• 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!

Minimum Phase vs Linear Phase

TuneInSoul

Member
Joined
Sep 20, 2018
Messages
76
Likes
5
If the filters setting affects your mood during listening , it is good to change it.
I can’t give you an answer for the question like which one is better : loose vs tight.
 

Matias

Master Contributor
Forum Donor
Joined
Jan 1, 2019
Messages
5,029
Likes
10,796
Location
São Paulo, Brazil
BTW, interesting your list of DAC with minimum phase and slow (my choice too). The logical thing is to assume that any modern DAC has the minimum and linear implemented in its two variants but...

https://yabb.jriver.com/interact/in...HPSESSID=d92m3jk4nn3sf3s1krtru594f5#msg785295

For that topic with JRiver I searched Stereophile for known brands of DACs that have the minimum phase, slow roll off so that I could argue the relevance of it. But actually, from my previous post, you can see that I came to these settings that have lower post ringing as the typical minimum phase slow roll off.

I invite everyone to choose a 44/16 lossless song of your choice with a loud and clear sounding drum kit, with a consistent snare rhythm. Then use foobar+resample-v to convert new files upsampling it to:
1. Linear phase, fast roll off.
2. Minimum phase, slow roll off.
3. The settings I have shown.

Listen to the 3 new files, paying attention to the snare sound, and see if you can hear the difference. Have fun! :)
 
OP
Ron Texas

Ron Texas

Master Contributor
Joined
Jun 10, 2018
Messages
6,076
Likes
8,911
For that topic with JRiver I searched Stereophile for known brands of DACs that have the minimum phase, slow roll off so that I could argue the relevance of it. But actually, from my previous post, you can see that I came to these settings that have lower post ringing as the typical minimum phase slow roll off.

Many recent DAC's allow user control over these parameters as this feature may be enabled in AKM and ESS chips. I do recall Bryston makes a DAC with an AKM chip which is permanently set to minimum phase, slow roll off.
 

scott wurcer

Major Contributor
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
1,501
Likes
2,821
I am sure you are aware that 12 bits at 44.1 kHz is enough for vinyl....
This is not true on several levels. The traditional SNR definition of an A/D does not tell the right story here and you need to consider capturing the noise spectral density (NSD) over the audio range. You then need to factor in dynamic headroom and equalization. The folks at Channel D (http://www.channld.com/puremusic/ ) have an AES presentation on the issues and I wrote a short article with some data and actual measurements. If you want to do digital RIAA 24/48 would be minimal IMO. If you want to do real time using IIR filters 24/96 is better since it reduces the frequency warping effects. There is a table with IIR coefficients that I computed for all the classical equalization curves at https://linearaudio.net/articles , you never need more than two basic bi-quad filters to get (I admit) absurdly close to perfect.

BTW, I'm not into the anti-vinyl vibe. Audio is a hobby too the more folks that enjoy whatever aspect of it the better as I see it, the nonsense and anti-science can be ignored.
 

Julf

Major Contributor
Forum Donor
Joined
Mar 1, 2016
Messages
3,004
Likes
3,998
Location
Amsterdam, The Netherlands
This is not true on several levels. The traditional SNR definition of an A/D does not tell the right story here and you need to consider capturing the noise spectral density (NSD) over the audio range. You then need to factor in dynamic headroom and equalization. The folks at Channel D (http://www.channld.com/puremusic/ ) have an AES presentation on the issues and I wrote a short article with some data and actual measurements. If you want to do digital RIAA 24/48 would be minimal IMO. If you want to do real time using IIR filters 24/96 is better since it reduces the frequency warping effects.

Sure, for recording and processing, you do want to use 24 bits and higher sample rates - but once you are done, you can definitely normalize to 16/44.1 without losing any audible information.
 

maty

Major Contributor
Joined
Dec 12, 2017
Messages
4,596
Likes
3,160
Location
Tarragona (Spain)
If you want to do digital RIAA 24/48 would be minimal IMO. If you want to do real time using IIR filters 24/96 is better since it reduces the frequency warping effects. There is a table with IIR coefficients that I computed for all the classical equalization curves at https://linearaudio.net/articles , you never need more than two basic bi-quad filters to get (I admit) absurdly close to perfect...

The choice of the 16 bits and 44 kHz was due to a compromise of what was known at the time and what the technology allowed. If they could change what was decided then they would have risen to 48 kHz. At that time 24-bit was not possible.

Years ago, in my resampler tests, starting from a 24/192 WAV vinyl rip, I liked it more 16/96 than 24/48. ALWAYS, without any doubt, 24/48 vs 24/44 -> then the logical option should be 24/96. Of course, the original rip at 24/192 WAV (someone at 32-bit) in case you want to do a cleaning via software, usually iZotope. If the vinyl is in good condition, new or NOS, better to avoid further processing, because sometimes it reduces the dynamic range significantly. Better withstand some sporadic clicks/noises, which are barely noticeable with speakers.

As with speaker and room equalization, it is better not to be too aggressive if software is used to optimize the sound.
 
Last edited:

adamd

Member
Joined
Sep 24, 2018
Messages
38
Likes
42
Archimago had the following conclusion in part III:

As we know last week, statistically, indeed there was a preference for minimum phase filtering, but not strongly significant based on statistical data.
Hmmmm. Sorry for being late to the party but this seems to me to be an odd summary. Out of three music samples, in two there was a mild non-statistically significant preference for MP and in one it was for LP. The nearest to a significant result was a preference in one track for LP amongst headphone listeners (I'm not sure that multi factor adjustment was made to the significance threshold). Again there was an overall (across the three tracks) non significant preference among headphone listners for LP and among loudspeaker listeners for MP.
All looking pretty damn random.
 

scott wurcer

Major Contributor
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
1,501
Likes
2,821
Sure, for recording and processing, you do want to use 24 bits and higher sample rates - but once you are done, you can definitely normalize to 16/44.1 without losing any audible information.

Of course, I did some old Takoma blues LP's for a friend and I and burned them as CD's long before we thought they would ever get a re-issue. Most of the cuts were actually noise limited by the ancient Ampex recorder they used.

A caution about bit depth, I don't know of any A/D with and ENOB near 24bits so 32 bits can mean two things 24 bits in a 32 bit integer container with the lower 8 bits 0 or meaningless, or it could be 24bit integers converted to 32bit floating point. Since the mantissa here is 23bits when normalized 32bit floats don't have much more real resolution than 24bit integers. Another problem with some software is that packed 24bit numbers are on odd word boundaries so if you want to support all processors it is a PITA so they force a conversion instead. When using something like a MiniDSP you always have think about the details, the order of two bi-quad filters can dramatically effect the numerical noise floor. Unfortunately this stuff is beyond a lot of users so canned solutions are what most people want.
 
Last edited:
OP
Ron Texas

Ron Texas

Master Contributor
Joined
Jun 10, 2018
Messages
6,076
Likes
8,911
The choice of the 16 bits and 44 kHz was due to a compromise of what was known at the time and what the technology allowed. If they could change what was decided then they would have risen to 48 kHz. At that time 24-bit was not possible.

Yeah, the old story is the Head of Sony wanted to be able to fit Beethoven's 9th on a single CD so he could listen to it in his car so we wound up with the odd number of 44.1. DVD's and most other cinema are done at 48k, although they use compressed formats.
 

maty

Major Contributor
Joined
Dec 12, 2017
Messages
4,596
Likes
3,160
Location
Tarragona (Spain)
To me is not an old story.

To finish the vinyl story, it is better to stay in the 24-bit to avoid subsequent dithering, which implies a new soft processing of the audio. The two typical are differentiable, I like more the sound with TPDF than RPDF. MBIT+ from iZotope is very good too and the dithering can be customized, I think I could not differentiate vs TPDF.

Curiosity. From a 24/96 source, if you want a MP3 (Lame 320 kbps) -> 16/48. But if OGG Vorbis -q9 -> ?/96. MediaInfo neither does specify! Must be 16-bit the two files.

foobar2000-ogg-vorbis-info.png foobar2000-ogg-mp3-info.png
 
Last edited:

Matias

Master Contributor
Forum Donor
Joined
Jan 1, 2019
Messages
5,029
Likes
10,796
Location
São Paulo, Brazil
Many recent DAC's allow user control over these parameters as this feature may be enabled in AKM and ESS chips. I do recall Bryston makes a DAC with an AKM chip which is permanently set to minimum phase, slow roll off.

Yes, many DACs have these filters for users to choose. Wyred 4 Sound and Matrix Audio too, for example.

But my point is that the DAC filters are too "binary", all or nothing, combinations of fast/slow roll off, linear/minimum phase. The beauty of Resampler-V is choosing intermediate values and seeing how they behave on the graphs.
 
OP
Ron Texas

Ron Texas

Master Contributor
Joined
Jun 10, 2018
Messages
6,076
Likes
8,911
Yes, many DACs have these filters for users to choose. Wyred 4 Sound and Matrix Audio too, for example.

But my point is that the DAC filters are too "binary", all or nothing, combinations of fast/slow roll off, linear/minimum phase. The beauty of Resampler-V is choosing intermediate values and seeing how they behave on the graphs.

I have experimented with resampler-V agree it is highly configurable. It just uses too many CPU cycles on my puny Pentium PC. Archimego uses a Linux implementation which runs on his Raspberry Pi endpoint. It's also fully configurable.
 

Julf

Major Contributor
Forum Donor
Joined
Mar 1, 2016
Messages
3,004
Likes
3,998
Location
Amsterdam, The Netherlands
To finish the vinyl story, it is better to stay in the 24-bit to avoid subsequent dithering, which implies a new soft processing of the audio.

That is a non-issue with vinyl - vinyl already has a noise floor that is so high that it is pretty much self dithered at a less-than-16-bit level.
 

Julf

Major Contributor
Forum Donor
Joined
Mar 1, 2016
Messages
3,004
Likes
3,998
Location
Amsterdam, The Netherlands
Yeah, the old story is the Head of Sony wanted to be able to fit Beethoven's 9th on a single CD so he could listen to it in his car so we wound up with the odd number of 44.1. DVD's and most other cinema are done at 48k, although they use compressed formats.

Actually the 44.1 kHz sample rate was based on the video horizontal line period / rate used in the U-matic tape system. The supposed Beethoven's 9th (as asked for by the wife of Sony's VP Norio Ohga) was used in an argument about the disc size, but the real reason was to not give Philips an advantage by adopting the slightly smaller format that Philips was already using.
 

Julf

Major Contributor
Forum Donor
Joined
Mar 1, 2016
Messages
3,004
Likes
3,998
Location
Amsterdam, The Netherlands
A caution about bit depth, I don't know of any A/D with and ENOB near 24bits so 32 bits can mean two things 24 bits in a 32 bit integer container with the lower 8 bits 0 or meaningless, or it could be 24bit integers converted to 32bit floating point. Since the mantissa here is 23bits when normalized 32bit floats don't have much more real resolution than 24bit integers.

Indeed, the "32 bit" stuff is pure marketing (except inside processors). Yes, the best ENOB (Effective Number of Bits) of current DACs is around 21 bits, and 32 bit floating point only has 24 bit precision.
 

scott wurcer

Major Contributor
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
1,501
Likes
2,821
That is a non-issue with vinyl - vinyl already has a noise floor that is so high that it is pretty much self dithered at a less-than-16-bit level.

Yes at least 6-7 bits of self dither via noise on the very best audiophile disks.
 

maty

Major Contributor
Joined
Dec 12, 2017
Messages
4,596
Likes
3,160
Location
Tarragona (Spain)
https://www.google.com/search?q=tpdf+versus+rpdf+mbit++dithering+audible

-> [Spanish] https://www.hispasonic.com/foros/comparativa-dither/427543

to English: https://translate.google.com/transl...ispasonic.com/foros/comparativa-dither/427543

There are a player with the file with the sound after:

What is heard in the attached file is the following:

1- Without "Dither", that is, the quantization errors to hair.
2- iZotope Ozone 5, MBIT + algorithm, noise shaping "ULTRA", dither amount "NORMAL".
3- PSP Xenon, noiseshape "B".
4- Voxengo Elephant, noise "GAUSS", shaping "EQUAL".
5- FabFilter Pro-L, noise shaping "Optimized".
6- Waves IDR.
7- Waves L2, type 1, shape "NORMAL".
8- Steinberg / Apogee UV22 HR.
9- Sonnox Limiter, type 3.
10- Kjaerhus audio MPL-1 Pro.

and the best of ALL, the file is MP3 320 kbps! And it is not from a vinyl rip I think.
 
Last edited:
  • Like
Reactions: MWC

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
... I invite everyone to choose a 44/16 lossless song of your choice with a loud and clear sounding drum kit, with a consistent snare rhythm. ...
Listen to the 3 new files, paying attention to the snare sound, and see if you can hear the difference. Have fun! :)
My DAC has a switch to select linear vs. minimum phase reconstruction filter. It selects either of 2 filters that are built into the WM8741 chip. With most music I can't hear any difference, much like the previously referenced Keith Howard article about the guys at Stereophile. It's possible that most music just doesn't highlight the difference and we're wasting our time trying to hear differences unless we're listening to tracks having the right kind of sounds.

So it would be useful to list tracks (music or test signals) that highlight the differences in these filters. If you've blind tested different DAC filters, what tracks have enabled you to hear differences?

Across my entire music collection, I've found only a couple of tracks that enable me to hear the difference. One is a high quality recording of castanets playing with a flute quintet. Track 8 of this: http://www.genuin.de/en/04_d.php?k=88
I also made a high fidelity recording of jangling keys in front of my mic, and I can hear the difference between the filters with this too. Happy to post a link if anyone's curious.
 

KSTR

Major Contributor
Joined
Sep 6, 2018
Messages
2,690
Likes
6,013
Location
Berlin, Germany
A caution about bit depth, ...
Within "academic only" relevance, I found that current 24-bit D/S-DACs like the AKM4490 actually DO output all 24bits with great linearity when extreme time-domain synchronuous averaging is used (look this thread for pics) to reduce all uncorrelated analog noise. Therefore I don't think it is impossible to get even more from a D/S-modulator even though this is utterly useless in normal practice, of course.
The confusing thing is that the chip makers never clearly state how many bits a DAC is effectively working with, in terms of resolution. It could actually well be that even the lowly AKM4490 already puts out more than 24bits (noise-removed) resolution when fed with 32-bit data (only few, if any at all, DAC frontends supply 32bit data to the DAC chip proper, via I²S. Does anyone know a DAC that does?).
 
  • Like
Reactions: MWC
Top Bottom