I think that while I can’t pinpoint to the exact cause(s ?) of the disparities between Resolve’s and SoundstageSolo’s results, I can, to a limited extent, reproduce the
trends observed between their measurements - even with the latest firmware, and with a single sample. That may be helpful for more knowledgeable people to start providing a more thorough explanation.
So, I went to have a listen. I was offered coffee and invited to sit in a worn out chair in a room full of old vinyls. God I hate these boutiques.
The sample they had came from Mark Levinson’s distributor here and freshly arrived. The pads were brand new (not ideal). Turns out the firmware wasn’t up to date,
I updated it myself to 1.5.0.4.
I had mixed expectations, but to be frank after no more than a minute I knew that something was quite a bit off, and that trying to enjoy myself listening to them wasn’t going to be particularly fruitful, so I took out the in-ear mics and started sweeping away.
From now on it’s all boring graphs. Sorry for the dump. I promise I’m much funnier at parties.
Some notes about the in-ear mics :
- They’re blocked ear canal entrance mics.
- The absolute values are incorrect. I didn’t bother to compensate the levels as I occasionally do, but I’ve often compared these mics with a UMIK-1 in near-field conditions against a speaker and it’s not that off either. Also, the mics aren’t calibrated : the absolute values on the y axis are meaningless.
- You shouldn’t compare the absolute values with results obtained from ear simulators (at the DRP), only the relative results between traces from the same sample on the same system (the sample I listen to with my in-ear mics on my own head, and the other samples on ear simulators).
- Relative results between headphones with blocked ear canal entrance mics may not be perfectly accurate in the 2-3kHz region and above 7kHz in particular. Don’t use these graphs outside of this post’s context.
- These results are only illustrating my own experience, with the sample I listened to, on my own head. Using the same sample, mics and method on your own head may yield more or less different results.
- I was in a rush. I usually prefer to repeat measurements several times and avoid measuring brand new pads. This in contrast was a quick and dirty job.
So, squiggles.
* The relative difference between some of the various modes available *
Right channel only, as they naturally sit on my head, blue trace ANC off, solid red traces ANC low or high, dotted red trace ANC adaptive, all wirelessly. Averages of 3 individual seatings, normalised at 3.15Khz (that value lands in a part of the spectrum where the SPL is less affected by position / pad compression and above the effects of the active filtering). Absolute values incorrect, only look at the relative values please.
I’ve elected to include the results above 7kHz but I repeat : please take them with a pinch of salt, I have reservations about this type of mics above that frequency. See that peak at 9700Hz or so ? Well in reality it might be located at a slightly different frequency for example. Its magnitude might also be quite different relative to other headphones than what these mics would show. Above that frequency, with these mics, it doesn’t really make sense to talk about “peaks” or “dips”.
I didn't test them in fully passive (wired, powered off) mode.
While I don’t think that we’re facing a feedback mechanism that needs a broader signal to properly work, like on the AirPods Pro and Max when ANC is turned off, or on the Airpods 3, given that the individual traces were quite consistent with low seatings to seatings variation, I’d still prefer to have these results confirmed with measurements using noise as a signal, as I believe that it’s generally good practice for headphones with a feedback mechanism. That wasn’t done (not enough time).
The shape of the ANC off trace, below 100-120Hz, might make you think that I’m not getting the same seal quality that you’d see on the sort of ear simulators that have been used so far to test them. We’ll see later on that, while to a moderate degree that might be the case (even though I couldn’t pinpoint to any obvious source of leakage when wearing them), even with a lot of pad compression, the bass levels when ANC is turned off never aligned with the ANC on results.
Now, let’s move on to where it gets interesting.
* Behaviour under pad compression *
A few notes first :
- Measuring frequency response under varying degrees of pad compression affects several variables at the same time (volume of air in the front volume, foam compression, the pad's side walls deforming, etc.). Therefore it can’t be used on its own to determine the exact cause of a phenomenon for someone like me, but can provide some clues to more knowledgeable people who would then be able to test each of these variables in isolation if possible or derive hypotheses. So think of it as a test that points towards directions for further investigation, not the cause in itself of the phenomena you obverse.
- I manually applied varying degrees of pad compression while running sweeps. Because of that method, the traces can be quite a bit noisy at lower frequencies. So I won’t show data below 70Hz.
- It isn’t possible to compress the pads by a precise amount each time. So what I tend to do is take a lot of individual measurements (north of 20) and then select 4-5 representative traces out of the total.
- A corollary of that caveat is that the way to read these graphs is to avoid looking at the magnitude (dB), but rather focus on trends / directions, the shape of the traces, and the relative changes in SPL for different parts of the spectrum.
- These are individual traces, not normalised, without averaging.
Let’s start with a long detour. We’ll circle back to the ML5909 eventually.
As a primer, that’s typically how a fully open, passive dynamic (HD650) behaves under varying degrees of pad compression :
The blue trace is how they perform when they naturally sit on my head. The red traces show increasing levels of pad compression.
@solderdude has published an interesting article on earpads where you’ll find plenty of compression traces, including for the HD650 :
home back to headphones published: July-09-2020, updated: Nov-25-2022 Ear-pads or ear-cushions are an important part of the headphone. They are consumables (in most cases) and wear. Pads can be on-…
diyaudioheaven.wordpress.com
A better way (for what follows) to represent these results, I think, is to compensate the blue trace to a flat line at 0, and only show the difference between it and the red traces :
If the lines stay flat, it means that the sound colouration / balance doesn't change as the pads are compressed, only the SPL rises.
These fully open dynamic headphones are quite constant below 3kHz or so in my limited experience.
A typical closed front volume / closed back like the K371 may show less linearity as the SPL rises under pad compression, but for this model it’s still quite linear below 3kHz, as long as you have a good seal (which I had for the blue trace) :
Now, this is how two ANC headphones (Bose QC45, Sony H910N) behave in ANC off or passive (wired) modes :
These two ANC headphones seem to have been designed in quite a similar way. There seems to be a null point where variation in pad compression doesn’t change the SPL much (around 1.5kHz, black arrow). Below that point the variation in SPL is quite smooth. Above that point the variation is quite wild. This is not the case for all ANC headphones when ANC is turned off or in passive mode, but I suspect that in particular all Bose ANC HPs stick to that approach quite religiously.
In both cases the raw measurements when ANC is turned off are quite poor.
What happens when you engage the ANC ? Well, the feedback mechanism springs into action :
I think that it tries to deliver an exact dB value at your eardrum for a specific input value. As you can see, in the range where it operates, it nullifies (in the case of the QC45) or at least makes more linear (H910N) the variation in SPL, and delivers a more or less constant SPL, regardless of the amount of pad compression that’s going on.
You might have noticed that the QC45 successfully maintained the SPL constant up to the “null point” mentioned above (which I guess, in the case of Bose headphones, is located very deliberately right above where the feedback range stops). This means that, when ANC is engaged, it can basically deliver a constant response up to 1-1.5kHz. On the other hand the H910N’s feedback mechanism seems to stop operating at around 500Hz.
And here’s the rather interesting bit for the H910N : in the range
above where the feedback mechanism stops operating, but
below 1.5kHz or so, when ANC is engaged, a similar amount of pad compression (which you can determine by matching the compression traces above 1.5kHz) results in
more SPL than when ANC is off, as if the active filtering was amplifying the effect of the compression, instead of reducing it :
This overshoot does not occur with the QC45 :
* So, how does the Mark Levinson 5909 behave - at least the sample I tried, on my head - under pad compression ? *
When ANC is turned off (wireless) :
It seems to behave more like the K371 than the QC45 and H910N. Perhaps as a by-product of the intent to have them perform well when used passively ?
You can also see that seal was only a minor issue.
When ANC is turned on, in the “low” mode (I didn’t have the time to check the other modes) :
I think you’re starting to see something that should already look familiar :
Now, we’re going to do a bit of Room EQ Wizard trace arithmetics acrobatics so please bear with me.
Let’s divide the ANC on results with the ANC off results, for either no compression (blue), or two matched (between ANC on and off, above 1.5kHz) pairs of compression traces (red) :
I normalised the traces at 4kHz as this is where the SPL varies the least with pad compression, and far above the range where active filtering occurs.
If there were no difference between the ANC on and off results, you’d see a flat line at zero.
That isn’t the case (cf first graph, even without pad compression), and you can see that as pad compression rises, the difference between the modes increases, following a certain trend :
- In the 150-600Hz region, as pad compression increases, ANC on reduces SPL relative to ANC off.
- In the 600-1500Hz region, as pad compression increases, ANC on increases the SPL relative to ANC off. In other words, the more the compression, the more the overshoot.
Now let’s take a look at Resolve and SoundstageSolo’s measurements, both performed, unless I’m mistaken, on the same ear simulator (GRAS 43AG + KB5000 pinna) :
Resolve’s in fancy fuchsia, Brent’s in fashionable turquoise. Solid lines either “passive” (Resolve) or ANC off (SoundstageSolo). Dotted lines ANC “high” (SoundstageSolo, similar FR for me as “low”), or ANC unknown (Resolve).
Plotting the difference between ANC on and ANC off for each set of measurements :
This isn’t a perfect match for the trends observed under increasing degrees of pad compression seen above, but I think that there might be something there. Among other factors (it’s entirely possible the firmware update did indeed affect the results), are we also seeing the effect of varying degrees of pad compression here - or at minima a different “value”, between Resolve and SoundstageSolo's tests, for at least one of the variables that are affected by pad compression ?
Besides, the difference between their measurements above 1.5kHz, ie above the range where active filtering occurs, seems to also share a fairly similar shape and trends to the ones I obtain when I increase pad compression, even when we’re dealing with totally different systems (ear simulators measuring at DRP with a flat plate around the pinna, vs. my own anatomy and blocked ear canal entrance mics). Which, theoretically we shouldn’t compare, particularly at higher frequencies, but let’s be unreasonable for a while :
If I divide my own results between the traces with the maximum amount of compression and the traces as they sit on my head, for both ANC off and ANC on, normalised at 4kHz, and compare the results with the ones obtained by dividing SoundstageSolo’s results by Resolve’s, we get this :
I’m probably really stretching it here I think. But is the trend towards a broad elevation around 6-8kHz, and the rather similar shape (albeit with frequencies offset a little bit) merely a coincidence, or indeed caused by one of the variables affected by pad compression ?
I would be surprised that this range would have been affected by the 1.5.0.4 firmware update, and let’s remember that both SoundstageSolo and Resolve used the same pinna.
So, to sum it up, while the 1.5.0.4 firmware update may have reduced the disparities between the various modes, it could also be that some of the difference observed between Resolve's and SoundstageSolo's results is the product of at least one of the variables affected by pad compression being different between their respective tests.
Given the design of the GRAS 43AG rig, it might simply be a question of different levels of pad compression during testing indeed, but I wouldn’t directly go to that conclusion without further details, and I've never manipulated such a fixture.
How I personally experienced them in terms of consistency between modes seems to be an in-between hybrid with some unique characteristics (here using the same data I used for the first graph in this post, without pad compression) :
As I wrote, pad compression merely is a proxy to test for several variables at the same time, so it’s difficult to know exactly why in my case I get these results. It might be the case, for example, that even with a similar amount of pad compression, the front volume of a pair of headphones may vary slightly between ear simulators and individuals because of geometry around the pinna. And perhaps we can't rule out sample variation. Or the pads' age. Etc.
This is why I don’t think that some of these measurements are “wrong”. Maybe they just simply illustrate the sort of variability we can expect between ear simulators and individuals or between individuals ?
And perhaps also tell us how mindbogglingly difficult it might be to design a pair of ANC over-ears ?