Minimum Phase vs Linear Phase

[danadam]

The extremely common choice of using linear phase filters in DACs

Among my (admittedly cheap) dongles it is not that extremely common:

Linear:

• Fiio K3 (ESS version)
• Fiio JA11 (polarity flipped)
• JCally JM6

Minimum:

• Apple EU
• Apple US
• Apple Lightning
• Dragonfly Red
• JCally JM20
• Samsung dongle

To me a linear filter in a DAC is "better" or "more correct" only in the sense that it preserves the shape of the waveform:

Post in thread 'Is Hi-Res Audio better because filter is outside audible range?'

Apr 29, 2025

My point was more that if 44.1kHz was adequate, we would just use the one and only correct solution, and we wouldn't need a choice of filters at all.

AFAICT we don't need it. Yes, it is offered but only because there is demand. Whether the demand is for rational reasons is arguable.

As to filter correctness, here's how I see it.

When capturing the music, the filters on the ADC side has to be fairly steep in order to avoid aliasing. I don't know if it is common to have selectable linear/minimum phase filter type there, but let's say it is. Here are two 1 kHz square waves "A" and...

• danadam

• 

Audibly there's no difference to me.

 

[danadam]

Unfortunately, I don't. I asked JJ in that thread whether he knows of any published studies that prove his assertions that phase distortion is audible, and he responded by giving me a file to play on my system. It was quite easy to hear. If you are interested, that file is somewhere in the thread. Unfortunately, that thread is rather long so it might be difficult to find.

Do you mean maybe the post with matlab code and then its description:

Post in thread 'Does Phase Distortion/Shift Matter in Audio? (no*)'

Oct 18, 2024

Thank you JJ, but remember who you are talking to. Medical scientist. Not an engineer. High school level understanding of maths. And no Matlab! My abilities are very limited.

Regardless, what I am interested in are studies, not demonstrating to myself if it is audible or not. I already think it is audible. But this is anecdote and not evidence. Nobody is convinced by anecdote, at least good scientists should not be. If I went around ASR loudly proclaiming that Toole is wrong and we can hear phase shift because I heard it, I would rightly be ridiculed because I have no evidence.

To...

• j_j

• 

The WAV files generated by that code are here in the attachments (am.flac.zip and fm.flac.zip).

I don't think that's true, either.

PMA was able successfully ABX DBT a rough square wave with and with out phase distortion:

View attachment 465412 View attachment 465413

In a similar vein I can add sawtooth-like signal (only 4 harmonics in total) and its polarity flipped version (also in attachment). Same frequency magnitude but easily audible difference.

 

[danadam]

Music signals aren't generally isolated square waves.

I'm not sure how common an isolated guitar string pluck is in music but its polarity flipped version is also fairly easy to distinguish, at least in a direct comparison of a short snippet:

Post in thread 'Does Phase Distortion/Shift Matter in Audio? (no*)'

Jun 8, 2021

What? I thought you said it was guitar string or something. I am not interested in test tones. Those are already in literature.

Bass guitar note attached, straight into ADI-2 Pro recorded by yours truly a moment ago.
Attack phase:

Apply a 90deg phase shift between fundamental and the 2nd harmonic will make it an asymmetrical waveform.

After 500ms (still settling):

(similar)
The spectrum here:

Fundamental and 2nd at almost same level, 3rd++ further down.

After 2s (steady state region):

Phase relationsships have settled by now...

• KSTR

• 

 

[Kiko1974]

Quite possibly.

Hopefully your amp is not adding coloration and/or distortion.

If you want that, proper EQ is a better solution. The reason vendors provide multiple filters is to make sure they cater for all customer preferences and beliefs.

The Marantz PM 6007 was the cheapest integrated amp on the Marantz catalogue at the time (I think It was released in 2020, I got mine on Summer 2022, with a good discount, I think It was about to be discontinued), and with the KEF Q550, I think they do a good match.
Despite its "low" power output, 45 Watts @ 8 Ohms, both channels driven simultaneously from 20 to 20000Hz.
I think the Marantz PM 6007 is not short on power to drive the KEF Q550, the combo can sound loud if I want It to, without distortion Sky rocketing.
For me is one of those "bargain products" that Marantz (and other brands) designes and markets from time to time. It has a Big toroidal power supply, high quality Elna condensers, and speaker terminals , two pairs, are made, I think, made of copper with a gold plate, something rarely seen at this price range.
Does It sound transparent? Maybe, maybe not. It sounds like a Marantz integrated amp, that is what I've had since I did my first "serious" HiFi system, except for a few years that I had a Pioneer AVR, an SC LX-76, the second model on the top range, only the SC LX-86 was on top of that.
Its sound wasn't really my cup of tea, It sounded fine with movies, not so fine with music, and It was 2000 € when I got It in early 2012. My inexpensive Marantz PM 6007 sounds better that the 2000 € Pioneer AVR, more revealing, more detailed, transparent and with an overall pleasing sound.

 

[KSTR]

That's different, that's not phase distortion. Again, the FFT would be different.

It fully represents phase distortion, by shifting the phase of the second harmonic vs the fundamental. If you are objecting the 0.5Hz offset to create the continuously increasing phase shift then you can used fixed phase distortion offsets in, say 30 degree steps (thus 12 variations, all having the exact same magnitude spectrum) and listen to those in an A/B always using sets which are 180 degrees apart. There a two sets that stand out in the compare, one is the set where the difference is strongest and another where there is no difference. They will be 90 degree apart. When looking at the waveform (at/near the ear) the set with the strongest perceived difference is the the set with the strongest asymmetry.

 

[KSTR]

Do you believe this represents the height of phase distortion audibility?

I'm not sure I understand the question.
Anyway, the effect I'm describing is just one aspect out of several aspects of phase distortion audibility, and the one that is probably the easiest to demonstrate, and it is reliably detected by almost everyone who tries.

 

[markanini]

I'm not sure I understand the question.
Anyway, the effect I'm describing is just one aspect out of several aspects of phase distortion audibility, and the one that is probably the easiest to demonstrate, and it is reliably detected by almost everyone who tries.

Great, it would count an example of phase distortion audilibility.

I don't think it can it serve as proof of audibile differences discussed between linear phase vs minimum phase filters in a resampler or DACs reconstruction filter. Or audible differences of absolute, syncronized, phase inversion. But it was a compelling example in the context of sound design.

Among my (admittedly cheap) dongles it is not that extremely common:

Linear:

• Fiio K3 (ESS version)
• Fiio JA11 (polarity flipped)
• JCally JM6

Minimum:

• Apple EU
• Apple US
• Apple Lightning
• Dragonfly Red
• JCally JM20
• Samsung dongle

Given the stink some raise to minimum phase implementations, and phase inversions, this should have set a lot of people off if it caused audible differences on such abundant devices.

I don't think it does disprove that minimum phase implementation in DACs are sidelined across the larger history of digital audio. Nevertheless, I enjoyed learning about these examples.

 

[Kiko1974]

Up to what point does High Rez material, like and album remastered at 96 kHz/24 bit or 192 kHz/24 bit IS affected by a Fast or Slow filter?
I own several recordings, all of them well known by me, analogue masters remastered at 96/24 and 192/24. Even at shuch high sample rates, let's say listening to the 192/24 Van Halen (Roth era) Bellman remasters, I can tell if a Fast or Slow or even Super Slow IS used.
With a sampling frequency, that has a theoretical frequency response Up to 96 kHz, very far from the audio band, I notice with my humble SMSL D400 PRO, what kind of filter is used.
I did the test with different recordings, like the aforementioned Van Halen albums, A BD-AUDIO I own Holts' The Planets (on Deutsche Gramophonne), I don't remember the conductor, de BD IS mastered at 192/24, the Jurassic Park soundtrack at 192/24, or Journey Greatest Hits, also at 192/24.
The test was a blind test. The player was a Sony UBP X-8OO M2 that has NO DISPLAY, and set to 0 the screen brightnes on the SMSL D400 PRO.
A good friend of mine was changing both the recordings, and the different digital filters.
I could always tell if a Slow or fast filter was selected.
How IS that possible? The potential side effects of those filters is supposed to be far away the audio band.

 

[KSTR]

Great, it would count an example of phase distortion audilibility.

I don't think it can it serve as proof of audibile differences discussed between linear phase vs minimum phase filters in a resampler or DACs reconstruction filter. Or audible differences of absolute, syncronized, phase inversion. But it was a compelling example in the context of sound design.

Fully agreed, reconstruction filters are a totally different matter. I'm having a incredibly hard time hearing any differences between linear and minimum phase filters with the same magnitude response. Badly designed linear phase filters with significant passband ripple (causing disturbing pre- and post echoes) are one of the few exceptions.

I have to disagree with the absolute polarity thing. It is a subset of phase distortion when it comes how it can deform waveforms. With certain phase offsets between the fundamental and the 2nd harmonic of a note, flipping the polarity gives the same waveform than offsetting the 2nd harmonic by 180 degrees and thus gives the same perceived difference in "fattness" of the note (of an electric bass guitar note's quasi-steady-state section, for example, but also for more transient stuff like kick drums or plucked upright bass).

 

[markanini]

I have to disagree with the absolute polarity thing. It is a subset of phase distortion when it comes how it can deform waveforms. With certain phase offsets between the fundamental and the 2nd harmonic of a note, flipping the polarity gives the same waveform than offsetting the 2nd harmonic by 180 degrees and thus gives the same perceived difference in "fattness" of the note (of an electric bass guitar note's quasi-steady-state section, for example, but also for more transient stuff like kick drums or plucked upright bass).

I'll entertain that is true and ask, how is the "correct" orientation determined at the production stage? What about the other 179 degrees of orientation?

 

[KSTR]

I'll entertain that is true and ask, how is the "correct" orientation determined at the production stage? What about the other 179 degrees of orientation?

???

The inner phase relationships are what they are in a given music signal, and the playback shall reproduce this faithfully.

As for "correct" absolute polarity, this obviously is only defined for a single (or stereo) microphone signal, basically unprocessed. Positive pressure (or pressure) gradient shall give positive voltage excitation.
The moment you mix differing polarity mic signals and/or electric signals there is no correct absolute polarity to speak of.

But at any rate the playback chain shall reproduce the signal in the same polarity than the source signal (but often a polarity switch is welcomed to allow the user to select the polarity that sounds better to him/her, also there are lists of mostly classical recordings which tell which recondings need to be inverted).

This obviously only has a reproducible meaning if the playback system is minimum phase, that is, any speaker crossovers are linear phase, otherwise the output signal has no clearly defined polarity. It is commonly agreed on, however, that the apparant polarity of the low frequency range shall match the source signal, that is, for the lowest way of a multiway speaker.

 

[markanini]

???

Your previous reply Implies that there is an optimal phase orientation of audio signals. We should assume audio works the same at the production and playback side. How do you determine the optimal phase orientation of isolated audio signals such as a bass line and kick drum? If consumer can take care not to reverse the phase at the playback stage for true playback, then equally the creator of the music can take care to orient the phase by the same principle? How would that be done?

Surely these are the types of question that can be answered easily if differences in "fatness" are real, despite looking the same to an FFT spectrogram?

 

[KSTR]

For recorded acoustic instruments the correct polarity is the one the ear would have seen at the position of the microphone, obviously. For purely electrical signals the correct polarity is arbitrarily the one the engineer chose, knowingly or not.
When I was recording bass guitar in the studio with a band we actually tested both polarities of the bass guitar and the kick to find out the positions that sounded better in the mix, the idea being strongly promoted by the engineer. You use the channel polarity switches all the time anyway to find the one that sounds better in the context, even for acoustic signals like multi-mic'd drums, individually per mic. This underlines that during production the technically correct polarity (if even existing) is seldom obeyed to, for good reasons.

Additionally, the correct, or better termed, the intended polarity is also not necessarily the one that the customer might prefer during actual playback.
In my experiments I found with many recordings (mostly rock, pop, some jazz and classical) there is no clear winner, often it goes like this (example): in the intended polarity the bass guitar is somewhat lean but kicks sound OK, and overall soundstage has focussed phantom sources. Inverted the bass guitar starts to shine, kicks are also a bit fatter but soundstage feels a bit worse, with less sharply defined phantom sources. Please remember we are talking nuances for nerd listeners here, not night-and-day differences.

 

[Jazigo]

Hi everyone!

I have tried to follow this debate. I am using a Topping E50 DAC, which have the options for minimum phase and linear phase. I also use the setup mostly for computer games.

Since I am changing these filter on the DAC, I think I am changing the filter for DACs reconstruction filter(?). As I understand it, we don't expect a difference in sound per se when doing this, but we do expect a difference in phase? (for high frequency)

I have tried to hear the difference myself and in most cases I can't tell a difference. A case where (I believe) I can tell a difference is where I mow down crash barriers in a car game. Minimum phase sounds slightly distorted compared to linear phase in this case. This may be bias, but it is extremely hard to capture two identical samples for me to produce a A/B log for this.

What I can easily differentiate in any scenario is the delay between minimum- and linear phase in games. It is a very slight difference, but can be easily spotted when you know what to look for.

 

[welwynnick]

Since I am changing these filter on the DAC, I think I am changing the filter for DACs reconstruction filter(?). As I understand it, we don't expect a difference in sound per se when doing this, but we do expect a difference in phase? (for high frequency)

Yes, you're changing the reconstruction filter.
The audible differences are very small.
If you don't hear a difference, then it doesn't matter what filter you use!

What I can easily differentiate in any scenario is the delay between minimum- and linear phase in games.

Can you explain what you mean by "the delay"?

 

[Jazigo]

Yes, you're changing the reconstruction filter.
The audible differences are very small.
If you don't hear a difference, then it doesn't matter what filter you use!

Can you explain what you mean by "the delay"?

By "the delay" I mean a delay between the video and the audio. Say I am firing a gun, which is a very instantaneous sound with a instantaneous visual feedback (muzzle flash).

 

[welwynnick]

By "the delay" I mean a delay between the video and the audio. Say I am firing a gun, which is a very instantaneous sound with a instantaneous visual feedback (muzzle flash).

Thanks.
It's a complex subject, and there are lots of hybrids, but essentially a linear phase filter is an FIR filter, which adds a small processing time delay.
I didn't expect it would be noticeable though.

 

[Jazigo]

Thanks.
It's a complex subject, and there are lots of hybrids, but essentially a linear phase filter is an FIR filter, which adds a small processing time delay.
I didn't expect it would be noticeable though.

Actually you are correct. It is not noticeable. I wanted to make sure, so I filmed in slow-motion with a cellphone a scenario where I was convinced there was a difference. I could not pass a simple A/B test between minimum- and linear delay.

 

[Keith_W]

Thanks.
It's a complex subject, and there are lots of hybrids, but essentially a linear phase filter is an FIR filter, which adds a small processing time delay.
I didn't expect it would be noticeable though.

Typically, how many taps would we find in those filters?

BTW, gamers are a different breed to us music listeners or casual TV watchers. I get motion sickness from watching my friend's son play. He thinks that 60fps is "not enough" and he talks about keyboard and mouse latency. These kids are superhumans with senses and reflexes that I can not imagine, so I suppose they might be able to detect very small audio latency. As for me ... I play Candy Crush, no split second reflexes required!

 

[welwynnick]

Gaming, like motorbikes, was an obsession that I never even started. I knew I would be terminally sucked in if I did (like my son, cousin, nephew, etc).

Typically, how many taps would we find in those filters?

It's a new area for me, but I think it's typically a few hundred taps.
The stand out DAC is the Chord DAVE, which has 164,00, but that's the exception.
It opened my eyes to the fact that a PC can do reconstruction filtering, rather like you do amplitude and phase room EQ.