Finding value in headphone measurements

[trl]

The "tilt" of a square wave reflects the low frequency phase shift;

Definitelly, this is why aiming to 50Hz should solve this.

However, the tilt itself shouldn't tell much about headphone's ability to reproduce the low-end, like you said it's caused by a phase shift, but a rather flat line between raise and fall might tell something instead.

 

[Mad_Economist]

Indeed. I also found that when using HATS, even after correction for FR, 'ringing' is shown in plots that doesn't appear to come from the driver but rather seems to be a resonance of the ear canal.

This one's pretty confusing to me. Do you have an example? When I've done input EQs to "notch out" the ear gain (from a crude approximation of DF-HRTF with IIR filters), the effect on all plots has been as you'd expect: "ringing" no more than it would if the ear were never there to begin with.

As a stand-alone measurement they do tell me whether bass is rolled-off, has a boost at certain frequencies, indicates a headphone sounds 'dull' or sharp. To me square waves and needle pulses thus do add some info (time related) when one evaluates a suite of measurements. Or should I say confirms some measurements.

I suppose so, but given how fast we can get an impulse for a proper frequency response these days, do we need another measure that's a proxy for amplitude and phase relationships of an arbitrary progression of frequencies? You can definitely get information from a square wave - it's not "noise" - but what does one get that isn't already accessible from a plot of ex. magnitude and phase vs. frequency?

 

[Mad_Economist]

Definitelly, this is why aiming to 50Hz should solve this.

However, the tilt itself shouldn't tell much about headphone's ability to reproduce the low-end, like you said it's caused by a phase shift, but a rather flat line between raise and fall might tell something instead.

Note that you'll still see a shift at 50hz for a lowpass at 10hz, or even lower:

The flatness of the line between the rise and the fall of the square wave is reflective of the relative intensities of the harmonics - this is a proxy for frequency response. Here's a 50hz wave with a -3dB peak filter on the fundamental:

And with the same filter on its harmonic at 150hz instead:

 

[JohnYang1997]

There will not be excessive ringing if you use minimum phase eq like Equalizer APO. You can basically eq all earphones to ideal square wave. I have done that 2 years ago. I thought it was mind blowing but later on it just only makes sense.

 

[Mad_Economist]

There will not be excessive ringing if you use minimum phase eq like Equalizer APO. You can basically eq all earphones to ideal square wave. I have done that 2 years ago. I thought it was mind blowing but later on it just only makes sense.

Because I am pedantic beyond reason, I'll pointlessly point out that it should rather be "eq all earphones to ideal bandwidth limited square wave"

 

[JohnYang1997]

Because I am pedantic beyond reason, I'll pointlessly point out that it should rather be "eq all earphones to ideal bandwidth limited square wave"

That's why I said basically. Or maybe I should have said "ideal" square wave. There will be variation between each run and there are some characteristics cannot be fully characterized as minimum phase.

 

[Robbo99999]

Note that you'll still see a shift at 50hz for a lowpass at 10hz, or even lower:
View attachment 79254
View attachment 79255
The flatness of the line between the rise and the fall of the square wave is reflective of the relative intensities of the harmonics - this is a proxy for frequency response. Here's a 50hz wave with a -3dB peak filter on the fundamental:
View attachment 79258
And with the same filter on its harmonic at 150hz instead:
View attachment 79259

Just a quick question about High Pass Filters seeing as you mention Phase Shifts. I use and have experimented with different High Pass Filters to remove very low bass with the theory (and to some extent personal listening tests) that back up the idea that it helps clear up the rest of the frequency range in terms of detail perceived in the music (headphones). I found that sharp High Pass Filters resulted in clipping appearing in PEACE, which I worked out with help of other forum users was due to phase shifts creating real changes in the frequency response as peaks in bass would be phase shifted on top of other peaks and therefore creating clipping.....clipping that "shouldn't be there" based on the total EQ curve and the negative preamp I was using. Since finding this out, I now use a very relaxed slow roll off High Pass Filter and it doesn't trip the clipping meter in PEACE, but would even the relaxed High Pass Filter that I'm using be creating Phase Shifts and how detrimental could they be? Here's the filter I'm using in Equaliser APO (and in the graph you can see the High Pass Filter starting to kick in at 20Hz):
Filter: ON HPQ Fc 12 Hz Q 0.75

 

[solderdude]

This one's pretty confusing to me. Do you have an example? When I've done input EQs to "notch out" the ear gain (from a crude approximation of DF-HRTF with IIR filters), the effect on all plots has been as you'd expect: "ringing" no more than it would if the ear were never there to begin with.

Basing this on the most 'known' squarewave plots which are from Tyll and we know he used an incorrect compensation and thus the shown squarewave is also not 'correct'. Not many folks publish squarewaves so don't have much to go on.

Also note that I wrote 'seems' indicating I don't know. I don't know as I have no access to HATS, just going on info I found on the web.
When you happen to have squarewave plots of the HD650 and can post them that would be helpful to change my mind.
Here's mine of the HD650 (440Hz) versus the actual stimulus (in green).

but what does one get that isn't already accessible from a plot of ex. magnitude and phase vs. frequency?

Easier to interpret by less educated people ?

 

[Mad_Economist]

Just a quick question about High Pass Filters seeing as you mention Phase Shifts. I use and have experimented with different High Pass Filters to remove very low bass with the theory (and to some extent personal listening tests) that back up the idea that it helps clear up the rest of the frequency range in terms of detail perceived in the music (headphones). I found that sharp High Pass Filters resulted in clipping appearing in PEACE, which I worked out with help of other forum users was due to phase shifts creating real changes in the frequency response as peaks in bass would be phase shifted on top of other peaks and therefore creating clipping.....clipping that "shouldn't be there" based on the total EQ curve and the negative preamp I was using. Since finding this out, I now use a very relaxed slow roll off High Pass Filter and it doesn't trip the clipping meter in PEACE, but would even the relaxed High Pass Filter that I'm using be creating Phase Shifts and how detrimental could they be? Here's the filter I'm using in Equaliser APO (and in the graph you can see the High Pass Filter starting to kick in at 20Hz):
Filter: ON HPQ Fc 12 Hz Q 0.75
View attachment 79260

Broadly speaking, yes a phase shift can cause clipping if you're really close to digital full scale - you could think of it as an analogue (ha!) to an intersample over insofar as nothing is "per se" telling your DAC to create an illegal value...but that's the actual instructions it's getting in the end. I can knock together some examples in ARTA with soundcard loopbacks it that's particularly useful, although I should probably use a software that lets me overlay square waves, I never thought I'd need that feature.

That aside, high Q high/lowpass filters also "ring" with passband amplitude variations, which may have been your issue in past.

Broadly speaking, the solution is simple - and fixes the intersample over issue as well, if that ever comes up: just don't get so close to digital full scale! A -6dB preamp should leave you with plenty of headroom.

 

[Mad_Economist]

Basing this on the most 'known' squarewave plots which are from Tyll and we know he used an incorrect compensation and thus the shown squarewave is also not 'correct'. Not many folks publish squarewaves so don't have much to go on.

Bear in mind, Tyll didn't EQ his inputs with his compensation, he applied it in his spreadsheet. The continuous waveforms are all "raw".

Also note that I wrote 'seems' indicating I don't know. I don't know as I have no access to HATS, just going on info I found on the web.
When you happen to have squarewave plots of the HD650 and can post them that would be helpful to change my mind.
Here's mine of the HD650 (440Hz) versus the actual stimulus (in green).

If the HD600 will do, I might be able to get that done before bed...

 

[Robbo99999]

Broadly speaking, yes a phase shift can cause clipping if you're really close to digital full scale - you could think of it as an analogue (ha!) to an intersample over insofar as nothing is "per se" telling your DAC to create an illegal value...but that's the actual instructions it's getting in the end. I can knock together some examples in ARTA with soundcard loopbacks it that's particularly useful, although I should probably use a software that lets me overlay square waves, I never thought I'd need that feature.

That aside, high Q high/lowpass filters also "ring" with passband amplitude variations, which may have been your issue in past.

Broadly speaking, the solution is simple - and fixes the intersample over issue as well, if that ever comes up: just don't get so close to digital full scale! A -6dB preamp should leave you with plenty of headroom.

Cool, I've already allowed for intersample overs by reducing Windows Volume by 2dB (but I do see that this is another/additional type of "intersample over"), which I did a few weeks ago after measuring some of my most digitally offensive music in Orban Loudness Meter - pretty much all my tracks are below +2dB Reconstructed Peaks. But I was kinda more asking about whether the relaxed High Pass Filter I'm using would create any negative effects in the music.....essentially creating stuff that's not supposed to be there, due to Phase Shift Effects?

EDIT: decided to transfer my Windows Volume 2dB reduction to an additional -2dB on the Equaliser APO preamp - as I think the 2dB reduction in Windows Volume Slider only counters intersample overs but not "Phase-Shifted peaks", whereas the Equaliser APO negative preamp would cover both phenomena......I don't know if I'm right in this distinction between the difference between changing the negative preamp in Equaliser APO vs lowering the overall Windows Volume using the Windows Volume Slider - it's the same loss in total perceived volume but I think this change covers both phenomena together rather than just one of them.

 

[solderdude]

Bear in mind, Tyll didn't EQ his inputs with his compensation, he applied it in his spreadsheet. The continuous waveforms are all "raw".

I don't believe this for one second. Think about this... a raw waveform of a squarewave of say 300Hz with raw input would show an emphasis of about 20dB at around 3kHz ? That would mean an overshoot of 10x the amplitude of the 300Hz would be seen. It would in no way even resemble a squarewave at all and the plots do look like squarwaves but with excessive ringing in time and amplitude. In his plots the overshoot/ringing however one looks at it is no more than 6dB so must have been the 'compensated' signal. As the compensation is incorrect the waveform will also be incorrect.
My plot also isn't EQ'ed but is corrected. The electrical output of my rig does not need additional 'compensations' it is built in hardware in the pre-amp.

Maybe you meant something else but that's the way my simple engineering/measurement mind sees it. Not from an academic level.

Can you post a plot of the raw output from a HATS where a squarewave was the input via a headphone ? We may be looking at this from too different angles. I would like to see a 'raw' squarewave from a HATS.

 

[Chocomel]

I don't believe this for one second. Think about this... a raw waveform of a squarewave of say 300Hz with raw input would show an emphasis of about 20dB at around 3kHz ? That would mean an overshoot of 10x the amplitude of the 300Hz would be seen. It would in no way even resemble a squarewave at all and the plots do look like squarwaves but with excessive ringing in time and amplitude. In his plots the overshoot/ringing however one looks at it is no more than 6dB so must have been the 'compensated' signal. As the compensation is incorrect the waveform will also be incorrect.
My plot also isn't EQ'ed but is corrected. The electrical output of my rig does not need additional 'compensations' it is built in hardware in the pre-amp.

Maybe you meant something else but that's the way my simple engineering/measurement mind sees it. Not from an academic level.

Can you post a plot of the raw output from a HATS where a squarewave was the input via a headphone ? We may be looking at this from too different angles. I would like to see a 'raw' squarewave from a HATS.

Idk about the exact decibel levels but the higher the harmonic in a square wave the lower it's amplitude.

 

[Mad_Economist]

Cool, I've already allowed for intersample overs by reducing Windows Volume by 2dB (but I do see that this is another/additional type of "intersample over"), which I did a few weeks ago after measuring some of my most digitally offensive music in Orban Loudness Meter - pretty much all my tracks are below +2dB Reconstructed Peaks. But I was kinda more asking about whether the relaxed High Pass Filter I'm using would create any negative effects in the music.....essentially creating stuff that's not supposed to be there, due to Phase Shift Effects?

EDIT: decided to transfer my Windows Volume 2dB reduction to an additional -2dB on the Equaliser APO preamp - as I think the 2dB reduction in Windows Volume Slider only counters intersample overs but not "Phase-Shifted peaks", whereas the Equaliser APO negative preamp would cover both phenomena......I don't know if I'm right in this distinction between the difference between changing the negative preamp in Equaliser APO vs lowering the overall Windows Volume using the Windows Volume Slider - it's the same loss in total perceived volume but I think this change covers the bases more.

In all cases, it's about the peak value being asked of your DAC being >0dBFS/max - doesn't really matter much where you do the subtraction, but I like all my preamps in one place. I'd suggest higher the -2, as said.

There's data on audibility thresholds for phase shift at LF out there somewhere, and if I were really motivated I'd go digging, but I'll say with confidence that the primary audible impact of your highpassing the subbass will be...highpassed subbass, which is apparently an effect you prefer subjectively, so keep at it.

 

[solderdude]

Idk about the exact decibel levels but the higher the harmonic in a square wave the lower it's amplitude.

I see the point.
I can easily test this myself and see the effect.

 

[Robbo99999]

In all cases, it's about the peak value being asked of your DAC being >0dBFS/max - doesn't really matter much where you do the subtraction, but I like all my preamps in one place. I'd suggest higher the -2, as said.

There's data on audibility thresholds for phase shift at LF out there somewhere, and if I were really motivated I'd go digging, but I'll say with confidence that the primary audible impact of your highpassing the subbass will be...highpassed subbass, which is apparently an effect you prefer subjectively, so keep at it.

Yep, thanks, I see your points, and I'll go do the digging if I want to find out that aspect.......good to know that the largest audible impact will actually be the high passing the subbass, so yeah makes sense to keep it.

 

[Mad_Economist]

I don't believe this for one second. Think about this... a raw waveform of a squarewave of say 300Hz with raw input would show an emphasis of about 20dB at around 3kHz ? That would mean an overshoot of 10x the amplitude of the 300Hz would be seen. It would in no way even resemble a squarewave at all and the plots do look like squarwaves but with excessive ringing in time and amplitude. In his plots the overshoot/ringing however one looks at it is no more than 6dB so must have been the 'compensated' signal. As the compensation is incorrect the waveform will also be incorrect.
My plot also isn't EQ'ed but is corrected. The electrical output of my rig does not need additional 'compensations' it is built in hardware in the pre-amp.

Maybe you meant something else but that's the way my simple engineering/measurement mind sees it. Not from an academic level.

Can you post a plot of the raw output from a HATS where a squarewave was the input via a headphone ? We may be looking at this from too different angles. I would like to see a 'raw' squarewave from a HATS.

The raw response of the HD600 at 3khz on Tyll's plots was roundabouts of 12-14dB above 300hz (-14~ to -26~):

The relative amplitude of the peak value of his 300hz square wave and the "shelf" of the real square is about .02 to .005 (12dB).

This doesn't seem in strong contrast to me, but perhaps I'm missing something.

Regardless, here's my HD600 in its raw form and with an EQ viewed in frequency response:

And in square waves:

My apologies for the lack of time alignment - I don't deal with ARTA's square wave outputs usually, but they appear to be from its actual time record, so differences in delay due to the EQ being on or off would need to be compensated either with a delaying filter on the "no EQ" measurement or by hand, and it is late and I am lazy. Ehhh it would have bugged me too much. Fixed.

Edit: And now in smell-o-vision zoomed in scale:

Edit 2: I didn't record this, but much as with multitones, the FFT of a square wave gives a (very limited) approximation of frequency response, in cases where its variation in the band of early harmonics is so extreme that it is more significant than the typical falling amplitude as a function of order.

 

[solderdude]

Yes, convinced me. Was about to simulate myself but don't have to.
In essence, it makes the squarewave response of Tylls measurements worthless.
I guess it would have been easy to run the results through some math and obtain similar results as mine.

 

[trl]

What would be the best way of measuring how fast a driver is? I'm also referring to the sounds the driver does when raising and falling too (would these be added harmonics that will reflect in the final THD?). Thank you!

 

[solderdude]

I noticed that almost regardless whether or not I use a headphone that has a very wide frequency range or one that just reaches say 15khz my mic. always shows about the same risetime when using a squarewave. It seems to be limited to an equivalent of 16kHz bandwidth, roughly.
I am sure there is an explanation for this (mad economist will probably know).

The risetime of a plucked bass note will probably not be much faster than 2kHz or so (highest meaningful harmonics).
The membranes thus have no problem accelerating.

I guess it has more to do with frequency response. SBAF showed quite different response between headphones with some long bursts. with some imagination this had good correlation with how I perceived the bass response of the same headphones I know but also with frequency response.

When I EQ headphones to my own target I often find bass sounds quite similar in 'speed' and 'impact' so I never took a plunge in experimenting further. It seems to be FR related and perhaps experience with how measured response is vs perceived response.