Home of the Squeezebox™ & Transporter® network music players.
Page 3 of 7 FirstFirst 12345 ... LastLast
Results 21 to 30 of 66
  1. #21
    Quote Originally Posted by pippin View Post
    Um. Sorry. Again, I fail to see it.
    Again: Why does interpolation filtering on the digital side improve the analog filtering and how is the interpolation even correlated to your desired filter response.

    I can find a lot of prosa about this on the internet, mainly by makers of such filters, but I haven't found a single explanation that makes sense. All I find is stuff like this:
    http://www.ni.com/white-paper/5515/en/
    Which - sorry - is nonsense (the digital interpolation part. The analog and imaging explanations are OK).
    Hi Pippin

    Let's have a look at this.

    Without oversampling, the analog anti-imaging filter requirements are obviously difficult - especially for the audio passband and 44.1khz sample rate.

    With oversampling however, the analog anti-imaging filter requirements are relaxed, because now the stopband begins at the original Nyquist multiplied by the oversampling ratio. The images in-between have been low-passed by the digital interpolation/oversampling filter, the exact parameters and usual filter tradeoffs of which are by design. This is what Figure 4 is trying to show in your link.

    The interpolation algorithm can therefore be compared to an analogue low-pass filter but in the digital domain. Neither of these need to analyse an hour in advance to perform reconstruction of the complex musical waveform according to sampling theory, although you are quite right, more than the one preceding sample would be necessary to interpolate, as your pictures nicely show.

    Regarding noise: oversampling actually increases the SNR as quantisation noise is spread over a wider frequency range, further enhanced by noise shaping and sigma delta process.

  2. #22
    Senior Member pippin's Avatar
    Join Date
    Oct 2007
    Location
    Berlin
    Posts
    14,656
    I didn't mean to question the sense behind upsampling. I fully understand why that makes the (analog) filter design easier and the result better.

    But John's argument was that you get a better result by using a NOS DAC and do the oversampling through sox because the interpolation filter in sox is simpler than the multiple ones in oversampling DAC designs and I was wondering why that is and what kind of filter sox is using.
    Intuitively I still feel just duplicating the samples as "normal" oversampling does it should give the best results but that's just a feeling in an area where I really don't have a lot of experience, that's why I asked.
    But to me it looks like this: while the "duplication" strategy is assured to create quite powerful images at the same time you can be sure to have all of that going on beyond your cutoff frequency while I feel arithmetic interpolation should create noise over a broader spectrum.
    ---
    learn more about iPeng, the iPhone and iPad remote for the Squeezebox and
    Logitech UE Smart Radio as well as iPeng Party, the free Party-App,
    at penguinlovesmusic.com
    New: iPeng 9, the Universal App for iPhone, iPad and Apple Watch

  3. #23
    Quote Originally Posted by pippin View Post
    Intuitively I still feel just duplicating the samples as "normal" oversampling does it should give the best results
    This is not how oversampling operates in a DAC (or ADC) - which would basically achieve nothing. Oversampling in a DAC operates exactly as described in the NI link you gave, and in my summary above: the new sample values are determined by an interpolation filter.
    Last edited by flimflam; 2013-07-09 at 09:48.

  4. #24
    Senior Member pippin's Avatar
    Join Date
    Oct 2007
    Location
    Berlin
    Posts
    14,656
    Ah. OK. So how does this interpolation filter work? In the NI link they show a filter curve that nicely follows the sine wave which will not be so simple for music.
    ---
    learn more about iPeng, the iPhone and iPad remote for the Squeezebox and
    Logitech UE Smart Radio as well as iPeng Party, the free Party-App,
    at penguinlovesmusic.com
    New: iPeng 9, the Universal App for iPhone, iPad and Apple Watch

  5. #25
    Senior Member
    Join Date
    Jun 2006
    Posts
    1,690
    Hi Pippin,
    to your underlying question: why do the filtering externally rather than in the DAC chip? The answer is "I don't know". When I bypass the internal filter and do it externally using a basic simple filter it sounds much better. And it's not just me. I've done this in blind fashion with a number of people and they all came to the same conclusion, the external filter sounded much better.

    Looking at the spec sheets for the DAC chips you can tell that they are using cascaded filters. When I use an external FPGA to implement a filter and program it to be cascaded multiple filters, I hear a similar sound, when I implement a simple filter (no cascading) I get the better sound. The conclusion seems to be that it is somehow the cascading of filters that causes the problem. Again no clue at all why this is so, what the mechanism is etc.

    They ARE different filter functions, they do output different bits, but they are very close to each other. Whatever it is, its not a big difference in the waveform.

    John S.

  6. #26
    Quote Originally Posted by pippin View Post
    Ah. OK. So how does this interpolation filter work? In the NI link they show a filter curve that nicely follows the sine wave which will not be so simple for music.
    The new samples are inserted and given zero-values. The result is then low-pass filtered.

    Remember, we have a special case of a bandlimited signal. Sampling theory means we can recover this signal - in all its complex, musical glory, no matter how unintuitive and surprising this first seems.

  7. #27
    Senior Member
    Join Date
    Nov 2009
    Location
    DUS
    Posts
    840
    Hi there/John.

    A lot of writing and reading.

    I tried resampling with SOX and other (reference) tools as discussed at Audio Asylum and elsewhere several times in the past.


    http://soundcheck-audio.blogspot.de/...esampling.html


    I never managed to get it working to my satisfaction. Neither realtime, nor offline. The SQ was always worse then the original.
    Maybe I missed something out.


    Issues:

    1. You need to attenuate the "digital" signal before you apply the filters to avoid clipping.
    2. Realtime SRC is causing higher load on the processor and higher traffic on the entire path. That might translate into audible losses on most of the systems out there.
    3. Offline SRC is kind of inflexible, if you change your DAC devices once in a while or if you feed different DACs at home. You always have to keep copies of the originals and the resampled materials.
    You might have a player which does offline SRC prior to playback and caches a full track/CD on e.g. a ramdisk. But that's not possible in a LMS environment.
    4. No filter is lossless.
    5. If you stay with 16bit you have to re-dither the already dithered material. E.g. You can't do 16 bit to 24bit conversions on LMS if you stream PCM. You need to re-encode PCM to flac again.

    6. I've never seen somebody offering an to me acceptable sample rate conversion setting. -- John. Shoot. I'm listening.


    To me SRC is a questionable workaround to cover the deficiency of rather poor or inflexible HW (firmware + HW) implementations. If your HW implementation is that bad, that SW SRC with all its issues as described above (maybe more) sounds better than your HW SRC, I'd look for a better HW.


    Please let me know your favorite SOX settings.

    Cheers

  8. #28
    Junior Member
    Join Date
    Apr 2012
    Posts
    9
    Quote Originally Posted by soundcheck View Post
    I tried resampling with SOX and other (reference) tools as discussed at Audio Asylum and elsewhere several times in the past.


    http://soundcheck-audio.blogspot.de/...esampling.html


    I never managed to get it working to my satisfaction. Neither realtime, nor offline. The SQ was always worse then the original.
    Maybe I missed something out.
    Klaus, I read your document, and unless I missed something, you were resampling from 16/44.1 to 24/96? Is that correct? Did you ever try going from 16/44.1 to 24/88.2 or 24/176.4? If so, did you hear any difference (compared to 24/96)?

    And I also tried SOX upsampling in LMS a while ago, but I remember at the time that I didn't like the idea that it only output FLAC (as all my music is WAV). But are you saying that was only because it was going from 16 bit to 24 bit? IOW, if I first convert some files to 24bit (offline) and then use SOX in LMS to upsample in realtime (say from 24/44.1 to 24/176.4), will it output PCM? Or is there any other method to use SOX in LMS and have it output PCM?

    Thanks.
    edward

  9. #29
    Senior Member Julf's Avatar
    Join Date
    Dec 2010
    Posts
    2,507
    Quote Originally Posted by edwardian View Post
    I didn't like the idea that it only output FLAC (as all my music is WAV)
    Any specific reasons to prefer WAV? I think the general view is that with lower-end processors (such as those in the squeezeboxes) that don't have dedicated I/O processors the additional network load caused by the wasted bits in WAV files causes more CPU load (and thus theoretically sound degradation) than the FLAC decoding, and the usual WAV problems of lack of tagging standards etc. are always a hassle.
    "To try to judge the real from the false will always be hard. In this fast-growing art of 'high fidelity' the quackery will bear a solid gilt edge that will fool many people" - Paul W Klipsch, 1953

  10. #30
    Senior Member
    Join Date
    Nov 2009
    Location
    DUS
    Posts
    840
    Quote Originally Posted by edwardian View Post
    Klaus, I read your document, and unless I missed something, you were resampling from 16/44.1 to 24/96? Is that correct? Did you ever try going from 16/44.1 to 24/88.2 or 24/176.4? If so, did you hear any difference (compared to 24/96)?

    And I also tried SOX upsampling in LMS a while ago, but I remember at the time that I didn't like the idea that it only output FLAC (as all my music is WAV). But are you saying that was only because it was going from 16 bit to 24 bit? IOW, if I first convert some files to 24bit (offline) and then use SOX in LMS to upsample in realtime (say from 24/44.1 to 24/176.4), will it output PCM? Or is there any other method to use SOX in LMS and have it output PCM?

    Thanks.
    edward
    I tried all kind of SR combinations.

    Beside that I was told that top quality algorithms do not need synchronous SRC.


    Have a look at http://src.infinitewave.ca/ to figure out the artefacts associated to SRC. If you look at the graphs of top quality converters, you wouldn't expect any "data" related losses.
    You don't see any artefacts on 44.1 -> 96 conversions.

    I btw also tried

    Izotzope
    Adobe
    r8brain pro

    beside SOX, since some of tha AA fellows swear Izoptope or Adobe would be the leading packages. I could not tell a difference. That's why I stayed with Sox as a free package.

    Asynchronous vs. synchronous plays a bigger role in HW reclocking implemtentations. You might catch intermodulation problems.


    However. You can simply change the numbers and try by yourself.


    If anybody has a better idea. Shoot. I'm very open to try new settings.


    ---

    Regarding 24bit.

    It's not just the bitdepth. It's also the sample rate. A different sample rate requires reencoding to flacs.

    I btw have converted all my tracks to 24bit. Because I'm running server-based offline volume control using replaygain (EBU R128 compliant, applied with r128gain tool) tags + system specific attenuation offset. Doing it this way I can continue to stream PCM. Nowadays I run my main system without manual volume control. What a relief.

    ---

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •