Saturday, 22 September 2018

MEASUREMENTS: RME ADI-2 Pro FS as DAC (Part I - output levels & digital filter settings)


After having explored a couple weeks back the ADC capabilities of the RME ADI-2 Pro FS, for this installment, let's start with evaluating the DAC output quality of the device and in the process examine the objective fidelity with audio playback.

Remember, with that previous article, I had already discussed my subjective opinions based on listening sessions. My opinion has not changed in the last number of weeks as I'm still very much enjoying the sound I hear hooked up to my main system. Subjectively, the output sounds very clean, has a neutral tonality and low noise level very much like the Oppo UDP-205 and other high quality DACs. As usual, it will be interesting to compare measured results among the different devices I put through the test bench.

But before we do that... Let's start with the basics and some of the "microscopic" measurements I usually begin with - things like output levels, filter settings, etc...

Before I start talking about measured findings, it's important to mention that I am testing the device with the current latest firmware - 180/88 released around August 20, 2018.

Technically, the DAC chip inside the RME is the AKM AK4490 "Verita" Series "Velvet Sound". Actually there are 2 of these AK4490's inside the machine, each one feeding one of the headphone outputs. I mentioned in the Preview that one of the chips drives the main outputs (XLR, TS) in the rear, as well as headphone "PH 1/2" up front while "PH 3/4" is served by the other DAC chip. This allows the opportunity for independent settings of the two headphone outputs as well as balanced headphone output mode.

Remember that RME also makes the ADI-2 DAC FS, which other than having lower output levels meant for the consumer/home market, and different circuitry for the front IEM output, will likely measure similar to what you see here.

I. Oscilloscope (1kHz 0dBFS square wave, 24/44.1)

As usual, we whip out the digital oscilloscope and have a peek at the output levels, and channel balance among various other things at a low level.


The tracing above is through the rear unbalanced TS outputs at full volume (0dB). We're seeing a +19dBu (19.53Vp-p) output level. Through the XLR balanced output, the signal would be higher at +24dBu (34.7Vp-p).

Notice a nice looking square waveform with a little bit of pre- and post-ringing here because I had set it to the linear phase "Sharp" filter. The default filter is "SD Sharp" which is AKM's name for their minimum phase steep filter (as an aside, funny that AKM uses the marketing term "Sound Color Digital Filter" for the different settings). Here's the headphone output using the default "SD Sharp" filter:


Notice that this is "Lo-Power" mode with +7dBu (~5Vp-p) output level. If we kick it into "Hi-Power" mode, check this out:


Wow. "Hi-Power" can put out +22dBu (9.75Vrms, almost 30Vp-p) output level. That should be enough for any pair of headphones. Coupled with a max output current of 260mA, that should be plenty of power for low or high impedance cans (there's potential here for 1.5W into 32-ohms!).

As usual, I estimated the headphone output impedance across a 20-ohm load using a 1kHz signal. Yup, headphone output impedance is clearly <1-ohm. In fact, the specs sheet lists the output impedance as 0.1-ohms unbalanced and 0.2-ohms in balanced mode. Fantastic... These headphone outputs can handle essentially anything you throw at them.

Notice the excellent channel balance for each of the tracings above.

For the record, the "square" waveforms I've been showing in these measurements above were created in 2013 and are somewhat bandwidth limited to show a cleaner square morphology so I can have a better look at peak levels. However the waveform isn't completely bandwidth limited below Nyquist, hence there is still a little bit of ringing so we can make out the type of filter being used.

Here's what a true unfiltered 0dBFS 1kHz square wave would look like with all the ringing through the ADI-2 Pro's TS output using the "Sharp" filter:


Pretty? :-) This does show us that the DAC has enough headroom such that the ringing peaks are not being attenuated in the analogue output.

II. Digital Filter Options

As per the AKM4490 DAC chip specifications, the ADI-2 Pro FS allows the user to choose from 4 digital filters plus the "NOS" option as well (Non-OverSampled, filterless):
1. SD Sharp
2. SD Slow
3. Sharp
4. Slow
5. NOS
"SD" stands for "Short Delay" which means "minimum phase", so the first 2 are minimum phase with either sharp or slower roll-off with the latter having more ultrasonic leakage beyond Nyquist.

As usual, we can examine digital filters in both the frequency domain and as impulse responses. So often we see in audiophile magazines only one without the other. As usual, for the frequency graph, I've used a variant of the suggestion by Juergen Reis back in 2013 - over the years I've called this the "Digital Filter Composite" as a composite of the various test tones including a rather strenuous wideband white noise which can tell us if intersample overloading is present. There are also 19 and 20kHz tones that can show intermodulation and imaging artifacts.

Here are the first 2 filter settings (note that I used the Focusrite Forte to grab these at 192kHz which is why you see that idle tone at 37kHz in the images):


As you can see, these are the minimum phase options. Notice that the higher amplitude wideband white noise with 0dB peaks demonstrates some intersample overloading particularly with the "SD Sharp" setting, but significantly less with "SD Slow".

Here are the other 3 filter options:


Again, we see some intersample overloading with the "Sharp" option and significantly less overloading with "Slow". As for "NOS", damn ugly frequencies, friends! As expected, we see significant ultrasonic imaging without a proper reconstruction filter. I'm amazed that some companies and audiophiles recommend listening without a filter as somehow being a "superior" listening experience (ahem... Audio Note). Having said this, there's of course nothing wrong with doing your own upsampling of 44.1/48kHz material with proper filtering to at least 88.2/96kHz then feeding the NOS DAC.

Notice that absolute polarity is maintained with each of the filters.

Despite the intersample overloading with the default AKM filters above, I noticed that in the newest firmware (180/88), RME has implemented intersample overload protection when the DAC is fed S/PDIF digital input and internal samplerate conversion (SRC) is activated. For example, here is TosLink input upsampled to 192kHz:


Nice. Alas, I neglected to record the impulse response but from the graph we can surmise that RME used a high quality "sharp" filter here (can't tell if minimum or linear phase) with great ultrasonic suppression and no intersample overload. Notice also that this graph was recorded with the RME ADC itself (hence the nice clean noise floor).

Addendum: Had a quick look, appears to be linear phase used for SRC to 192kHz - nice, no phase shift:



III. Summary, for now...

Well guys and gals, that's all for the measurements this time :-). Now you've seen a few of the microscopic oscilloscope tracings and have an idea of not only what kind of digital filters the ADI-2 Pro FS offers but by extension the filters in the AKM AK4490 "Verita" DAC chip itself.

As I've said before, I appreciate the fact that we have all these filter choices to round out the feature set. I'm happy with the sound of good, strong, "brick wall" filtering and a little bit of intermediate phase filtering as discussed in the past.

There's obviously more to come. We'll continue next time and have a look at the DAC measurements like general characteristics such as the frequency response, noise level, distortions and jitter. Of course along the way we'll examine comparisons with other DACs I have around here.

One last thing for fun :-). Since I had the oscilloscope hooked up, I can measure the signal "peak time" (I previously used the phrase "rise time" which would not be accurate as that is usually a measure of 10% to 90% of the transition to/from the mean voltages) for square waves sent to the DAC. Remember my friend's demonstration of increased signal rise times with MQA's filter awhile back? Well, we can do something like that with each of the filters of the ADI-2 Pro FS:

RME ADI-2 Pro FS - Square wave edge rise times with the different digital filter settings.
That's cool, right? As expected, the NOS filter will allow a ringless peak-to-peak wave transition in the shortest time but of course in the absence of frequency graphs, it doesn't paint a full picture of actual sound quality with all those artifacts shown higher up using the DFC plot of the device set to NOS.

If we just looked at the square wave edge plot with time measurements, we'd be tempted to say "for sure - NOS is best!" This is not much different than companies like MQA pushing their claims of "time-domain accuracy" and articles like this one from John Atkinson recently presenting a skewed view as if these impulse diagrams alone are that important! Regarding the Atkinson article, seriously, when was drawing a fantasy, unipolar, arbitrary "impulse" in a wave editor and looking for ringless reproduction of such considered reasonable test methodology for audio equipment fidelity!? What kind of real-world audio presents itself like that "impulse" he drew!?

Remember, to claim that impulse response squiggles are somehow "ideal" simply because of a lack of pre-ringing or of short duration is ridiculous. At the end of the day, as I discussed previously about filters, the rational audiophile should recognize that there's no need to place any significant amount of weight on this. NOS (absence of filter), MQA, and Ayre "Listen" settings perform suboptimal filtering. This is why there is little ringing and allows more of the ultrasonic artifacts to seep through - allowing an "illegal" impulse wave to "look good". There is no magic here and subjectively, I have no qualms about some audiophiles preferring the sound of these filters with their associated frequency characteristics. Having said this, I must say that in the other extreme, I also see no need to go insane with hyper-extended filters like the 1M-tap Hugo M Scaler because practically, that appears to be way more processing than is needed!

Okay... We'll continue discussing the ADI-2 Pro FS as DAC next time with more measurements and comparisons.

Hope you're all enjoying the music! Happy autumn / spring - whichever hemisphere you belong to...

16 comments:

  1. Yes as I wrote in some thread before, mscaler and other Rob Watts-like filtering, although interesting and technically advanced, seem overkill to me. Yes theoretically they are getting close to the optimum but the main frequency where this has sense to think about is 44.1 kHz. With higher sample rates even more relaxed filters start far beyond audible frequencies. If we want to solve the problem of filtering further (than it is now), it would be better to use 48 kHz or 96 kHz for recordings and not to try to squeeze even more into 44.1 kHz format. That said, filtering at 44.1 is a real field for audio enthusiasts, I am currently using 92.8 passband in Resampler (92.4 in SoX), sounds very nice, but for sure there are other nice combinations, like 95.2 (95.4), etc.

    NOS DACs would be good if the filtering were not that bad. Closer to better performance are Schiit Multibit DACs but I have not heard them personally they are pretty expensive. But as far as I know they do NOS at 192 kHz (for example Modi Multibit).

    Nice measurements, by the way, thank you ...

    ReplyDelete
    Replies
    1. Hey there Honza,
      Yeah, exactly. We can play with the filtering parameters on SoX all day long and achieve very high quality results already.

      I've seen all kinds of comments and "previews" that are really IMO way over-hyped! Like this:
      https://audiobacon.net/2018/07/21/the-chord-electronics-hugo-m-scaler-your-prayers-have-been-answered/

      I mean, seriously... Changing the upsampling routine made the Chord DAVE "sound like a plastic toy"!? I mean, come on... By any standard, the DAVE already has an excellent 164,000-tap ultra-steep linear phase brick-wall filter implemented. Just what does anyone think going up to 1Megataps is supposed to "buy" in terms of sound quality?

      "Watt's" it supposed to do except cost money and add to one's electric bill? :-)

      Delete
    2. I think they build on that "we are approaching the megamillion best filter in the world" approach. While initially for sure it made sense to do high quality filtering (which is highly reasonable), on the other hand, change from 164000 to one million taps hardly can change the resulting sound more than e.g. changing the passband cutoff frequency in SoX within interval 20000-21500 Hz.
      When filtering 44.1 we have to simply accept that the result will always have some very small variations given the filtering is close to audible band and steep if we wanna preserve what we can hear.
      I would be happy if more music be distributed in 48 kHz (or 96 kHz), but 44.1 is also very good and of course usually not ABXable.

      Delete
    3. Well, there's always bragging rights to say that one has created a "ONE MILLION TAP FILTER"!!! Great number/specification for the advertising folks to play with and likewise wonderful for the consumer who owns the "best of the best of the best" to now have such a device on his audio rack :-).

      Delete
  2. Hi Archie.
    Thanks for the measurement.
    I haven’t measured AKM chips lately, so I am again surprised, that also here, those chips do not care about / prevent intersample overloads. All ESS and TI/BB doesn’t do either.
    As mentioned in earlier replies, I have some RME Interfaces for different recordings scenario, and use a good amount of headroom during recording, so this is no issue for me,
    and for mixing / mastering, I use the very nice MAAT Level meter to watch for and avoid sample and intersample overs, so also in the this case this is no issue for me,
    but for listening commercial releases, that most of them have nowadays intersample overs, I have, and it would be good to have, DACs without intersample overs.
    Juergen

    ReplyDelete
    Replies
    1. Hi Juergen,
      Thanks for the note and tips.

      Yup, still a bit of intersample overload present across the board except with the Oppo UDP-205:
      https://archimago.blogspot.com/2018/06/measurements-oppo-udp-205-part-1-output.html

      I would not be surprised if the Oppo engineers addressed this by sacrificing a little bit of the ESS ES9038Pro's massive SNR when they programmed the filters.

      I'll have a closer look into the amount of overloading...

      Delete
    2. When you do software upsampling in Resampler, you can guard for clipping and thus partially also for intersample peaks, or even arbitrarily reduce gain e.g. to 98 percent of original. If upsampling to x4 (176.4/192), further upsampling in DAC makes less intersample peaks since the "heavy lifting" is already done and level adjusted.

      Delete
    3. Nobody listens to white noise at 0 dBFS, and no music equals that kind of signal. The fact that there is no DAC passing this test might mean that it is not a valid test. Maybe it needs a different test signal to see differences between DACs. I read in the RME forum that their units have about 2.5 dB headroom for DA intersample peaks. If true Archimago's test would not show this, although in real-world usage the unit would play any CD without added distortion.

      Delete
    4. To be more precise: it is bandwidth unlimited white noise, digitally generated - an invalid signal. So even less people listening to that ;)

      Delete
    5. True Techland,
      Nobody should be listening to unlimited digitally generated white noise! The problem is that it does happen in the real world.

      I've tested my white noise signal and it will detect up to +3.01dB overload level as per Nielsen & Lund:
      https://toneprints.com/media/1018176/nielsen_lund_2003_overload.pdf

      As you can see in that paper from 2003, "hot spots" happen all the time in modern mastered music (see Table 1). Whether we can actually hear the distortion or not is up for debate of course, but it certainly does happen if one is into pop/rock. In truth, this should not happen and really should be dealt with in the studio with better audio engineering; alas, this is what we have...

      I think as "perfectionist" audiophiles, there's nothing wrong with knowing this in testing and promoting that our devices protect against adding even more distortion to the playback if possible.

      I'll look more into the RME overhead in a future post. Certainly with some filters like the "Slow" settings, it seems to handle the peaks better (will specifically have a peek at the difference there).

      Delete
  3. Hi Techland.

    Please use a “good” level meter (for example as mentioned above) und play you favorite music through it and watch the level meter, you will see, that actually all “modern” mixed music have thousand of intersample overs in each songs. Not hundreds, they are thousands.

    Or, if you have the Mastered for iTunes test tool, just drop the file into this test and you will see how many intersample overs do occur and where they are.

    Yes, white noise is by far not a “typical” music, but is just one test method to test this and can be used and reproduced all over the world.

    Juergen

    ReplyDelete
  4. PS: And to make it more clear – I haven‘t measured any DAC chip (natively as they are) in the last 10 years, that have any headroom for intersample overs, but I have measured some DACs (here I mean complete DA converters), that have headroom for intersample overs.

    ReplyDelete
    Replies
    1. Yup. And I see you've been making sure this is the case in your designs :-).

      Delete
  5. Just an example for intersample overloads. This is a Pop ballad, so not very loud, and not any Rock, Metal or Techno or Hip Hop song.

    And even there are no sample overloads, meaning that there is no data in 44k1 16 Bit, that does clipp, but the intersamples do clipp.

    Here is an excerpt from this song where you can see in the table that it shows no sample overload. But when looking at the analysis, there are a few thousand of Intersample Overloads:

    www.juergenreis.de/ISOC-Table.jpg

    And when you look at the special view of the Intersample Overload analysis graph, you see in the upper graph the level of the song, and in the lower graph the Inter-sample Overloads.

    www.juergenreis.de/ISOC-Graph.jpg

    I hope this makes clear, that you do not need to search for very loud songs. Also in modern Pop Ballad (this was a #1 Hit in 2015) you find enough intersample overloads.

    ReplyDelete
    Replies
    1. Hello Juergen, I think you misunderstood me a bit. I am well aware of intersample peaks in my music collection, and I use RME's own Digicheck level meters that show True Peak (levels above 0 dBFS) to check and analyze. My point was that many single overloads might not equal white noise, a massive simultaneous overload attack. A DAC might be able to handle typical overloads better. And what I don't get from your links is the real level value of those peaks. The second graphic seems misleading as the peak diagram in the lower part is no longer following the level scale. My music collection shows many peaks up to +1 dBFS, but only few up to +2 dBFS. +3 dBFS is very rare. So when RME claims to handle +2.5 dBFS without distortion (something that Archimago did not test here) then I feel no need to worry...

      Delete
    2. Hi Techland.

      Now I know what you mean. I am a bit short of time, because I am busy and will be “on the road” for the next days and so I do not have much time to be Online.

      Even white noise is by far from being representing “music”, you will have similar numbers concerning intersample overload headroom, as when using appropriate sinusoidal tones (yes I know, they are also far from being representing “music”), but same / similar goes with THD tests or Jitter tests, and most of the internationally used tests.

      But when you have a DAC, where 0 dBFS white noise does not produce intersample overs, these DAC will then also not produce intersample overs with appropriate sinusoidal tones. So with both methods you can be sure you are on the “safe side”.

      Your question is, how far is your music, from this upper limit.

      So when you see the numbers and levels of the possible intersample overs (= true peak, measured typically with 4-times linear phase oversample simulation), compared to “sample peak” (= the 44k1 16 Bit data), you can compare with the measured headroom.

      With both, the white noise and the appropriate sinusoidal tones, you can reduce them stepwise in 0.5 dB steps until no intersample overload will occur, as their 0 dBFS stands roughly for 3 dB intersample overload headroom.

      And my last comment, before I have to go is the MAAT Dynamic Range analysis of the above used song. Here you see the maximum level of the sample peak and the true peak.

      www.juergenreis.de/ISOC-MAAT.jpeg

      The table and the graph in my former reply are from the Mastered for iTunes test program. As I am also a MfiT certified mastering engineer, I have to fulfill, that all my masters don’t have any sample peaks and intersample peaks (even after converting into and from AAC).

      Juergen

      Delete