Electrostatic Headphone Measurements

[arnaud]

Hm. So it looks like the current spreadsheets have reasonably good data, or at least data thats as good as I can get it. I could take data over a longer period of time if that would help by setting the sweep to say 5 or 10 mSec, but there's not much going on out there signalwize. Would that be helpful to look at?

Hi Tyll, have a nice ride . In regards to the " there's not much going on out there signalize", there's more than meets the eye I'm afraid, especially if you're referring to the lin scale impulse response. We want to see at least 30dB worth of decay and Purrin's data has shown that many headphones take more than 4ms to decay by that amount. I am not sure yet why I don't see it with your data but I suspect one of these menus forces the data to zero toward the end of the time window (exponential decay), even though it doesn't say so.

Nice result. Looks like there's some correlation. Let me know if you need any raw impulse response files of other headphones from me for comparison / tweaking Tyll's settings.

Surely, could do with more comparisons, please send me if you have!

Edited September 4, 2011 by arnaud

[Tyll Hertsens]

Alrighty then, as soon as I get back I'll take some measurements with a longer window. See ya!!!

[arnaud]

So I did another test with the longer MLS word (128K) and the next highest sampling rate:

...

THis shit is pretty complicated, so I figure you might need to look at the manual . Chapter 13 is about MLS and the impulse response etc. Alrighty, I'm listening for what to do next.

I just started looking at the manual but clearly, some things are fishy just from the graph you posted:

- The standard measurement specifies a 160 micro sec. delay (for subtracting from the phase response) and you see in the impulse response that the signal ramps up at around that time

- Yet your 2nd graph at higher sample rate with 128k sequence shows an impulse response starting at 750micro seconds or so.

- Yet again, the resulting impulse responses in the spreadsheets your sent are virtually identical and effectively ramp up at 160 micro seconds

Edited September 5, 2011 by arnaud

[arnaud]

Wow, I've seen a few data acquisition front ends in my life, but this one is a winner in the confusion department!! Looks like the person behind the GUI was not a transpose from Apple's human interface team, lol . Maybe it's because it a full featured analyzer which caters to all kinds of crowds, but still allowing the user to input different sample rates in various parts of the GUI, while at the same time specifying in the user guide that one should not do that is pretty crazy!

End of rant, going back to trying to decipher this but I am a feeling this is beyond my limited IQ...

[purrin]

Looks like the person behind the GUI was not a transpose from Apple's human interface team, lol .

Hahaha. 10 different ways of doing the same thing with most of them not working. Sounds like MS.

[anetode]

If Tyll isn't too busy with the double-o hotness, I've got a 404LE I could send so there's at least one Lambda representative.

[TMoney]

^

Sweet! Thanks.

I'm sure he'll let you know when to send them once he's done with the 009 extravaganza. In the meantime, don't forget to put your info on the spreadsheet on the first page.

[arnaud]

Hahaha. 10 different ways of doing the same thing with most of them not working. Sounds like MS.

Yeah, this is a reminder of the worst years of Microsoft when they used to cram all sorts of options in some sub-menus. It's getting much better recently though...

As for the user manual at hand, there is one place I found which may be a potential culprit (see exponential settling in page 431 as part of the sweep testing settings). Maybe it's set to perform some exp. windowing in the last ms or so?

Tyll, I will need you to clarify how you are exporting / generating the MLS result as this interface is an utter confusing mess (I now understand why you had to run away with your bike for a week lol ). My (poor) understanding of how this thing is setup goes as follows:

• The MLS sequence (output of your front end, feeding the headphone amp) is generated by a DSP as opposed to traditional analog signals so you must assign it a D/A sample rate and it must match that of the signal analyzer (you're using 65kHz)
  
• The microphone data is recorded back and can by viewed as is in analog domain (going through a 10Hz-22kHz bandpass filter though)
  
• However for MLS processing, the system actually needs to perform FFTs so it's digitizing the analog inputs, in which case you must set the sample rate to match that of the MLS signal generation (not exactly sure why, but I guess it's to guarantee that the MLS sequence does correspond to the acquisition window (as it's not quite fully random signal). In your case, you don't only want to see the result on the screen but also record the data so it must be digitized regardless of the post-processing or not.
  
• For some reason, you're then having to setup a time sweep recording in order to perform the MLS acquisition. In that panel, you're actually overwriting all previous sampling settings above as you're specifying the time step (=1/sample rate) and start / stop time (acquisition period = 1/ frequency resolution of the resulting fft). From trying to decipher the manual, what's happening in the background is that the thing is actually decimating the data and sampling at much higher rate than you're specifying in the various menus.
  
• As a result of this mumbo-jumbo, it's coming up with a 3ms impulse response with an apparent 170kHz sample rate.
  
  The only recommendation I have after reading the manual is no different than the first one - you should try to modify the sweep panel settings which really is the only thing that controls the estimated system impulse response from MLS:
  • Keep the MLS sequence to 28k length, 65kHz D/A
    
  • Keep the digital analyzer to 65kHz A/D
    
  • Get at the very least 6ms of data in the time sweep panel, which is sufficient to see the ringing of most headphones (this will give a resolution of 165Hz (1/0.006) but this can circumvented by zero padding the data). I would go with 12ms (we can always zero out room reflections later if need be).
    
  • You could probably reduce the number of steps as the current settings are a bit overkill (apparent 170kHz sample rate, I was using 1 in 5 samples to generate the CSD graphs). But since we're better safe than sorry, I'd say go with 2047 steps for 12ms which will be the same rate as currently (for a fast headphone like HD800, it might be nice to have that kind of bandwidth). BTW, my current spreadsheet reads 2048 samples.
    
  • I doubt there will be issues with reflections from the walls of the test cell, you're using pretty thick liner I believe? In any case, as mentioned above these reflections can be removed from the impulse response afterwards (except if they're swamped in the middle of headphone reflections, in which case you gotta accept this or move to larger test cell).
    

  hope this will help you...

[Horio]

Sounds like the repair on my 007's should finally be finished by sometime next week (still not holding my breath since it's Yamas). It would be nice to listen to them for a couple weeks since I've been without them for the past month. I can probably send them along your way after you've had a chance to measure the SR-009's.

[screaming oranges]

Horio, you can send your pair later. I sent my O2 MKI to Tyll and it made it there yesterday along with the BHSE (though the BHSE wasn't left at the door and needs to go get picked up). This project will be epic!

Again, as for your cans, enjoy them for a bit. Too many cans and some will be sitting idle. I'm sure Tyll will let you know when he's ready for more measurements. Yours is MKII, no?

[Horio]

Perfect. My O2's are the MK2's (or some people call them the MK2.5's) and not the SR-007A. I figured it's going to take Tyll a good couple weeks to get measure the 009's and O2 mk1's. I can send mine along after that.

I'm very interested to see how the different O2 makes compare. We need to find someone with the 007A's to send them in to Tyll for measuring, so we'll have the full O2 lineup.

[screaming oranges]

Mine might be 007A and not MKI. Frankly, I don't know. I thought MKI meant champagne housing and brown leather, and mine matches that description, but the case I got them in (used) says 007A. Maybe someone can clear that up when Tyll posts pics of all the gear...

...which he better do!

Edited September 9, 2011 by screaming oranges

[arnaud]

I thought 007A was the naming for domestic market (mkII overseas)? In any case, I have no issue sending mine as another point of reference (late SZ2 series, SZ2-16xx). Shouldn't cost more than 50USD to ship to, problem is shipping back...

[screaming oranges]

Maybe mine are 007A after all... came in that box anyway. Never heard another version to tell the difference.

[spritzer]

Show me pics and I can tell you for certain. Hell, just the serial number will tell me what version it is.

The SR-007A should just be the Japanese market version of the SR-007Mk2 but I have never seen a SZ3-xxxx SR-007A so perhaps Stax didn't change the Japanese model and only the export version...

[Spychedelic Whale]

Show me pics and I can tell you for certain. Hell, just the serial number will tell me what version it is.

The SR-007A should just be the Japanese market version of the SR-007Mk2 but I have never seen a SZ3-xxxx SR-007A so perhaps Stax didn't change the Japanese model and only the export version...

Mine had a SRZ3 serial and the box SR-007A with the 'A' being a sticker.

Edited September 10, 2011 by Spychedelic Whale

[Horio]

The O2's that I am sending to Tyll, are definitely the domestic mk2's. They were purchased in the US, have the black housing, and a SRZ3 serial number. It would be great to compare the SRZ2-xxxx and SRZ2-xxxx models, to see how the subtle changes Stax made impact the measurements.

[Tyll Hertsens]

Still on the road, but thanks for your thoughts on this. I'll look at it closely when I get home early next week. Thanks mang!

Yeah, this is a reminder of the worst years of Microsoft when they used to cram all sorts of options in some sub-menus. It's getting much better recently though...

As for the user manual at hand, there is one place I found which may be a potential culprit (see exponential settling in page 431 as part of the sweep testing settings). Maybe it's set to perform some exp. windowing in the last ms or so?

Tyll, I will need you to clarify how you are exporting / generating the MLS result as this interface is an utter confusing mess (I now understand why you had to run away with your bike for a week lol ). My (poor) understanding of how this thing is setup goes as follows:

• The MLS sequence (output of your front end, feeding the headphone amp) is generated by a DSP as opposed to traditional analog signals so you must assign it a D/A sample rate and it must match that of the signal analyzer (you're using 65kHz)
  
• The microphone data is recorded back and can by viewed as is in analog domain (going through a 10Hz-22kHz bandpass filter though)
  
• However for MLS processing, the system actually needs to perform FFTs so it's digitizing the analog inputs, in which case you must set the sample rate to match that of the MLS signal generation (not exactly sure why, but I guess it's to guarantee that the MLS sequence does correspond to the acquisition window (as it's not quite fully random signal). In your case, you don't only want to see the result on the screen but also record the data so it must be digitized regardless of the post-processing or not.
  
• For some reason, you're then having to setup a time sweep recording in order to perform the MLS acquisition. In that panel, you're actually overwriting all previous sampling settings above as you're specifying the time step (=1/sample rate) and start / stop time (acquisition period = 1/ frequency resolution of the resulting fft). From trying to decipher the manual, what's happening in the background is that the thing is actually decimating the data and sampling at much higher rate than you're specifying in the various menus.
  
• As a result of this mumbo-jumbo, it's coming up with a 3ms impulse response with an apparent 170kHz sample rate.
  
  The only recommendation I have after reading the manual is no different than the first one - you should try to modify the sweep panel settings which really is the only thing that controls the estimated system impulse response from MLS:
  • Keep the MLS sequence to 28k length, 65kHz D/A
    
  • Keep the digital analyzer to 65kHz A/D
    
  • Get at the very least 6ms of data in the time sweep panel, which is sufficient to see the ringing of most headphones (this will give a resolution of 165Hz (1/0.006) but this can circumvented by zero padding the data). I would go with 12ms (we can always zero out room reflections later if need be).
    
  • You could probably reduce the number of steps as the current settings are a bit overkill (apparent 170kHz sample rate, I was using 1 in 5 samples to generate the CSD graphs). But since we're better safe than sorry, I'd say go with 2047 steps for 12ms which will be the same rate as currently (for a fast headphone like HD800, it might be nice to have that kind of bandwidth). BTW, my current spreadsheet reads 2048 samples.
    
  • I doubt there will be issues with reflections from the walls of the test cell, you're using pretty thick liner I believe? In any case, as mentioned above these reflections can be removed from the impulse response afterwards (except if they're swamped in the middle of headphone reflections, in which case you gotta accept this or move to larger test cell).
    

  hope this will help you...

[spritzer]

While the testing should easily show the difference between a SZ2 and SZ3 then the results of these sets could be wrong given how demanding they are of a correct fit. My mods to the Mk2/A's can be skipped simply by getting the fit right (well about 90% of the way) which would cause issues with measuring.

Mine had a SRZ3 serial and the box SR-007A with the 'A' being a sticker.

Well that saves me buying a set. I had been watching all auctions in Japan looking for the serial numbers for a while now.

[arnaud]

While the testing should easily show the difference between a SZ2 and SZ3 then the results of these sets could be wrong given how demanding they are of a correct fit. My mods to the Mk2/A's can be skipped simply by getting the fit right (well about 90% of the way) which would cause issues with measuring.

That explains... I always wondered how much the typical issues with 007A's bass overbloom were coming from improper seal. At least for me, I felt the proper seal made very significant difference in how clean / tight the bass sounds - regardless of using a non-linear amplifier . Good news with the 009 is that I don't experience such sensitivity to fit and it basically fits like a glove...

[arnaud]

Still on the road, but thanks for your thoughts on this. I'll look at it closely when I get home early next week. Thanks mang!

Sounds good Tyll. BTW, I know I am not "supposed to" but I paste her a discussion with Purrin's from the HF thread on CSD measurements, in case you're not visiting there. Bottom line is that, after thinking about Marv's setup, I believe there's very little chance for your and his CSD results to have much resemblance:

------------------------------------------------------------------------------------------------

Marv, I reckon you started doing these tests on regular dummy head data and then found out you got cleaner results with your current anechoic termination test rig.

While I thought this was a great idea at first, I am not sure of the meaning after thinking a bit about it:

• There does not seem to be a fundamental issue with using a standard dummy head (see my CSD post-processing of Tyll's data)
  
• Headphones are (for the most part, grant you the K-1000 is a different beast) designed to feel an acoustic load forward of the driver (e.g. the acoustic chamber between the driver the ear due to earpad spacing).
  
  
  I see several major issues with the anechoic termination test:
  • As you pointed out, the low frequency part of your CSD plot has little to do with actual performance. (that's regardless of open or closed design, being in the acoustic near field of the driver, e.g. it's almost like doing an IEM test without a proper seal)
    
  • The anechoic termination completely changes the acoustic resonances of the enclosure (from a sort of rigid boundary condition to anechoic one, acoustic resonances will shift down in frequency VERY significantly, by a factor of 2 for example). This will be seen up to very high frequency because such acoustic resonances occur above 2kHz.
    
  • The anechoic termination adds significant acoustic absorption in the front chamber (ear to driver, bounded by the earpad) so the acoustic resonances of this region will not only be shifted in frequencies but also significantly damped so you pretty much won't see them in the CSD plot.
    

  Have you thought about going back to standard dummy head measurement now that you got more experience with CSD testing? I understand you might want to isolated the driver resonances from the acoustic chamber but in this case, extracting the driver (like you did with the SR80?) or better yet, doing a laser vibrometer measurement on the diaphragm (can you get your hands on that ? ) would be suitable.

  ------------------------------------------------------------------------------------------------

[purrin]

But yes, I've started doing the "dummy head" type tests too. The results with open phones are similar between anechoic an dummy head tests. Closed type phones are different story.

See my reply on other site.

Edited September 21, 2011 by purrin

[spritzer]

That explains... I always wondered how much the typical issues with 007A's bass overbloom were coming from improper seal. At least for me, I felt the proper seal made very significant difference in how clean / tight the bass sounds - regardless of using a non-linear amplifier . Good news with the 009 is that I don't experience such sensitivity to fit and it basically fits like a glove...

Even with 100% correct fit the lowest registers will always be slightly overblown. What Stax tried to do with this unit is to have the earcups breath just a little bit so once the phones sit on the head the port is supposed to be almost closed. As with so many other brilliant ideas it doesn't really work in practice. The force on the exerted on the earpads will always depend on the size of the users head and the tension/angle of attack of the arc's. Same issue as with the Mk1 but now with the added benefit of the ports not closing properly and thus messing up the bass and midrange.

I'm not sure what Stax were thinking with the Mk2.5 (SZ3) though. Rumor has it that it was a test bed for the new diaphragm material found in the SR-009 which could make sense as the drivers are certainly different. My set has been on loan for a couple of months now but the plan is to get it back and see if I can "fix" them. I've already install Mk1 springs with no real difference so it is time to plug the ports and install Mk1 earpads.

[Robbo1802]

Hi,

A total newb here but not to electrostatic drivers. I am a bit curious as to what you are trying to achieve with these measurements. Is this an attempt to correlate possible audible effects to measured effects, i.e. resonances to colouration? Alternatively are you trying to relate the design of the headphones to their measured performance? You guys are going to quite a lot of effort to get together a large collection of ES phones but have you worked out exactly what tests you are going to perform and how you are going to do it? This is not criticism, I am genuinely curious and quite excited by the prospect.

Since this is all about electrostatics (as an addendum to the standard CSD etc tests), have you considered going back to basics and use the methods employed by Peter Walker etal when assessing their speaker designs? One of these would be to exploit the reciprocity of the electrostatic transducer and use it both as speaker and microphone. This is fairly easily done by measuring the blocked impedance (electrical) (with signals of choice eg. sweeps, MLS, impulse),measuring the unblocked impedance (electrical), subtracting the two to give an almost exact representation of the acoustic power being radiated by the diaphragm. This power will change depending on the acoustic impedance applied to the diaphragm by the real world and will show up resonances in the stators, casing, stator hole impedances, reflections from the stators, diaphragm primary resonances (but not many modal resonances - more later), even reflections from the room boundaries. For the uninitiated, the blocked impedance is simply the impedance of the system when the diaphragm is blocked from moving, in the case of electrostatics this is easily achieved by turning off the diaphragm bias voltage.

One of the problems with this method is that it gives best results (from a designers point of view) when the transducer is radiating into an empty room - well at least one member of this forum seems to like pulling his expensive ES headphones apart. >;-)

Other tests worth considering are the distortion tests, especially a swept Intermodular distortion test (IMD is sometimes referred to in loudspeakers as Doppler distortion). This test can often reveal unexpected diaphragm behaviour like the modal patterns shown by Streng in his papers (something often ignored by ES designers or their marketing departments - I notice that Wiki still implies but does not state the fallacy of pure planar behaviour). This is very important in ESs as unlike most other drivers an ES has its radiating surface inside the motor ( Ok, there are others such as AMTs and planar magnetic, in the former modal breakup would be very different if it occurs at all and I haven't much studied the later). So, in ES speakers, modal anomalies can significantly affect the 'motor behaviour' regarding distortion and overall available output levels even if they are not clearly audible (many modes have little net output as the positive nodes and negative nodes tend to cancel each other out [see Olsen, Berenak etc] and may not show up very prominently on CSD plots).

Unfortunately, I am not in a position to help. I am currently traveling around Australia in the a motorhome (year 7) and have none of my library on ESs with me. Indeed, I have had to revert to dynamic headphones as I didn't deem the ESs sufficiently robust or their amplifier practical with our limited (off the grid) power resources.

I should have brought them with me as a disaster last year has seen them and a great deal of other HiFi gear (which I had in storage) completely destroyed - some of it was practically irreplaceable - older, collectible stuff - often impressive to look at, some of it quite beautiful, and the odd bit here and there occasionally even sounded good.

There are copies of PDF copies of Borwick 3rd ed floating around on the internet so Peter Baxandall's work on ES is available but this seems to have changed considerably from what I remember of the 2nd ed. I have not been able to source the Walker, Streng, etal papers off the internet. The Streng ones originally published in the Philips Technical Review are most interesting as they led to an experimental segmented ES loudspeaker that was shaped (from memory) as an isosceles triangle.

I was tempted to post this in the CSD tread on HF but this one is specific to electrostatics so I thought it more appropriate.

Regards,

Bob

BTW

Sorry about the rambling post but my english skills as well as abstract mathematics skills have abanded me these days. My first electrostatic - Micro Seiki (Stax in disguise), bought new circa 1975. From memory( 25+ years), mine seem to be different to the one sometimes shown on the internet. The diaphrapms (which eventually failed) were not tensioned until the cell was assembled. There was a ridge on the stator molding which pressed against the diaphragm locating it centrally in the gap while at the same time tensioning it, when dissassembled the diaphragms were quite loose. I haven't seem this technique mentioned elsewhere although it may have been something to do with the diaphragm material which did not appear to be mylar - it seemed more elastic. From memory, the bias voltages were also quite different from what you would expect, maybe someone can shed some light on this. Ohh yes, the Micro Seikis sounded quite good as long as you were a bass atheist, even then I has hooked on ESs (and Heil AMTs).

[Hopstretch]

Welcome, Bob!