[dwk]

This was hotly debated in the speaker world when driver testing first rose to prominence. I sit a bit in the middle. Mathmatically, xnor is correct - that for any given impulse response measurement, it's possible to correct it to a near-ideal impulse.  However, I do think it's critical to remember that nothing has been done to actually alter the mechanics of the system - we haven't added damping to change the mechanical resonance. All we can do is reduce the energy exciting that resonance, so that *over the measurement interval* we net out to the same overall energy as for other parts of the spectrum.
 What this means in my mind is that
- how the resonance shows up in the impulse response/ freq response is going to depend on how long your measurement interval is.
- ultimately as the damping of the resonance goes to zero, the correction filter will tend towards an infinite notch.
 
What this means is that the degree of peak associated with a resonance is going to depend on how long you measure. For a very undamped resonance you'll see a bigger peak in a 20ms window than a 10ms window for example. This is going to alter the correction filter you apply - a deeper notch will show up in the longer filter than the shorter.  For the limit condition consider something like a tuning fork that rings for like a minute after being struck - if we could somehow alter the spectrum of the impulse that excites it, we'd have to completely notch out it's fundamental to the point of injecting zero energy in order to have it not ring.
 
So, for 'relatively benign' resonances that decay 'a bit slower' than the rest of the spectrum, I think xnor's approach should work well. The worse the resonance gets though, the more problematic it will become.

 

[arnaud]

Dwk, what you're referring to is a bit of an extreme case (which simply is a result of undersampling, too short digital filter). The smyth guys can handle it for 8 channels with their realiser (something like 800ms worth of decay per channel as I recall), so digital eq. technology as much evolved it appears. You can do anything you want from a computer if you're porcessing the data once and playing it back offline (i.e, no realtime filtering like the realiser).

Furthermore, it seems like headphones are "usually" well behaved with only few resonances to control with rather large Q. On the other hand, as discussed before, some aren't visible until you look at a CSD, so you'd would need some bravery to tacke these (param. Eq. with small delay)?

Then, as mentioned the corrections above 10kHz with sharp notches and dips may be silly because of the sensitivity to both can and in-ear mic positioning (so are those below 100Hz unless you really know you had a good seal for the test). Just like when trying to measure or interpret a headphone measurement, care must be taken. I wonder if xnor is referring to this when he says raw data (he doesn't answer to me for some reason, so hopefully he answers to you ).

 

[xnor]

^ Yes, that's what I meant. My inverse filter was actually calculated from DC to Fs/2 on the raw data (no smoothing, no limit for boosts/cuts ..). I'm sorry arnaud but my measurements were done quickly and are too noisy for more detailed analysis. Maybe purring can help out with clean data. As I've shown it's not hard and only takes a couple of minutes.

 

[Joe Bloggs]

Quote:

^ Yes, that's what I meant. My inverse filter was actually calculated from DC to Fs/2 on the raw data (no smoothing, no limit for boosts/cuts ..). I'm sorry arnaud but my measurements were done quickly and are too noisy for more detailed analysis. Maybe purring can help out with clean data. As I've shown it's not hard and only takes a couple of minutes.

 
Wouldn't that actually make the CSD 'look' more perfect than if you did smoothing and limited boosts/cuts?
 
Your inverse filter was strictly minimum phase yes?
 
Oh and which phones were the inverse filter calculated for?

 

[arnaud]

I am also curious on headphones fitting the minimum phase assumption (we discussed it before, and xnor shows it appears to be that way). I have an idea to investigate this ...

 

[ZeNmAc]

Quote:

It seems like you can hide the flaw, though. It's much more difficult to perceive a dip, compared to a peak.  It seems as long as you keep the correction within reasonable terms, it should mostly, although not solve, hide a mechanical issue while not having a great change to your sound quality because of how your brain fills in the gaps. 

 
Quote:

I'm sorry to disappoint but the decay is as fast as it can get. A 'artificially' created perfectly minimum phase FIR filter with the same magnitude response shows the same decay. So the FUD's inappropriate here.
 
Also, the slow/fast music sentence doesn't make any sense to me.

 
Not sure we're talking about the same thing.  I quoted from a few pages back when we were talking about using only eq.
 
Let me rephrase what I meant with slow/fast music example.  With hps with poor decay; slow music, a cymbal is struck once and it doesn't sound too terrible.  Fast music, cymbal is struck 300bpm and it sounds like a blurred mess.

 

[Joe Bloggs]

@ZeNmAc:
 
Quote:

It's not perfect but that's not surprising considering I used the raw data:
 

 
This is the CSD xnor obtained by applying "only eq" to the same phones you were complaining about at 4300Hz;  it's so perfect at the main audible frequencies (including 4300Hz) as to be unrecognizable as a waterfall plot; there's ZERO decay time.  This demonstrates that the "decay" you saw in the original CSD was by and large only a direct consequence of its uneven frequency response.
 
I can confirm with my own extensive EQ experience that with the right EQ EQing out ear canal resonances, my $10 Philips SHE3580 gains phenomenal speed, snap and PRAT.  Admittedly not all phones at this price level benefit as much from EQ.

 

[keanex]

Isn't it bad to EQ a frequency over 0dB? You're essentially pushing a driver past it's specifications which could lead to distortion and potentially damage, depending on listening volume, from my understanding.

 

[Joe Bloggs]

Quote:

Isn't it bad to EQ a frequency over 0dB?

 
Yes but where are we doing that?
 
Quote:

You're essentially pushing a driver past it's specifications which could lead to distortion and potentially damage, depending on listening volume, from my understanding.

 
No, EQing past 0dB is essentially pushing the sound file format past its specifications.  Unfortunate combinations of existing sound frequencies in the sound file and frequency boosts could drive the resulting output past the full scale of the output format, which leads to clipping, which then leads to distorted sound.  But it's not because the driver is pushed past its specs, it's because the audio data has been essentially corrupted.  The driver is just faithfully replaying the corrupted data--although too much of this distorted sound played too loud can indeed potentially damage a driver.
 
EQing past 0dB can be avoided by using the preamp setting provided with most EQs to lower the signal before applying EQ boost.

 

[keanex]

This picture of an EQ that you posted not too long ago, if I'm right in assuming this is an EQ: 
 

 
Thanks for the info though.
 
Also I'm considering EQing my Ad2000, what program should I use?

 

[Joe Bloggs]

Oh, that graph is centred for illustration purposes.  When actually using the EQ to listen to music I can lower the whole curve below 0dB but then parts of the curve would go off the chart.
 
As for your EQing, depends on how committed you want to be.  That graph is from Electri-Q which you can download here
http://www.aixcoustic.com/index.php/posihfopit_edition/30/0/
and you can use it directly in foobar2000 (with a VST plugin) or winamp (native support)
 
If you want to go the whole nine yards you could start following my tutorial
www.head-fi.org/t/615417/how-to-equalize-your-headphones-advanced-tutorial-in-progress
 
which links software you need to equalize headphones as well as software that makes it easier to test the frequency response of your phone-ear combination to determine a good EQ.
 

 

[keanex]

Thanks for the links, I'm going to look into them. Ideally I would like to reduce the hump around 4k in my Audio Technica Ad2000 and raise the mid-bass simply a tad. By raising it I mean lowering other frequencies. 

 

[ZeNmAc]

Quote:

@ZeNmAc:
 
 
This is the CSD xnor obtained by applying "only eq" to the same phones you were complaining about at 4300Hz;  it's so perfect at the main audible frequencies (including 4300Hz) as to be unrecognizable as a waterfall plot; there's ZERO decay time.  This demonstrates that the "decay" you saw in the original CSD was by and large only a direct consequence of its uneven frequency response.
 
I can confirm with my own extensive EQ experience that with the right EQ EQing out ear canal resonances, my $10 Philips SHE3580 gains phenomenal speed, snap and PRAT.  Admittedly not all phones at this price level benefit as much from EQ.

 
So you're saying that by getting a perfectly flat frequency response the decay time automatically goes to zero.  Makes sense

.
 
By "only eq" I meant only a basic equalizer (like the one in foobar, itunes etc), no other dsp.
 
Quote:

This was hotly debated in the speaker world when driver testing first rose to prominence. I sit a bit in the middle. Mathmatically, xnor is correct - that for any given impulse response measurement, it's possible to correct it to a near-ideal impulse.  However, I do think it's critical to remember that nothing has been done to actually alter the mechanics of the system - we haven't added damping to change the mechanical resonance. All we can do is reduce the energy exciting that resonance, so that *over the measurement interval* we net out to the same overall energy as for other parts of the spectrum.
 What this means in my mind is that
- how the resonance shows up in the impulse response/ freq response is going to depend on how long your measurement interval is.
- ultimately as the damping of the resonance goes to zero, the correction filter will tend towards an infinite notch.
 
What this means is that the degree of peak associated with a resonance is going to depend on how long you measure. For a very undamped resonance you'll see a bigger peak in a 20ms window than a 10ms window for example. This is going to alter the correction filter you apply - a deeper notch will show up in the longer filter than the shorter.  For the limit condition consider something like a tuning fork that rings for like a minute after being struck - if we could somehow alter the spectrum of the impulse that excites it, we'd have to completely notch out it's fundamental to the point of injecting zero energy in order to have it not ring.
 
So, for 'relatively benign' resonances that decay 'a bit slower' than the rest of the spectrum, I think xnor's approach should work well. The worse the resonance gets though, the more problematic it will become.

 
I'll just quote this.

 

[Reticuli2]

That is clearly a defective pair of 1350s.  You should have Beyer repair/replace them.  Neither mine nor Tyll's have that problem.  I'd also be curious how the 990 Pros stack up to the 990 250 non-pro you tested.  And the Ortofon Headphones.
Quote:

Beyer DT1350

 

Subjective impressions - stream of consciousness

 
Voice comes from left of center. Imaging is kind of wierd. Right channel lacks bass and mids. Oh oh. Possibly another driver mismatch a la T1. 
 
The tonality of the headphone doesn't have anything horrible sounding (in terms of nasty spikes that destroy our hearing), but it just kinda sounds "not good," but it's not bad either. There's definitely some trouble in the treble too - it's uneven. Some dips in FR in the mids and high mids and mid-bass seems slightly recessed. Mid-bass dip is never good. IMO, a little bit of warmth and even bass bump (at least with my measurements) from 100-200Hz is required for headphones to sound right. Think SRH840: it lacks a bump in this region - so it too sounds a little cold and antisceptic - at least for me. "Thin" would definitely not be the word to describe these. "Cold" would be too much. Maybe "cool?"
 

FR

 

 
The channel imbalance thing is odd, I re-calibrated and measured several times to confirm. It's hard to pick up because the higher frequencies are decently matched - compared to the lower frequencies which are totally off. Most spatial information (and the all-important harmonics) is contained in the higher frequencies. The imbalance very obvious with test tones. There's no doubt in my mind that these headphones would sound less "cool" and "off" if the channels were matched properly (that is with the one with the better FR.) It would bring out the overall bass - since bass seems especially missing on the right channel. It's very hard for me (and most others) to detect channel imbalances with headphones anyways when compared to speakers. I know LFF is good at it, probably because he's trained himself to do this.
 

Waterfall Plots

 
CSDs below are very clean other than the spike at ~14k. The noise floor is actually a VERY LOW -40db, so the results are impressive. Other than the tone, it's very clean sounding and is able to reproduce real musical information.
 
As always, note the frequency, time, and amplitude ranges on the axis for these plots before comparing to others

 

 
Finally, what's up with these Beyer QC issues?  I'm sure I would like these headphones if the green (left channel) frequency response line were replicated on the right channel. The right channel response is a little bit too lean for me. But for now, the V-Moda M80 reigns supreme for the supra-aurals - at least for me.

 

[Nevod]

Quote:

Originally Posted by dwk /img/forum/go_quote.gif
All we can do is reduce the energy exciting that resonance, so that *over the measurement interval* we net out to the same overall energy as for other parts of the spectrum.

Yep, so EQing does not improve SNR at the particular frequency, and probably even reduces it, as there is generally less signal energy going into phones.
 
 
Quote:

 (the HRTF is the pressure level at the ear canal entrance normalized by the spl measured without head/torso at that location).

Really? Not at the eardrum? That renders them quite unuseful for portraying real headphone FR, which depends not only on these blocked-canal FR measurements, but also on the acoustic impedance, so it seems from the papers on the headphones' FEC and PDR research. Apparently ear channel resonances are dependend on the radiation impedance and peaks and troughs appearing from impedance mismatch can be +-15dB, which probably is what contributes to unrealisticness of headphone sound and soundstaging problems.