Intersample clipping

Discussion in 'Audio Science' started by Riotvan, Jan 12, 2025.

  1. Riotvan

    Riotvan Snoofer in the Woofer

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    4,600
    Trophy Points:
    113
    Location:
    The Netherlands
    This is definitely worth a watch.


    Copy pasta from my profile post(might rewrite later) https://www.superbestaudiofriends.org/index.php?members/riotvan.70/

    I randomly came across it when reading changes for the new firmware for shd studio. I reduced the input into the asrc by 3dB and it was very noticeable. Someone at the minidsp forum linked the video and it confirmed I was not hearing things.
    Roon does the same thing, it reduces digital gain by 3dB before sending to the endpoint. I always found it to sound superior to tidal connect. Now they are equal.

    Also toslink from the TV into the shd sounds better. So I applied it to every input. Some DACs have the digital headroom built in already.
    Going to do some testing with the other inputs on Yggdrasil a2.
    And you have to change it at the input before it enters the DAC. Once the noise is added the stream or even the file through shoddy resampling like 44.1 to 96khz for a hires remaster it's permanent.

    Distortion can easily reach double digit %. And is non harmonic odd order and can
    spread over the whole frequency range.

    Would love to see some input from @atomicbob @purr1n and others.

    Edit: I think it's funny and sad that people so religious about bit-perfect sound might actually be getting worse sound quality than if they prevented ISP issues. They might not be aware as i was so i think this definitely needs more exploration and testing of various devices.

    Screenshot_20250112_120805.png
    Screenshot_20250112_121752.png
     
    Last edited: Jan 12, 2025
  2. fastfwd

    fastfwd Friend

    Pyrate
    Joined:
    Aug 29, 2019
    Likes Received:
    1,068
    Trophy Points:
    93
    Location:
    Silicon Valley
  3. Eric Rosenfield

    Eric Rosenfield Facebook Friend

    Joined:
    Mar 20, 2023
    Likes Received:
    231
    Trophy Points:
    43
    Location:
    New York City
    Hans Beekhuysen did a video on this same subject a year ago fwiw:
     
  4. Riotvan

    Riotvan Snoofer in the Woofer

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    4,600
    Trophy Points:
    113
    Location:
    The Netherlands
    He also made a follow up video:
     
  5. haywood

    haywood Friend

    Pyrate
    Joined:
    Oct 22, 2015
    Likes Received:
    799
    Trophy Points:
    93
    • Agreed, ditto, +1 Agreed, ditto, +1 x 1
    • List
  6. Riotvan

    Riotvan Snoofer in the Woofer

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    4,600
    Trophy Points:
    113
    Location:
    The Netherlands
    In the video it's demonstrated that it happens with a pristine recording as well since it's a digital headroom problem in the reconstruction filter.
    I can't guesstimate how many dacs are affected but with the many cheap chifi dacs going around the past several years one has to wonder if there are corners being cut or mistakes made out of ignorance.
     
  7. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    92,822
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Calling bullshit on this, especially in regards to pop music. Anyone who has loaded WAV files into audio editors can see the clipping in a lot of pop music. Leave some headroom and ISC will never be an issue. Most of the clipping is in the music, not caused by ISC (although it will make it worse)
    upload_2025-1-12_10-17-50.png

    ISC clipping only occurs in those samples where it happens, which depend on how badly brickwalled or close to brickwalled the recording is maybe happens 1 out of 3843 samples or so. The below is misleading. It's NOT a constant 10-15% THD that is added to the recording throughout.
    upload_2025-1-12_10-26-45.png

    1. Yup. The ISC issue arises with brickwall type records where the interpolation filters (oversamplling) overshoot full scale output, particular on the higher frequencies where the slope is steeper.
    2. Not all DACs are affected. Applying some negative digital gain before the interpolation can fix this
    That PPTs above vasty overstate the issue.

    THOUGHT EXERCISE:

    Also, if ISC has been such a great issue, how come no one has noticed it? Did everyone's Taylor Swift, Metallica, and Red Hot Chili Peppers suddenly sound worse?
    • Does Taylor Swift, Metallica, and RHCPs sound better on Benchmark DACs than on our own favorite DACs (that probably do F up on ISC)?
    • F Benchmark. F John Seau. Seriously F that guy. Dude makes shitty sounding gear.
    • Fing listen to music instead of listening for the odd samples (1 out of 3423) where ISC occurs (where you can't hear it anyway).
    P.S. I put on Motu on -1db gain for output.
     
    • Like Like x 8
    • Epic Epic x 4
    • List
    Last edited: Jan 12, 2025
  8. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    92,822
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    That dude needs to spend more time with family, wife, kids, friends, go out more, smoke weed, work out at the gym, ride a bike, smoke a brisket, walk a mile to get a donut. Godammit. I really hate the audio hobby some days.

    We go soooo far into the weeds we forget why we are in the hobby in the first place.

    I mean, I'd more more amenable to this if Benchmark made great sounding DACs and gear. But they don't.
     
    • Like Like x 7
    • Agreed, ditto, +1 Agreed, ditto, +1 x 2
    • List
    Last edited: Jan 12, 2025
  9. Riotvan

    Riotvan Snoofer in the Woofer

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    4,600
    Trophy Points:
    113
    Location:
    The Netherlands
    LOL, i was happy for many months. That firmware update was a trap!
     
  10. Wilewarer

    Wilewarer Almost "Made"

    Joined:
    Sep 22, 2021
    Likes Received:
    361
    Trophy Points:
    63
    Location:
    Chicago
    Interestingly, the Callia has a setting for this, which they blame on the SACD specification at some point changing its maximum peak level from 0 dB to 3.1 dB. I think this only impacts the DSD conversion, which (maybe) implies that they perhaps did not think this was a real issue worth worrying about for PCM, or at least that there was a clear always correct answer.
     
  11. Thad E Ginathom

    Thad E Ginathom Friend

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    15,068
    Trophy Points:
    113
    Location:
    India
    First, the anti-digitals tried to persuade us that we were not listening to smooth music at all, but to a stair-step-graph output. They were hardly to blame for this, as most explantions for non-mathematicians actually showed this.

    My attempts to understand digital music theory were limited by my innumeracy, +-/* arithmetic is about my limit. But I recall someone explaining how some mysterious sinX (function) actually produces curves, not stairsteps. I had no hope of understanding but it was a breakthrough to at least know that this thing existed!

    The world has moved on. The "natural" (for want of a better word: I'm sure the literal meaning can be argued) curves output is accepted ... and has become a problem. Because that limit 0dB, beyond which we all know that there is no music, just a farting sound, can be exceeded by the creation of the curve output itself. In practice there is no 0dB hard limit.

    But why wouldn't some marketing guy pretend there is a problem, and that his product has the answer? That's what marketing men do. Never let truth get in the way of a good sales pitch!

    Thanks to purr1n and co for putting the record straight.
     
  12. fastfwd

    fastfwd Friend

    Pyrate
    Joined:
    Aug 29, 2019
    Likes Received:
    1,068
    Trophy Points:
    93
    Location:
    Silicon Valley
    Haven't listened to all his gear, but his DAC3 HGC sounds fine to me in my system.
     
  13. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    92,822
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    LOL hyperbole on my part. It sounds fine.

    --

    This ISC stuff is the audio nerd equivalent of the weirdos at the Stanford IT department who wrote that guide on how not to be offensive to other people and always walk on eggshells , e.g. can't say "firewall whitelist / blacklist" (as if that will cure racism or the even more subtle colorism), can't say "balls to the wall" (because balls is male, even though the balls refer to the balls on the top of the throttle in airplanes).

    People looking for offense are gonna find it. People looking for oddball measurement artifacts on an APx555 are gonna find it.

    I really don't care about "bitpefect" or intersample clipping errors either way.
     
    • Like Like x 4
    • Dislike Dislike x 1
    • List
  14. Psalmanazar

    Psalmanazar Most improved member; A+

    Pyrate Slaytanic Cliff Clavin
    Joined:
    Sep 27, 2015
    Likes Received:
    5,436
    Trophy Points:
    113
    ISC is usually brief analog clipping. It is not nerd stuff. The peaks are much higher than the samples, especially at 44.1 and 48 khz. The higher the sample values, the more likely the filtered sinc pulses are to clip. It's only a real problem for broadcast standards and garbage disposable popular music.
     
    Last edited: Jan 13, 2025
  15. Psalmanazar

    Psalmanazar Most improved member; A+

    Pyrate Slaytanic Cliff Clavin
    Joined:
    Sep 27, 2015
    Likes Received:
    5,436
    Trophy Points:
    113
    bitperfect is bs. You just want the samples to do what you want and not have the OS filter them and mix with with bullshit beep boop computer sounds. This is easy with Core Audio on Mac and ASIO on Windows. It turns out that dsp does things cleaner than analog including reasonable volume control and eq. the samples are already being manipulated crazily in production. digital volume control is just fitting the fixed point 16-bit or 24-bit sample into a 32-bit or 64-bit float, rescaling it, adding a dither, and sending it to the device. Most rooms and recordings have a noise floor above 16-bit from tape and resistor noise. Especially all of these audiophile recordings that have quiet distance mics amplified a ton by a quiet pre. Then all the compression in silly distorted pop music using silly distorted noisy vintage equipment might have a noise floor above a cassette and then use gates, expanders, and Izotope RX to clean it up.

    What detail is left after denoising? Why are people so scared of digital? Now we have all the special filter bs not realizing all of their music already went through the normal linear phase brickwall (sometimes half band) filters when recorded or digitized. Or early to mid 80s CDS that went through god awful bs and you had better hope they used converters with the Apogee filters that were de rigueur for the old Sony converters and digital tape machines to not have thousands of degrees of phase shift. Now of course you can buy an Apogee interface for a few hundred to few thousand dollars that sounds much better than 5 figure 80s gear.
     
  16. Psalmanazar

    Psalmanazar Most improved member; A+

    Pyrate Slaytanic Cliff Clavin
    Joined:
    Sep 27, 2015
    Likes Received:
    5,436
    Trophy Points:
    113
    The earlier products were not so fine.
     
  17. Syzygy

    Syzygy Friend

    Pyrate
    Joined:
    Jun 13, 2018
    Likes Received:
    2,186
    Trophy Points:
    93
    Location:
    DFW, Texas
    We either had a thread, or discussed inter-sample overs at some length, IIRC in regards to the DAC2541 and/or DSHA-3F testing. AtomicBob mentioned that he'd set the DAC to -3dB output to accommodate it at the time, and posted graphs.

    That was awhile ago, but I remember investigating a particular song track to find & count the overs in the file. Can't remember which song and/or track.

    I always kept my DAC2541 set at -3dB output into the DSHA-3F and never worried about it again after that discussion.
     
  18. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    19,971
    Trophy Points:
    113
    Location:
    On planet
    This topic has been covered in many threads on the forum. Here are some useful links:

    Windows WDM Audio System Limiter
    Tutorial - Lies, Damn Lies and USB DAC technical measurements
    Windows audio system can introduce noticable distortion in the digital stream when the signal is greater than -0.131 dBFS. This affects all DACs and a good reason to use WASAPI, ASIO or avoid Windows USB audio completely. Having standardized on Dante I don't use USB except when operating transportably. Even then I usually use a portable Dante endpoint such as RDL SF-ND2.

    DAC filter design issues
    AMB Gamma 2 DAC (or how NWAVGUY fooled us all)
    ODAC digital filter design demonstrates what happens when one doesn't consider digital music with 0 dBFS peaks with fast transients.

    DAC gain staging vs recovery filter type
    Soekris dac1541 technical measurements
    Many DAC designs set 0 dBFS = Max Vpp based on a sinusoid. When presented a fast transient such as percussive sounds the recovery filter can overshoot thus clipping the output. Soekris has a digital gain control and a good clipping indicator allowing reduction of the digital signal to avoid such clipping when presented music with 0 dBFS peaks containing such transients.

    AES presentation covering many aspects of running digital music streams close to digital maximum, 0 dBFS and why setting to -3 dBFS avoids unpleasant listener expriences for DACs with such issues.
    https://www.aes-media.org/sections/pnw/pnwrecaps/2018/oct2018/
    Download the powerpoints to see the presentation.
    May want to do so very soon as this website will be migrating to a new server at some future time TBD.
     
    • Like Like x 12
    • Epic Epic x 7
    • List
  19. Riotvan

    Riotvan Snoofer in the Woofer

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    4,600
    Trophy Points:
    113
    Location:
    The Netherlands
    Thanks! I'll have a look.
     
  20. Biodegraded

    Biodegraded Friend

    Pyrate Contributor
    Joined:
    May 28, 2017
    Likes Received:
    8,917
    Trophy Points:
    113
    Location:
    Vancouver BC
    Although not strictly to do with clipping/overshoot, also to consider wrt filter design is the influence of sample rate. See more stellar @atomicbob work here, and audible effects (which shouldn't be due to clipping, as gain at both rates was reduced in the player and at the DAC as per earlier recommendations).
     

Share This Page