CHANGSTAR: Audiophile Headphone Reviews and Early 90s Style BBS
Lobby => Headphone Measurements => Topic started by: ihasmario on August 09, 2012, 09:18:35 AM
-
I'd like to make several requests for measurement types.
1. I'd like the inclusion of RAW output if you are able to do so (I do not currently have knowledge of your measurement methods).
2. I'd like to see a scientific discussion of the validity of speaker-centric measurements for headphones, particularly those that are modified by treble
To clarify number 2, I'd like to see square wave responses looked at in a once off (if you are capable of doing this). Firstly, I'd like to see them produced traditionally. Secondly, I'd like to see the same square wave reproduced with an EQ placed that is the inverse of the raw response of the headphone (or the HRTF you are/probably are subtracting for the compensated response). If possible I'd prefer actual measurements rather than a transform from an impulse response, or whatever.
I say this because, it is my direct understanding that square waves are affected by treble; scooped/sloped by an excess of treble, and made rounder/sinusoidal with a lack of it. As there is a trend for darker headphones to produce more accurate square waves on most sites, I'd like to see the results.
I'd also like to see more explanations of impulse responses. Although I've only looked at a few measurements, I've seen one (possibly two) of them. For example, ringing that is a result of excess treble is easily identfiiable.
(I'd actually like to see the raw vs compensated measurement for impulse response as well, just to tickle a curiosity)
-
Will try to include more measurements. I don't do this for a living, so more visualizations means more time away from family and work, which are of course more important to me. I've got a quick-and-dirty workflow down, so it's hard to make changes. The other reason I don't make post SQs and IRs is that Tyll already does so.
As far as square waves and treble, why not look at the FR graph or CSD to see the effect of treble? Those visualizations are much better are discerning the quality of treble than a SQ wave at X frequency. Although I assume you may be interested because it's another measurement technique and want to see SQ data corroborate with other measurements.
-
I'll try to explain my curiosity as best I can; currently working on no sleep because of university.
Assumption 1: Ideal for speakers is to reproduce real life
Assumption 2: The ideal square wave, in real life, though impossible is a perfect square wave
Conclusion 1: Therefore the ideal speaker should produce as close to a perfect square wave (and therefore impulse response??)
Assumption 3: The goal of headphones is to sound like real life, and therefore to produce the same square wave as a speaker AT the ear.
I would say this conclusion is correct, and it shows in our speaker and headphone measurement standards: speakers are measured with a flat microphone (or as close to flat as possible), and plotted more or less as-is. Headphones are designed in a manner that tries to reproduce the same sound at the ear as a flat speaker.
The problem is that this makes headphones a non-flat speaker in practise. Therefore, you will have distortions to the square wave and impulse response, no matter how perfect other aspects of the design are. Of the top of my head the problems are as follows;
1. Treble deficiency causes square waves to become rounded (more sinusoidal).
-> The reason for this is that a perfect square wave is a stack of an infinite number of odd harmonic sine waves.
2. Treble excess causes a "downwards slope" and ringing in the square wave
3. Treble excess causes ringing (and overshoot?) in the impulse response.
Now, as is my understanding; you can create the squarewave response and the frequency response from the impulse response. Meaning that, with the fourier transforms, they have a strong correlation.
I noticed, in my many hours of analysing headphone graphs a trend in headphones; they all slope downwards, something you'd expect for excess treble.
Now, someone might pipe up and say that it's because headphones are typically bad at producing bass, because of their small surface area. And that's fine. Allow me to post some headroom graphs (Feel free to remove them if you don't want them posted here). Please feel free to paruse them.
Now, based on these graphs: it is my contention that this method headphone analysis is FUNDAMENTALLY WRONG. A bassy speaker should produce a rounded square wave, and a trebly speaker shoudl produce a "sloped" (and ringy) square wave. I feel I've demonstrated the strong correlation between the two.
So, why then, do bassier headphones produce nicer looking square waves than ones that look flat? The answer I have come up with is Equalisation. I'll repost that moller graph from another thread.
Look at the curves (specifically the diffuse field for headphones, because Tyll uses a ID equalisation for his HRTF). Unless I am blind, the difference between the treble spike, and the bass is 105-96=~ About 9dB.
Now, if you'd kindly look at the graph with the LCD-2 (headphonemeasurements3.png). The square waves look pretty good, don't they? I'd like you to work out the difference between the bass and the treble. To my eyes, it looks like the LCD-2's bass curve is centred around about +4 dB, while the higher treble (2.5khz) is what,about -3dB. A 7dB difference, close to the moller curve. There is still a little more treble than the moller curve, which is why the square wave still has some ringing and downward slope. Notice that every headphone in that graph has a similar square wave. It's not a unique property of amazing LCD-2 technology In fact, it looks like the LCD-2 doesn't qutie have the best square wave response there. Why? Because it has the least bass, and therefore, it is further away from the Moller graphs than one of the others. Note that some go over 9dB difference, and there is some rounding in those, instead of the slope.
I hope you can see my reasoning, and are interested in this too.
Let me be perfectly clear: It is my contention that people have been mislead by dodgy measurement practise that the LCD-2 and other headphones produce "good bass" based on the square wave, when what it actually produces, at the ear is a slightly rounded square wave compared to a square wave produced by a flat speaker.
That is not to say that the square wave responses we receive are wrong. I think they're pretty close to correct, given the consistency. What I do believe is that Tyll, and others are mixing methods; for frequency response they present headphones as "how they sound compared to a speaker in the field we recorded in", for square wave, I believe they're presenting "This is just what it puts out". If theyre going to do one or the either, they should do it for ALL measurements, and not one. I'd personally prefer RAW responses for all of them because it just makes more sense to me, as I can identify which headphone is designed to which method/standard/field more easily.
I'd really love someone to try and publish the square waves produced by headphones, with their frequency response compensated (via EQ, to save changing the methods of measure).
But to be honest, I haven't really though seriously about this in a long time, so feel free to chime in with corrections/thoughts. I could be completely wrong, and would like to be proven so (to avoid the upset of others). But honestly, I expect that, correctly compensated, we will see that Neutral (i.e. diffuse field, free field etc) headphones produce exactly the same square wave pattern as a speaker that is flat in the same listening conditions (diffuse field, free field). I hope someone does try it, the vindication of being correct, or the humilation of being wrong. I'd still like to see it. Of course true vindication won't be possible unless someone with a flat speaker set (I believe you have nearfield monitors, they're close enough) can demonstrate that they can reproduce, say the square/impulse response (or close enough to it), on their speaker by modifying the frequency balance to be equal to the raw response of a particular headphone.
If you ask me, we are on the cusp of a revolution. Maybe not of the entire measurement practice, but of people at least thinking before they trust anyones graph.
popcorn
-
It is my contention that people have been mislead by dodgy measurement practise that the LCD-2 and other headphones produce "good bass" based on the square wave, when what it actually produces, at the ear is a slightly rounded square wave compared to a square wave produced by a flat speaker.
Hah. Funny you mention this because we have discussed square waves as a potential BS measurement when trying to correlate what we were hearing w/ present data available. I think you have said a lot of what purrin has been thinking on the matter but hasn't really wanted to go there yet due to the amount effort required to swim upstream on the matter. The amount of times I have been able to trip up an LCD's bass response w/ my test tracks simply didn't correlate to the perfection I've seen rendered in various data sets.
-
Funnily enough, I started investigating when I heard the LCd-2s round, heavy sound.
Nice to hear at first (after the wtf no treble phase) but I wouldn't (and didn't) buy it
-
Square waves are the sum of odd order harmonics with less and less influence for each successive harmonic. By the 9th harmonic, the influence is minimal. The reason LCD2/3 have nice square waves at 50Hz is because it measures flat from 50Hz to 450Hz (9x50) and even beyond to 1kHz or more.
Mathematically, you can calculate with reasonable accuracy the shape of a square wave from frequency response, but not necessarily the other way!
For example, a 50Hz square wave will tell us what is going on at 150Hz, 250Hz(somewhat), 350Hz (a little bit), and 450Hz (very little). However it will tell us jack shit at 100Hz, 200Hz, 300Hz, because those signals are not even part of a square wave. Although it should be noted that we will be able to see distortions and ringing.
In regards to the LCD2, people have different opinions of what good bass means - good square wave and all. And even then, at what frequencies do we examine the square wave? 20Hz, 50Hz, 75Hz, 200Hz?
I'll explore the SQ thing if I have time (which means next winter at earliest or never), but honestly we should be using the right visualizations in the first place to correlate measurements with sound: The trifecta of 1) FR; 2) CSD; 3) Distortion (of which there are many ways to visualize this.)
BTW. I didn't not invent this "trifecta" shit either. JA of Stereophile has been doing two of them (FR+CSD) for quite some time now. Zaph (www.zaphaudio.com) has been doing all three. And Mark K (http://www.audioheuristics.org/ (http://www.audioheuristics.org/)) has concentrated a lot of non-linear distortion measurements.
Square waves or step responses are useful for me as a speaker builder for one thing: time aligning drivers / crossovers.
But yeah, the whole LCD2 thing was what inspired to me present other sets of data. Too many people masturbating to those LCD2 measurements. I'll see if I can dig up a square wave or two for other headphones.
-
I like to play a game with Tyll's measurements where I try to guess what the square waves will look like from the slope changes in the FR. The resulting amusement justifies their inclusion.
-
The reason LCD2/3 have nice square waves at 50Hz is because it measures flat from 50Hz to 450Hz (9x50) and even beyond to 1kHz or more.
Could you please look at the HE6 graph?
The square wave is measured, heavily sloped as though there is an excess of treble, when the bass from 50 to even the 10th harmonic is qutie flat - in fact, there is more bass, and less of the other frequencies.
When taken 50-20k, the only thing I can see that the LCD-2 has that the HE6 doesn't is a roll-off of about 10db in the treble If what you say is correct (it definitely seems correct), then is it safe to say that these measurements are just wrong?
-
Interesting. Just looked at LCD2 and LCD3 IF square waves graphs, and LCD2 actually has a better square wave than LCD3! LOL, I think this is why I never gave SWs that much thought nor cared that much about them.
I've got HE400s on hand which have similar SW measurement attributes on Tyll's data sheet to the HE6. Will take a quick look with HE400. BRB.
-
What is not shown at IF is the FR phase of the headphone which can mishappen a square wave. I had an exchange with xnor at HF that got me motivated to write some code:
http://www.head-fi.org/t/612665/how-far-can-eq-really-go-towards-truly-equalizing-headphones/75#post_8455637 (http://www.head-fi.org/t/612665/how-far-can-eq-really-go-towards-truly-equalizing-headphones/75#post_8455637)
At the bottom of the code there are a set of plots related to 3 different filters: A 5th order Butterworth IIR, a 5th order Chevyshev, and a windowed-based (Hamming) linear phase FIR. Shown are the FR magnitude response of the 3 filters which are flat from 200Hz and above. Only the linear phase filter lets the 300Hz square wave through virtually untouched. The rest look rolled off.
The square wave high frequencies are not absent, they are just linearly re-combined with it's low frequencies with a modified phase due to the filter FR phase.
EDIT: I also agree with Purrin regarding the questionable value of the square wave response. Square waves have no frequency components bellow their fundamental and only contain odd harmonics with attenuated gain (as a function of frequency.) Compound the problem by the phase response signature imparted by the headphone... Difficult to read. The impulse response and it's frequency domain dual, the frequency response are much more valuable IME.
-
Yup - changes in phase can really result in some screwy square waves. This is why square waves (or step responses) are more suited for crossover design where filters will have a serious influence on said square wave.
I did make a bad assumption: while each successive harmonic does have less influence, the decibel scale is logarithmic, so the influence of the higher harmonics is much more significant than I thought.
For our viewing and masturbatory pleasure: HE400 square wave @ 30Hz with spectrum plot attached.
-
O M G. Look at the curves on that square wave!
(http://2.bp.blogspot.com/_Ne5Lb2SiFHg/Rqi0mAP7OmI/AAAAAAAAFCE/MrJm8moWBRM/s320/homer+hottie.jpg)
-
10/10 post ultrabike. Could you link me to some literature to read up on phase, as it's not something I have looked at. Could you explain how the Hamming creates the ideal square wave pictured? Because I don't understand that part at the moment (Obviously I need to learn more).
I'm liking how this thread is shaping up. Thanks for the post, purrin!
I've heard all manner of absurd claims with square wave, including "showing you how fast the headphone is" )( )(
By the way, returning to the tile subject; how do we read inverted impulse responses? I've always wondered. Is it read exactly the same way? For example, if you had a dipole speaker, would the impulse response be inverted if you stood behind the speaker? Or is it simply an issue of polarity?
This begs the question further: Does it matter for dipole speakers (for example, a lot of stax phones seem to have an inverted impulse response)
-
Could you explain how the Hamming creates the ideal square wave pictured? Because I don't understand that part at the moment (Obviously I need to learn more).
You and I both :)p17. As far as I know, the requirement for a nice looking SW response are: filter/headphone/linear-system FR flat magnitude and linear phase across the SW frequencies. There are many FIR filters that satisfy those requirements (Like that Hamming FIR), but that is not what headphones necessarily are. Like Purrin said, your ears, FR, CSD, and Distortion plots will likely better serve you...
-
By the way, returning to the tile subject; how do we read inverted impulse responses? I've always wondered. Is it read exactly the same way? For example, if you had a dipole speaker, would the impulse response be inverted if you stood behind the speaker? Or is it simply an issue of polarity?
This begs the question further: Does it matter for dipole speakers (for example, a lot of stax phones seem to have an inverted impulse response)
We just flip the inverted response over - it's just polarity - switching the +/- on the wires will do. It doesn't matter. Some amps will invert polarity.
-
I did make a bad assumption: while each successive harmonic does have less influence, the decibel scale is logarithmic, so the influence of the higher harmonics is much more significant than I thought.
Care to comment/demonstrate how much when you have the chance? Ty
-
The full spectrum of 30Hz SW shows that even past the 9th harmonic, the higher frequencies still have a good amount of influence on the shape of the square wave. (I thought the higher harmonics would just die, but I mistakenly did not account for the log nature of our hearing.)
Analysis of several (more than one) square waves to determine FR is not necessary wrong since as I mentioned before, a SW can be obtained from FR (but not the other way around). A much better way to get a grasp of a headphone's tonal characteristics (or frequency response) would be by looking at the FR graph instead of a set of two or three square waves. Don't ya think? ;D
P.S. One minor issue which I forgot to mention is that if the FR of a headphone has a narrow bump at an even harmonic (which does not extend to the odd harmonics) of a square wave, the square wave won't catch it. This is very unlikely though.
-
I've been meaning for some time to extract IR, SW, FR, CSD from idealized headphones to illustrate effect of roll off at both ends of the spectrum, presence of resonances and what not. The main objective was to highlight that SW don't bring much to the table and probably best left to where they come from (checking bandwidth / phase of electronic components).
I agree that SW on a headphone measurement takes some expertise to interpret because I believe it is an uncompensated result. Tyll's ID correction is magnitude as function of frequency only, you can't apply it to a time domain response (IR, step or SW) without creating an artificial filter under some assumption, minimum phase for example even though I can't imagine an HRTF is minimum phase. Anyhow phase itself takes an interesting meaning when the equalization curve is not 1 direction (FF) but an average of several (DF or Tyll's ID).
Personally, I can't interpret a SW although I can imagine how some basic headphone traights will make it deviate from the ideal (which isn't a prefect swuare as discussed above). In particular:
- HF roll off should be visible as rounded off SW (provided the SW fundamental is high enough, say 1kHz)
- Low Q resonances with cause large overshoot and ringing at every corners of the SW (provided the SW harmonic content is such that it meshes reasonably well with the headphone ringing frequency(ies)
- Large phase variation across the frequency range of the SW must indeed cause interesting changes, the possibilities are endless here
As you can see, there are too many provisions in the usage of the SW data to make it reliable. It's a crap shoot at best and one is better off looking at FR, CSD, IR.
Im(not so humble)o...
-
What is not shown at IF is the FR phase of the headphone which can mishappen a square wave. I had an exchange with xnor at HF that got me motivated to write some code:
http://www.head-fi.org/t/612665/how-far-can-eq-really-go-towards-truly-equalizing-headphones/75#post_8455637 (http://www.head-fi.org/t/612665/how-far-can-eq-really-go-towards-truly-equalizing-headphones/75#post_8455637)
At the bottom of the code there are a set of plots related to 3 different filters: A 5th order Butterworth IIR, a 5th order Chevyshev, and a windowed-based (Hamming) linear phase FIR. Shown are the FR magnitude response of the 3 filters which are flat from 200Hz and above. Only the linear phase filter lets the 300Hz square wave through virtually untouched. The rest look rolled off.
The square wave high frequencies are not absent, they are just linearly re-combined with it's low frequencies with a modified phase due to the filter FR phase.
.
Ultrabike, nice, this is part of what I meant to do, but never did (what a lazy b. i am :) )
-
Hey dudes! Nice yabber. Gonna have to look around a bit.
-
Hey, welcome!! Finally someone not talking out of his behind on this board (kidding, this really only applies to me).
Saw your article on SW soon after I posted and want to congratulate you on a neat article! Glad you found the time to write in spite of the limited audience. I had a few other examples in mind when we discussed it back then (you probably forgot since i've been slacking so bad), but I like the illustrations you made!
-
I could dig a bit more around this and here are some preliminary results of SQW and Impulse Response of the Tyll's equalization curves (assuming minimum phase characteristics).
Tyll, in case you're ok with it, I can also plot the target curves and fitted filters, let me know. :)p1
Note: DF for diffuse field, ID for independent of direction
Square wave 30Hz:
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/SQW30DF.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/SQW30ID.jpg)
Square wave 300Hz:
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/SQW300DF.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/SQW300ID.jpg)
Impulse response:
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/IRDF.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/IRID.jpg)
-
I thing I just realized: these should be inverted as these curves are what should be subtracted from the raw data.
-
^^^Geek. I bet you enjoy doing it, even when you do it wrong.
-
Went to bed at 2AM to do this while the kids are asleep, I guess that qualifies as geek borderline insane.
And in French, we say the important is to participate rather than win, so yeah, I am definitely enjoying it!
Let me know if I can show the eq. curve or not, extracting the inverse response is straightforward at this point actually.
-
Let me know if I can show the eq. curve or not, extracting the inverse response is straightforward at this point actually.
I don't see why not. It is specific to my particular head though.
Oh, and I would love to see the square wave post filter. TIA, mate.
-
Tyll, there you go :)p12:
Overview:
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_Overview.jpg)
Curve fitting results:
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_Fitted_DF.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_Fitted_ID.jpg)
Reconstructed filters (FIR, minimum phase):
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_FIRFilters_DF.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_FIRFilters_ID.jpg)
30Hz Square Wave response:
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_SQW30Hz_DF.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_SQW30Hz_ID.jpg)
300Hz Square Wave response:
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_SQW300Hz_DF.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_SQW300Hz_ID.jpg)
Impulse response:
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_ImpulseResponse_DF.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Compensation%20curves/EqCurves_ImpulseResponse_ID.jpg)
-
Thanks mang.
Always knew those curves weren't really right for headphones. Gonna have to work on that.
-
Next step is actually to attempt to extract step response with the pinnae effect equalized out (using the same minimum phase hypothesis). Not sure how this will turn out but worth a shot...
-
Me thinks there might be a plan afoot, matey.
:)p1
Perhaps we should jump in the dingy and chat?
:)p8
-
Sure captain!
Let me give it a shot tonight with say, the 009 (it'd be nice to explain the 300Hz SW response...). I'll check how much the raw impulse response deviates from minimum phase at the same time since this has been puzzling me for a while (what a strange hobby indeed)...
-
Hoy! What do the D'x' values in harmonic distortion measurements represent exactly? Steps in fed signal as in different orders?
-
D2 = 2nd harmonic
D3 = 3rd harmonic
D4 = 4th harmonic
-
... Thanks :)
Without sounding more stupid, where I can read more into this?
I understand this kind of measurement: http://www.head-fi.org/t/595683/fostex-th900-impressions-discussion-thread/1350#post_8550368
How do the harmonics compare to this?
Since the whole frequency range is covered, does it mean that you use four different signals to excite the driver?
Edit: Nevermind. Found this thread: http://www.changstar.com/index.php/topic,370.0.html
Thanks!
-
Edit: updated links
So, I went through with the idea to remove the measuring head/torso effect from all of Tyll's data (i.e. not just the FRF magnitude but also impulse and square wave responses) using minimum phase inverse filter of the Diffuse Field and Independent of Direction HRTFs.
It seems it went somewhere after all... From what I can see, the SR009 seems to have been designed to follow DF target curves. Furthermore, once you remove the head effect from the impulse and square wave responses, you end up with much more reasonable looking results. For instance, the overshoot in the 300Hz SW response is no longer present after equalizing out DF response curve of the measuring head...
And now some pictures, because we like pretty pictures on this site :)p1 :)p1 :)p1:
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_SR009_FRF_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_SR009_FRF_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_SR009_SW30_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_SR009_SW30_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_SR009_SW300_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_SR009_SW300_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_SR009_IR_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_SR009_IR_R.jpg)
-
I like it.
-
YES. This is exactly what I've been waiting for. Did you construct the inverse filters from Tyll's Diffuse Field Equalization Curve and ID Field Equalization Curve?
-
Very nice!
-
Great stuff, Arnaud!
Would love to see graphs of more headphones if possible.
-
Great stuff Arnaud.
I'm wondering if we could work backwards to get a better HRTF compensation. For example, let's just assume the HD800 is perfect. Then use it's raw data to create an HRTF. Then have a look at it's square waves to see how clean they are. Thank run a, let's say, LCD3 through that HRTF and see how it comes out.
Maybe it's possible to take three or four headphones we think are very, very good; make an average of their raw FR; then create an HRTF from that. I don't know...I'd really like to have something better than the stock ID or DF HRTF, but I really don't know how to get there.
-
Thanks maties! :)p3
Yes, the inverse filter was constructed from the narrow band DF and ID equalization curves Tyll has for his measuring head, see the filters response vs target in previous post from august.
Anything could potentially be used as reference basis (1 or more headphones average) and I can give it a shot. In any case, it should be an average of a few positions just like your compensated response on Inner Fidelity graphs is. I can try with HD800, although the SR009 is the more neutral of the two in my biased opinion :)p17
Also, in case you're not fully satisfied with your current eq. Curves Tyll, the suggestion I have is to simply record your own HRTF using suitable headings / headings average. You really only need a full range speaker and typical omni mic and perform the measurement in anechoic conditions (for example outside with thick foam pads on the floor).
It's then a matter of recording the speaker response at using your dummy head at target headings and the speaker response at the center of the 2 ears (dummy head is removed) such that speaker response can be excluded from the HRTF. Strictly speaking the HRTF is the L/R ears response divided by the center mic response for a given heading.
Only reason I am suggesting a proper HRTF is that it's difficult to rule any one headphone as the reference anything else should be rated against. i feel the HD800 is excellent, but has its issues with peaks in upper mids. Maybe an average of 3-4 reference headphones could get us somewhere indeed. Let me know the candidate list and I will give it a shot...
-
Same processing applied to the HD800, it measures well indeed...
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_HD800_FRF_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_HD800_FRF_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_HD800_SW30_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_HD800_SW30_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_HD800_SW300_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_HD800_SW300_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_HD800_IR_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_HD800_IR_R.jpg)
-
Edit: corrected graphs, I was plotting the HD800 again...
Well, the LCD2r2 ain't too shabby either:
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD2r2_FRF_L2.jpg)(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD2r2_FRF_R2.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD2r2_SW30_L2.jpg)(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD2r2_SW30_R2.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD2r2_SW300_L2.jpg)(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD2r2_SW300_R2.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD2r2_IR_L2.jpg)(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD2r2_IR_R2.jpg)
-
I am stopping here with one of the LCD3 measurements because it's gonna get boring. Now, I ask you, which is to be the reference there?
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD3sn2312260_FRF_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD3sn2312260_FRF_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD3sn2312260_SW30_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD3sn2312260_SW30_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD3sn2312260_SW300_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD3sn2312260_SW300_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD3sn2312260_IR_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_LCD3sn2312260_IR_R.jpg)
-
I am stopping here with one of the LCD3 measurements because it's gonna get boring. Now, I ask you, which is to be the reference there?
SR009 = Win
-
Yes, the inverse filter was constructed from the narrow band DF and ID equalization curves Tyll has for his measuring head, see the filters response vs target in previous post from august.
Ah, I missed that part. More readings to do now. :) Do you have any suggested reading materials on constructing a minimum phase FIR filter? I'm still trying to wrap my head around that topic...
-
Ah, I missed that part. More readings to do now. :) Do you have any suggested reading materials on constructing a minimum phase FIR filter? I'm still trying to wrap my head around that topic...
Hi ujamerstand,
Using a math oriented programming language (I used Octave), here are the steps:
1. Polynomial fit of the target curve (magnitude response)
2. Create a linear phase FIR filter (symmetric in the time domain so has that weird preringing) use least square fit of sorm sort
3. Extract a minimum phase version of the filter (using cepstrum analysis)
These are typically single call functions in a math package so fairly accessible to anyone motivated, esp. if you've done a little bit of matlab programming at school.
cheers, arnaud
-
Yes! That is all that I need to start hacking! If only I could applaud you a thousand times. Un grand Merci!
-
Yes! That is all that I need to start hacking! If only I could applaud you a thousand times. Un grand Merci!
Cool, looking forward to what you come up with!
-
I am stopping here with one of the LCD3 measurements because it's gonna get boring. Now, I ask you, which is to be the reference there?
SR009 = Win
I see... So, you mean that the Edition 8 does not apply to the club?
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_Ed8_FRF_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_Ed8_FRF_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_Ed8_SW30_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_Ed8_SW30_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_Ed8_SW300_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_Ed8_SW300_R.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_Ed8_IR_L.jpg)
(http://i374.photobucket.com/albums/oo181/acharpen/Headphones_PinnaeEffectRemoved/DeEQ_Ed8_IR_R.jpg)
-
I see... So, you mean that the Edition 8 does not apply to the club?
Hehe. I think the bouncers should probably turn it away at the door.
-
I'd like to resurrect this thread, this time with a discussion on the merits of Impulse Responses, CSD and other such measurements that seem to give an impression of "clean-ness" (whether or not this is the case). How valid can an impulse response/csd be as a measurement of headphone technical fluency in a piece of music? While the measurement has scientific and mathematical merit, looking at things like damping, I can't recall more than a sngle piece of music that had a full spike to immediate silence, without the presence of reverb or echo. Essentially, what I understand therefore, is that the measurement doesn't actually provide real world application for musicians or indeed people listening to music. It is my contention, further, that Planar technologies with even force distribution (and true and even push/pull for a moment across the whole surface) will perform better in a test that shows continued motion than a dynamic headphone. In my experience, a lot of stats and planars don't have a lot of mechanical damping.
Just for my own curiosity; is anyone able to measure the relative performance of three different technologies (dynamic, electrostatic, and planar) with two impulses played with inverse polarity with respect to eachother? That is to say, a force followed by a normalising force, and then treating the measurement in the same manner an impulse response is typically considered after the second force. If possible, can we experiment also with distances (time) between the two, and the relative intensity of the normalising (inverted) force?
-
Impulse response measurements fully characterize linear systems that do not change over time. Linear systems are those that do not add frequencies on their output that where not there in the first place in their input. Real world systems are not fully linear or fully static over time, but many are very close. The more linear a system the better. For those non-linear issues there are other types of measurements that deal with noise, harmonic distortion, among other things.
Things like damping are linear properties which are captured by the impulse response. Frequency response and CSD plots display information gathered by the impulse response in a much more readable and convenient way.
Seldom is an actual spike used to obtain the impulse response. Instead, a pseudo-random noise signal is generated whose auto-correlation function is an impulse (spike). That signal sounds like white noise, because it is white noise. There are other methods, but those don't use a spike either for practical reasons.
It is my understanding that if if two headphones exhibit similar impulse response, then they will perform similarly in test with continued motion regardless of driver technology. There are however differences in non-linear behavior. AFAIK, some stats and orthos exhibit very low non-linear behavior relative to dynamic and BA technologies. If there are perceived differences not shown by the impulse response, probably best to look at non-linear issues.
Note also that when characterizing systems, the pseudo-random signal is usually repeated back to back. This results in an impulse train after auto-correlation. So in most cases the equivalent of multiple impulses already go through the system. Furthermore, the pseudo-random signal itself already changes polarity rapidly before auto-correlation post-processing.
-
Maybe to satisfy those in a similar boat as IHasMario, Tyll (and other measure-ers) could consider whether or not a double impulse chart might be more telling? I.E.-- looks the same, except two spikes ~.001 seconds apart.
(great description of linear vs. non-linear systems, though popcorn)
-
If one sees the raw results from a pseudo-random noise generator, IME it is hard to tell what's going on. I've done it when I developed one of these things for a DSL modem in order to detect bridge taps, load-coils, opens, and shorts. It's all there, but it's like having time shifted, random-polarity mini-impulse responses riding on top of each other.
When you apply the auto-correlation, then you get a nice clear impulse response that is far more telling of what is going on. If we were to apply two (or more) impulses with all kinds of crazy polarity shifts we would get closer to the raw results from the pseudo-random generator signal. Hard to tell, specially if the system exhibits some type of reverberation funkyness a la ultrasone ED 10.
-
Thanks for the great reply ultrabike.
It is my understanding that if if two headphones exhibit similar impulse response, then they will perform similarly in test with continued motion regardless of driver technology.
Is this truly the case? It is my understanding that the force for dynamic headphones is applied at the voice coil, which is for most speakers is behind and at the centre of the cone (hence moving cone), whereas a planar devices forces can be considered as occurring at all points along the system.
If we consider reconstructions of the system, taking for example a wobble board for planar designs, and a circular piece of paper with a string (knotted at either ends) pushing back and forth at the centre, it seems to me that the restoration force is more effective. Or, to put it another way, there is less of the original force present at the moment of impulse in planar designs, because the force is evenly distributed, whereas the circular piece of paper would still be moving forward at it's outer most points provided they aren't restrained or damped in some fashion, which would infer an artifact during playback, in my opinion (but not during a single impulse test).
Could be wrong though, just curious
How relevant do you consider an impulse test in demonstrating the musical qualities of headphone systems (where frequency response would be rated highly, in most cases).
-
I'm not an acoustic engineer, but this post was very helpful to me and it addresses the effect of the shape of the driver on the pressure wave: http://www.head-fi.org/t/223263/the-stax-thread-new/20355#post_8967417 (http://www.head-fi.org/t/223263/the-stax-thread-new/20355#post_8967417)
When we talk about the impulse response of a headphone, we are really talking about the impulse response of the pressure wave that results from it's driver. It is therefore IMO very relevant.
I'm also of the opinion that you don't have to understand or believe in impulse and frequency responses to enjoy music, or to evaluate if certain audio equipment is performing to expectations. I believe your ears come first.
-
Hm, looks like I need to do more research.
I'm not an acoustic engineer either (obviously....), just a curious musician who engineers acoustics (poorly). and a primary school teacher popcorn
Edit: Hot damn, head-fi finally has a following of people who don't like the LCD-2? Well, damn. The LCD-2 was bad enough for me to start wondering about square waves, which started my research! :)p1
-
No worries mang. I'm just posting what I think I know, ... and there is definitively loads of stuff I don't know. Relatively speaking, I recently started to become very curious about audio. I love music, but I'm not a musician. I'm a fan of the bass guitar and the accordion. Maybe one day I'll get formal about it :)
-
Are there any audio books you've read you'd like to recommend, particularly in terms of headphone design and measurement types? I'd like to read more than what I have already, which have mostly dealt with design and equalisation, most of the measurement stuff has been speculation and my own understandings from simple phyisics/maths
-
I'm currently reading this one: http://www.changstar.com/index.php/topic,565.msg14086.html#msg14086 (http://www.changstar.com/index.php/topic,565.msg14086.html#msg14086). Highly recommended by LFF.
For impulse response and linear systems stuff a Schaum's Outline on Signals and Systems might give you a brush up. But that might be too general and may be more math stuff than necessary.