Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: "Bending" a integer-multiple resampling algorithm to fit existing sample points. (Read 3063 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

"Bending" a integer-multiple resampling algorithm to fit existing sample points.

See here:
http://www.head-fi.org/t/784602/chord-mojo-the-official-thread-please-read-the-3rd-post/16095#post_12511348

Although I posted in the Chord Mojo thread, this post is about Schiit and their claim that their "megaburrito" filter preserves the original samples in a "closed form" "final solution" of upsampling.

My reasoning is as follows:
1. The correct solution to exactly preserving the original sample points in an upsampled bitstream calls for an infinite length reconstruction filter to produce a perfectly sharp transition band, total preservation of the passband and complete rejection of the stopband.

2. We can safely assume that Schiit did not find a way to turn an infinite time operation into one that can be computed by their onboard CPU.

3. Alternatively one can compute a finite length reconstruction filter and bend the interpolated points to fit with the existing samples.

My questions are as follows:

1. Is it really true that only an infinite length reconstruction filter would naturally fit with the existing data points?  Are there other possibilities?  Say for example a filter with an unorthodox (albeit suboptimal) placement of the transition band--perhaps with the -3dB point placed exactly at the Nyquist frequency?  Or not rolling off at all before Nyquist?

3. I'm assuming that bending the interpolated points to fit with existing samples would result in degradation of the output rather than an improvement (in signal processing terms, if not marketing terms).  Does anybody see a particular way of bending the data that would NOT degrade the output in terms of distortion or compromising passband / stopband performance?

Finally, is there anyone here who thinks Schiit's claims about their megaburrito filter is not full of... well?

Re: "Bending" a integer-multiple resampling algorithm to fit existing sample points.

Reply #1 –
No filter at all can preserve the points but imaging is a problem.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: "Bending" a integer-multiple resampling algorithm to fit existing sample points.

Reply #2 –
Quote from: Joe Bloggs on
Although I posted in the Chord Mojo thread, this post is about Schiit and their claim that their "megaburrito" filter preserves the original samples in a "closed form" "final solution" of upsampling.

I only read a few posts in that thread, but I think that guy doesn't know what many of those words mean.  Or at least he is not using them correctly. 

Quote from: Joe Bloggs on
1. Is it really true that only an infinite length reconstruction filter would naturally fit with the existing data points?  Are there other possibilities?  Say for example a filter with an unorthodox (albeit suboptimal) placement of the transition band--perhaps with the -3dB point placed exactly at the Nyquist frequency?  Or not rolling off at all before Nyquist?

You need a filter that has perfect unity frequency response at all non-zero input frequencies.  So if the input waveform is low pass filtered at 10 khz you can probably hit each point more or less exactly, at least within your quantization step size. 

Quote from: Joe Bloggs on
3. I'm assuming that bending the interpolated points to fit with existing samples would result in degradation of the output rather than an improvement (in signal processing terms, if not marketing terms).

You can come up with interpolators that are constrained to go through all sample points.  Actually, many spline interpolators do this automatically.  They tend to have poor quality though since the optimal interpolator isn't necessarily constrained in such a way.

Quote from: Joe Bloggs on
Finally, is there anyone here who thinks Schiit's claims about their megaburrito filter is not full of... well?

This is easily testable.  Resample a 10 kHz tone from 44.1 to 48k (or any other non-integer, closely spaced sampling rates) and see what the distortion is like.  If its terrible, then its probably just a crappy polynomial interpolator.

Re: "Bending" a integer-multiple resampling algorithm to fit existing sample points.

Reply #3 –
I remember TDA 1543 designs of paralelled DACs that created additional points. You may search what this effort was called.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: "Bending" a integer-multiple resampling algorithm to fit existing sample points.

Reply #4 –
I don't see anything new here, just people confused by S(c)hit marketing?

edit: oh gawd, digital filters truly are the new audiophile fad ... and these clueless people again have something new to justify what they want to hear with
"I hear it when I see it."

Re: "Bending" a integer-multiple resampling algorithm to fit existing sample points.

Reply #5 –
You can create 8x oversampling without digital filter followed by some simple analog filter. If that is it, it is old voodoo.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: "Bending" a integer-multiple resampling algorithm to fit existing sample points.

Reply #6 –
A half band filter will exactly preserve the input samples and "correctly" interpolate the in between samples.  But it also aliases near the transition (at pi/4 it's value is 1/2.)  I believe that Schitt's megaburrito filter preserves the original points like a half band filter but also avoids aliasing.  But they are proud of having a "closed form" solution and I don't know if they are merely referring to not using iterative methods to derive their filter (e.g. no Parks–McClellan or Remez exchange) or if they are trying to make some other point too.  To my way of thinking a windowed sinc is closed form for most sane windows but I don't know if they eschew trig too.

Re: "Bending" a integer-multiple resampling algorithm to fit existing sample points.

Reply #7 –
Quote from: saratoga on
You need a filter that has perfect unity frequency response at all non-zero input frequencies.  So if the input waveform is low pass filtered at 10 khz you can probably hit each point more or less exactly, at least within your quantization step size. 
If it's lowpassed, even to 20 kHz i.e. full audible bandwidth, then nothing special is needed (e.g. sox with default settings will retain original samples with peak noise less than –130dB).

Re: "Bending" a integer-multiple resampling algorithm to fit existing sample points.

Reply #8 –
Thats the problem with the marketing speech. Somehow it is interpreted that no lowpass is needed so retaining everything to 22.05kHz.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!