Discussion in 'Measurement Techniques Discussion' started by purr1n, Jan 8, 2017.
Different setups, slightly different headphones (JAR600 vs HD650), but some similarities.
Including the "chunked" first half-wave on the 50Hz signal of the HD600. Note that neither log or abs wasn't applied to the first plot, so it's less phallic in appearance. I've noted this to be a characteristic of the HD600 driver and have not seen it in others.
I'll see if I can repeat that with my HD600. Seems weird to me. WIll update this post later.
Okay, so here are my results. Didn't get such a glitch.
I thought it'd either be because of a dip at 50Hz like we sometimes see on Tyll's dummy head measurements or because of some HF ringing, which is why I took multiple measurements on different couplers. Each of these have a slightly different FR. The V1 flat plate coupler resting on a soft material has a distinct dip at 50Hz, which is exactly what I had hoped for, yet the only thing that seems to change is that the decay now takes longer. The measurement on the head is usually very peaky, but it seems the slow rise time of the 50Hz sine doesn't excite any of the peaks here. This is despite a very audible "click" sound at the beginning and end of the bursts when listening.
Overall I'm not sure what's going on with the above result. I can't reproduce it with my HD600.
BTW, I took measurements in REW for all of these configurations so it's also possible to generate results for other burst frequencies.
I prefer the top display (%, no abs applied, unfilled curve) - it's easier to see the changes in amplitude and wavelength at both ends. Can you easily add a time scale on the x-axes? Sure it's different at every frequency, but to aid comparison with impulse response and CSD patterns it'd be nice to see it directly rather than having to calculate/guesstimate.
The idea is to tweak the visualizations / presentations for what they are meant for.
Straight values with no abs applied for attack and successive wave analysis.
Log values with abs for decay analysis.
The time scale is easy. Just calculate time = 1 / frequency for each sine wave. However, these cannot be compared to CSD though because we are plotting against cycles, not time. So in a way, maybe we need to think differently when it comes to a 4800HZ decay (relatively fast, even it does vibrate on and on) vs 50Hz (relatively long).
@Biodegraded Since these results are generated from the sine burst waveform, it would be pretty easy to include both results. Just doing proof of concept stuff for now and was more interested in trying my hand at the decay visualization.
@purr1n I have been playing around with ARTA's burst decay plots lately to see if they might have value, since they report on cycles rather than time. Trying to see how closely it matches the sine burst decay results.
Maybe there's some value there, but I don't think it presents information in a way that's immediately as tangible as the visualization you've been using lately.
And, yeah, for an initial round of experimentation, I was pleased that results weren't too far off. Still plenty of work I need to do. I'll have to test my stock HD650 and the RS2e while I have it. I do expect my measurement methods will no doubt cause some differences.
On that old topic of grain, prickliness, jumpiness, etc. I've included Ether which I feel has a jumpiness, but not so grainy.
Ok I like those displays!
These are windowed transforms of the burst response, energy - integration under the curve, magnified somewhat to accentuate the differences.
Yes, these are interesting and easier to compare!
Here is ESP95x. It's a stat. What we do think about stats? Flaccid or soft low bass, but otherwise "fast" or sharp transients throughout, e.g. great separation of cymbals and percussion. Could the attack characteristics in the highs facilitate this effect?
SR1a. I love the transients of the SR1a. Tactile, fast, some edge, yet not unnaturally so.
I really like these plots in helping to depict the time domain behavior for the attack. What is the y-axis? (I can't seem to find a reference in the previous posts). I'm thinking its amplitude, but I'm not sure.
The Y-axis in the latest plots is amplitude in "voltage" scale. The plot is a signal envelope of the burst. The level of signal varies depending upon frequency response of the headphone, ~70db SPL give or take. There are no demarcations - something I need to work on - even if there were, it would be a relative scale with the steady state level at 1.0. The sine waves are normalized to each other according to their peaks after the waves stabilize (waves 6-9).
Did some tweaks, such as set the window to a full wave rather than half wave as before. This eliminates much the jaggies, a result of driver offset, without losing out too much if the first one half wave of the attack is stronger than the back half.
50Hz ESP-95X graph is interesting...among others.
I know I am getting ahead of myself here, but I am really excited to see if different amplifiers visibly affect these graphs.
I think they shouldn't, as all these graphs really show with (most) headphones is FR. That is unless the FR magnitude or phase change considerably, say through dynamic driver impedance interaction (bass bump) or severely limited bandwidth.
REW can generate these graphs using the trace arithmetic function. Headphones will look boring, but speakers should definitely be interesting with their not-even-close-to-zero excess phase.
This is the HD600 on V1 coupler resting on soft foam, same as above. Black is the loopback graph and red is with the HD600 and the microphone added to the signal chain.
For fun I also generated plots for two speakers from measurements I had. Note that they were not taken in the best conditions and will show reflections, too much ringing, too uneven FR, but the general idea is there. I arbitrarily aligned the top of the third sine.
Of course it's possible to generate these graphs for any possible input signal. Different bursts, music, whatever. But generally I feel it's easier to look at the measurements directly.
Not really. I wouldn't assume, and thereby dismiss the work I've done here so rapidly. I can give you a ton of counterexamples or thought exercises why this doesn't work. FR does not give you this information.**
Can you look at a frequency response graph and tell me if a driver has tight or mushy in the bass, or has sharp delineated attacks in the highs, or could have grain or prickliness? If you answer yes, I will tell you that you are full of shit.
Or you are wanting to be technically right* again, missing the forest for the trees?
*You would still be technically wrong as while these be convolved from impulse response, FR plots by themselves do not contain phase information (and may not have the resolution and will be smoothed), therefore an inverse FFT won't work totally right.
** Regardless, using your line of reasoning, why even both with generating FR when all it really shows is impulse response?
I wasn't trying to dismiss your efforts. As you've said, we could look at the impulse response or step response as it contains the same information, but it's a hell of a lot more convenient to look at the FR magnitude and deviation from minimum phase / excess phase. In the same vein this visualization is interesting, aswell.
Smoothing is an issue, that's true. But we really shouldn't need the phase information for headphones, which is why I included the 'most'. Well, for the most part. It seems that with my measurements the bass rolloff isn't quite minimum phase. With the HD600 measurement from above I'm at around 4.6° of phase shift at 50Hz or about 23° at 10Hz. That seems to be a common theme for every measurement, actually. I'm not sure yet why that is.
Also some guys on reddit claim the M1060 isn't minimum phase, as Ultrabike does of the KSC75. Could be the enclosure or measurement rig, though since measuring with a microphone at the ear canal gives non-minimum phase results aswell. Severe cancellation seems to do that.
What I was trying to say is that
1) As ultrabike has demonstrated, we can generate these burst signals from the measurements, no need to take additional measurements. I think you're already doing this, aren't you?
2) With a simple parametric EQ we can get great FR/IR/burst measurements, yet I don't think the subjective qualities of transient speed change with the EQ much, if at all. This is why I'd trust your subjective impressions more than looking at graphs. And there have been headphones which got really good measurements that don't sound particularly impressive to me.
As far as if this visualization is helpful or not, I can't say for sure with these headphones. I've not heard the Verite. I've only heard the ESP950, not the Massdrop version with different pads. I haven't heard a final version of the Borealis. The Ether didn't sound like the graphs look like to me, but maybe I'm misinterpreting them. Sounded soft, rather. HD660S sounded nothing like the Elex in terms of transient response, yet up to 1500Hz they look pretty much identical. Their FR is very similar, though.
Overall, I'm not sure. I don't think they hurt, either.
Ultrasonic and subbass extension as well as low distortion and low driver response irregularities seemed to correlate better to some of these subjective qualities for me, but YMMV. Also @Bill-P told me that he thinks a 1 to 1.5 kHz emphasis can make something sound "fast". The STAX SR009, Focal Utopia and Abyss all have an emphasis in this region and they sound very fast to me (each in their own way). To me the HD800 does seem to sound faster once modded (1-4kHz gets emphasized by the mods). HD600 has some emphasis around 1kHz (subjectively) and you mentioned it sounding faster than an HD800 to you.
Then again the HE90 doesn't have much of the typical 1kHz stat emphasis, yet it sounds pretty fast to me. Or take a HFM HE-6: Pleather pads give it more of a 1kHz emphasis vs velour, yet I don't think it changes subjective transient response that much to me. Maybe here the ultrasonic extension comes into play: the HE90 (as I guess most stats) probably doesn't roll off much up to 100kHz.
EDIT: I forgot to mention, but the reason why I posted the above graph was also because I was pleased to see that the recorded and simulated bursts look pretty much the same, as they should.
Also the reason I mentioned that they likely only change when FR changes (or DAC/ADC changes) is to emphasize that the coupler plays a role, as it doe for every measurement. I'm not quite happy with the current couplers (dummy head or flat plate), so maybe with a better coupler the bursts would match subjective impressions better.
Pardon the shitpost (that's also just my 2c) but I just want to say that I find it easier or at least more intuitive to read the old burst responses (traces and phalluses) than the new multicolour things so have been going back and forth trying to figure them out.
Yes there is an explanation posted and the demarcations do help, but ¯\_(ツ)_/¯
I get where you're coming from @Serious since based on my own limited experiences with headphones (non-EQed) I can tentatively agree that emphases in the midrange region do seem to sound "more immediate", but not necessarily "faster". IDK, is it the more upfront staging giving the presentation of immediacy, i.e. less space between performer and audience? Asking this cuz some of the faster dynamics I've heard, mostly Fostex and the Klipsch yes, have midrange troughs but don't sound slow to me.
Focal Elear definitely immediate AND fast (not heard Clear, hardly any time with Utopia), but inferior to the biodynas IMO. Still, limited time to compare so could be wrong, might ask someone if they wanna trade headphones for a bit post-lockdown.
I just think going to FR to explain this seems overly reductionist. True there's a lot to be gleaned from it, but also way more that can't be easily sussed, or gleaned at all, from the same.
I said this in the Zoom meet but I still have my tinfoil hat on about why the HE90 pads come mis-matched, haha. Might have to do with helping modify sonic presentation somehow!
We need more data to arrive at conclusions, but I am seeing stuff that you probably aren't because I'm sitting on more data / visualizations than you have seen. From where I sit I'm seeing a lot more recognizable patterns. Also, we need to take into account the decay characteristics. Again, there is no ONE measurement that tells us everything. And if we think about transient response, it's both attack and decay.
For instance, let's take the ESP950 and Ether - both have similar FR in the lows. I agree the that Ether likewise sounds soft or more precisely the lows sound mushy and murky. Normally, this may be distortion, but Ether measures decent in distortion. The ESP950 deep lows hit like a wet noodle.
So far, the Ether is the only one that has a long lived overshoot at 50Hz like this. Could this be the cause of its murky, blurry, bass? I could disagree that the bass hits soft, but I also don't think it's solid and hefty either. Is there something here? Hard to say, we need more data to see if this is a recurring pattern.
In contrast, ESP950 is like this for 50Hz
Yup, totally describes how people feel about 'stats low bass performance. "limp dick". It's tries to get it going at first, but it's too soft, and almost takes 7-8 cycles to get up to the proper amplitude.
Now let's take a look at decay. I'd argue that the ESP950X has more noticeable faster decay, at least in the short term, the second, third, fourth half-cycles (which is long when it comes to bass).
Separate names with a comma.