the review from Bob Katz on this new ‘multi-path’ DAC design, because people in this thread seem to think the measurements of the unnamed 'standard' DAC were from an ADI-2 DAC FS. As you will see this is definitely not the case. Furthermore the review measurements show very unusual, bad results, that need both clarification and correction. Quotations from that review start with ‘BK’, as now:

BK: Next, I lowered the test tone level to -70 dBFS, which produces -56 dBu analog level. I amplified the input level 60 dB with a high-performance microphone preamplifier to overcome the analyzer ADC performance. Then I recalibrated the reading to reflect the actual dBu levels from the DAC. As you can see (Fig. 3), at this low level, the sine wave is corrupted by noise and distortion within the single path DAC. At this level, the single path DAC produces 2.18%/2.24% THD (without noise), Left/Right channel. The noise is significant enough so that THD+N rises to 3.25%/3.47%. These unimpressive low level characteristics are similar to that of all standard (single-path) DACs.

There is so much objectively wrong here, so let’s get into details. I tried to replicate the setup as good as possible: sending a -70 dBFS 1 kHz sine at 96 kHz sample rate from the SPDIF output of the APx555B to the ADI-2 DAC, the analog output goes back to the APx. Without that 60 dB gain amp, because – as shown later – it doesn’t make any difference for these measurements (it would do in specific others). I also set the APx to an (unusual) 1M FFT size, and made sure the Analyzer results are capped to SR/2 bandwidth, like in the review.

Pic 1 shows the result: This ‘single path DAC’ produces -51 dBu output level (quite similar to the review), but THD+N is a mere 0.4% instead of 3.3%. That is 8 times better than the ‘standard DAC’ used in the test. And I do not claim that this is an unusual result – DACs with top AKM and ESS converters (as tested all the time here at ASR) can easily show even better results (noise floor in FFT around 8 dB lower, so even lower THD+N values).

View attachment 481601

As the FFT also clearly shows there is 0 (zero) distortion. If there is it would be hidden in the noise floor, but even with averaging none became visible. That means THD here is at least -100 dB, or below 0.001%. This is in stark contrast to the continued mentioning of distortion in the review, where there might have been no distortion at all but only noise, or the computed distortion values do not make any sense as SpectraPlus was not able to differ noise from harmonics. Modern ‘standard DACs’ do not distort at lower levels, not at all.

While the review does not include FFT screenshots from this measurement , we can grasp what’s going on with the next measurement at -90 dBFS. Once again my results in pic 2 show a very different picture of today’s ‘single path DACs’. The difference in THD+N is so big (3.8% to 38%) that it becomes obvious something is very wrong in the review. The respective FFTs have nothing in common. The noise floor from the ADI-2 DAC’s APx measurement is about 30 dB (!) lower, and there is (again) zero distortion. THD is unknown, but at least -65 dB, or lower than 0.056% - world’s apart from the stated 6.6% (not to mention that strange 5 Hz needle in the review’s FFT).

View attachment 481602

BK: As you can see, the single path DAC's waveform is hardly recognizable as a sine wave, as there is so much noise and distortion

There is no distortion visible – only noise. The APx Scope of the ADI-2 DAC signal shows a bit noisy waveform, but still a good sine. I wanted to know what it looks like using an ADI-2/4 Pro as ADC to capture this signal, similar to the setup in the review. The result matches the APx, see pic 3.

View attachment 481603

So while the D1 shows a perfectly clean sine (it should), any other DAC will show a bit noise on the signal – but nothing like what is shown in the review.

BK: Never Before Seen in Print. No one, I mean no one, has ever seen a clean -90 dBFS sine waveform from any DAC, until today!

I would have to disagree. All DACs reaching 120 dB SNR are capable of that – and that is quite the norm these days. Maybe the D1 looks even more clean and smooth, but the notion that they all look like the mess shown in the review is wrong.

I also did a screenshot of what the APx could analyze, see pic 4. For evaluating the D1 I expect values like THD, THD+N and SNR/DR are limited on the AP side, and need either an external preamp or cannot be measured correctly at all. But the FFT itself seems more than capable to show a difference between a single path and multi-path DAC (BTW, I would have preferred the term dual-path). Here is the FFT with 3x averaging and inputs terminated with 150 Ohms. The noise floor is then around -165 dBu. That is more than enough to see any of the suggested advantages of the D1 (and also if the multiple spuriae from the review measurements are from the D1 or originate somewhere else).

View attachment 481604

Also note that I tried to use my standard +34 dB preamp and was not able to get any different (improved) results from the -70 dBFS and -90 dBFS tests above. For this kind of stuff the APx seems sufficiently equipped already.

Thanks for that.

Absolutely vital that these manufacturer based tests are held to account.

 

[pkane]

BK: Never Before Seen in Print. No one, I mean no one, has ever seen a clean -90 dBFS sine waveform from any DAC, until today!

I would have to disagree. All DACs reaching 120 dB SNR are capable of that – and that is quite the norm these days. Maybe the D1 looks even more clean and smooth, but the notion that they all look like the mess shown in the review is wrong.

Agreed! Here's a -100dBFS generated waveform captured at the output of a 60dB amplifier from a DO300 DAC. Looks nothing like the mess in the article:

 

[markstein]

Hello everyone,

I feel that I have to comment on the review from Bob Katz on this new ‘multi-path’ DAC design, because people in this thread seem to think the measurements of the unnamed 'standard' DAC were from an ADI-2 DAC FS. As you will see this is definitely not the case. Furthermore the review measurements show very unusual, bad results, that need both clarification and correction. Quotations from that review start with ‘BK’, as now:

BK: Next, I lowered the test tone level to -70 dBFS, which produces -56 dBu analog level. I amplified the input level 60 dB with a high-performance microphone preamplifier to overcome the analyzer ADC performance. Then I recalibrated the reading to reflect the actual dBu levels from the DAC. As you can see (Fig. 3), at this low level, the sine wave is corrupted by noise and distortion within the single path DAC. At this level, the single path DAC produces 2.18%/2.24% THD (without noise), Left/Right channel. The noise is significant enough so that THD+N rises to 3.25%/3.47%. These unimpressive low level characteristics are similar to that of all standard (single-path) DACs.

There is so much objectively wrong here, so let’s get into details. I tried to replicate the setup as good as possible: sending a -70 dBFS 1 kHz sine at 96 kHz sample rate from the SPDIF output of the APx555B to the ADI-2 DAC, the analog output goes back to the APx. Without that 60 dB gain amp, because – as shown later – it doesn’t make any difference for these measurements (it would do in specific others). I also set the APx to an (unusual) 1M FFT size, and made sure the Analyzer results are capped to SR/2 bandwidth, like in the review.

Pic 1 shows the result: This ‘single path DAC’ produces -51 dBu output level (quite similar to the review), but THD+N is a mere 0.4% instead of 3.3%. That is 8 times better than the ‘standard DAC’ used in the test. And I do not claim that this is an unusual result – DACs with top AKM and ESS converters (as tested all the time here at ASR) can easily show even better results (noise floor in FFT around 8 dB lower, so even lower THD+N values).

View attachment 481601

As the FFT also clearly shows there is 0 (zero) distortion. If there is it would be hidden in the noise floor, but even with averaging none became visible. That means THD here is at least -100 dB, or below 0.001%. This is in stark contrast to the continued mentioning of distortion in the review, where there might have been no distortion at all but only noise, or the computed distortion values do not make any sense as SpectraPlus was not able to differ noise from harmonics. Modern ‘standard DACs’ do not distort at lower levels, not at all.

While the review does not include FFT screenshots from this measurement , we can grasp what’s going on with the next measurement at -90 dBFS. Once again my results in pic 2 show a very different picture of today’s ‘single path DACs’. The difference in THD+N is so big (3.8% to 38%) that it becomes obvious something is very wrong in the review. The respective FFTs have nothing in common. The noise floor from the ADI-2 DAC’s APx measurement is about 30 dB (!) lower, and there is (again) zero distortion. THD is unknown, but at least -65 dB, or lower than 0.056% - world’s apart from the stated 6.6% (not to mention that strange 5 Hz needle in the review’s FFT).

View attachment 481602

BK: As you can see, the single path DAC's waveform is hardly recognizable as a sine wave, as there is so much noise and distortion

There is no distortion visible – only noise. The APx Scope of the ADI-2 DAC signal shows a bit noisy waveform, but still a good sine. I wanted to know what it looks like using an ADI-2/4 Pro as ADC to capture this signal, similar to the setup in the review. The result matches the APx, see pic 3.

View attachment 481603

So while the D1 shows a perfectly clean sine (it should), any other DAC will show a bit noise on the signal – but nothing like what is shown in the review.

BK: Never Before Seen in Print. No one, I mean no one, has ever seen a clean -90 dBFS sine waveform from any DAC, until today!

I would have to disagree. All DACs reaching 120 dB SNR are capable of that – and that is quite the norm these days. Maybe the D1 looks even more clean and smooth, but the notion that they all look like the mess shown in the review is wrong.

I also did a screenshot of what the APx could analyze, see pic 4. For evaluating the D1 I expect values like THD, THD+N and SNR/DR are limited on the AP side, and need either an external preamp or cannot be measured correctly at all. But the FFT itself seems more than capable to show a difference between a single path and multi-path DAC (BTW, I would have preferred the term dual-path). Here is the FFT with 3x averaging and inputs terminated with 150 Ohms. The noise floor is then around -165 dBu. That is more than enough to see any of the suggested advantages of the D1 (and also if the multiple spuriae from the review measurements are from the D1 or originate somewhere else).

View attachment 481604

Also note that I tried to use my standard +34 dB preamp and was not able to get any different (improved) results from the -70 dBFS and -90 dBFS tests above. For this kind of stuff the APx seems sufficiently equipped already.

Very nice to see this kind of fact based, no marketing blabla expert feedback. Much appreciated!

And what a surprise: tests and opinions in "professional" circles are not necessarily more professional (and scientific) than those by audiophile golden ears.

 

[danadam]

BK: Never Before Seen in Print. No one, I mean no one, has ever seen a clean -90 dBFS sine waveform from any DAC, until today!

I would have to disagree. All DACs reaching 120 dB SNR are capable of that – and that is quite the norm these days. Maybe the D1 looks even more clean and smooth, but the notion that they all look like the mess shown in the review is wrong.

It's so bad that I wonder what it was (maybe that was 16-bit DAC and something added dither?)

Here is:

• -90 dBFS 1 kHz at 48 kHz sampling rate played through several DACs,
• captured through E1DA Scaler at max amplification (+26.7 dB)
• to ADCiso +13 dBu at 96 kHz sampling rate.

The DACs are:

• RME Adi-2 Pro, +13 dBu
• Apple dongle, US version
• Dragonfly Red
• FiiO K3, hi gain
• JCally JM20
• Tanchjim Space

(waveform plots amplified to -6 dBFS RMS)
, , , , ,

128K FFT of the captured files (8 averages):
, , , , ,

 

[IVX]

16 bits -90dbfs looks ugly indeed.

 

[Hayabusa]

Still the elephant in the room: Any one tried to capture the 'switching' level where the signal is using both paths? How does this look like?

 

[KSTR]

Still the elephant in the room: Any one tried to capture the 'switching' level where the signal is using both paths? How does this look like?

I would assume when you slowly decrease the level of a test signal (like a sparse multitone with known spectrum and enough holes to see the noise floor) there will be one point where the noise floor starts to drop suddenly and significantly because the main DAC is disconnected from the output, removing its noise impact -- which is the core idea of this DAC. It might also produce some artifacts in the signal, showing signs of remaining DC offset, and also the distortion profile will change a bit.

This is the only hard switching point with an actual analog switching process and therefore probably the most critical aspect. Above that point, and depending on the signal itself and its history, there will be cross-fading going on which also will have some impact on measurement notably if there is residual gain mismatch.

I fully agree these are the tests we want to see.

 

[pma]

I did similar tests, with very low digital domain signals, DAC output amplified by low noise gain 30-40 dB low distortion amplifiers. It reveals some issues that remain hidden in the usual measuring system noise, and also low level THD/level nonlinearity. However, potential audibility of such issues is highly questionable.

 

[pinger]

Has Stereophile "reviewed" this yet? I'm sure they will find sonic differences LOL

 

[antcollinet]

16 bits -90dbfs looks ugly indeed.

The ugliness is too small to hear though.

Try turning down your music by 90dB and see what you can hear - I've just gone 79.5dB down (*as much as my AVR will turn down) from a volume about 10dB louder than I normally listen - and it is basically inaudible. Then the "ugliness" even on the worst examples above is probably another 10dB lower.

 

[Soniclife]

The ugliness is too small to hear though.

FWIW I read that as sarcasm.

 

[antcollinet]

FWIW I read that as sarcasm.

In the absence of smilies or equivalent - Poe's law applies.

 

[signalpath]

I would assume when you slowly decrease the level of a test signal (like a sparse multitone with known spectrum and enough holes to see the noise floor) there will be one point where the noise floor starts to drop suddenly and significantly because the main DAC is disconnected from the output, removing its noise impact -- which is the core idea of this DAC.

Correct! Somewhere around -40dBFS, when the paths change, we measure a marked THD+N step function (see graph). This is what is meant by "improvements in the atmospheric region" --which can be significant. Is it audible? A number of pro audio engineers say yes (2/3 of our 50 beta testers purchased their D-1). I personally heard an immediate and significant "space and imaging" improvement way back on my original prototype (probably 2017). At some point, we want to do a clinical ABX trial with a suitable number of trained listeners. Maybe ASR can sponsor it.

A perceptive reader will ask "why not add additional paths to continually reduce THD+N ratios over the entire perceptual band?" Of course, we tried this, and it's written up in the patents. But we found that a single, dramatic path transition strategically sited in the atmospheric listening region gave (by far) the largest perceptual "ah ha!" improvement. Additional path transitions delivered diminishing sonic returns.

I also wanted to share a bit of good news -- the imersiv D-1 multi-path DAC just won the Grand Prize in the SAE Media Create the Future Design Contest, from a field of 630 vetted inventions from 67 countries. There were over 50 judges (in total), mostly engineers and scientists from Intel, NVIDIA, Ford, Airbus, Northrup, Analog Devices, etc.. You can watch my pitch in the final round last week In NYC https://contest.techbriefs.com/

Would also add that we just completed a project with NASA Goddard, applying HDR-A architecture to laser measurement for the 2034 LISA Gravity Wave Detection space mission, achieving a stable low-path broadband/unweighted noise of -145dBuEIN.

The AES 2021 peer-reviewed Multi-Path engineering paper just went open-source! You can access here: https://aes2.org/publications/elibrary-page/?id=21106

 

[phofman]

Would also add that we just completed a project with NASA Goddard, applying HDR-A architecture to laser measurement for the 2034 LISA Gravity Wave Detection space mission, achieving a stable low-path broadband/unweighted noise of -145dBuEIN.

Well, that is a great use of your technology indeed, hats off. It makes perfect sense there (less so in human audio with inefficient human senses, IMO).

 

[Ropeburn]

Most of the content I see on streaming services these days is 16-bit (96 dB DNR). Very little of it is released at Hi-Rez 24-bit (144 dB DNR), and I'm not sure if it was recorded at that resolution in the past, or just remastered now to 24-bit.

The only possible way to get true 24bit signals without noise and distortion filling the LSBs is directly generating everything in software anyway. So theoretically, any real use of this quality is limited to electronic music, pretty much. As soon as you use anything recorded, it becomes useless in theory and practice both. I really don't see the point of this.

 

[signalpath]

I really don't see the point of this.

See graph, two comments up.

 

[EERecordist]

This is what is meant by "improvements in the atmospheric region"

ASR would benefit from posting Fiedler's and other papers or a summaries on ASR in our psychoacoustics section. Having listened to a lot of classical music live and in the control room, the theory sounds rational. For ASR "atmospheric" needs a more scientific name, like reducing electronic noise in quiet passages.

Do you see your technology applicable to digital microphones? Schoeps is pushing into that space with their colette series. Is your technology applicable to Sennheiser's RF design?

 

[signalpath]

ASR would benefit from posting Fiedler's and other papers or a summaries on ASR in our psychoacoustics section. Having listened to a lot of classical music live and in the control room, the theory sounds rational. For ASR "atmospheric" needs a more scientific name, like reducing electronic noise in quiet passages.

Do you see your technology applicable to digital microphones? Schoeps is pushing into that space with their colette series. Is your technology applicable to Sennheiser's RF design?

We've referenced Fielder's AES papers from the 70s thru 90s extensively in our R/D, especially those dealing with noise perception vis signals. In fact, I was invited to meet with him a few years ago at Dolby HQ to describe our work. Dolby's VP of Advanced Technology now sits on our imersiv advisory board, and was the first person outside of the company to test and validate our design performance (2017-18).

Yes, HDR-A architecture ("multi-path") can be applied to just about any transducer or A/D or D/A function. The concept is well-known, but it's expensive and complex compared to traditional single-path topology. Salzbrenner has had a multi-path ADC in market for 30 years. Many low-end preamp/ADC interfaces are now using a two-path "clip-protection" variant. Sony recently developed a two-path video CCD pixel with a high-dynamic path and low-dynamic path, with 20dB improved visual dynamic range. In my CTF contest pitch, I describe a range of multi-path applications (link above), including medical, seismic, gravity wave sensing, etc..

In R/D, we're now working on a fundamental two-path 170dB microphone. Conceptually very simple. But when you start digging, it gets hard. It's why the D-1 DAC took us 10+ years of R/D. This is new engineering, a lot of fun, but terribly expensive and time-consuming. Also working on multi-path power amplification, based on one of our patents. It's important to keep in mind -- this is ALL about reducing systemic THD+noise, not "making things louder".

 

[popej]

See graph, two comments up.

Don't you have any real world measurements? The kind that validate the implementation.