This is the most likely cause of your preference. Maybe the X32 doesn't have the best analogue stages or could be the gain structure is off, an audible rise in the noise floor points to either of these.
Measuring with the gain control is using an attenuator at the input which means that the amp isn't really amplifying. The best way to measure the performance of an amp is with the attenuator at the output so you can see the effect the amp has on the output at different power levels. For your comparison it is not a problem as it allows you to capture the analogue conversion stage in your measurement.
The icepower amps have a very different THD profile vs frequency at different powers. The noise goes up significantly in the higher frequencies as power rises. Measuring with the gain turned down will make the the amp look much better than it really is.
No issue with you preferring one over the other, using the OpenDRC in pure digital mode is the best way to use it if possible.
Record a sweep in REW through both chains, make sure you align the impulses to the same position then you can compare the difference. Becomes hard to get a true comparison as the analogue chain will need an A/D conversion to be measured and the OpenDRC doesn't. To include the A/D you need another D/A which will throw it off again.
You could also do a THD comparison and a basic electrical transfer function test to see the difference in the noise floor. The REW help file or ARTA manual explains how to measure these if you haven't done them before.
Seems to me like the reduced tap count is preferable to you over the issues that adding the extra analogue circuits in the chain causes. If you can't remove those from the chain it will be impossible to compare the tap counts directly. Maybe no need to actually measure it
Sorry that most of this has nothing to do with rephase
To do a digital recording of the open DRC, you'll want a soundcard with a digital input. AES, SPDIF, whatevrz. I've done this with my M-Audio USB card. Not all soundcards have digital inputs, you'll need to borrow one that does. When recording I used one laptop as the source, another to record with. Never could get it right using the same computer for both because of driver conflicts. That may not be an issue with later version of Windows. The recordings were bit perfect, so no worries there.
Barring a soundcard with digital input, do as suggested in the posts above and make analog recordings of both. Be careful about recording from the Icepower amps, tho. Being class D there will be a good bit of RF carrier in the outputs. Although the Icepower modules do a very good job of cleaning that up, it still might do odd things to the ADC and soundcard inputs.
Barring a soundcard with digital input, do as suggested in the posts above and make analog recordings of both. Be careful about recording from the Icepower amps, tho. Being class D there will be a good bit of RF carrier in the outputs. Although the Icepower modules do a very good job of cleaning that up, it still might do odd things to the ADC and soundcard inputs.
you said when going drc route you went x-32 digital into drc. can the x-32 not go digital to the ice amp?
I am assuming that going analog to the ice amp involves an extra ad conversion?
(Pre)ringing length and volume are dependent on filter slope and frequency.
Using a lower number of taps (and appropriate windowing algorithm) will (gently) shop them at a given length, and as a result reduce slope.
This is the way slope is controlled with brickwall filters: the more taps and the sharper the window, the steeper the slope and as a consequence the more (pre)ringing you get.
But the important thing is that what matters here is the resulting slope.
@Wesayso and Pano, thx for the direction on how to begin comparing files. Very helpful....given me good start learning something new.
@fluid, yes I believe the cause of my preference is as you stated. Thx for pointers on measuring amps, and REW measurement technique. And I think you're right, there may be no need to measure, when AD / DA variables are accounted for
@1201, the x-32 has only 1 AES out. I need two AES output to feed 2 ICE amps, 2 channels each. I'd like to find a soundcard that could take at least 6 usb channels from JR and assign them however needed, to at least 3 AES outs.
You assume correct....I think the x-32's DA conversion of JR's usb outputs, simply isn't as clean as the DRCs.
@ Pos and BYRTT, sorry for all the discussion not more directly related to rePhase.
Getting back on topic, I skimmed the linked paper.
I was struck by the very high filter orders being discussed....
Do FIR filters really get described/used as a 500th or 700th order, with an order being the usual 6 dB per octave ? that degree of steepness doesn't seem to make sense ...
@all THANKS
mark
@fluid, yes I believe the cause of my preference is as you stated. Thx for pointers on measuring amps, and REW measurement technique. And I think you're right, there may be no need to measure, when AD / DA variables are accounted for
@1201, the x-32 has only 1 AES out. I need two AES output to feed 2 ICE amps, 2 channels each. I'd like to find a soundcard that could take at least 6 usb channels from JR and assign them however needed, to at least 3 AES outs.
You assume correct....I think the x-32's DA conversion of JR's usb outputs, simply isn't as clean as the DRCs.
@ Pos and BYRTT, sorry for all the discussion not more directly related to rePhase.
Getting back on topic, I skimmed the linked paper.
I was struck by the very high filter orders being discussed....
Do FIR filters really get described/used as a 500th or 700th order, with an order being the usual 6 dB per octave ? that degree of steepness doesn't seem to make sense ...
@all THANKS
mark
@fluid, yes I believe the cause of my preference is as you stated. Thx for pointers on measuring amps, and REW measurement technique. And I think you're right, there may be no need to measure, when AD / DA variables are accounted for
@1201, the x-32 has only 1 AES out. I need two AES output to feed 2 ICE amps, 2 channels each. I'd like to find a soundcard that could take at least 6 usb channels from JR and assign them however needed, to at least 3 AES outs.
You assume correct....I think the x-32's DA conversion of JR's usb outputs, simply isn't as clean as the DRCs.
There are a few interfaces that have multiple AES outputs but they are not cheap. These are two that I know of there are probably others.
AES16e
MOTU.com - Overview
If you splashed out for the MOTU one you could even do a pretty intense multichannel active system!
@BYRTT. To me it seems like the paper refers to FIR order as number of TAPS not the order of the filter as in analog or IIR. Also my simulation uses 96k samplerate, think the paper uses lower. (My bad)
@POS. The paper shows that it is not the magnitude error in the frequency domain that is annoying (inside limits). It is the combined timeshifted preringing of the filters that gives artifacts at 3k Hz and more than 700 taps.
So more taps/longer FIR filter gives longer preringing and therefore MORE audible artifacts.
(That is the suprising findings of the paper)
@silverprout. I know you are not suprised
@POS. The paper shows that it is not the magnitude error in the frequency domain that is annoying (inside limits). It is the combined timeshifted preringing of the filters that gives artifacts at 3k Hz and more than 700 taps.
So more taps/longer FIR filter gives longer preringing and therefore MORE audible artifacts.
(That is the suprising findings of the paper)
@silverprout. I know you are not suprised
Last edited:
Not same resulting slope, of course. Somewhere down the decibels it will deviate.
The strange ting is of course that the shorter FIR filters performed better in listening tests than the longer.
The filters used are listed in the enclosed picture from the thesis. I think the FIR filters are long enough to mimic the LR IIR filters to a good approximation
It's all described very well in the thesis. (I have no affiliation with the authors or the university, but have an acoustics masterdegree years ago, myself)
Attachments
Last edited:
If there are differences in the slopes and one is preferred over the other it wouldn't be fair to blame it on the length of the filter in my humble opinion.
There is nothing wrong with using long filters as long as you know what the filter is doing.
A measured result would be more convincing.
Like with anything, there is a right way and a wrong way to apply FIR filters. Personally I try to minimise any pré-ringing. In all honesty, I don't have crossovers and don't linearize the phase but make my system behave like the true band pass device that it is.
With crossovers you can achieve the same, as long as off axis response isn't deviating from the desired response. A Synergy concept is a good example of that. Wherever you measure them within their coverage the results will hold up.
Blaming it on long FIR filters isn't fair i.m.o.
There is nothing wrong with using long filters as long as you know what the filter is doing.
A measured result would be more convincing.
Like with anything, there is a right way and a wrong way to apply FIR filters. Personally I try to minimise any pré-ringing. In all honesty, I don't have crossovers and don't linearize the phase but make my system behave like the true band pass device that it is.
With crossovers you can achieve the same, as long as off axis response isn't deviating from the desired response. A Synergy concept is a good example of that. Wherever you measure them within their coverage the results will hold up.
Blaming it on long FIR filters isn't fair i.m.o.
They used the same speaker and elements and the same room and program material.
The only change was the FIR filter length. So either was their experiment flawed or the FIR filter length mattered.
Can anyone point me to of axis controlled listening tests that shows that longer filterlengths is better?
The only change was the FIR filter length. So either was their experiment flawed or the FIR filter length mattered.
Can anyone point me to of axis controlled listening tests that shows that longer filterlengths is better?
I just skimmed the thesis quickly.
Page 17/18 and 45/46 it appears the author is clearly using brickwall filters, and what he calls order in that regard is the length of the FIR.
When designing brickwall filters the longer the FIR the steeper the slope, and of course the longer the (pre)ringing as a result.
Again, the slope is responsible here, not the length of the FIR.
Brickwall filters are undoubtedly the easiest type of filter to build in FIR (windowed sinc, it does not get any simpler than this...), but certainly not the only one, and clearly not representative of what can be done in FIR.
Fortunately directly associating FIR to brickwall filters, albeit still a common shortcut 10 years ago, has become an outdated practice today
Page 17/18 and 45/46 it appears the author is clearly using brickwall filters, and what he calls order in that regard is the length of the FIR.
When designing brickwall filters the longer the FIR the steeper the slope, and of course the longer the (pre)ringing as a result.
Again, the slope is responsible here, not the length of the FIR.
Brickwall filters are undoubtedly the easiest type of filter to build in FIR (windowed sinc, it does not get any simpler than this...), but certainly not the only one, and clearly not representative of what can be done in FIR.
Fortunately directly associating FIR to brickwall filters, albeit still a common shortcut 10 years ago, has become an outdated practice today
Last edited:
Yes, it's pretty save to say the FIR length had nothing to do with the results of this test.
They could have linearized the phase of a Linkwitz-Riley crossover (like RePhase lets you do), instead they just increased the slope, the longer the filter became. So the warning here would be: don't use brick wall filters with the steepest slope (just because you can). You could get audible pré-ringing off axis if the slope is steep enough.
Don't blame the filter, blame what they did with it.
They could have linearized the phase of a Linkwitz-Riley crossover (like RePhase lets you do), instead they just increased the slope, the longer the filter became. So the warning here would be: don't use brick wall filters with the steepest slope (just because you can). You could get audible pré-ringing off axis if the slope is steep enough.
Don't blame the filter, blame what they did with it.
Last edited:
I agree about the brickwall filter and the higher slope for longer FIR filter.
Still I don't understand why the ear/brain should react to high slopes? The study had high scores for long filters on axis/no delay
It still must be the ringing because of the the time shift, that makes the artifacts.
And even LR linear phase filters with time shift generates huge ringing effects and no ringing with no time shift between the filters.
So I think the conclution still stands even if the FIR length must be longer ( e.g. steeper slope) when non brickwall filters is generated, to get the same length of ringing.
Luckily REW can be used to generate timeshift on one filter and sum with another to look at the ringing effects.
Anyways, it is easy to run a 10 Hz square wave through a FIR filter speaker to check for off axis artifacts compared to on axis sound signature.
Still I don't understand why the ear/brain should react to high slopes? The study had high scores for long filters on axis/no delay
It still must be the ringing because of the the time shift, that makes the artifacts.
And even LR linear phase filters with time shift generates huge ringing effects and no ringing with no time shift between the filters.
So I think the conclution still stands even if the FIR length must be longer ( e.g. steeper slope) when non brickwall filters is generated, to get the same length of ringing.
Luckily REW can be used to generate timeshift on one filter and sum with another to look at the ringing effects.
Anyways, it is easy to run a 10 Hz square wave through a FIR filter speaker to check for off axis artifacts compared to on axis sound signature.
Last edited:
Just realized that my initial question probably was answered by POS earlier.
For a given filter, longer FIR gives longer ringing because the impulse is not so much truncated by the window/FIRlength. But, is does not matter.
Because that happen longer from the main lobe the longer the FIR filter.
From the paper it seems like the strong ringing close (0.1 to 1 ms) to the main lobe is most critical for artifacts with normal audio filters and off axis listening.
Maybe I'm wrong and it was not proposed by the paper...
For a given filter, longer FIR gives longer ringing because the impulse is not so much truncated by the window/FIRlength. But, is does not matter.
Because that happen longer from the main lobe the longer the FIR filter.
From the paper it seems like the strong ringing close (0.1 to 1 ms) to the main lobe is most critical for artifacts with normal audio filters and off axis listening.
Maybe I'm wrong and it was not proposed by the paper...
torgeirs, the steeper the slope the longer the ringing, that is all there is to it, and this is true for IIR and FIR.
FIR let you design steeper slope, increasing possible ringing.
It also let you design linear-phase filters, wich moves hald of the ringing in front of the impulse (preringing).
A 24dB/oct FIR filter will have the same amount of ringing than the equivalent IIR filter. Less in fact, as the IIR is infinite...
FIR let you design steeper slope, increasing possible ringing.
It also let you design linear-phase filters, wich moves hald of the ringing in front of the impulse (preringing).
A 24dB/oct FIR filter will have the same amount of ringing than the equivalent IIR filter. Less in fact, as the IIR is infinite...
Here's the best paper on pre-ringing I've seen so far.
http://www.bodziosoftware.com.au/Pre_Post_Ringing_IR_And_Pulses.pdf
