• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Reasonably priced good quality 6 or 8 channel USB DAC?

If one then turns that exact same filter with the same Q into a linear phase filter the graph shows that much of the pre-ringing will no longer be masked.
So masking indeed says that linear phase is more audible than minimal phase.

Ok sure :)

But are you certain that all linear phase FIR filters will exceed the masking thresholds at all frequencies (have never tried to calculate this myself)?

And you need to keep in mind that the pre-ringing will only occur off-axis, which means that to apply the masking thresholds to it you'll need to compare the SPL of off-axis reflections against the SPL of the direct sound, taking into account the delay caused by the longer path length of the reflected sound.

So let's say the reflection arrives 10 ms later than the direct sound (which would be the case for a path length difference of about 3 metres): pre-ringing occurring less than 10ms before the signal will arrive at the listener coincident with and/or following the signal, and so will be subject to simultaneous and/or post-masking thresholds (depending on the duration of the signal) rather than pre-masking thresholds.

Also, masking says that linear phase FIR ringing is more likely to be audible than minimum phase ringing, all else being equal. But this statement doesn't take group delay into account. There are still too many unknowns IMHO to make a definitive statement as to which is more likely to be audible overall, or indeed even as to whether the typical types of filters used in both cases are likely to be audible at all...
 
Last edited:
Thanks!
Actually, I was simplifying things by saying minimal phase IIR has only post-ringing.
Minimal phase IIR actually has short pre-ringing and long post-ringing. Pretty much like the masking graph shows!
One can use the masking graph to construct the maximum Q of a minimum phase filter which is still masked.
If one then turns that exact same filter with the same Q into a linear phase filter the graph shows that much of the pre-ringing will no longer be masked.
So masking indeed says that linear phase is more audible than minimal phase.

I think the error is thinking that 2 drivers and a crossover should have a linear phase as if its one driver for it to be perfect and that nature did a bad trick on music coming from multi-driver loudspeakers :) Nature didn't do a trick, minimal phase LR crossovers are perfect. And nature itself uses minimal phase EQ, be it analogue EQ or the filtering of sound traveling through air etc.

I have done this, you can make the filter minimum phase. I have done this as a test to try and reduce latency for theatre duties. Short version is It really doesn't sound as good. Poorer imaging and less clarity.

Yes The minidsp asrc converts to a 96KHz sample rate. My own tests show it is not a good performer., but you would need to check how it is operating in the sharc dsp. Saying "running at" is simplistic.
 
o_O:eek::eek:o_Oo_Oo_O

I need to brush up on Digital Signal Processing .. I am utterly lost on the above discussions...
 
Yes The minidsp asrc converts to a 96KHz sample rate. My own tests show it is not a good performer., but you would need to check how it is operating in the sharc dsp. Saying "running at" is simplistic.

Have you posted your tests somewhere by any chance? :)
 
I have done this, you can make the filter minimum phase. I have done this as a test to try and reduce latency for theatre duties. Short version is It really doesn't sound as good. Poorer imaging and less clarity.

Yes The minidsp asrc converts to a 96KHz sample rate. My own tests show it is not a good performer., but you would need to check how it is operating in the sharc dsp. Saying "running at" is simplistic.
Aah ok.. hmm. Thank you I will do the same test once my speakers are finished. Who knows, perhaps headphones are not an ideal test for this, but I would find it very strange..

No I meant that the ASRC is running at 96kHz on the MiniDSP.
I thought it was common knowledge for a very long time already that an ASRC needs to run at a different sample rate than the incoming sample rates. But apparently MiniDSP isn't aware of this.. (Hypex is, and my very old Lavry DA10 already knew too)
 
Aah ok.. hmm. Thank you I will do the same test once my speakers are finished. Who knows, perhaps headphones are not an ideal test for this, but I would find it very strange..

No I meant that the ASRC is running at 96kHz on the MiniDSP.
I thought it was common knowledge for a very long time already that an ASRC needs to run at a different sample rate than the incoming sample rates. But apparently MiniDSP isn't aware of this.. (Hypex is, and my very old Lavry DA10 already knew too)

It's a sharc dsp so it would use their asrc block

https://www.analog.com/media/en/tec...sheets/ADSP-21483_21486_21487_21488_21489.pdf

https://www.google.com/url?q=https:...iBH0QFggTMAM&usg=AOvVaw3rHsdEs-GUWqG7w6xAPBu-
 
Last edited:
Regarding ASRC and minidsp, it is done differently in the nanoSHARC/2x4HD (inside the DSP as March Audio noted, and measured as a bad performer) and in the miniSHARC/openDRC (SRC4382 chip, pretty common and solid).

Regarding ASRC to a different sampling rate or not, this is rarely a problem in practice, and 99% of the harware DSP out there do this on all incoming signals, regardless of the incoming sampling rate and its relation to the internal one.
Hypex' choice remain uniq AFAIK. By the way, Bruno Putzeys did implement a linear-phase acoustical crossover in the Grimm audio LS1, as well as in the Kii ;)
 
No, unless you mean linear-phase, and then it depend of the crossover frequency and order.
Yes, this was a comparison of linear phase FIR vs minimal phase IIR :)
It depends on crossover frequency and order but also on quality if I'm correct?
With minimal phase for instance 24dB/oct Butterworth lowpass there is a truly finite time of pre-ringing and an infinite time of post ringing if the "bit depth"/dynamic would be infinte?
With linear phase FIR 24dB/oct Butterworth lowpass both the pre and post ringing would be infinite if the "bit depth"/dynamic range would be infinite?
 
Regarding ASRC and minidsp, it is done differently in the nanoSHARC/2x4HD (inside the DSP as March Audio noted, and measured as a bad performer) and in the miniSHARC/openDRC (SRC4382 chip, pretty common and solid).

Regarding ASRC to a different sampling rate or not, this is rarely a problem in practice, and 99% of the harware DSP out there do this on all incoming signals, regardless of the incoming sampling rate and its relation to the internal one.
Hypex' choice remain uniq AFAIK. By the way, Bruno Putzeys did implement a linear-phase acoustical crossover in the Grimm audio LS1, as well as in the Kii ;)
Aah I didn't know. Is the MiniSHARC any good? What does it use for the ASRC sample rate?

I've never seen an ASRC set to the same sample rate as an incoming sample rate (before MiniDSP). This is common now? That's just wrong right?
And Bruno put linear-phase in as an option to woo misinformed buyers I think :p Haha.
 
There is no pre ringing in a minimum-phase filter.

As far as FIR is concerned, you of course need to truncate the response, so nothing infinite there ;)
IIR is infinite, and errors do pill up as it goes.

For the specific case of linear-phase FIR filter, there is no "latency vs quality" trade-off per se: what you get with too short FIRs is linear distortion, ie a magnitude and/or phase response that does not match the target.
This can all be simulated and the choice of a given length for a given correction will give this or that amount of deviation, also depending on the chosen windowing algorithm.
 
Aah I didn't know. Is the MiniSHARC any good? What does it use for the ASRC sample rate?
48kHZ with the openDRC firmware, 96kHz with the miniSHARC one.

I've never seen an ASRC set to the same sample rate as an incoming sample rate. This is common now? That's just wrong right?
ASRC do resample regardless of the incoming sampling frequency, because they have to sync the signal on their own internal clock anyway: 96kHz from the source will never exactly be the same as the one in the processor.

And Bruno put linear-phase in as an option to woo misinformed buyers I think :p Haha.
Same reasoning as Hypex' 93.75Hz sampling rate ;)
 
48kHZ with the openDRC firmware, 96kHz with the miniSHARC one.


ASRC do resample regardless of the incoming sampling frequency, because they have to sync the signal on their own internal clock anyway: 96kHz from the source will never exactly be the same as the one in the processor.

Yes I know but the thing is when the resampler is the same as the incoming sample rate or a close multiple or division of it then you will get all kinds of audible artifacts from the clocks not being in sync. These artifacts of ASRC are very much reduced by choosing a sample rate that isn't a multiple of 44.1 or 48kHz (like 93.75kHz). To my knowledge this was well known from the start of ASRC.. Done by Lavry and many other DAC makers many years ago.
 
There is no pre ringing in a minimum-phase filter.

As far as FIR is concerned, you of course need to truncate the response, so nothing infinite there ;)
IIR is infinite, and errors do pill up as it goes.

For the specific case of linear-phase FIR filter, there is no "latency vs quality" trade-off per se: what you get with too short FIRs is linear distortion, ie a magnitude and/or phase response that does not match the target.
This can all be simulated and the choice of a given length for a given correction will give this or that amount of deviation, also depending on the chosen windowing algorithm.
I thought minimum phase pre-ringing = group delay?
The trucating of the response of FIR is related to dynamic range correct? So for theoretical inifinte dynamic range / quality there is infinite pre and post ringing? The distortion that does not match the target is ever further down -dB the less one trucates. So this sounds very much like a latency vs quality trade off?
 
Yes I know but the thing is when the resampler is the same as the incoming sample rate or a close multiple or division of it then you will get all kinds of audible artifacts from the clocks not being in sync. These artifacts of ASRC are very much reduced by choosing a sample rate that isn't a multiple of 44.1 or 48kHz (like 93.75kHz). To my knowledge this was well known from the start of ASRC.. Done by Lavry and many other DAC makers many years ago.
Yet, with the exception of Bruno Putzeys's designs, AFAIK all existing hardware DSP solutions use these common sampling rates.
Anyway, I don't know the gory details of all these ASRC implementations, and who would I be to question his choices.

By the way, here is an interesting and very informative thread by an ASRC designer: https://www.diyaudio.com/forums/digital-source/28814-asynchronous-sample-rate-conversion.html
 
As for errors piling up with IIR, this is only for feedback / Q right? And one can easily make the argument that simply increasing the bit depth solves this, and probably still more efficiently than a FIR of equal quality?
 
I thought minimum phase pre-ringing = group delay?
The trucating of the response of FIR is related to dynamic range correct? So for theoretical inifinte dynamic range / quality there is infinite pre and post ringing? The distortion that does not match the target is ever further down -dB the less one trucates. So this sounds very much like a latency vs quality trade off?
You can tailor the way linear-distortion will happen with your choice of windowing algorithm.
You can try for yourself using rephase for example.
 
As for errors piling up with IIR, this is only for feedback / Q right? And one can easily make the argument that simply increasing the bit depth solves this, and probably still more efficiently than a FIR of equal quality?
The "infinite" in IIR means infinite recursion, with infinite errors as a result :D
More seriously IIR can be made to work, of course, but there are other advantages to FIR outside sound quality (algorithmic "purity" :p) or phase-linearity: arbitrary magnitude and phase response, predictability, simplicity.

When you measure a device you actually record a FIR (finite), and you accept that it is an accurate representation of what you are going to correct, down to a given precision. This is only logical to address its correction with a FIR (some even use an inversion of the measurement FIR for this purpose ;) ).
Nothing is infinite here, everything has to be truncated. Doing this right in the measurement (already done) and correction (only done in the case of a FIR) process using a careful choice of windowing is only logical and sound, rather than letting the device applying the chosen correction decide how to truncate/approximate a given target.
 
Yet, with the exception of Bruno Putzeys's designs, AFAIK all existing hardware DSP solutions use these common sampling rates.
Anyway, I don't know the gory details of all these ASRC implementations, and who would I be to question his choices.

There are DACs that ARSC everything to arbitrary sample rates the same way the Kii does.
 
  • Like
Reactions: pos
Back
Top Bottom