Saturday, 8 September 2018

MEASUREMENTS: RME ADI-2 Pro FS ADC performance and as measurement hardware.


I discussed the features and some listening impressions of the RME ADI-2 Pro FS ADC/DAC a couple weeks ago. As I noted at that time, my main aim in owning this device is for the purpose of using it as an ADC for measurements here on the blog going forward.

For the purpose of measurements, we want a tool that can allow us to obtain reproducible results and good accuracy. Over the years, I've achieved reproducible results with consistently minimal inter-test variation by standardizing the way I run most measurements with the digital sources, cables, standard procedures, and types of tests I run. What I want is better accuracy - an ADC that has lower noise floor for improved resolution, doesn't add as much of its own distortions, have higher timing accuracy (eg. for jitter tests), and perhaps more features to expand the measurement quality (eg. higher sample rate to capture impulse responses more accurately, and can handle wider input levels for the range of devices tested).

Given that this is the priority, instead of measurements of DAC performance first, let's get straight to describing the characteristics of and using the ADI-2 Pro as a measurement tool which means we'll need to get a taste of how well the analogue-to-digital capability performs. Let's compare the results I'm getting from the RME to my previous measurement ADC device over the last couple years, the Focusrite Forte for a sense of the changes in resolution I can expect from the new upgrade.

Remember that over the years, I have used 2 other ADCs on a regular basis for measurements here. For a cameo, I dug out the old Creative E-MU 0404USB for the line up:


This would then make the RME ADI-2 Pro FS the "3rd Generation" tool for the measurements here. :-) With all the features highlighted last time plus the ability to handle very high sample rates (384 & 768kHz), assuming I don't run into any issues, this could be the ADC device to last me for awhile!

Let's see what it can do...

I. Noise Floor & Stability

An extremely important basic measurement to start looking at is the noise floor. After all, it is out of the noise floor that we can measure minute changes and determine potential resolutions for the devices being tested. Calibrated to the same input peak level (+12dBu, this is the max input level for the Focusrite Forte), tested using the tried-and-true WaveSpectra using ASIO at 192kHz, 131,072 FFT bins, this is what the noise floor looks like connected to the Oppo UDP-205 (previously measured to be a very high resolution and quiet DAC) playing digital silence through XLR output to the various devices:


As you can see, compared to the other devices, the RME ADC has the lowest noise level in the audio frequencies, at least up to 30kHz. It has no significant noise spikes unlike the Focusrite Forte which unfortunately has a issue around 37kHz. Furthermore we see that the "Gen1" E-MU 0404USB tended to have numerous noise peaks from 10kHz onward and overall had higher baseline noise levels as well. Notice also that with the ADI-2 Pro, the high frequency modulator noise remains lower with a smoother gradient than the other ADCs; particularly the Focusrite Forte which tended to rise quite steeply by 50-60kHz - while obviously this is not audible, it's nice to keep noise to a minimum as a measurement device. This is a fantastic start.

The noise floor speaks highly of the device overall but perhaps specifically to the choice of the AKM AK5574 "Verita Series" AD converter first released in 2016. According to the datasheet, it is capable of a 121dB A-weighted dynamic range, increasing to 124dB with the "4-to-2" mode since it is a 4-channel part that can be summed down into 2-channels. As with the DAC side of the ADI-2 Pro FS, it can convert analogue to 768kHz PCM and DSD256.

Back in 2015, I tested out the Tascam UH-7000 as a potential upgrade ADC. Unfortunately, I found that there was quite a bit of noise floor pollution as the device warmed up as a result of its power supply. So, what happened to the noise floor of the ADI-2 Pro FS over 5 hours of intermittent checking?


Nothing happened to the noise floor. :-) Excellent! The noise floor at each time point essentially overlays on each other. Certainly by 5 hours the device will have reached a steady temperature. Furthermore I doubt I'll be sitting around the machine doing measurements for longer than 5 hours at any one time! Remember that this device does get quite warm in regular use, so make sure there's reasonable air flow.

It's always nice to have a look at the lower frequencies, particularly ~60Hz here in North America to see if any mains hum is seeping through into the ADC or DAC. To look at this with higher resolution, I've employed the use of SpectraPLUS-SC (version 5.2.0.20 at time of testing) with the ability to do up to 1M bin FFTs at 192kHz sampling rate through ASIO:


Beautiful 0-250Hz noise floor. Nice and clean which tells us that neither the Oppo DAC playing digital silence nor the RME ADI-2 Pro doing the recording is introducing 60Hz hum or its harmonics into the system. Remember folks, the RME device is powered by a switching power supply... Clearly this is not an issue given no evidence of noise across the spectrum at least to 96kHz in the plots above, nor around 60Hz.

One could check the noise floor even further out since the ADI-2 Pro can sample all the way to 768kHz. WaveSpectra can handle up to 384kHz using ASIO in realtime. Here's what the noise level out to 192kHz (384kHz sample rate, 131k FFT) looks like:


Again, nice and clean all the way to 192kHz with no gross idle tones sticking out.

Finally, even though I don't have a realtime FFT that can handle 768kHz, we can do this by recording at 32/768kHz through Sound It! Pro 8 and then run it through the FFT (65536 points) in Adobe Audition which accepts files of this bit-depth and samplerate:


Adobe Audition scanned 5 seconds of the 32/768kHz digital silence recording to come up with that noise floor. Again, nice to see the absence of any unusual noise - Raspberry Pi 3 USB to Oppo UDP-205 to RME ADI-2 Pro FS with generic USB and XLR cables.

Other than extreme oversampling in the home, research into ultrasonics, or perhaps professionals wanting to archive something at 2xDXD, I think it's highly unlikely than anyone would want/need to perform home recordings or vinyl rips at such a high samplerate!

For perspective, remember that bats with all their amazing echo-location capabilities are said to have their most sensitive hearing between 15-90kHz, maybe at the extreme, hear up to 200kHz. Marine mammals like dolphins can emit sounds up to 150kHz, also useful for echo-location. So, I think an ADC like this could be great for animal research into recording ultrasonic vocalizations and animal hearing with proper microphones like these! :-)

II. -90.3dBF 1.1025kHz non-dithered 16-bit LSB Waveform

One measurement I have not done in awhile has been the -90.3dBFS 1kHz non-dithered 16-bit LSB "waveform peep". In principle, if the device being tested is of high resolution, and our ADC is likewise of excellent quality, we should see a beautifully stable signal on both channels even at the very low -90dB amplitude zoomed in for visualization... Here's the Oppo UDP-205 (sharp filter) recorded with the ADI-2 Pro FS (sharp filter) compared to idealized performance:


That "ideal performance" image uses linear sinc interpolation so I put the ADI-2 Pro FS ADC into the "Sharp" linear phase filter mode for comparison purposes. By default, the device is actually set to "SD Sharp" which is a minimum phase setting and the one I used for the remainder of these tests.

For the record, the RME ADI-2 Pro FS's ADC has 4 digital filter settings to choose from: minimum phase "SD Sharp" and "SD Slow", plus linear phase "Sharp" and "Slow". At some point, I'm sure I'll use this ability to demonstrate filter ringing on the ADC side like what we've done with impulse responses on the DAC side.

In any event, the comparison image above looks great! I set the RME ADC to capture the peak input level (+13dBu) appropriate for the Oppo without clipping and just recorded the -90dB signal. Nothing was done to amplify that -90dB signal when fed into the ADC to increase SNR. The only thing done to visualize the waveform was normalization the low amplitude signal to ~85% in the Adobe Audition software. Remember, the waveform shown is of the lowest bit of a 16-bit digital signal flipping back and forth to create a 1kHz undithered "sine wave" (that looks square for obvious reasons due to the 16-bit resolution). For a DAC (Oppo) to produce that and on the other end, the ADC (RME) to capture that detail obviously meant excellent fidelity with very low noise level on both ends.

BTW, notice that unlike how Stereophile plots this test, I've created the 2 channels to be in inverse polarity so the waves don't fully overlap on top of each other. One will need to instead appreciate the symmetry of the "square box" formed by the two channels.

With these low level signals, we could set the ADI-2 Pro's settings to +4dBu reference level and capture that -90dB signal utilizing more of the dynamic range available to the ADC. The result is an even nicer looking (almost ideal) waveform:


Let's not short change the Focusrite Forte, because it's also a high-resolution capable ADC. Here's what the same waveform looks like through that device:


Not bad at all. The morphology of the peaks are not as consistent as the RME and overall not looking as nice, but certainly it's showing good channel balance with no anomalies like a DC offset that we see sometimes when looking at these -90dB waves.

III. RightMark Comparison Between RME ADI-2 Pro FS vs. Focusrite Forte

Now let's do some relative comparisons between results I obtained previously with the Forte versus the RME in RightMark. While the software does have limitations and still has a number of unfortunate bugs, it does work and allow for a nice suite of tests which over the years I've found to be replicable and accurate within the limits of test hardware of course.

As I mentioned above, by default, the RME uses "SD Sharp" (SD = Short Delay = minimum phase) filter setting. This was what I kept the device at for these measurements. While not shown here, changing this to the linear phase "Sharp" filter does not change the results to any significant degree; this is to be expected since the measurement system does not introduce non-bandwidth-limited signals to the ADC to induce any kind of ringing that might be picked up. Since changing the filter made no difference, going forward with my measurements, I will leave the ADC filter to "Sharp".

The hardware measurement chain looks like this:
Device (PonoPlayer / Oppo) --> 6' RCA/XLR (for RCA, an RCA-to-XLR adaptor used with RME) --> RME ADI-2 Pro FS / Focusrite Forte --> 12' USB2.0 cable to measurement laptop (Windows 10)
Obviously the PonoPlayer is its own source. For the Oppo DAC, I used my Raspberry Pi 3 B+ "Touch" through USB2.0. The Pi 3 was connected to my home ethernet system (10Gbps capable though not at this speed for the Pi 3 of course). All cables are generic, nothing expensive. As discussed and measured years ago, the high-priced audiophile cables are really quite unnecessary for high quality signal conduction. Of course, enjoy them if they look nice to you.

From now on, all RightMark testing will be done with RightMark PRO (currently version 6.4.5) and one change I've also made is to adjust the frequency response parameters to 20Hz-20kHz instead of the default of 40Hz-15kHz which is numerically reflected in the "Frequency response (multitone), dB" row in the summary chart. With high quality gear, reporting the full "20-20" spectrum just made more sense to me.

A. PonoPlayer!

It's fun to start with this mobile device as we know that it has some idiosyncrasies we should be looking for. A few years ago, I also compared the PonoPlayer between the Forte with E-MU 0404USB.


As expected, noise floor lower and dynamic range higher with the RME.

One of the obvious idiosyncrasies with the PonoPlayer is the Ayre "listen" minimum phase, slow roll-off filter employed which at 20kHz causes a ~4.5dB dip demonstrated using both the RME and Forte. Notice also the concordance in the calculated results for distortion measures for THD and IMD. While similar, we see the RME numbers being lower than the Forte, likely due to the improved accuracy of the device overall contributing less distortion.

Some graphs to consider...

16/44.1 graphs. Both the RME and Forte demonstrate the early frequency roll-off from the Ayre "listen" filter.

24/96 graphs. RME able to measure the frequency response a bit further, lower noise levels overall.

Grrr... RightMark Pro still has issues with graphing at 192kHz so crosstalk and IMD+N sweeps not shown. Notice for the noise level that both the RME and Forte are picking up noise with the PonoPlayer at ~80kHz.
One curious difference I saw between the RME and Focusrite device is that the stereo crosstalk result is a little higher with the RME (still looking at about -90dB). Remember that crosstalk measurements are related to the cables used and the RME measurements were done with a different set of RCA-to-XLR adaptors than the Focusrite which may have had an effect; perhaps an RCA-to-TS adaptor could have performed better.

B. Oppo UDP-205
Although I did grab some results for the TEAC UD-501 which has been a reference for me since 2013, these days, the Oppo UDP-205 and its internal ES9038Pro DAC is my reference for accurate conversion. Looking around, it's clear that for technical accuracy, it's going to be tough to beat the resolution of what this Oppo can do, even down to the intersample overload protection of the digital filters which is not the case with all ES9038Pro-based DACs. It seems the Oppo engineers have taken some extra effort in this regard.

Here are some results from the Oppo UDP-205's XLR balanced output, with digital filter set to "Linear Phase Fast":


Like with the PonoPlayer measurements above, we're seeing consistently better noise level and remarkably lower harmonic and intermodulation distortion results through the RME indicating a more accurate ADC that's not adding as much distortion to the signal.

With the lower noise floor, we're seeing a fantastic ~20-bit measured resolution with the Oppo UDP-205 used as a USB DAC!

This time around with the same 6' balanced XLR cable with no adaptors used, the results are better with the RME compared to the Focusrite Forte across the board including stereo crosstalk.

16/44.1 graphs.

24/96 graphs.

24/192 graphs.
Notice with the 16/44 test the degree of concordance between the RME and Focusrite with THD and IMD results. These days, 16-bit audio doesn't really present a challenge anymore with modern devices and at 16-bits, the Focusrite Forte's ADC is of good enough resolution such that the RME's superior performance doesn't improve the result significantly. However, once we start going into the 24-bit tests and higher sample rate, using the low noise balanced connection, then the superiority of the RME shines through!

As mentioned 2 weeks ago, RME has improved the smoothness of frequency response with the recent firmware updates. The ADI-2 Pro FS ADC is clearly capable of significantly flatter and markedly extended frequency response than the Forte as witnessed with the 192kHz measurement.

Reduced ripple with new firmware on ADI-2 Pro FS.
With the ADI-2 Pro FS, I could use RightMark to measure all the way to 384kHz with a device like the Oppo (not that I feel there's any benefit for 192+kHz music). Maybe I'll do that and show the results another time just 'cuz the device can do it.

IV. SpectraPLUS THD+N & IMD

Using the SpectraPLUS software, we can do a simple 1kHz measurement of THD(+N) in high resolution. Here is the Raspberry Pi 3 B+ USB --> Oppo UDP-205 XLR --> ADI-2 Pro FS measured with 96kHz bandwidth (ie. ADC operating at 192kHz, only 20Hz to 20kHz shown in graphs) with a 0dBFS 1kHz tone (click on image for enlarged view) sourced at 16/44 and 24/48:



We can see that the results from the two stereo channels show good consistency. Looking at the 24-bit results, there's a THD distortion factor of around 0.00015% for both channels or -116dB distortion attenuation, and THD+N of around 0.00023% or less than -112dB. SINAD (signal-to-noise and distortion) likewise of over 112dB; excellent.

Here's a peek at a standard IMD test using SpectraPLUS (again, analogue output from the Oppo, bandwidth of 96kHz, ADC running at 192kHz) with 16/44 and 24/48 signals as source:



Note that this is using a test tone set of 250Hz and 8020Hz with a 4:1 amplitude ratio (-12dB) for some variation. The RightMark test above uses the standard SMPTERP120-1994 signals at 60Hz and 7kHz when measuring IMD. Again we see that the two stereo channels are the same, both reporting around 0.00005% IMD (a tiny -126dB) - essentially no intermodulation products (this is not IMD+N, so noise not accounted for in the measurement) with the 24-bit signal.

As usual, 16-bit audio presents no problem to modern DAC or ADC hardware as it doesn't strain the limits; nonetheless I think a 16/44 signal is worth reporting on due to the vast amount of music in this bit-depth and sample rate.

Another nice measurement that can be done but not shown here is SpectraPLUS's THD+N vs. Frequency sweep which is automated by the software. I'll remember to show that when we measure the ADI-2 Pro FS's DAC output quality.

To end off this segment, here's the THD+N comparison using a -5dBFS 1kHz signal on the ADI-2 Pro FS compared to the Focusrite Forte. Distortion with a -5dBFS signal is probably more in line with actual music levels and the Oppo's +13dBu output level overwhelmed the Focusrite's maximum XLR input amount of +12dBu...



Because of the fact that the Forte cannot handle the full output level of the Oppo UDP-205, I'm about 1dB short of being able to achieve -5dB peak to reflect the Oppo's true output (without some kind of attenuator of course). In any event, the result from the RME ADI-2 Pro is about a 10dB improvement in SINAD over the Focusrite Forte; about the same amount of improvement in the THD(+N) between the two devices as well. Remember that because my ADC was set to 192kHz sample rate for the 96kHz bandwidth used in the calculations, some of the difference between the RME and Focusrite in THD+N is likely the ultrasonic noise shown in the comparison between the devices as per part I above.

V. Jitter Contributions

Speaking of distortions contributed by the ADC, we can also consider whether the ADC could be adding some jitter into the J-Test.

We already know that the Oppo UDP-205 has excellent jitter performance especially over the asynchronous USB interface. Using the 1M-point FFT from SpectraPLUS again, this is what the 16-bit J-Test looks like with the Forte compared to the ADI-2 Pro FS (same input levels into both ADCs):


And here is the 24-bit J-Test using SpectraPLUS, again comparing the result from the Focusrite Forte and the RME ADI-2 Pro using the same analogue input signal from the Oppo UDP-205:


In both instances, we can easily see the superior lower noise floor achieved with the RME. We can also better appreciate that there are much fewer sidebands than the tracings obtained with the Focusrite Forte. Note that for the 16-bit J-Test, the sidebands are all below the level of the odd harmonic low-frequency LSB square wave. With the RME J-Test result, we can make out the presence of what looks like a Gaussian bandlimited random component ("skirt") at the base of the primary signal along with some low frequency periodic sidebands rather than a mass of sidebands with the Focusrite - especially noticeable with the 16-bit test.

Remember that jitter anomalies in the frequency domain are more evident at higher input frequencies. Over the years, to strain the system even more, I've used the 24-bit (24/48) J-Test accelerated to 96kHz. I've rarely shown this test with the Focusrite Forte because the noise floor around that primary signal of 28kHz isn't as clean as I would like. So, to drive home how much better the RME is, notice how this test is not a problem with the ADI-2 Pro FS!


Impressive how well controlled noise and potential jitter is with the Oppo UDP-205 using the USB input as shown with the RME. With the primary signal at -6dBFS, even the strongest sidebands (particularly proximal to the primary signal thus easily masked even if one could hear them!) are at -130dB. Interesting that with the primary signal at both 12kHz and 24kHz, the sidebands are of the same amplitude and general morphology. I've seen Amir on Audio Science Review comment that this finding in the Oppo UDP-205 is reflective of reference voltage modulation rather than actual jitter since jitter should significantly increase in amplitude with frequency. BTW, you can also compare what I have here with his results using the Audio Precision APx555 (remember, that's a >$20k machine; nope, don't think I'll need that :-). Secrets of Home Theater and High Fidelity likewise published some UDP-205 results with the AP2700-series device but with older Oppo firmware which likely had an effect on the output results.

Want to get really extreme with these J-Test measurement? How about we accelerate the 24-bit J-Test even further to 192kHz? Here's the analogue output of the Oppo fed by my Raspberry Pi 3 B+ "Touch":


Again, a beautifully clean noise floor from which the -6dBFS 48kHz primary signal rises! Notice that the graph includes +/-20kHz on either side to look for any significant symmetrical sidebands. Remember that there is a 1kHz 24th bit square wave buried under the noise floor. Needless to say, humans would not be able to hear such a high primary frequency... It's really just a "torture test" for the digital audio gear and as you can see, the Oppo DAC fed into the RME ADC results in an amazingly clean signal. It's also a reminder as to how well a modern asynchronous USB interface is able to maintain jitter-free performance these days even when fed by an inexpensive (<$40) Raspberry Pi 3 single board computer! Make sure to let that sink in when hearing talk about jitter-this-and-that with people claiming that multi-thousand-dollar server and player systems often without even a DAC component can "improve" audible jitter performance. [Remember to have a listen to the jitter simulation yourself and consider whether you really think jitter is even worth worrying about!]

VI. Conclusion...

The RME ADI-2 Pro FS performed wonderfully as an ADC!

Comparing the ADC performance of the Focusrite Forte vs. RME ADI-2 Pro FS is important as a way to make sure the upgrade is going in the expected direction for the measurements I intend to use the hardware for. The ADI-2 Pro ADC is capable of very low noise performance that is unhindered by hum with a noise floor that remains stable over hours of operation. It is obviously capable of accurate conversion even of very low input levels such as demonstrated with the undithered -90dB 16th bit of a 16/44 signal.

Furthermore, using this ADC, I can reach down to around 20-bits of dynamic range as calculated in the RightMark tests; a nice improvement from the previous hardware with at least an extra half bit of resolution in the 20-20kHz audible spectrum. More importantly, the measured distortion levels are also significantly lower suggesting minimal distortion added by the ADC itself; realistically achieving a THD+N level better than -112dB operating at 192kHz using the stock power supply plugged into my usual power conditioner (Belkin PureAV PF60, see some measurements here using the old Gen1 hardware - maybe I'll look at this again with the better gear ;-) where I also plug in my audio system. It would be interesting trying the RME battery powered. The improvements shown in the J-Test over the Focusrite Forte likewise are substantial and suggests a significant upgrade in the accuracy of time-domain performance provided by the RME's ADC clock as well as freedom from noise intrusions. All this befitting of a professional level device.

At this point I haven't even started using higher sample rates like 384 and 768kHz for measurements but they're available. Unfortunately the SpectraPLUS-SC software currently cannot go beyond 192kHz... Hmmm, SpectraPLUS-SC developers, how about opening up the software all the way to 768kHz? As more ADCs are released with this extended sample rate, it would be nice to provide the option with a commensurate increase in the maximum FFT size to 4M points...

Despite the clear superiority of the RME, one convenient feature of the Focusrite Forte is that it can be completely USB bus-powered. As such, there might come the occasion where I would use the Forte simply for the convenience if/when I do some testing away from home for example.

With that, let's get going with using the "Generation 3" measurement system :-).

Next time, we'll start looking at the characteristics and quality of the RME ADI-2 Pro FS's own DAC output...

Autumn's here. Hope you're all enjoy the music!

Addendum - September 11, 2018:
I was reviewing the results and realized that for the SpectraPLUS data, although I had shown only the 20Hz - 20kHz FFTs, in fact the software was calculating THD(+N) and IMD for the full bandwidth that the ADC was operating at. Since I had the RME ADI-2 Pro FS and Focusrite Forte operating at 192kHz, the results, especially the THD+N would have incorporated the ultrasonic noise from the ADCs into the calculations up to the 96kHz bandwidth!

Typically when we see manufacturers publish THD+N specs, they're using a lower bandwidth like 22kHz and the number will be better looking than what I've got here (all else being equal like number of FFT bins used).

I'll do another post looking at the RME compared to Focusrite with more typical bandwidths while operating the ADC at 44.1kHz or 48kHz to correspond with the signals used to demonstrate and for comparison with what's typically seen in published results.

40 comments:

  1. Another thorough and entertaining read.
    Well done.
    If you don't mind a question unrelated to the article.

    Reading about the excellent measurements of the 205's USB input put this in my mind.
    A lot is made in certain circles regarding galvanic isolation in the USB signal chain.
    Is there a method to measure this?
    Like your jitter samples is there a way to introduce this into the USB signal chain and capture it?
    Thanks much for your writing.

    ReplyDelete
    Replies
    1. Hey Douglas,
      Yes, a lot is made about galvanic isolation. Probably mostly by folks who want to sell something in the audiophile world.

      But when was the last time we saw any measurements from an audio manufacturer showing what beneficial effect this provides; and with what equipment? Or when a reviewer claims they hear a difference, like this:
      https://www.audiostream.com/content/intona-technology-usb-20-high-speed-isolator

      Who of them can actually back up the belief with actual objective facts or demonstrations of their apparent hearing acuity to "immediately recognize a drop in the noise floor with a deeper black background"? (And why don't they!?)

      Obviously, galvanic isolation is "a thing" and of course at times we may see stray currents and noise affecting electrically connected devices. I guess it could be a problem if I were still running an electrically noisy computer connected through USB to the DAC and the DAC being susceptible to this (have not noticed this to be a problem with the stuff I've measured over the last 3 or 4 years at least!).

      These days though, I'm more likely running a low power machine like the Pi to my DAC. As far as I can tell, there's no evidence this poses a noise problem at the analogue outputs with any of the decent DACs I have. And as we see in the post today, the Pi --> Oppo is USB plus I have another RME --> Windows 10 laptop USB; neither seem to be contributing anything of significance. All the more interesting because I can look at that RME noise floor all the way to 384kHz and still not see any issues there!

      As usual, the onus of proof IMO is in the hands of those who would claim significant audible changes.

      Delete
  2. Thanks for the reply.
    Over the last several months I heavily researched a "sophisticated" file-based playback as I am growing tired of the simple WD passport drive connected via usb to my Oppo and having to navigate via a Windows based directory structure from an iPad mini.
    What I quickly discovered was that all matter of bespoke auxiliary devices needed to be inserted into the signal chain to defeat all manner of demons that exist there.
    To be honest the only items that intrigued me were custom power supplies & this galvanic isolation thing.
    All of the other supposed gremlins that inhabit the digital world, mainly in transport protocols I ignored. Having a BS in Computer Science & 30+ years in several disciplines of same told me that things that many people obsess over don't exist despite their sincere testimony otherwise.
    Quite simply & succinctly, there's a perfect reason that analog transmission protocols haven't existed in the global communications world in forever.
    They suck. They're noisy, limited in bandwidth, sensitive to external factors & unreliable in transmitting the data intact from the source to the target.
    Sorry for the soapbox but I like to explore topics with which I'm unfamiliar.

    ReplyDelete
    Replies
    1. Oh yeah Douglas,
      No question digital transmission heads and shoulders better for integrity and resolution than analogue these days! As an aside, I find it funny when the hardcore vinyl folks claim that an LP has higher resolution/fidelity than digital with absolutely zero supportable rationale for this other than idiosyncratic opinion.

      I certainly appreciate your desire to move on from using the WD Passport feeding your Oppo!

      Now as for things like custom power supplies and galvanic isolation. Sure, maybe these things will improve noise level and overall resolution of the analogue output; there's of course the possibility that they worsen fidelity which we always must remember regardless of cost or subjective hype.

      As for linear power supplies, beyond being more costly, let's also not forget the size and inefficiency (ie. more heat, wasted energy). I would suggest looking for evidence of improved sound quality before investing in something expensive.

      Remember to check out Benchmark's page on switching vs. linear power supplies if you haven't seen it:
      https://benchmarkmedia.com/blogs/application_notes/152143111-audio-myth-switching-power-supplies-are-noisy

      Delete
    2. Yes properly done digital today is superior to analog. Especially 24 bit audio has very little limitations if any.

      Delete
    3. Indeed Honza,
      I really don't think there's any debate in this regard at all. Certainly ain't the 80's anymore with 8-bit digitized audio or even early 16-bit ADC/DACs :-).

      Having said this, of course some analogue/LP's sound fantastic. Certain vinyl masterings sound better than the digital release... For those, I happily rip from vinyl in 24/96 and more likely than not, downsample to 16/48 for my music server.

      Delete
    4. Yes :-)
      I would downsample those rips to 24/48 without dithering, though ;-)

      Delete
    5. I did the same thing as an experiment: I'm totally unable to distinguish among 24/96 and 16/48 (for sure I'm deaf:)).
      Apart that what I don't understand why streaming providers (e.g. Tidal and Qobuz) don't offer 16/48 instead of 16/44.1. Marketing reasons?

      Delete
    6. Isn't that obvious? Source materal is in 44.1. Why upsample that to 48? Does more damage than good at that ratio.

      Delete
    7. Hey Honza,
      Yeah, 24-bit vinyl rips sound great of course :-). Alas, I have not been able to convince myself that keeping those lower 8-bits have been necessary given the noise level in vinyl...

      Delete
    8. It is not about noise level of vinyl, but quantization noise of the digital and necessity to dither. When you save 24/48 you needn't dither at all (or you can apply plain TPDF 24 bit dither but difference is at -140 dB). When you use 16 bit dither you should dither, preferably light noise shaping like modified-e-weighted. Also not a big deal or change, but bigger that at 24 bit.

      Delete
    9. Yeah, that's exactly what I do Honza... The dithering for my vinyl stuff is the lightest and lowest amount of noise shaping that iZotope RX 6 offers.

      Delete
  3. Great article, glad to see ADC goes first. Waiting for more measured ADCs including some DAT dinosaurs, please :) . It’s little sad to see you are waisting your talent measuring identical DACs all the time.
    Not sure it’s technically difficult, but please check the possibility to measure distortion vs signal amplitude not just frequency. Reading your blog regularly and enjoying your approach a lot.

    ReplyDelete
    Replies
    1. Hey Nobody,
      Nah, I don't think it's wasting time checking the same hardware/DACs because like I noted to Helloworld below, the goal here for me is different than reviewing a bunch of hardware. When not just listening to the music, I'm more interested in the systematic attempt to understand and in that way build on the knowledge base and develop hopefully in a rational way the philosophy around all of this. Spending money is easy, but acquiring "things" isn't what I desire in my hobby. This is why I have a number of "MUSINGS" posts, not just to show the technical stuff but try to express a broader vision of my personal beliefs in the hobby with hopefully the data and experience to back it up...

      For something like the post today, it's good to use a "standard" DAC like the Oppo and show how the ADC's differ. Also, knowing that AP results exist out there helps. Being systematic and disciplined is important to me, not just with audiophile stuff but in other domains of life. Only then can I say with some certainty that I "know" something or "understand" something well enough to feel confident.

      Delete
    2. BTW: Good point about the distortion vs. amplitude. Will look around to see what I can find!

      Delete
    3. It’s so easy to change/upgrade DACs today, but “perfect” ADCs will stay with us “perfect forewer” in recordings we love. My point, that measuring couple popular historical ADCs from the last 30 years could tell us more about music reproduction qualities than measuring newlly issued DACs.
      Your systematic and consistent aprouch is so rare in this hobby, I just trying to propose more entertaining paths :). About amplitude .... what could be more entertaining like observing splashes when signal “goes under the surface of noice floor sea” or “hitting brick wall”. You have fenomenall tools on hands, forget AP, entertain us :)

      Delete
    4. Hi Nobody,
      Yeah, I like how you think. This is very true about the almost unspoken piece among audiophiles of just how important the ADC side of digital is to the quality of music. Spending >$100,000 on a top of the line sound system is meaningless if the music being experienced originated in a half-hearted production with sub-par ADCs and studio tools.

      From the consumer side, this is I think why it's important to talk about and comment on obvious deficits like poor dynamic range and clipping in the music we listen to.

      Will certainly keep an eye out on "classic" ADC devices that would be worth examining :-).

      Delete

  4. Well, It will be much greater to borrow one Apx555 analyzer to test the real performace instead of using software which is not very accurate. Just my 2 cents

    ReplyDelete
    Replies
    1. The so called 'detailed reviews' of your favorite website miss so many details that Archimago's reviews are 20 times more worth even without an Audio Precision analyzer.

      Delete
    2. Hey Archi,

      I think you do a wonderful job here by NOT using an AP, but by teaching the forumsters how to use affordable and free tools to get the maximum of information out of them. So much can be done even as hobbyist, but often the tools are used in a wrong or invalid way. That includes self-acclaimed experts having AP and D-Scope in their hands, spitting out lots of nonsense diagrams because of wrong settings in the hard- and software. Stay on your way, it suits you well and seems to be good for us all!

      Delete
    3. Hey there Helloworld,
      Yeah, it would be nice to have the AP gear I suppose... The truth is if I really wanted one, it would be no problem to acquire. But where's the fun in that since my primary business isn't trying to design DACs? :-)

      My interest/goal is as an audiophile hobbyist - the process of discovery done in a way that other hobbyists also could and without many dollars. To seek knowledge and hopefully in the process some wisdom to speak and show things different than what the audiophile world has done in the press and typical forums. Besides, an AP machine would look terrible in the sound room, and I'm not sure whether I'd be able to listen to good music with it nor do vinyl rips when I want. :-)

      Remember that the AP devices are also very much tied to the software. They run through a Windows computer with drivers and such. IMO software that I use can be just as accurate for what I need - obviously AP will have all kinds of cool features I won't have. The limiting factor is the ADC accuracy hence the upgrades and explorations over the years!

      Delete
  5. Hi Archi. Great read. A nice and painstakingly walk through of the ADC performance of the RME ADI 2 Pro. Great to see this. Looking forward, seeing your DAC measurements. It is a pity, that I don’t have / don’t need (right now) a ADI 2 Pro for my recordings, otherwise I could have backup-ed your findings with another set of AP measurements. Juergen

    ReplyDelete
    Replies
    1. Independent AP measurements can be found in the review from German magazine StudioMagazin. The pdf is linked to the left on RME's product page, or here:
      http://www.rme-audio.de/download/StudioMagazin_ADI2Pro_d.pdf
      The German version includes pictures of all the measurements, an English translation without pictures is also there:
      http://www.rme-audio.de/download/StudioMagazin_ADI2Pro_e.pdf

      Delete
    2. Thanks for the note JR and Techland.

      Cool, nice seeing those measurements of the RME ADI-2 Pro. So, that's what the inside of the machine looks like :-).

      Delete
    3. @Techland: Thank you very much for the link to the test of this device in the Studio Magazin. Looks good.

      @Archi: Please have a look at the phase behavior in NOS mode (that we have discussed some time ago).

      Delete
  6. Let me think that my Audiolab 8200CD is becoming obsolete.
    Thanks to you I discovered the RME ADI-2 DAC, with very good declared characteristics and good price (here in Italy).
    Given that it supports DSD direct it can be easily used to compare the two architectures I mentioned some post ago:
    1. native 16/44.1 input with the up-sampling done by the DAC
    2. upsampling done by software (HQPlayer or JRiver and Sox) and DSD direct (or a NOS DAC)
    Waiting for your next post ...
    Thanks

    ReplyDelete
    Replies
    1. Sound good Teodoro,
      You'll have to let us know how the ADI-2 DAC works out for your needs! I'm pretty sure that the ADI-2 DAC will perform very close if not in many ways identical to what I'm seeing so far with the Pro. So far certainly looking very nice :-).

      Delete
    2. Upsampling by software has usually bigger potential, since we can supress aliasing and set the cutoff. Also double precision filters can be used. The question for 44.1 kHz is where the cutoff frequency should be. I used 95.2, 97.8, 92 and 93 (with ferocious resampler, with Sox it is similar but not the same since the parameters in ferocious resampler represent some slight level of attenutation at the cutoff so the fitler "starts" a little earlier) during recent years, and all of them sound well, but although not ABXable I think that they sound differently ....

      Delete
    3. Just for completists, also 92.8 cutoff sounds nice :-)

      Delete
    4. Hey Honza.
      I've been fooling around recently with 99% passband and 100% stop band in SoX to 192kHz. Just a little more CPU power. What do you think of that?!

      Delete
    5. Thanks for answer Archimago. Well … my experiences from upsampling are pretty subjective, because I am not able to prove the differences reliably in ABX.
      That said I would fully support not allowing any aliasing, therefore 100 percent stopband is highly recommended.

      Passband (cutoff) is as I wrote very, very debatable. My experience is: all the passbands 90-99 are good enough. 95.2 (Resampler) or 95.4 (SoX) is the only passband around 95 I preferred, in my opinion "pure" 95 Sox sounds a little bit "dirty". Currently I use as I wrote nice cutoff 92.8 (Resampler) and I am very satisfied :-). As for 99 yes that also sounds clean and nice, that means passband length of 200 Hz approx. But offline upsampling to 176.4/192 which I use lasts a bit long with passband 99 and even if we know that ringing is not a signficant problem with real life signal still that filter is really very steep. Rob Watts would be happy with such a filter, but he says his multi tap filter needs also other adjustments that increasing tap and decreasing length to achieve high performance.
      But yes if you have properly bandlimited signal and like almost full brickwall sound then why not you are getting close to theoretical optimum :-)

      By the way, I found out that OptimFROG software compresses offline upsampled files far better than FLAC usually, that way offline upsampling occupies less space ( and also spaces is cheap today).

      But yes online upsampling also has its strong merits and reasons to do. I use it for upsampling AAC/MP3 playback.

      Delete
  7. Would be interesting to see results when compared with Antelope's Pure 2, it's a easy win when your competitors are 3 class under...

    ReplyDelete
    Replies
    1. Hi Grof,
      Interesting looking device. Beautiful in fact!

      While I'm sure the Pure 2 must have impeccable temporal resolution, I am not sure the noise level is as good as the RME device. The sample rate is also "limited" to 192kHz and no DSD for those who want to use it as a DSD DAC (I'm thinking more of the ADI-2 DAC consumer perhaps).

      Looking around the web, I see that the AD is being done by the TI/BB PCM4222 which is actually an older component from around the 2006/2007 timeframe. I suspect the AKM AK5574 in the RME will surpass that when doing things like THD+N measurements (it is almost a decade newer!).

      Hard to know unless experimented side by side of course but I doubt "3 class under" is true!

      Delete
  8. Metric Halo has been using AKM ADC/DAC chips for many years. Their boxes are supposed to be a very good value. I hope they'll release new interface products soon with the latest AKM chips. Metric Halo boxes are upgradable, and great sounding mic preamps are optional. BJ Buchalter is a very competent designer. Would be a good comparison.

    ReplyDelete
  9. Thanks for the report, I enjoyed it.
    As you mentioned, it would be interesting to see how ADI-2 Pro behaves with a powerbank battery as a measuring device. Any plan to test it?

    ReplyDelete
  10. Stellar report, thank you! I almost bought the RME for its DSD recording features, but after checking out the datasheet of the AKM A/D chip I decided against it. For 'native DSD' recording, it appears that the ADC first converts the analog signal to PCM (including decimation, naturally) before it goes about its business. I can't remember, though, at what point the signal is converted to DSD. Anyway, it would appear that the RME is, strictly speaking, not capable of native DSD recording. Instead, I went for an ADC that does the A/D conversion natively in DSD with a Burr-Brown ADC chip. Sadly, though, it's limited to DSD128, unlike the RME. Otherwise, the RME is a truly fantastic machine.

    ReplyDelete
    Replies
    1. You might have been fallen for the wrong device then. The only block diagram in the AK5574 data sheet is only showing PCM mode. For example in DSD mode HPF (High Pass Filter) is disabled, because that would destroy the DSD signal. So at that point there is already a DSD signal present. I wouldn't wonder if the A to DSD conversion of the ADI-2 Pro outperforms your chosen BB chip by a margin. RME claims that PCM and DSD are fully identical in both modes, so you get the exact same SNR, DR and THD/THD+N either way. And the PCM values are stellar...

      Delete
    2. I just knew beforehand that this would be contested...
      And I'm impressed that you would know that I chose the wrong device without knowing what it is!
      Anyway, I had it confirmed by RME's technical staff (in a lengthy discussion about many things other than only DSD conversion), because I couldn't believe it myself either. Yes, the AKM clearly outperforms the BB, as it should as the BB is a much older design (AKM 2015 vs BB 2004/2007). The problem with these datasheets (and hence any block diagram therein) that they are far from complete and even always 100% correct. Manufacturers are not likely to divulge any trade/technology secrets or compromises in their publicly available datasheets.

      As to what you say about RME's claims that PCM and DSD are fully identical as to SNR/DR, THD+N etc. only confirms the point: both formats follow the exact same path. There's absolutely no way that with separate conversions (analog to PCM or to native DSD) you get identical results. Just think about the out-of-band (from about 23Khz onward) noise that DSD generates! Furthermore, check out the path the analog input (AINxx) follows in the block diagram: all through the decimation filter (!); further explained on page 46. Using a decimation filter is one of the reasons DSD buffs reject the DSD-to-PCM method and a sure fire indication of DSD to PCM conversion. But who knows for sure?

      It must be said, though, by following the DSD-PCM route allows AKM to implement features that would be impossible with straight Analog to DSD. It also allows RME to implement a wealth of brilliant features in their device; it is a very impressive machine. These wouldn't be possible with any of the BB alternatives. Once the signal is native DSD, there's very little you can do with it. This is extensively described in the excellent (though not for non-techies) manual that comes with the RME.

      And the specs of, say, the BB 4222, are impressive enough for a 2007 design. For my purposes, the spec differences between the AKM and BB are of no consequence whatsoever. And tell me, what analog source could even come close to either the AKM or BB in terms of specs? The farther you open the window, the more crap is likely to creep in.

      The device I ended up buying, is way cheaper that the RME, and offers 2 fully configurable MM/MC phono inputs as well as a line input. It sounds wonderful in any mode, but cannot be compared to the RME, which is much more geared towards the pro-market, whereas mine is a more or less typical consumer type device. If I had gotten the RME, I would have had to buy additional equipment to create the necessary setup.

      The reason I would like to have had DSD256 capability is the nasty out-of-band noise that can wreak havoc in wide-band amplification stages. The higher the DSD sampling rate, the less it will be a problem: out-of-band noise will be reduced and pushed higher up the frequency scale.

      Oh, and my favorite DAC (I own quite a few) is equipped with an AK4497. Talking about stellar performance...

      Delete
    3. No reason to hit me with words as I didn't do that in my reply to you!

      You wrote: Just think about the out-of-band (from about 23Khz onward) noise that DSD generates!

      That noise won't change the THD, SNR or THD+N specs at all as they are taken with audio-band limitation (20 kHz).

      For the rest I fully agree with you!

      Delete
    4. I'm sorry if I offended you, it was not my intention at all.

      I'm aware of the measurement limitations to within the audio band, though I'm not sure if that's the case here. Many measurements in the AK5574 datasheet at least, have been made over the entire frequency band the AK is capable of, starting on page 9 of the datasheet.

      But even within the limited audio band, I maintain the point that it is impossible to achieve identical measurements IF the conversion paths are truly separate (one for PCM and a separate one for native DSD). The inherent characteristics of PCM and DSD differ so much that it makes it
      impossible. And, out-of-band noise can, and often does, influence in-band behavior, for instance in the time domain and other areas.

      Either way, all this takes nothing away of the sheer brilliance of the RME machine. And let's face it, the converter chip, while important, is "just" one element of many that make a complete device what it is.

      Delete