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

Digital Distortion

Discussion in 'Digital Audio' started by Wombat, Nov 13, 2017.

  1. Wombat

    Wombat Member

    Joined:
    Nov 5, 2017
    Messages:
    84
    Likes Received:
    25
    Location:
    Australia
    Are there any common standards for measuring and evaluating distortion in digital audio circuits?

    Can the the consequent results be published in a standard way that allows meaningful comparison between devices, as with analogue audio circuits?

    Can a level of inaudibility be set for summated digital distortions?
     
  2. DonH56

    DonH56 Addicted to Fun and Learning Technical Expert

    Joined:
    Mar 15, 2016
    Messages:
    919
    Likes Received:
    337
    Location:
    Monument, CO
    What do you define as "digital audio circuits"? And what makes them different than any other digital circuits? Not trying to be an arse, trying to define the scope of your question.

    There are a myriad of things that can go wrong with the transmission and reception of digital data, but in the end it all boils down to bit errors, correctable or not, how many, and what they do to the end result. For digital audio I suppose you could add clipping in the digital domain.

    You can look up standards for AES3, S/PDIF, I2C, SAS. SATA. PCIe, FC, etc. etc. etc. to see how "digital" testing is performed and compared.

    Audibility happens when the bits are converted to audio, so you'd have to translate audible errors back to the bits. Things like jitter are relatively easy, assuming somehow the jitter makes it to ADC or DAC (that is not technically a "digital" thing), but I do not know if there are studies showing how large and how long an uncorrected bit error (series) must be to be audible.
     
  3. Wombat

    Wombat Member

    Joined:
    Nov 5, 2017
    Messages:
    84
    Likes Received:
    25
    Location:
    Australia
    Audio circuits as used in audio systems and devices. I would usually accept the traditional distortion measurements of the eventual converted-to-analogue output but I often read that listeners hear digital 'artifacts' in the music. How are the various artifacts quantified for comparison purposes? Audibility assessment?
     
  4. fas42

    fas42 Major Contributor

    Joined:
    Mar 21, 2016
    Messages:
    2,649
    Likes Received:
    100
    Location:
    Australia
    It is extremely rare to hear pure, digital 'artifacts' - just by chance, an early opera CD I bought has them encoded on the very first track - someone screwed up the mastering, and in the first minute or so the levels fed to the ADC were completely wrong - you can hear the bits flipping in the 'silent' gaps. Then they realised what was going on, corrected it - but never went back and redid the conversion.
     
  5. Blumlein 88

    Blumlein 88 Major Contributor

    Joined:
    Feb 23, 2016
    Messages:
    2,071
    Likes Received:
    537
    Well you need to ask those listeners. Mostly I think that is all in their mind. Assuming maybe the old ideas digital is harsh, cold, hard, metallic sounding. The 'digital' sound.

    Digital has some kinds of distortion that might be uniquely digital with artifacts in the analog result. Jitter if high enough in level has been mentioned. Aliasing is another. This is when frequencies above the cutoff (say 22.05 khz for CD) are reflected back down in the below 20 khz band at low levels. Low level linearity was a problem with multi-bit conversion which is when very low signal levels don't taper off with level because the lower bits reproduce the wrong values. Linearity isn't much of an issue with modern sigma-delta architecture. Sample rate conversion done without proper dither can have some residual distortion unique to digital, but there is no good reason this should go on. And that would be so low with 24 bit it probably doesn't matter at all. Digital gear sometimes has idle tones at low levels in various places. Usually that would be at inaudible levels or frequencies.

    I can't claim to be an expert on digital, but those are the only artifacts I can come up with.

    Mostly jitter levels are inaudible. Low level linearity is solved with modern ADC and DAC chips. Aliasing or imaging is mostly a problem with 'audiophile minimum phase' filtering or very inexpensive chips that use half band filters which even then is not obviously audible. No reason not to dither correctly. I wouldn't say digital is perfect, but it seems better than any other method for recording and playback. When someone dots their i's and crosses their t's not trying to shortcut the design digital can be extraordinarily clean.
     
  6. RayDunzl

    RayDunzl Major Contributor Central Scrutinizer

    Joined:
    Mar 9, 2016
    Messages:
    3,726
    Likes Received:
    569
    Location:
    Riverview, Fl
    Unless I'm mistaken, the analog output is re-digitized for analysis. That happens here...

    If you want to see what changes after a digital stage, you'd feed those numbers to the analyzer.
     
  7. DonH56

    DonH56 Addicted to Fun and Learning Technical Expert

    Joined:
    Mar 15, 2016
    Messages:
    919
    Likes Received:
    337
    Location:
    Monument, CO
    Jitter impacts ADCs and DACs. Random jitter raises the noise floor; colored jitter even if random can raise close-in noise around signal tones, and correlated jitter can add harmonic and non-harmonic spurs (distortion).

    Aliasing folds HF signals into baseband during the sampling (ADC) process. Imaging at the output of the DAC creates HF signals at multiples of the sampling rate/2 (Nyquist frequency).

    Digital filters introduce phase shifts, transition bands, and such just as analog filters in some cases, but can also add additional latency and ringing pre- and post-signal, and restrict the dynamic range (as can analog filters, plus active analog filters add noise and distortion as well as more significant frequency variations over PVT).

    Tones are rare these days... Higher-order delta-sigma designs and the addition of noise decorrelation (dither) has essentially eliminated them.

    Nonlinearities from INL/DNL caused by non-ideal step sizes in the AC and/or DAC adds distortion.

    DACs often exhibit switching glitches that add HF content.

    A myriad of other things but I'm tired...
     
    oivavoi likes this.
  8. Cosmik

    Cosmik Major Contributor

    Joined:
    Apr 24, 2016
    Messages:
    1,088
    Likes Received:
    299
    Location:
    UK
    As always, the list of possible issues seems huge and frightening. But the difference between digital and analogue is that all the digital errors are known about theoretically - because they are mathematical. They can be plotted on a graph and zoomed in on until they fill the page - and would look exactly the same whether the system was 4-bit 22 kHz, or 4000-bit, 128 GHz. If, instead, the only access to the system was measurements of its input and output, it would be viewed as quite superb compared to the best analogue system; the so-called pre-ringing 'issue', for example, would never be identified because it doesn't show up on real music - as well as being ultrasonic and 'correct'.

    There is no way to dispel the audiophile fear of such theoretical errors. Personally, I would be convinced if it could be shown that the errors, when isolated from the main signal (i.e. unmasked) were inaudibly quiet at normal listening levels, but in audiophilia such a result would be regarded as too simple to believe.

    You can still listen to albums from the digital stone age - the 70s and early 80s. On every measure they must be thousands of times inferior to today's state-of-the-art digital and, relatively speaking, must be dripping in the very worst of all kinds of supposed digital distortion, yet some are even cited as audiophile classics.
     
    Blumlein 88 and oivavoi like this.
  9. bennetng

    bennetng Member

    Joined:
    Nov 15, 2017
    Messages:
    21
    Likes Received:
    2
    How about measurable jitter without converting to analog?

    I don't know if this phenomenon is common or not, I have a very old (2001) non class compliant USB 1.1 audio interface with MIDI and synthesizer capability (Roland SC-D70). It uses SRC on SPDIF input, there is no selectable master/slave mode and the recorded audio is never bit-perfect. The unit has a CS8420 SRC chip inside.

    SC-D70 only supports 44.1 and 48k sample rate, changing sample rate requires power cycling the unit. The funny thing is it is less susceptible to jitter when the unit's operating sample rate is different from the incoming SPDIF signal's sample rate. Which means a 44.1k > 48k or 48k > 44.1k SPDIF recording will have less jitter than a 44.1k > 44.1k or 48k > 48k recording.

    All files in the screenshots are played and recorded at 24-bit. Test tones are at 1/4 of playback sample rate so the frequency of 44.1k test and 48k test is different. The 16-bit dither is a visual aid to show that some jitter components are above 16-bit limit. Personally I can't hear any difference but IMO the amount of jitter is tremendous for a purely digital transfer.

    44rec.PNG 48rec.PNG
     
  10. DonH56

    DonH56 Addicted to Fun and Learning Technical Expert

    Joined:
    Mar 15, 2016
    Messages:
    919
    Likes Received:
    337
    Location:
    Monument, CO
    Uh, to see the jitter, you are converting to analog, yes? Or are these purely digital file results?

    You'd have to know the clock circuits for the different tests. There may be reclocking in one case, helping to clean up the incoming jitter, whilst the data stream is applied directly to the DAC when no resampling is used and thus incoming jitter directly translates to output jitter. SRCs can also add their own distortion and noise since they are by definition manipulating the data in a nonlinear manner.
     
  11. bennetng

    bennetng Member

    Joined:
    Nov 15, 2017
    Messages:
    21
    Likes Received:
    2
    No analog conversion is involved and that's why I found it interesting. There is a similar result in this review as well:
    http://ixbtlabs.com/articles2/m-audio-revolution71/index.html
    Find the phrase "the same jitter, not that bad" in the report. The graphs (png files) are incompatible with some web browsers and displayed as blank but right-click the blank area can save the png file and open it with other photo viewers. I resaved the png file and attached it here.
    imd1.png
     
    Last edited: Nov 15, 2017
  12. DonH56

    DonH56 Addicted to Fun and Learning Technical Expert

    Joined:
    Mar 15, 2016
    Messages:
    919
    Likes Received:
    337
    Location:
    Monument, CO
    The "jitter" in the purely digital domain looks to me like issues with the resampling algorithms introducing correlated sampling spurs (related to signal and clock frequencies with level determined by the resampling algorithm and FFT bins and whatever windowing is applied to suppress discontinuities at the "ends" of the data).
     
  13. bennetng

    bennetng Member

    Joined:
    Nov 15, 2017
    Messages:
    21
    Likes Received:
    2
    The files in my previous post were played by my X-Fi soundcard and recorded by Roland SC-D70. Now I do it in reverse, SC-D70 as the playback device and X-Fi as recording device, using the same optical SPDIF cable.

    The X-Fi has a bit-perfect slave recording mode but I am not going to test that since there is nothing interesting to see. The card supports 44.1/48/88.2/96k and I tested all of them in SRC mode. The jitter-like artifacts in matched sample rate are suppressed in a much better way.
    44p44r.PNG 44p48r.PNG 48p88r.PNG 48p96r.PNG

    Another interesting thing is X-Fi's resampler uses minimum phase filter, may have something to do with the card's main DSP (Emu20k1) since it is also a wavetable synthesizer and needs to handle some low sample rate resampling and pitch shifting stuff. Here is the impulse response in 48k play (SC-D70) 44.1k rec (X-Fi) mode.
    imp48p44r.png
     
  14. Wombat

    Wombat Member

    Joined:
    Nov 5, 2017
    Messages:
    84
    Likes Received:
    25
    Location:
    Australia
    I'm looking for, standardised, types of distortion and values(numbers), as in analogue measurements, that can be published for comparison.
     
  15. DonH56

    DonH56 Addicted to Fun and Learning Technical Expert

    Joined:
    Mar 15, 2016
    Messages:
    919
    Likes Received:
    337
    Location:
    Monument, CO
    I still do not understand exactly what you are asking. Distortion of the raw digital data, or the analog output of the DAC? For the former, it essentially comes down to dynamic range, number of bits used at each processing stage and ensuring no clipping through the processing, and perhaps sample rate if you include that in the digital domain. For the latter, same as for analog.
     
  16. Wombat

    Wombat Member

    Joined:
    Nov 5, 2017
    Messages:
    84
    Likes Received:
    25
    Location:
    Australia

    When dealing with the performance of analogue audio equipment there are established performance parameters that can be measured, quantified, referenced and compared.

    Do digital shortcomings present in a way that there are such performance measures?

    In simple terms, how the heck can I compare digital performance of audio devices without analysing graphs published by independent testers.
    NUMBERS please if that is possible. e.g. percent jitter distortion if that is real, sampling errors, etc that would be audible after 'ideal' conversion to an analogue signal.

    Or do the digital shortcomings translate neatly into the traditional analogue audio performance parameters?
     
    Last edited: Nov 17, 2017 at 5:02 AM
  17. bennetng

    bennetng Member

    Joined:
    Nov 15, 2017
    Messages:
    21
    Likes Received:
    2
    Audio file = full information, can be analyzed in many different ways, including listening.

    Graph = a lot of numbers represented in a visual way, but still incomplete since data which are unrelated to the analysis method will be discarded.

    Number = even less information than graph.

    The closest things I know are something like PSNR and SSIM in evaluating video quality, but there are also counter examples to show that they are unreliable.
     
  18. Wombat

    Wombat Member

    Joined:
    Nov 5, 2017
    Messages:
    84
    Likes Received:
    25
    Location:
    Australia
    The traditional measurements at the analogue audio circuit output terminals are all that is needed to quantify performance of DACs - any negative contribution from the digital stage will be revealed thus?
     
    Frank Dernie likes this.
  19. Blumlein 88

    Blumlein 88 Major Contributor

    Joined:
    Feb 23, 2016
    Messages:
    2,071
    Likes Received:
    537
    Is this what you are looking for?

    Yes all the traditional analog performance parameters apply.

    In addition parameters peculiar to digital systems.

    Jitter which is noise from imperfect timing of samples. Comes from ADC and DAC use. Investigations into how much is audible put the number above a nanosecond. Other than HDMI or other odd situations of poorly done gear you are unlikely to run into digital gear with a nanosecond or more of jitter.

    Low level linearity was a problem for multi-bit digital gear. The lowest bits weren't always translated into the right size signal. For instance a signal might drop in a file, but the output actually went up. That is also just nearly never an issue anymore.

    Aliasing. This is when a frequency above the nyquist rate or half sample rate is reflected down into the audible band. It is caused by poor filtering. As an example maybe there is a high tone at 37 khz in a recording done with a 48 khz ADC. The ADC should filter out everything above 24 khz. If it has a half band filter or some audiophile minimum phase filter that isn't steep it may not. That 37 khz tone could show up at a low level as an 11 khz signal in the recording. There actually is a standard for measuring this in the AES 17 guidelines, but I have never seen anyone use it. I don't know what an acceptable number for a cutoff on audibility would be other than if this is low enough you can't hear it. Usually it is rather low or even below the noise floor.

    Imaging is similar to aliasing only upon playback. It occurs with poor filtering in DACs. A high level tone at 11 khz with poor filtering might leak out a mirror imaged tone at 37 khz. This is less of a problem as such a tone would be low in level, you can't hear it at that frequency, and at low levels it is unlikely to cause other issues with your playback gear. Again there is a standard and I have never seen anyone list the results of this test.

    Hopefully if I bungled any of this and said something stupid Don or Amir can correct me. There are a few other things a bit esoteric and just generally a non-problem. You couldn't find that info without doing your own tests I don't think.

    So not sure how helpful this is other than saying if it hits good numbers on your regular analog parameters it isn't likely to be plagued by these other issues at audible levels.
     
  20. bennetng

    bennetng Member

    Joined:
    Nov 15, 2017
    Messages:
    21
    Likes Received:
    2
    Such as clipping caused by fixed point DSP without adequate headroom, or poor resamplers with a lot of aliasing and HF roll off? If such stuff are done wrong analog conversion is not necessary to reveal the problem.
     

Share This Page