Let's test the output of our DACs with RightMark Audio Analyzer.

Discussion in 'Audio Hardware' started by Robert C, May 12, 2016.

  1. Robert C

    Robert C Forum Resident Thread Starter

    Location:
    London, UK
    133 dB dynamic range?! Surely there must be an inaccurate measurement there! :D
     
  2. Rolltide

    Rolltide Forum Resident

    Location:
    Vallejo, CA
    This is a question that's been on my mind lately, and I think the question highlights the need for a sober approach to measurements. You can measure something relative to another device across a range of objective criteria, but usually what we learn is that one device has a noise floor even further below the audible range then the other, THD at even further decimal place then another, etc.

    Furthermore, is there anything available today that actually measures poorly? I'd expect a $10 DAC to measure fine.
     
  3. daglesj

    daglesj Forum Resident

    Location:
    Norfolk, UK
    I just took the output and plugged it into the Line in and ran the test.

    I have to say that software is needlessly over complicated and has a crappy UI.
     
    Strat-Mangler and Jimi Floyd like this.
  4. daglesj

    daglesj Forum Resident

    Location:
    Norfolk, UK
    Tried it again, same results.

    [​IMG]

    Just to show how I selected the input/output.

    I might test again but this time without the ferrites, input short and the USB power isolator just to see if that changes anything.
     
  5. daglesj

    daglesj Forum Resident

    Location:
    Norfolk, UK
    Tried it again without the isolator and shorty, same results exactly.
     
  6. Robert C

    Robert C Forum Resident Thread Starter

    Location:
    London, UK
    Incredible :) which product is it?
     
  7. daglesj

    daglesj Forum Resident

    Location:
    Norfolk, UK
  8. Robert C

    Robert C Forum Resident Thread Starter

    Location:
    London, UK
    Shouldn't you be selecting the DAC in the playback menu?
     
  9. daglesj

    daglesj Forum Resident

    Location:
    Norfolk, UK
    The DAC doesn't appear. Its the Toslink out of the PC into the DAC and then into the Line in of the PC.
     
  10. daglesj

    daglesj Forum Resident

    Location:
    Norfolk, UK
    The DAC doesnt appear. Its the Toslink out of the PC into the DAC and then into the Line in of the PC.

    [​IMG]

    This is the setup.

    [​IMG]

    I have no idea. :confused:
     
  11. Robert C

    Robert C Forum Resident Thread Starter

    Location:
    London, UK
    Just got the renowned E-MU 0202 in for doing needledrops and testing more digital equipment :D (in case anyone's googling, I am on Windows 10 and the 0202 appears to work fine with the Win 7 64-bit driver :))

    Here are the RMAA results, check out the 24-bit crosstalk! The frequency response is a little uneven but hopefully this won't affect my transfers.

    [​IMG]

    [​IMG]
     
  12. back2vinyl

    back2vinyl Forum Resident

    Location:
    London, UK
    The crosstalk is truly amazing. And there's nothing wrong with that frequency response, in my opinion. The scale is the important thing. OK, so there's a very slight channel imbalance, but it's a fraction of a fraction of a dB and certainly inaudible. Same with frequency - so the left channel is 0.2 dB down at 30 Hz. As if the human ear could ever detect that! I reckon those results are highly impressive.
     
    Robert C likes this.
  13. darkmass

    darkmass Well-Known Member

    I've picked up a new external soundcard, an RME "ADI-2 Pro" AD/DA converter. I still have my Sound Devices USBPre 2 that was initially noted in post #47 of this thread. But I stay open to equipment upgrades where they make sense to me.

    This is what the RME unit looks like:
    [​IMG]

    Besides the knobs and buttons on the front, there are two headphone jacks that, with the right adapter, can be used together for balanced headphones. The back has an input for the supplied external power adapter; USB; S/PDIF optical in and out; via a supplied "breakout" adapter for the multipin connector, S/PDIF coax in and out, and two XLR AES/EBU jacks for input and output. For analog, there are balanced XLR outputs, unbalanced 1/4 inch TS jacks for output, and combination balanced TRS jack/XLR, unbalanced TS jack inputs. There is no provision for mic in, and there is no microphone phantom power. If digital mic recording is needed, an appropriate mic preamp would be needed. (In my case, the USBPre 2 would be a workable mic interface.) The ADI-2 Pro is designed for both studio and home use according to RME, but there are no analog RCA ins and outs. The manual (available: here), covers easy RCA adaption on page 86, and follows that with paragraphs detailing how to make best use of the four available reference levels so there should be no loss of signal quality when going the RCA route.

    Now I actually have the limited production "Anniversary Edition" of the ADI-2 Pro. It has black front and case, "hi-fi" style feet, no holes on the sides for mounting rack ears, it has gold plating on the XLR connectors, and there's additional shielding around the XLRs. There's one small additional difference, which can be seen in this page of a 20 page Super Best Audio Friends forum posting primarily covering using the ADI-2 Pro as a DAC/headphone amp. If anyone is interested, the person who provided the two photographs also wrote this review. And there's a Sound On Sound review of the ADI-2 Pro here.

    That's some background, but this thread is about using RMAA to test DACs and ADCs. Let's do that. Because of the SH Forum's budget of no more than five photographs per post, I am making use of three posts.

    Oh, though the RME is capable of up to 768 kHz AD/DA, RMAA and I can only can get up to 384 kHz. Though I've long had the type of RCA adapters RME recommends, for all the ADI-2 Pro tests I'm posting I used balanced XLR terminated cables.


    Here are test results:
    [​IMG]


    The 44.1k/16 frequency response:
    [​IMG]

    The Frequency Response window had a scroll bar below it, and I used the scroll bar to screen cap the frequency range then Photoshopped the range into a single graph.

    Throughout all tests, I used the RME recommended "SD Sharp" (Short Delay Sharp) default filter, which RME states, "offers the widest and most linear frequency response and lowest latency". The SD Sharp filter has to be the cause of the slight ripples seen in the upper frequency response. I don't much like ripples, but the amplitude of the ripples is fairly slight. The "Sharp" filter keeps the ripples, but the "SD Slow" and "Slow" filters reduce them. Also, higher sampling rates work to reduce the ripples. Oh, the SD Sharp filter has all ringing after an impulse.


    The 192k/24 frequency response:
    [​IMG]


    And, the 384k/24 frequency response:
    [​IMG]

    The 384k RMAA plot provides no way to see what is happening around 20 Hz, but a two-way AD/DA plot on page 79 of the manual indicates that for 384k sampling the response at 2 Hz (yes, that's two Hz), the response is down roughly 1/4 dB. And, yes, that same graph shows ripples at the lower sampling rates.


    That's PCM...next two posts for some DSD. The ADI-2 Pro allows the recording and playback of up to DSD256 ("quad" DSD).
     
    Ham Sandwich and Robert C like this.
  14. darkmass

    darkmass Well-Known Member

    It probably won't surprise anyone that working with DSD presents some problems. Actually editing DSD is essentially impossible if someone wishes to not leave the DSD realm during the course of the edits. The only workstation software attainable by mere, if somewhat well-heeled, mortals that even knows about DSD, Pyramix (though not the cheaper versions of Pyramix), can read and write DSD. But for editing purposes Pyramix internally uses DXD, 352.8k/24 PCM, and converts that back and forth to DSD. Us mere mortals can kind of fake the Pyramix approach...assuming reading and writing DSD is possible, that it is possible to convert DSD <--> PCM, and that someone is willing to live with whatever listening losses are expected to go along with DSD <--> PCM conversions. Naturally Pyramix itself most certainly must induce the same kind of listening losses.

    But some of us, maybe even some of us in this thread, are kind of nerds about all this...and we are sometimes in the mood to just poke around and investigate things.

    So the ADI-2 Pro can convert analog back and forth to DSD. DSD needle drops and stuff? Hey, why not! But actually recording DSD is a first problem. If Pyramix can do that, I'm not there yet. Fortunately some attainable DSD, and also high rate PCM, recording software has recently been released--for both PCs and Macs. It's even software the ADI-2 Pro manual mentions. Sound-it (in the manual), but more exactly "Sound it! 8 Pro"...for Windows, as seen here (they really should work on the name). I find the editing capabilities (which only work for PCM) rather rudimentary compared to other software I have, but I've recorded up to DSD quad and up to PCM 384k with it and that's fine. But please, please don't format convert with "Sound it..."! It's fair when converting DSD to PCM, but when converting PCM to DSD "Sound it..." rolls off the high and low frequencies atrociously. They should probably fix format conversions before they take on correcting the software's name.

    Now the "gold standard" for sample rate conversion is iZotope SRC, but it only works within PCM, and in my experience only works up to 192k Hz. (Though I think iZotope can downsample 384k to 192k, and quite likely to lower sampling rates). So a sampling higher rate, and DSD aware, SRC is needed. For that, I've gone with "XiSRC - Sample Rate Converter" (for Windows and Mac)--as seen here. And it works quite well, even if it may not reach iZotope SRC quality within PCM.

    There's a "gotcha" in the interoperation of "Sound it..." and XiSRC. "Sound it" happily writes out DSD data in the pretty much standard '.dsf' format--and XiSRC certainly converts '.dsf' to near anything; however, XiSRC does not successfully perform a conversion from .dsf when it's a DSD recording made by "Sound it" that has been written out as a .dsf file. The easy workaround is to have "Sound it" write out recorded DSD data in the ".dff" format and let XiSRC convert the .dff file to whatever output format is desired. That works fine. XiSRC can even convert the "Sound it" .dff file to .dsf, and XiSRC can then readily convert the .dsf file it produced to some other format or sample rate. But (except that .dsf can contain metadata and .dff cannot), it's more straightforward to have "Sound it" write out .dff and let XiSRC just work with the .dff format.


    Almost to the point where I can put up pictures, but there's one more thing...

    I was not able to use strictly the ADI-2 Pro to stand alone, via balanced analog, convert DSD to analog at the same time the ADI-2 Pro was converting the analog back to DSD. Nor could I use the ADI-2 Pro to convert PCM to analog at the same time it was converting the analog to DSD. I won't detail what I tried, but I could not get that to work. However, if I used my DAC to convert PCM to analog and had the DAC feed the analog to the RME while the RME was converting the analog to DSD, that worked fine. So that's what I did. I'd write PCM to the DAC, for D/A, with one piece of software, then record the A/D DSD from the RME with "Sound it". Then I'd use XiSRC to convert from the recorded .dff format to .wav format. That's a fair number of conversions, but what are you going to do?

    In any test involving my DAC, I used balanced analog cables connecting my DAC's XLR outs to the RME's XLR ins.

    In this post, I'm providing resulting images for RMAA 96k/24 test file -> DAC analog -> RME DSD in DSD1, DSD2, and DSD4 formats -> Sound it '.dff' -> XiSRC 96k/24 for RMAA results analysis. To provide some baselines, the same RMAA images will contain 96k/24 -> RME analog -> RME 96k/24, as well as 96k/24 -> DAC analog -> RME 96k/24.
    In the next post I will do the same, but in the next post I'm subbing 384k/24 for the 96k/24 used in this post.

    Note that the DAC has its own set of filters. Throughout, I had the DAC in its "SRMP" (slow roll-off minimum phase) filter setting, which provides some roll-off of high frequencies, but is generally quite listenable.

    First, the test results table:
    [​IMG]


    The frequency response results:
    [​IMG]

    In the frequency response, above, I again used the lower scroll bar and Photoshop to get the RMAA frequency range in a single image.


    Noise level results:
    [​IMG]

    You can see the DSD "single" noise beginning a rapid increase at about 20 kHz.


    Dynamic range results:
    [​IMG]

    Very similar to noise level, but some additional context.


    And, lastly, crosstalk as a function of frequency:
    [​IMG]


    Next, opening up to 384k/24 source data...
     
    Last edited: Dec 9, 2017
  15. darkmass

    darkmass Well-Known Member

    Because of all the groundwork that's already been described, this post of 384k/24 sourced data can be mostly pictures.

    First, the summary of test results:
    [​IMG]


    Frequency response:
    [​IMG]


    Noise level:
    [​IMG]


    Dynamic range:
    [​IMG]


    And, lastly, crosstalk:
    [​IMG]


    For me, it has been a lot of fun, not to mention educational in a lot of different ways, to put these three posts together. If it's all been even somewhat educational/useful to any readers...so much the better.
     
  16. Robert C

    Robert C Forum Resident Thread Starter

    Location:
    London, UK
    Great post, @darkmass ! Check out the high-res dynamic range, almost 20 bits of resolution!
     
    darkmass likes this.
  17. darkmass

    darkmass Well-Known Member

    Thank you for your comment on my post. My desire was to be reasonably thorough while staying within the purpose of your thread...a thread I've truly enjoyed. The dynamic range of the ADI-2 Pro at high resolution is pretty amazing. In a few ways, the RME box is a beast!

    Perhaps there are things in the new posts that @back2vinyl will find interesting.
     
    Robert C likes this.
  18. back2vinyl

    back2vinyl Forum Resident

    Location:
    London, UK
    Darkmass, I think that's an extraordinary piece of work on an extraordinary piece of kit. Seriously, the RME looks really nice especially if in its limited edition form and it tests amazingly well.

    I know what you mean about that ripple - it obviously doesn't matter in the slightest and wouldn't have any discernible effect whatsoever on tonality but given a choice between having it and not having it, I know which I'd prefer, given my somewhat OCD tendencies! My question is, and I hardly dare ask after you've done such an incredible amount of work on different sample rates, is whether you were tempted at all to do a comparison between all the different filters, of which there seem to be quite a few? I don't even pretend to understand the difference between all these filters - I don't even understand the difference between linear phase and minimum phase - but it might be interesting to see whether or to what extent the filters make a difference to EQ.

    The ingenuity you put into measuring the DSD performance is awesome.

    Just going back to the first test, it's interesting to see the difference in measured performance between 16-bit and 24-bit. This was evident in Robert C's last test, too. Again, probably not audible but it does perhaps suggest some rational basis for preferring 24-bit to 16-bit where there's a choice.
     
    Robert C and darkmass like this.
  19. darkmass

    darkmass Well-Known Member

    Back2vinyl, I very much appreciate your comments. Thank you. I was hoping you would again find your way to this thread--you were certainly in my mind while I was working up my text and images. I knew the posts would touch on a few of your areas of interest. The same is true for Robert C of course. I think at this time, however, everyone else in the forum probably bailed out before they got through the first page. :)

    I had a lot of thoughts and a lot of explorations before I settled on the form the three posts above would take. Though I thought some of the unpublished thoughts were worth passing on, I was attempting to keep the three posts to a single thrust in an effort to minimize complications.

    I did take a glancing look at the various ADI-2 Pro filters--one reason I was able to state the "SD Slow" and the "Slow" filters reduced the ripples. I did no detailed comparisons of the filters however. But I get some obscure satisfaction from explorations, and from writing up posts distilling the explorations, so I think I'll cook up something to tack onto this thread. One complication is that all this RMAA testing is D/A -> A/D, and the RME permits (I think, but have not tested) the D/A and A/D filters to be different from each other. At least the filter settings look like there are discrete selections permitted for D/A and A/D. Maybe there are reasons to bring my DAC into play once more. Maybe. I'll have to do some thinking.

    Also, a fuller understanding of linear phase and minimum phase is beginning to dawn on me, so perhaps I'll be able to venture an explanation as I am starting to understand things.


    I'll add one more thing I just might fold in--most likely, will be a separate post in this thread...

    Say an end result of analog source recording would be to burn to regular CD, 44.1k/16 files. The rippled RME 44.1k/16 files could well be optimum, but I like linear stuff as much as you. (You hardly need to tell me about OCD!) My conjecture is that RME recording at 192k/24 then downrezzing to 44.1k/16 might be a splendid way to go--with no ripples in the results. But that certainly depends on the SRC. It would probably be fun to put in the same frequency response plot: direct RME 44.1k/16 recording versus RME 192k/24 recording -> iZotope 44.1k/16 versus RME 192k/24 recording -> XiSRC 44.1k/16. That working with different base sampling rates kind of stresses RMAA (it doesn't want to do it...and the iZotope and XiSRC results would each involve two sampling rates in the course of things). But in the process of considering my initial three RME posts, I successfully carried out a personal "demonstration of concept" that seemed to result in being able to put correct frequency response results for different sampling rates in the same physical graph (without Photoshop). I'd explain if this does result in the post I'm hoping for.


    Stay tuned.
     
    Robert C likes this.
  20. back2vinyl

    back2vinyl Forum Resident

    Location:
    London, UK
    I agree, it would be fun to compare a downsampled 192k recording with a true 44.1k recording but it sounds as if it will require another large dose of your ingenuity. Good luck with that! I will watch and wait.

    I don't think my old Lynx Hilo offers me any choice of filters at all, but I'm almost glad of it in a way since I'd have no clue which one to choose!
     
  21. darkmass

    darkmass Well-Known Member

    Time for a new set of three posts. I'll take on the quoted idea first--posting the results of different PCM sampling rates in the same physical RMAA graph. The two posts following this one will be: ADI-2 Pro D/A and A/D filters...and after that, sample rate converter strategies for analog to PCM recording with the ADI-2 Pro (but likely just as useful, or unuseful, when recording with any A/D).

    The approach for putting multiple sampling rates in the same graph occurred to me when I was thinking about using foobar2000 DSD to PCM as part of my RME DSD evaluation. The problem is that the PCM sampling rates that foobar DSD to PCM uses and the sampling rates that RMAA uses don't quite mesh. Foobar likes DSD -> PCM sampling rates like 44.1 kHz, 88.2 kHz, 176.4 kHz, and 352.8 kHz. RMAA is happy with 44.1 kHz, 48 kHz, 96 kHz, 192 kHz, and 384 kHz. Yes, 44.1 kHz is in common, but in my mind that's really too low a sampling rate to be useful for full explorations. RMAA can write out 88.2 kHz and 176.4 kHz test files, but RMAA cannot successfully read those same test files back in--even if they have not been modified in any way. RMAA can read back in a 352.8 kHz test file it has generated, but even when I ran a 352.8 kHz RMAA file through a very simple process, RMAA refused to read the 352.8 kHz file back in. Swell.

    Somehow it occurred to me to take a fresh 48 kHz test file, sample rate convert the file to 44.1 kHz, then run that file though (SD Sharp) RME D/A/D and sample rate convert the result back to 48 kHz--just to see what RMAA might do with the result. RMAA not only accepted the result, it showed data running to only 22.05 kHz (half the 44.1 kHz sample rate, natch). And using the RMAA plot's cursor reading capability, the SD Sharp frequency response ripples displayed at the correct 44.1k sample rate locations...and with the correct amplitudes!

    Ultimately, I decided to not use foobar, I wanted to keep any sample rate conversions minimized when I was looking at DSD, so I used natural RMAA frequencies and the XiSRC DSD <--> PCM format converter--which had no problem using the natural RMAA frequencies. But what I had discovered was still intriguing, and assuming the extra fuzziness arising from additional SRC modifications was understood and acceptable, there might be situations when putting multiple sampling rates in the same plot would help with understanding.

    Now I was not able to take something like a 96k RMAA test file, SRC it to a different frequency, and have RMAA read the result in at the new frequency. No, I had to SRC the test file to one frequency then SRC it back to the original test file frequency before RMAA would read it. Okay, got it.

    This is all hardly a guarantee of useful results, even if RMAA will read the final file back in and plot things. Some of the RMAA tests will work out better than others...and there are those two extra (at least) SRC conversions which should certainly make things fuzzier.

    Still, let me show everyone how this works...


    From post #87, here is the test results table once more:
    [​IMG]


    For comparison, here is the equivalent test results table when I started with a 192k/24 file for each table column, SRC'd to the appropriate frequency before running the file through RME D/A/D, then SRC'd each result back to 192k/24. I chose 192k because the frequency response plot included 20 Hz...and it also showed when a sampling rate like 384k "ran into the wall" on the far right side of the frequency response plot.

    Interestingly, everything in the "Frequency response (multitone)" row is an exact match for the table above:
    [​IMG]

    Some of the other rows do well, some rows don't quite.


    The resulting Frequency Response plot:
    [​IMG]

    The 44.1k/16 and 44.1k/24 lines completely overlie each other, which should be the case. SD Sharp ripples show in the same sampling rates that they were visible in during my previous RME posts.


    The Dynamic Range results:
    [​IMG]

    The 44.1k/16 Dynamic Range data is kind of messed up--I think because the original 24 bit file was knocked down to 16 bits for running through RME, then changed back to 24 bits. But it gets the idea across of how a 16 bit depth suffers in dynamic range compared to a 24 bit depth. You can also see that the meaningful data for each sampling rate ends at the right place. (For the record, however, natural 44.1k/16 RME D/A/D results can readily be posted in the same RMAA graphs as 44.1k/24 D/A/D results. The identical sample rate is the key. Doing this provides a more natural, and easier, way to look at 16 bit versus 24 bit dynamic range results.)
     
  22. darkmass

    darkmass Well-Known Member

    There is certainly much, much more I could, and probably should, do to arrive at a deeper understanding of filters than the understanding I have. But I'll give you my general, if largely unfounded, impressions...I now have used RMAA to get a pretty good look at the EQ effects of different ADI-2 Pro filters--both as they apply to D/A and as they apply to A/D. Let me also suggest you give a look at manual pages 76 through the top half of page 79 for some detailed information specifically relating to the ADI-2 Pro. (For convenience, I'll repeat the manual location.)


    Due to the Nyquisty nature of PCM audio, all audio frequencies above half the sample rate have to be filtered out before PCM sampling. If that filtering out is not done, frequencies above the half sample rate will fool any analog producing "reconstruction filter" into inventing audio frequencies below the half sample rate that are not in the original source audio content. So some form of audio filtering before PCM samples are a necessary fact of life. The most prominent filter types, as general classes, have become "liner phase" and "minimum phase". Such filtering is generally familiar in DACs. Also, it seems no filter is perfect, any filter type can have side effects.

    Given my lack of filtering knowledge, my perception is that filtering continuous sine waves is no particularly big deal. However, any filter works on everything presented to it, and a situation where unintended filtering effects are particularly visible is with "pulses"--and that certainly includes pulses down to the width of a single sample. In the top portion of RME manual page 77, the D/A effects of single sample pulses are shown for four RME D/A filters. Not one of the four filters is able to show a pure single pulse, each shows some variation of "ringing" effects. What is typically described as "ringing" is really an artifice of the digital filtering process. (As an additional note, what RME calls "short delay" filters are usually described as "minimum phase" filters by other DAC makers. Likewise, the RME "Sharp" and "Slow" filters are elsewhere described as "linear phase" filters. My DAC's "SRMP" [Slow Roll-off Minimum Phase] filter is equivalent to the RME "Short Delay Slow" filter--though the actual frequency response results of the two respective filters differs by degree of effect.)

    Now such filters can be pretty good, and certainly can do their job effectively, but the side effects are there. The filters are effectively special purpose algorithms that are designed (while the designer keeps in mind the ultimate audio goal) to provide the designer's favored balance of desirable characteristics and undesired side effects. In a fashion that is still a mystery to me, some frequencies of the set making it through a given filter can be "delayed" (by somehow adjusting "phasing") to end up starting a very short time later than other frequencies in the set that pass through filtering. If frequencies pass through the filter in a manner that is phase-neutral to all frequencies, this is "linear phase" and the resulting filtered residue of the source "pulse" will look symmetrical on the audio timeline--even though in real nature nothing should happen before the burst of a "pulse". A "minimum phase" filter is when the various frequencies of the ringing effects produced by the filter are adjusted just enough so that no ringing artifacts occur before the spike of the filtered pulse. This "minimum phase" phase adjustment produces ringing artifacts that go on for more time than they should (even if it's really a quite brief time), but many feel this is a more natural sounding, and less digital, effect than when any ringing artifacts occur before a pulse.

    The duration of ringing (both for "linear phase" filters and for "minimum phase" filters) can be reduced, but this seems to require amplitude reduction of high sample rate frequencies...and this tends to roll off the higher frequencies in any final frequency range. You will certainly see examples of reduced high audible frequencies in the RME frequency response plots I am posting.

    Why don't we move on to some pictures?


    Primarily, there are two classes of filters in the AMI-2 Pro:

    1) SD Sharp, and Sharp
    2) SD Slow, and Slow

    Any of these four filters can be selected for the RME D/A, and, independently, any of these four can be selected for the RME A/D. That is, you can have RME D/A/D that is something like Slow -> SD Sharp as easily as you can have something like SD Slow -> SD Slow. There is a fifth filter, "NOS", which can only be used on the D/A side, but we'll get to NOS after first looking at the Sharp and Slow sets.


    Here's a Test Results table for the various Sharp and Slow filters:
    [​IMG]

    I've used 44.1k/24 sampling throughout because frequency response ripples are very prominent at 44.1 kHz, and I chose 24 bits deep just for greater noise suppression. For the moment, just look at the first four columns in the table...for each of these four columns, the same filter was used for A/D as was used for D/A. For any table row, the results across the first four columns are very similar--except for Frequency Response, which falls into a "Sharp" group and a "Slow" group.


    Here's the corresponding Frequency Response graph:
    [​IMG]

    As it turns out, the "sharp -> sharp" frequency response completely overlies the "SD sharp -> SD sharp" frequency response, and the "slow -> slow" response completely overlies the "SD slow -> SD slow" response. The RME manual does say the Sharp and SD Sharp filters have the same frequency response as each other and that the Slow and SD Slow filters have the same response as each other--the RMAA graph certainly demonstrates that. What is much less clear from the graph, and no more clear from manual statements, is: does the Slow D/A filter have the same frequency response as the Slow A/D filter...and so on?


    For further insight, Test Result table columns 5 and 6 come into play. Since the frequency response for "Slow" is identical to that for "SD Slow", and so on, column 5 is "SD Sharp -> SD Slow" D/A/D filtering and column 6 reverses that to "SD Slow -> SD Sharp" D/A/D filtering. If column 5 frequency response is identical to column 6 frequency response, then the D/A frequency response for any given filter will be identical to the A/D response for the same filter.


    The frequency responses for column 5 and 6 don't look like each other in the table...here's the graph for more clarity.

    For reference, columns 1 and 2 have been included:
    [​IMG]

    This graph tells us a lot!

    For one, a careful look reveals the more ripply responses belong to column 1 and column 6 traces; what these two traces have in common is "SD Sharp" on the A/D side of things. There may be a slight, slow period ripple for column 5, when SD Sharp filtering is used in the D/A, but it seems much less significant than the SD Sharp A/D effect. So RME D/A/D rippling is significantly due to either of the "Sharp" filters during A/D. On the other hand, columns 2, 5, and 6 make clear that one of the "Slow" filters anywhere in the D/A/D chain will cause significant roll off of high frequencies. Of course any of the even most extreme RME ripples is very slight in amplitude and likely cannot be heard. If I were to offer a judgment based on this graph, I'd say that the RME D/A characteristics probably means it a pretty fair DAC if the Sharp or SD Sharp filters are used for PCM. I'd also venture that the RME A/D characteristics make it quite reasonable for recording analog to PCM if Sharp or SD Sharp filters are used for the A/D.


    Let's get back to looking at RME's use of "SD" in some of the filter names, as well as RME's use of "Sharp" versus "Slow". In the four illustrations in the top portion of manual page 77, the two response illustrations with "Short Delay" in the titles ("Short Delay Sharp" and "Short Delay Slow") have all ringing after the pulse. That is, the Short Delay filters delay all ringing just enough that the pulse precedes all ringing effects. While the two "Sharp" filters are illustrated on the page with a lot of ringing, the two "Slow" filters show almost no ringing...this is done by adding a range of increasing, high frequency volume reduction to the "Slow" filter characteristics. The "Sharp" filters, in effect, don't kick in till approximately 20 kHz is reached, then they have to act "sharply" to filter out all frequencies before 22.05 kHz is reached. The two "Slow" filters start their actions earlier in the frequency response range (starting at about 10 kHz, from looking at the RMAA frequency responses), because they taper down a larger range of frequencies than is necessary for the Sharp filters.


    Now, only the "NOS" filter to go...

    Here is RME's statement from page 77 of the manual:

    "The DAC includes another filter which is called Super Slow in its data sheet. The impulse response looks perfect, but checking the output signal with an Oscilloscope reveals steps that are more typical for so called Non-OverSampling (NOS) devices, so we renamed it NOS within the DAC filter menu. Note that there is no audible distortion, the steps equal high frequency harmonics that are mostly higher than 20 kHz. Please also note that Slow and NOS filters cause much more aliasing into the audio band and out-of-band noise than Sharp filters."


    Here's a Test Results table that includes NOS filter D/A columns, along with two previous reference columns:
    [​IMG]


    And the corresponding Frequency Response plot:
    [​IMG]


    No comment.
     
    Last edited: Dec 24, 2017
  23. darkmass

    darkmass Well-Known Member

    It's time to take a look at that.

    My first consideration was to make a high quality analog signal available to the A/D side of the ADI-2 Pro. While the evidence so far is that the D/A side of the ADI-2 Pro is more than capable of providing high quality analog output, I had not tried running the two sides of the RME box at different sample rates. (And that particular investigation is still in the future.) Also, PCM audio D/A involves filtering, and I wanted the "high quality analog" signal to not have passed through any D/A PCM filtering. The RME A/D was going to be filtered, and working to minimize the effects of the A/D filtering was part of the investigation...but I wanted the ultimate results to be as pertinent to the A/D filtering as possible. I also knew that RMAA would properly analyze 44.1k/16 results if the original RMAA test file was also 44.1k/16.

    So I decided to take an RMAA 44.1k/16 test file and use my XiSRC format converter to convert the RMAA file to double DSD, then use my DAC to convert the double DSD file to analog for RME PCM A/D use. DSD processing generally reduces the filtration needed since there are no worries with DSD about Nyquist limit aliasing. DSD DACs might typically have high frequency filtering to reduce, or eliminate, the energy of the high frequency "DSD hump", but by using double DSD I figured that such DSD filtering would be well away from the audio range 44.1k sampling was concerned about.


    My initial check was to use XiSRC to first convert the RMAA test file to double DSD, then use XiSRC once more to convert the double DSD back to 44.1k/16 for RMAA analysis. This procedure made no use of my DAC or the RME, it was just a look at the software conversion process. However, when I got the result back in RMAA, the RMAA frequency response showed a slight bit of "grittiness" at high frequencies. The amplitude of the grittiness was lower than the amplitude of RME SD Sharp ripples, but it seemed the analog signal generated by my DAC from XiSRC results wouldn't be quite as clean as I was hoping.

    So I turned to a DSD/PCM format converter I had purchased shortly after getting the ADI-2 Pro: AuI ConverteR--specifically, "AuI ConverteR PROduce-RD" as can be found somewhere on this complicated page. I had originally intended using this quite good quality converter for my RME DSD write-up in posts #88 and #89, but in the process of that I discovered a significant AuI complication bearing on that particular write-up: by design, AuI filtered out everything above 100 kHz in converter output. (And even that depended on the right option setting. The software's author truly believes any audio above 20 kHz is not reasonable, and the AuI default filter option eliminates all audio output above 20 kHz--no matter the PCM sample rate or DSD rate desired.) Now a 100 kHz audio limit is hardly a matter of real concern, but when I was trying to examine high frequency effects when using DSD at up to quad DSD, well, that 100 kHz limit made me kind of unhappy. Fortunately, I later found XiSRC, and XiSRC saved the day for full-range DSD frequency response and noise examinations.

    So I now tried AuI to convert the RMAA 44.1k/16 test file to double DSD, then back once more to 44.1k/16 for RMAA analysis...and RMAA showed the AuI results to be a dead flat line--which ended at the proper 20 kHz for 44.1k sampling. So I could use AuI produced double DSD and my DAC to serve as a high quality analog signal generator for use by RME A/D.

    Now that I had that, I was able to make one RME A/D pass with 44.1k/16 sampling of the DAC's double DSD -> analog, and one RME pass with 192k/24 sampling of the DAC's analog. Then I took the RME 192k/24 output file and separately used XiSRC, iZotope SRC (within "Sound Forge Pro 11"), and also AuI to downsample the RME 192k/24 output results to 44.1k/16 for RMAA comparisons.


    Here's the Test Results table:
    [​IMG]

    - The first column in the table is the analysis of a completely unprocessed RMAA 44.1k/16 test file.

    - The second column is the result of using AuI to convert the RMAA test file to double DSD then back to 44.1k/16 with no additional processing.

    - The third column is the result (from my very first RME post) of using SD Sharp RME D/A/D to pass through the 44.1k/16 test file. The first three table columns serve as baselines for the various SRC concepts being looked at.


    On to the right-hand side of the table...

    Data column four is the result of RME 44.1k/16 sampling of the double DSD -> DAC analog signal. Naturally, this corresponds to just using the RME A/D side for a file suitable for straight CD burning of any analog source.

    Data columns five, six, and seven are the result of RME 192k/24 sampling downrezzed to 44.1k/16 by, respectively, XiSRC, iZotope SRC, and AuI ConverteR.


    RMAA frequency response graphs provide some more complete information.

    This is the Frequency Response plot of original RME D/A/D two-way 44.1K/16 sampling in comparison with having RME A/D 44.1k/16 sample the double DSD/DAC "clean signal":
    [​IMG]

    You can see the two curves agree up to 400 Hz, then they slowly diverge on up to 20 kHz. I measured the maximum separation to be in the vicinity of 0.02 to 0.03 dB. It's reasonable to consider 0.03 dB to be the amount the DAC DSD D/A rolls-off in comparison with RME PCM D/A (at least at a 44.1k RME sample rate). That's probably not a very serious roll-off. The above graph helps serve as baseline information.


    The next Frequency Response graph is where things get interesting...and it's the reason for this post:
    [​IMG]

    I gave the graph the maximum vertical expansion allowed by RMAA (as I did for the plot just above). I had first seen that the AuI curve exactly overlaid the iZotope curve throughout the displayed frequency range, so there was no point in including AuI in this particular graph.

    You can see the slight XiSRC "grittiness" I was referring to, but it's truly of small amplitude--and of significantly less amplitude than the RME ripple. However, the RME ripple is only about 0.06 dB peak to trough, so the RME ripple seems like not much of a concern. iZotope SRC (and AuI) output is smooth, with no grittiness or ripples--though it looks like there is a low amplitude slow roll in the iZotope SRC results.


    Based on pure visual frequency response results, high rate RME A/D sampling followed by iZotope (or AuI) downrezzing would be the best way to go...while admitting the distinctions between any of the listed approaches is probably not discernible to human ears. However, it's also worth noting from the Test Results table that in the various categories of noise, iZotope downrezzing tends to be worse than for straight 44.1k RME sampling or for 192k RME sampling followed by XiSRC or AuI downrezzing to 44.1k. RME 44.1k sampling and AuI downrezzing to 44.1k/16 are essentially tied with each other for the best noise results throughout.


    Here's the dynamic range graph:
    [​IMG]


    And the crosstalk graph:
    [​IMG]

    Though iZotope shows a notable, but strange, hump in the dynamic range and the crosstalk graphs (and other noise graphs) beginning at about 15 kHz, I'd believe any of the four approaches I tried (and undoubtedly many SRC approaches not tried for my post, but shown here) would lead to completely satisfactory results. If any digital editing was to be involved it would probably make the most sense to have RME A/D sample at a high rate, then not downrez till all editing was completed. Of course, if no CD burning was to be involved it could make sense to keep the results at a high sample rate.
     
    Last edited: Dec 24, 2017
  24. back2vinyl

    back2vinyl Forum Resident

    Location:
    London, UK
    Darkmass, once again I can only gasp with admiration at the ingenuity you've employed, not only in finding ways to carry out these tests but also in finding ways to display the results in an accessible manner.

    My head is spinning with all this information but for the moment, just in response to your second post, about the filters:

    I suppose in one sense RMA is a tough test because it involves a D/A as well as an A/D conversion and therefore any effect of the filter is going to be doubled up compared with normal use, when you'd usually only be going in one direction. That aside, while still struggling to get my head around this whole filter business, I very much enjoyed your explanation for the difference in frequency response between the slow and the sharp filters and it certainly seems to be supported by your test results. I also enjoyed the remorseless logic in attributing the ripple to the SD Sharp A/D filter! As you point out, however, an inaudible ripple is a lot preferable to a potentially audible roll-off so the manual seems to make a good call in recommending use of the SD Sharp filer. As for the NOS - yes, no comment.

    I think it's useful doing these tests because one could waste a lot of time trying to detect barely audible differences in frequency response between the filters - and possibly making mistakes - while there is no arguing with the test results. What's less clear to me is the degree to which the other effects of filters may be important, notably the ringing. What I don't know is whether or to what extent this is genuinely audible.

    I'm also wondering what filters other manufacturers use, such as Lynx (as per my Hilo, which offers no choice of filter) or your Sound Devices USBPre 2. Is this incredibly minor ripple actually commonplace and is it always found with any half-decent filter, or do other manufacturers find some way of avoiding the ripple while still minimising the ringing and getting a flat frequency response? I doubt whether that's possible - there are always trade-offs.

    I will leave it there just for the moment and have another look at your amazingly thorough researches later.
     
  25. darkmass

    darkmass Well-Known Member

    Back2vinyl, your personal chops are rather far from inconsequential. There are threads of yours where I do not have useful knowledge to add (clear example) where, even so, I can only stand in awe of your discipline, logic, science, and the equipment experiments you readily bring to bear as you work to increase your knowledge and to solve problems. If people conclude I'm a fan of yours, well, chips will fall where chips like to fall. That you and I have worked on some situations in concert, and with fortune could do more of that in the future, is something that makes the SH fora a very happy place for me to frequent. And, I am always grateful for your responses to my infrequent posts.


    Okay, genuine audibility is a tough knot to unravel. If some sound characteristic is immediately audible to all listeners, quite likely it's a very, very pronounced effect indeed. A cymbal struck during a violin solo. Even within the somewhat rarified reaches of the members of the Steve Hoffman site, there have been innumerable threads (often enough ultimately locked down) devoted to arguments over whether one aspect or another of some digital or analog characteristic results in "significant audibility". Not to mention that the human psyche takes great joy in getting into the mix. "Expectation bias", as one important phenomenon, can induce some to hear things that (may or) may not be there in some instances, as well as induce the same people, or others, to not hear things that may (or may not) be there. Experience and ear training can enable some to hear things they once did not notice, fatigue or certain age related issues can decrease sensitivity to some subtle sound matters. My broad (and perhaps weaseling away from the issue) advice is that each person, if they care to, should learn what matters to them, then should build up a skill set that will bring them the happiest results.

    If this is an example, in the case of my Sound Forge Pro's iZotope SRC implementation, for the tests above I used the "Simplified Quality" setting, with the "Quality" slider at "Highest" quality (which is what I always personally use for uprezzing). For my own personal iZotope downrezzing I once experimented with the "Prering %" slider (which is locked at 100%, i.e. full "linear phase" filtering during "Simplified Quality"--while 0% is full "minimum phase"), and ultimately decided I liked 25% "Prering" the best. I now use 25% Prering for my personal iZotope downrezzing. That said, I wanted to keep the tests above as repeatable, and as "generally used", as possible so "Simplified Quality" is what I used here for up or down rezzing. Along those same lines, there has been one lengthy "Computer Audiophile" forum thread where users would provide their own preferred iZotope settings (and if the "Simplified quality" check box is not selected, five iZotope SRC attributes can be modified to taste.) There was no universal agreement to what the best settings might be.

    Of course, you specifically asked about the hearability of "ringing", and my answer to that particular filter side-effect is: I do not know. I know that 25% "prering" worked (or I imagine it worked) for me in iZotope, and I very much like the philosophy of "minimum phase"...but that may well be how I wish to set my "confirmation bias". For hearability by others (and, to be honest, for hearability for myself), I have no firm conclusions just yet.


    On other manufacturers finding ways to avoid a "ripple"... Possibly some do. But along those lines, I read statements made by Charles Hansen (of Ayre Acoustics) not long before he passed to the effect that for Ayre (including the Pono player), he would personally spend significant time adjusting filter parameters (for the effectively "off-the-shelf" chipsets he and other manufacturers used) to get things optimized to his liking. He also felt it was exceptionally rare for others to do any such adjusting at all.

    So for what manufacturers in general do to tune D/A or A/D characteristics... Such tuning has to have a development cost...and I'd guess that some few do such tuning, but that Charles Hansen likely had been correct: many, many do no tuning at all.

    Indeed, there are always trade-offs.
     
    Ham Sandwich likes this.
  26. darkmass

    darkmass Well-Known Member

    I received a PM that has given me additional information concerning the ADI-2 Pro and additional information for one other matter.

    I have the author's permission to post the following extraction from the PM...

    The ripple is part of the used ADC chip, the AK557x family. There is no way around it than using a different chip. In an interview in a German magazine RME stated that they did not like the ripple (like anyone they would have preferred less ripple as found with other ADCs), but still used the chip for the Pro as it was so 'damn good in THD and SNR - and allows to convert to DSD'. Meanwhile the AK557x chips are also used in other, partly very expensive units, for example Merging, and show the exact same ripple there.

    Then this:

    "I was not able to use strictly the ADI-2 Pro to stand alone, via balanced analog, convert DSD to analog at the same time the ADI-2 Pro was converting the analog back to DSD. Nor could I use the ADI-2 Pro to convert PCM to analog at the same time it was converting the analog to DSD."

    The manual includes a note about DSD operation, it is basically independent on AD and DA. But you won't find a software that can record DSD and does playback at the same time as PCM, and the Windows ASIO driver also can not do that. The unit should be in Stereo/2-channel mode, then you can use the DAC with digital to DSD conversion on output 1/2 and digital to PCM conversion at output 3/4, simultaneously. This mode comes into effect when you change the source of output 3/4 in the display's menu to Analog In or Digital In with a PCM signal as source. At least that's what I remember, maybe I have to try that again...but I never need that when I use the unit...

    And allow me to add the final point from the PM, this one concerning the quandary that was initiated by post #75...

    There was a digital loopback active within the Realtek driver/soundchip [for the Fiio DAC], from digital out to record in. The analog Line input was not measured at all.
     
    Robert C likes this.

Share This Page