Extreme filtering software upscaling

[Steveo888]

5 hours ago, legend said:

Sounds good. Could also compare the Brooklyn MQA with the Qutest plus M-Scaler and with extreme-filtered files that @Ittaku has sent me.

 

If not I have to come up to Canberra on 3/4 April going to a CSO concert with a brother there but will be with my wife so time will be limited!

We can probably do this when I bring my Dave to @legend s place. We can use @Sime V2 mscaler and experiment a bit. I need to clear up a suitable sunday

[Guest rmpfyf]

21 hours ago, Eggcup The Daft said:

Group delay is a time delay that is a function of frequency... in the case of digital filters, where the filter is not linear phase.

The threshold for directly hearing it is pretty wide - measured in milliseconds. It shouldn't be a problem. However, when the function is linear (so that group delay increases with frequency) some people claim that it can be heard, and that is supposed to be the case with some non-linear phase filters. I've seen this discussed in various places, but as is so often the case, there's never cast iron proof

 

I'll back @davewantsmoore here. 

 

'Group delay increases with frequency' - not so sure here. 

 

If you're looking to hear individual bits of pre or post ring then no, that's quite hard. 

 

If you're looking to hear a redistribution of energy owing to different filter characteristics then yes, that's quite audible, and if it wasn't we wouldn't be able to differentiate sounds (that are themselves different shapes of pressure energy distribution). 

 

There's no surprise that some uber uber DACs are running significant upsampling filters at ridiculous frequency to limit temporal effects of filters. Particularly where phase effects of the filter are significant (e.g. min phase) running the filter to max Fs possible gives a near-as-dammit-linear phase response in the audible band.

 

I have a little script here that uses SoX to upsample a Redbook track to half volume then double frequency min phase, double frequency linear phase, same frequency dithered. Never had someone come over that couldn't pick differences, and 88.2kHz is the tip of the iceberg on possibilities here.

[legend]

My most listened to audio files are generally 88.4 or 96 native (ripped from SACDs, DVDAs or downloaded as such) and then upsampled x2 or x4 by Amarra or Jriver before sending to the Project S2 DAC. With these I can hear little difference between the S2's min phase and linear filters.  Maybe I should listen to Redbook tracks with no upsampling to the DAC?

 

 

[davewantsmoore]

4 hours ago, rmpfyf said:

SoX to upsample min phase

What is the frequency response of the filter in the audio-band?

 

[davewantsmoore]

58 minutes ago, legend said:

Maybe I should listen to Redbook tracks with no upsampling to the DAC?

It's quite a complicated question .... more so for your DAC.

 

Your DAC has 2 stages of resampling, eventually get the rate up to Mclk/64 ...... should you 'avoid' these to provide you own resampled digital data?   What should that data look like?     That question requires super-detailed information about the the reconstruction that the DAC is doing at circa 1.5 mhz.      i.e you would change the digital data, not to confirm to some theoretical ideal .... but change it in whatever way improved the final output.

[Guest rmpfyf]

5 hours ago, davewantsmoore said:

What is the frequency response of the filter in the audio-band?

 

Depends on the filter Dave. And you can tailor this in SoX and many other tools.

 

There's a few sites giving response curves across a range of filters e.g. http://src.infinitewave.ca/

 

(Though you know all this!)

[Nada]

On 29/01/2019 at 11:17 AM, Ittaku said:

..... The fun part was I did exactly what Sox told me not to do "note that band-width values greater than 99% are not recommended for normal use as they can cause excessive transient echo" which is of course echo above the limits of human hearing .....

Are you sure about that?  Filter maths isnt trivial and there's some misleading info out there.

[Ittaku]

14 minutes ago, Nada said:

Are you sure about that?  Filter maths isnt trivial and there's some misleading info out there.

Am I sure about what?

[davewantsmoore]

4 hours ago, rmpfyf said:

Depends on the filter Dave.

Of course, I mean the mp filter which you used.

 

Maybe not have flat frequency (phase) below ~20khz (?!)

[legend]

On ‎31‎/‎01‎/‎2019 at 3:46 PM, Guest Eggcup The Daft said:

 

Why exclude group delay? It's certainly mentioned in any number of discussions of filters, both digital and analogue. I do know that it is more common in analogue filters, but isn't it common with very steep digital filtering as well?

 

 Below is the group delay of the Kantu Be speakers I have been using over the past few weeks to test the effects of @Ittaku extreme upsampling and also that of the Chord M-Scaler.

 

Note it is very flat in the treble but varies by about 1 ms through the midrange probably due to driver & Xover non-linearity and then starts to rise in the bass mainly due to physical effects of mic being closer to the mid & treble divers (j ust under1 m to limit room reflections) but therefore being further away from the bass drivers.  Of course the latter largely disappears at the listening position of 2-3 m.

 

The passive Xover filters are 400Hz 12 dB/octave and 2.5 kHz 18 dB/octave.

 

 

 

Edited February 2, 2019 by legend
additional info

[Guest rmpfyf]

9 hours ago, davewantsmoore said:

Of course, I mean the mp filter which you used.

 

Maybe not have flat frequency (phase) below ~20khz (?!)

 

Still waiting on my high-freq interface for my DAC; it was delivered through it and the DAC are off to the DAC's maker for more playtime and an all-in-one integration - short signal paths and all that. 

 

When back we'll have a tinker. I wasn't completely happy at my previous max (96kHz); if taking both extremes (linear/min) honestly my preferences changed per music type. 

[Nada]

15 hours ago, Ittaku said:

Am I sure about what?

..... The fun part was I did exactly what Sox told me not to do "note that band-width values greater than 99% are not recommended for normal use as they can cause excessive transient echo" which is of course echo above the limits of human hearing .....

 

I think the modern term is pre and post ringing. You seem to imply it cant be heard?

[Ittaku]

Just now, Nada said:

..... The fun part was I did exactly what Sox told me not to do "note that band-width values greater than 99% are not recommended for normal use as they can cause excessive transient echo" which is of course echo above the limits of human hearing .....

 

I think the modern term is pre and post ringing. You seem to imply it cant be heard?

That's right, it happens at a frequency just below the Nyquist frequency which even at redbook sample rates is 22kHz.

[Nada]

1 minute ago, Ittaku said:

That's right, it happens at a frequency just below the Nyquist frequency which even at redbook sample rates is 22kHz.

Sorry but are you saying pre and post ringing from filters upsampling digital PCM are inaudible when listening to music? 

[acg]

9 minutes ago, Nada said:

Sorry but are you saying pre and post ringing from filters upsampling digital PCM are inaudible when listening to music? 

 

As you know Nada, my playback software (XXHE) upsamples 16x to 705kHz/768kHz and sends it to a NOS dac that has zero onboard filters...it just plays what it gets.  As far as I am aware that software still has the only filter on the planet that is completely devoid of pre or post ringing (Keith Johnson thought it was impossible to ever achieve).  Let me put this clearly...pre and post ringing can be heard...or more importantly people that use XXHE tend to prefer the zero ringing filter over the others.  

 

@Ittaku, nice stuff!  I love seeing this kind of initiative.

 

[Ittaku]

22 minutes ago, Nada said:

Sorry but are you saying pre and post ringing from filters upsampling digital PCM are inaudible when listening to music? 

Yes I am saying that. As I said the ringing occurs just below the Nyquist frequency. Do you have something that shows otherwise? However the effects of the filter on the rest of the frequencies in the audible range do alter the sound of it. The measurable ringing is an easy thing to demonstrate and measure though.

[Nada]

1 hour ago, Ittaku said:

Yes I am saying that. As I said the ringing occurs just below the Nyquist frequency. Do you have something that shows otherwise? However the effects of the filter on the rest of the frequencies in the audible range do alter the sound of it. The measurable ringing is an easy thing to demonstrate and measure though.

 In what way do " the effects of the filter on the rest of the frequencies in the audible range do alter the sound of it." Thanks.
 

[Ittaku]

27 minutes ago, Nada said:

 In what way do " the effects of the filter on the rest of the frequencies in the audible range do alter the sound of it." Thanks.

There are changes to amplitude, phase, and latency within the audible spectrum depending on the filter design. It's my postulation that the ringing itself as used in digital filtering is a complete red herring and conducted some experiments with obscene ringing simulations and posted some of my results here a few pages back where I injected 22kHz tones into music at levels louder than the music itself without changing anything else.

[Guest rmpfyf]

3 hours ago, Ittaku said:

That's right, it happens at a frequency just below the Nyquist frequency which even at redbook sample rates is 22kHz.

Still looking at it the wrong way

[Ittaku]

Just now, rmpfyf said:

Still looking at it the wrong way

Feel free to correct me then...

[Guest rmpfyf]

1 minute ago, Ittaku said:

Feel free to correct me then...

You’re looking at the oscillations in the impulse response and making inferences on the audibility of their frequency; what you’ll actually hear is the general energy redistribution - the change in maximum amplitude response and the temporal magnitude of any frequency distribution.

 

At mega Fs an impulse response is going to look a lot more like an impulse regards overall energy distribution, and a min phase implementation is likely to guarantee a cymbal sounds a lot more like a cymbal.

 

You’ll hear a change in shape of energy distribution through the same mechanisms you’ll discern different sounds. Nyquist limits and applies with equal measure... being that it’s not the whole story.

[Ittaku]

1 minute ago, rmpfyf said:

You’re looking at the oscillations in the impulse response and making inferences on the audibility of their frequency; what you’ll actually hear is the general energy redistribution - the change in maximum amplitude response and the temporal magnitude of any frequency distribution.

 

At mega Fs an impulse response is going to look a lot more like an impulse regards overall energy distribution, and a min phase implementation is likely to guarantee a cymbal sounds a lot more like a cymbal.

 

You’ll hear a change in shape of energy distribution through the same mechanisms you’ll discern different sounds. Nyquist limits and applies with equal measure... being that it’s not the whole story.

I disagree because all filters to date change the amplitude of any impulse due to the approximations they do. Min phase ones change the amplitude the most. A critically steep filter as I'm applying here inspired by RW's work is the first that ever maintains the same amplitude impulse after filtering so there is no energy redistribution, only the addition of the ringing, plus a constant latency addition across the entire frequency response. An average filter usually has 95% impulse height on reconstruction. An average minimum phase filter has only 65% impulse height. Yes I know impulses don't happen in nature, but there is something in this from my audible experiments that far outweighs the subtle differences between min phase and linear filters which only sound "different". This just sounds universally better to me irrespective of the material (so far.)

[Guest rmpfyf]

32 minutes ago, Ittaku said:

I disagree because all filters to date change the amplitude of any impulse due to the approximations they do. Min phase ones change the amplitude the most. A critically steep filter as I'm applying here inspired by RW's work is the first that ever maintains the same amplitude impulse after filtering so there is no energy redistribution, only the addition of the ringing, plus a constant latency addition across the entire frequency response. An average filter usually has 95% impulse height on reconstruction. An average minimum phase filter has only 65% impulse height. Yes I know impulses don't happen in nature, but there is something in this from my audible experiments that far outweighs the subtle differences between min phase and linear filters which only sound "different". This just sounds universally better to me irrespective of the material (so far.)

 

I get where you’re going and I’d stress there is nothing new here - higher order filter design has been around a long while - it’s just computationally expensive.

 

You have negligible or minimised energy redistribution, not no redistribution.

 

The ring is small as energy is preserved in the peak; there is no energy added. Temporally the time magnitude of the ring is Fs dependent.

 

We live in an age where the computations involved aren’t too taxing anymore. Code it into a product and ship it  

[Ittaku]

Just now, rmpfyf said:

 

I get where you’re going and I’d stress there is nothing new here - higher order filter design has been around a long while - it’s just computationally expensive.

 

You have negligible or minimised energy redistribution, not no redistribution.

 

The ring is small as energy is preserved in the peak; there is no energy added. Temporally the time magnitude of the ring is Fs dependent.

 

We live in an age where the computations involved aren’t too taxing anymore. Code it into a product and ship it  

Gotcha. Alas I'm not a finalising kind of person. My software projects are full of half baked products that work and that's about it without any polish. I even know how I would create something that could be a standalone device without too much latency to rival the mscaler at maybe 1/3-1/2 the cost, but I'm not business nor manufacturing minded. In its current form, the 430 second latency probably might not cut it for a real time device (lol.)

[Nada]

33 minutes ago, Ittaku said:

I disagree because all filters to date change the amplitude of any impulse due to the approximations they do.....

Your reply suggests to me you have entirely missed the point made by  rmpfyf

 

Ringing affects signal in the audible spectrum. Just because ringing is modulated at high frequencies doesn't mean its inaudible.  

 

For an analogy think of ClassD amps running at 120kHz and someone claiming they are therefore inaudible.