Paul, we all owe you for your relentless work in creating new filters to try!
And I haven't even tried a single one since I just finished my breadboard build of the DAM...
Will try in the weeks to come.
On another note: have Blue - Solar Fields playing while I'm typing... nice ambient stuff. Have you tried Connect.Ohm - 9980?
And I haven't even tried a single one since I just finished my breadboard build of the DAM...
Will try in the weeks to come.On another note: have Blue - Solar Fields playing while I'm typing... nice ambient stuff. Have you tried Connect.Ohm - 9980?
No. Impulse peak time is simply the filter shape and the coefficients number. More coefficients greater delay. But I see 2 ms delay is no important.
What is interesting is the slope time length of ascent and descent. It would have been ideal if the graph soften the peaks and valleys but the program does not make.
And if some filter spoiling phase too and can tweak to improve like 1021SA2F2v1.
Last edited:
Yes I have. I've a bit of an Ultimae fan since I discovered their Fahrenheit Project releases in the mid-2000's.
On the filter R&D front:
I've done the filter co-efficents using the v3 settings but sweeping phase for Minimum (p=0) through to Linear (p=50) in increments of 5. v1-3 are all done using p=30. I'll get around to making up filter files for these later today.
I had a quick play around with Octave's FIR2 routine and found I can get very similar but non-identical results to what SoX produces using b= to set kaiser-window beta.
Attachments
Last edited:
I was about to post up batch of v3 filters when I had a listen to the -p0 version...
I had a suspicion that the minimum and intermediate phase weren't quite as sexy as the cheers leaders would have you believe. You only have to have a peek at the step response to see what is going to happen.
Changing phase doesn't remove ringing from the filter, it simply shifts it. And there-in lies the rub. If you move all the pre-ringing energy to after the impulse you double the amplitude of the post ringing, and that means the overshoot doubles too. Barf.
This plot shows the step response of the filter coefficients.
Orange is p0 (-M) , green p30 , and blue p50 (-L).
-p30 sounds ok using volume control, -p0 was just plain awful.
It would be possible to drop the gain on these filter but the output is going to end up very low.
I've attached the folder of filters but be aware that anything less that p25-30 is likely to distort badly.
I had a suspicion that the minimum and intermediate phase weren't quite as sexy as the cheers leaders would have you believe. You only have to have a peek at the step response to see what is going to happen.
Changing phase doesn't remove ringing from the filter, it simply shifts it. And there-in lies the rub. If you move all the pre-ringing energy to after the impulse you double the amplitude of the post ringing, and that means the overshoot doubles too. Barf.
This plot shows the step response of the filter coefficients.
Orange is p0 (-M) , green p30 , and blue p50 (-L).
-p30 sounds ok using volume control, -p0 was just plain awful.
It would be possible to drop the gain on these filter but the output is going to end up very low.
I've attached the folder of filters but be aware that anything less that p25-30 is likely to distort badly.
Attachments
Last edited:
Controlling Phase with Sox on
Once again an excellent post Paul. I been meaning to test exactly what you just did there. But can’t for the life of me figure out how to do those impulse plots?
From the SoX manual we can read:
Is –p given in samples or milliseconds?
Once again an excellent post Paul. I been meaning to test exactly what you just did there. But can’t for the life of me figure out how to do those impulse plots?
From the SoX manual we can read:
Code:
A phase response setting may be used to control the distribution of any
transient echo between ‘pre’ and ‘post’: with minimum phase, there is
no pre-echo but the longest post-echo; with linear phase, pre and post
echo are in equal amounts (in signal terms, but not audibility terms);
the intermediate phase setting attempts to find the best compromise
by selecting a small length (and level) of pre-echo and a medium
lengthed post-echo.
Minimum, intermediate, or linear phase response is selected using
the −M, −I, or −L option; a custom phase response can be created
with the −p option. Note that phase responses between ‘linear’
and ‘maximum’ (greater than 50) are rarely useful.Is –p given in samples or milliseconds?
Yes I have. I've a bit of an Ultimae fan since I discovered their Fahrenheit Project releases in the mid-2000's.
Speaking of Ultimae Records, the track "Alone" by Jaia from the Fahrenheit Five always comes to mind as the most deep, spacious and dimensional one from the entire series... Even on my plain old 1616m it sounds incredible.
Would be interesting to supply the most "dimensional" filter for that track, brewed by an Ultimae fan
Hi, Got a quick prototype up and running while awaiting remaining parts to arrive. Tried just about all the filters on here - thank you - having read all 50-something pages with interest. Likely will have a go myself at some point once I figure out how I can measure the performance of the filters with my Tek TDS2024. I prefer the 'Crap V2 filters', the original and the latest one posted a few hours back. In my mind they offer greater weight and scale to the sound, dynamics if you like, and I think the most transparent of the lot. My aim, simply to reporduce what is in the original mix without colouration or bias of any kind. Guess that's what we all look for, kind of....??
Thanks for all the brilliant work and helping me enjoy playing electronics & music once again!
Andrew
Thanks for all the brilliant work and helping me enjoy playing electronics & music once again!
Andrew
neither!
0 gives 100% post-ringing and 100 gives 100% pre-ringing. 50 gives an equal balance between pre and post ringing or your standard linear phase impulse.
The step response was done in fuzzmeasure.
Ahh, gotcha we just shift the null-point. Make sense. What idiot other than me would order a filter with ringing?
What files do you feed fuzzmeasure? I have REW and Holm installed. But they both laughs at me when I try to import files.
Holm can’t make sense of anything above 192kHz. In my mind we need to simulate with intended coefficient table?
I can have Holm do impulse plot with
But I’m not sure if that 44.1k fs plot would equal to? edit: fs would of course, I ment impulse respons with 44.1 fs?
Is it possbile to do impulse plots with Octave by altering this code?
What files do you feed fuzzmeasure? I have REW and Holm installed. But they both laughs at me when I try to import files.
Holm can’t make sense of anything above 192kHz. In my mind we need to simulate with intended coefficient table?
I can have Holm do impulse plot with
Code:
sox --plot octave -r 44.1k -n output.wav synth 1 noise sinc -b15 -L -20.5k -t3800 > filterDM_Linear.wav
Code:
sox --plot octave -r 352.8k -n output.wav synth 1 noise sinc -b15 -L -20.5k -t3800 > filterDM_Linear.wavAn externally hosted image should be here but it was not working when we last tested it.
Is it possbile to do impulse plots with Octave by altering this code?
Code:
[h,w]=freqz(b,1,4096);
plot(176400*w/pi,20*log10(h))
Last edited:
Here's some comments from Mike Moffat @Schiit regarding the filter design of Yggdrasil DAC, which is claimed by those who heard it as the next best thing.
" That the DSP filter in the Yggy is closed form, and preserves all of the original samples is commonly known. The filter is also time domain optimized which means the phase info in the original samples are averaged in the time domain with the filter generated interpolated samples to for minimum phase shift as a function of frequency from DC to the percentage of nyquist - in our case .968. Time domain is well defined at DC - the playback device behaves as a window fan at DC - it either blows (in phase) or sucks (out). It is our time domain optimization that gives the uncanny sonic hologram that only Thetas and Yggys do. (It also allows the filter to disappear. Has to be heard to understand.) This is combined with a frequency domain optimization which does not otherwise affect the phase optimization. The .968 of nyquist also gives us a small advantage that none of the off-the shelf FIR filters provide: frequency response out to 21.344KHz, 42.688KHz, 85.3776KHz, and 170.5772KHz bandwidth for native 1,2,4, and 8x 44.1KHz SR multiple recordings - the 48KHz table is 23.232, 46.464, 92.868, and 185.856KHz respectively for 1,2,4, and 8x.
The Pacific Microsonics PM100 filter (HDCD) was a variation of all of the off-the shelf cookbook filters of the late 80's with the standard Frequency Domain successive approximation coefficient calculation found in the Burr-Brown, Phillips, Analog Devices, and are still used on the front ends of most delta-sigma DACs today. I cannot comment beyond that because I was not interested in a format that had already been condemned to stillbirth because of its required encode - exactly what's wrong with DSD. You have to get all of the studios on board. As I have said before, this leads to recordings like: The Folsom Prison Orchestra playing Regional Polkas of the Balkan States, or better yet, the folkloric Orkney Island Shepherds Cries of Ecstasy upon Summer Moonlight. The latter was so well recorded, I've heard, that the listener could hear the shepherds gently guiding the young kids' rear legs into their high topped boots as the objects of the rapturous ecstasy gently bleat. Those recordings are not for me - my musical tastes are quite ordinary - my products support the 99.999 per cent of recordings. Again, I digress. Since lower frequency wavelengths are measured in tens of feet, placement in image gets increasingly wrong as a function of decreasing frequency in non time domain optimized recordings - these keep the listener's ability to hear the venue - not to mention the sum of all of the phase errors in the microphones, mixing boards, eq, etc on the record side. An absolute phase switch is of little to no value in a stochastic time domain replay system."
" That the DSP filter in the Yggy is closed form, and preserves all of the original samples is commonly known. The filter is also time domain optimized which means the phase info in the original samples are averaged in the time domain with the filter generated interpolated samples to for minimum phase shift as a function of frequency from DC to the percentage of nyquist - in our case .968. Time domain is well defined at DC - the playback device behaves as a window fan at DC - it either blows (in phase) or sucks (out). It is our time domain optimization that gives the uncanny sonic hologram that only Thetas and Yggys do. (It also allows the filter to disappear. Has to be heard to understand.) This is combined with a frequency domain optimization which does not otherwise affect the phase optimization. The .968 of nyquist also gives us a small advantage that none of the off-the shelf FIR filters provide: frequency response out to 21.344KHz, 42.688KHz, 85.3776KHz, and 170.5772KHz bandwidth for native 1,2,4, and 8x 44.1KHz SR multiple recordings - the 48KHz table is 23.232, 46.464, 92.868, and 185.856KHz respectively for 1,2,4, and 8x.
The Pacific Microsonics PM100 filter (HDCD) was a variation of all of the off-the shelf cookbook filters of the late 80's with the standard Frequency Domain successive approximation coefficient calculation found in the Burr-Brown, Phillips, Analog Devices, and are still used on the front ends of most delta-sigma DACs today. I cannot comment beyond that because I was not interested in a format that had already been condemned to stillbirth because of its required encode - exactly what's wrong with DSD. You have to get all of the studios on board. As I have said before, this leads to recordings like: The Folsom Prison Orchestra playing Regional Polkas of the Balkan States, or better yet, the folkloric Orkney Island Shepherds Cries of Ecstasy upon Summer Moonlight. The latter was so well recorded, I've heard, that the listener could hear the shepherds gently guiding the young kids' rear legs into their high topped boots as the objects of the rapturous ecstasy gently bleat. Those recordings are not for me - my musical tastes are quite ordinary - my products support the 99.999 per cent of recordings. Again, I digress. Since lower frequency wavelengths are measured in tens of feet, placement in image gets increasingly wrong as a function of decreasing frequency in non time domain optimized recordings - these keep the listener's ability to hear the venue - not to mention the sum of all of the phase errors in the microphones, mixing boards, eq, etc on the record side. An absolute phase switch is of little to no value in a stochastic time domain replay system."
Code:
fs=352800;
[H,f] = freqz (b);
figure(1);
subplot(211);
plot((f/(2*pi))*fs,20*log10(abs(H)));
xlabel('Frequency (hz)');
ylabel('Amplitude Response (dB)');
axis([17000, 19000, -0.2, .1]);
grid('on');
subplot(212);
plot(f/(2*pi)*fs,20*log10(abs(H)));
xlabel('Frequency (hz)');
ylabel('Amplitude Response (dB)');
axis([0, fs/2, -160 5]);
grid('on');
print -FHelvetica -r72 -dpng fir_response.png;
figure(2);
subplot(111);
[Hi, Ti] = impz(b);
plot(Ti,Hi);
grid('on');
axis([0,size(b,1),-0.05,0.13]);
print -FHelvetica -r72 -dpng fir_impulse.png;
dlmwrite("imp441_test.txt", b, "delimiter", "\\r");What I do is run the plot file which creates array "b". Then I run the above which is saved as SoxPlot.m
Octave has a impz() function that will generate the impulse plot.
I wish people would turn on their Bullschitt detectors more often.
The entire paragraph on HDCD is so full of opinionated disinformation and garbage that it is pretty obvious the "Fulla" was named in honour of Mike.
They might have some magic happening but they aren't giving much away about how it's done with the broad brush strokes and "I'm a celebrity, listen to me" presentation. Sounds like that little infomercial would have given the guys at ComputerAudioPhool a boner. Ooooohhhh "uncanny sonic hologram", best sell my "last months fashion DAC" and buy one.
You'll have to excuse me, I have a virus at the moment and I'm feeling pretty schitt.
Last edited:
Speaking of Ultimae Records, the track "Alone" by Jaia from the Fahrenheit Five always comes to mind as the most deep, spacious and dimensional one from the entire series... Even on my plain old 1616m it sounds incredible.
Would be interesting to supply the most "dimensional" filter for that track, brewed by an Ultimae fan
That has forced me to upgrade my old emusic mp3 copy of Vol 5 to lossless. They'll probably bring out a 24bit version next week
I'll have a play around and add that as a reference track.
No, the whole thing is just marketing crap. If an engineer wrote it they should be totally ashamed. Importantly, we are not very sensitive to phase, we do use phase for placement information up to about 800Hz, above that the the size of the head means that we need to use other cues (ILDs, pinna effects etc). Importantly, the last statement "placement in image gets increasingly wrong as a function of decreasing frequency in non time domain optimised recordings" is fundamentally wrong and is against the laws of physics and how we hear.
I could go on but it's pointless in the face of a lack of logic and science.
It's a bit hard to tell what the technical details are. Maybe zfe can make sense of it, has he a much better grasp on the technical/math than I do.
BTW this is from Goodwin's High End - HDCD
Goat Herds and Prison Choirs must have been big in the lead up to the Millennium.
There is a listing of 5000 HDCD titles here: Goodwin's High End - HDCD Recordings
It would have been easy enough to dismiss HDCD on the basis that Micro$oft purchased the HDCD IP in 2000 when they bought Pacific Microsonics, and the studio mastering side was basically left to die. Even so the Pacific Microsonics Model II ADC/DAC was regarded as one of the best studio convertors available even as late as 2010-11.
DEQ,
What I'm doing is processing a 1 sample chirp using SoX's fir option, then using the resulting .wav in fuzzmeasure.
The chirp is 5 seconds of silence with a single 0dBFS sample in the middle.
What I'm doing is processing a 1 sample chirp using SoX's fir option, then using the resulting .wav in fuzzmeasure.
The chirp is 5 seconds of silence with a single 0dBFS sample in the middle.
Code:
sox /path/to/TestChirp441.wav -b24 -r352.8k /path/to/output.wav upsample 8 fir /path/to/filter_coefficients.txt vol 8Sorry I missed your post in the flurry....
Glad you are enjoying.
The CRaP filters seem to reasonably good. The original CRaP Magic has a fir2 filter that rolls off a bit too early, and is probably what was TNT was commenting on - mind you it was above 20kHz so shouldn't be too audible in theory.
My preference is for the second revision FIR2 filter as this stays "out of the way" more, and I'm using these on all the recent filters.
I'm enjoying the latest round of filters - the IP3p30->50 are all pretty listenable. However in comparison to a filter that is flat to 19.5kHz they sound slightly "dark" to my ears.
I was just experimenting with a filter that rolled off fairly slowly above 19kHz but didn't fully attenuate until 28.5kHz. This combination didn't work especially well - there seemed to be distortion in piano notes* that isn't apparent using the IPv3p35 for example.
As a point of reference the CRaP Magic filter was only 0.1dB down at 20kHz but attenuated fully by 25kHz.
Perhaps I need to do an IPv4 with the corner frequency lifted further again...
cheers
Paul
* The piano track was Bernie Worrell's "I'd Rather Be With You" from Elevation: The upper air; I'd Rather Be With You | Bernie Worrell
Other selected listening matter today:
- The Pleiadians - Seven Sisters from Family of Light: https://www.youtube.com/watch?v=kBh9ZRHAxBg
- Bill Laswell "Nothing" from City of Light: https://www.youtube.com/watch?v=4i3Acuoshgc
- Paul Schütze - The Surgery of Touch: The Surgery of Touch | Paul Schütze
I may have some problems with the language, but are you not doing both the same statement?:
"The correct phase of low frequencies is important (the lower the more important) for correct placement".
"time domain optimized recordings" seems to be their term for trying to preserve phase.