Driver story, the acoustic system and the limits of EQ

Some here have looked at Fourier analysis and have the basic concept of it well entrenched in their heads and a few of those have a lot of the details well-understood. I’m guessing, though, that there are >1 people in the thread here who don’t and don’t see the relationship between things like an impulse response or square wave and frequency response.

What I do for a living often centers around MRI imaging. We make pictures of parts of your bodies (like ear canals) by making different parts of our image resonate at slightly different frequencies and phases so that we can take a little squiggle that comes out of an antenna and turn it into a picture. This whole thing relies on Fourier. There are many nuances that I’m surely unaware of, but I have presented the gist of it more than once in my time and getting someone able to play with it themselves can often be eye opening.

Fortunately, with REW and Audacity, you can do this yourself. Here’s a brief how-to:

You’ll generate some sound files and use the spectral analysis tools in REQ or Audacity to see how things like random noise or impulses are made up of weightings of all the possible frequencies (based on your sampling) and how applying EQ “slows” the impulse response (and that a “slowed” impulse response will affect the frequency response).

I’m not trying to put this out as some perfect method (we’re working with my limited mastery of both bits of software for starters), but it gives you a starting point and a way to explore yourself how these two domains can really be interchanged.

2 Likes

What do you mean it ‘ slows the impulse response’?
Do you have an example of an impulse response and on with EQ for the differences?

Agree that this could probably use some clarification. I suspect there’s more nuance in Blaine’s comment than we’re probably taking away from it though.

Blaine’s knowledge of audio is much vaster than mine. But maybe you could think of it this way… The minimum phase characteristics of EQ are desirable, because they make on the fly FR adjustments of minimum phase devices (such as headphones) possible, without more significant forms of time distortion. That does not necessarily mean that the phase changes in a min phase EQ are always a good thing though, from the standpoint of sound quality and imaging.

Most processing in digital audio is lossy in some way, And I suspect that the phase changes in digital EQs are also lossy in some ways. Though in many cases the loss may be rather small, and difficult to perceive or quantify for an average listener. And the benefits and ease of use of digital EQ might outway most of the negative effects, especially if you know how to mitigate them.

There are other ways of adjusting FR that don’t involve EQs though, as mentioned in my previous post. Physical mods and filters are one option. And there are also amps that can add their own degrees and forms of coloration. (I don’t use it, but my Bellari HA543 amp has an “Enhance” button that appears to add a “smile” to its normally more linear output. Z seemed to like the idea though, for headphones that lack enough zing in the treble.) Other enthusiasts may use the nonlinear impedance of their dynamic headphones with higher impedance amps to color their sound. Or tube-roll to make finer adjustments to sound quality.

I’m still catching up on the previous posts here btw, after a few days of work elsewhere, so may have to backtrack a bit on some of this, after reading others comments here. From what I’ve been reading on this subject though, the phase changes are likely to be most pronounced on steeper “higher Q” EQ curves than on EQ curves that are more gradual and broader in bandwidth. This might be a good argument for using mostly lower Q PEQs. A potential downside to this is that it would probably allow less precision when it comes to matching the FR of different headphones, the OP’s stated goal.

1 Like

From what I read, for headphones a min phase EQ is best.
Some people also say to try and avoid high EQ values as well. Unless there is a steep resonance like in my 1990. Don’t know how much that matters.

Maybe someone else can chime in on that.

3 Likes

I watched the Canjam presentations where Rob Watts talked about these things, and he was making some pretty outlandish claims. If I recall correctly, he was saying he heard things like noise or distortion from rounding errors being audible, or creating audible effects (I remember he said ‘the soundstage collapsed’), at -300db. And that he could perceive this in his listening sessions.

I mean, I like some of his designs (I own a Chord Mojo 2 - it’s excellent but for the lack of a volume knob), but until his claims are backed up by third party blind testing I don’t think anyone should be treating them as credible. The ‘lossless EQ’ included in the Chord Mojo 2 sounds like a great solution to the problem of lossy EQ that he alone seems to be claiming exists. That said, I would also be interested in whether anyone can offer a reasoned basis for why those things aren’t (or aren’t likely to be) issues.

2 Likes

This is where I get hung up as well.

Does a headphone’s “frequency response” represent all it can do? In a sense, yes. Waves have two characteristics, amplitude and frequency, and all sounds are waves. So a headphone doesn’t do anything more than produce waves with amplitude and frequency, which, in theory, we could represent graphically.

But from that premise to conclude that the measured frequency response of a headphone tells you all you need to know, or that once you adjust with EQ you can recreate essentially whatever you like . . . no, I’m not on board with that. And I don’t think that’s what anyone is claiming. Or maybe someone is? Regardless, there’s a big missing step there which encompasses an assumption about our measurement capacity, which I don’t think is warranted, even if we could fully adjust for individual anatomy, because, as you put it, “busy mixes aren’t sine waves.”

So, I don’t think anyone is claiming exactly that, for several reasons. One would be there definitely are other things to consider like harmonic distortion and potentially acoustic Z. But on our end at least, I’m not aware of anyone saying that if those other things are kept equal, the measured frequency response as people currently think of it is all that matters, or that you can easily recreate whatever you like with EQ. I think that’s kind of still confusing the position a bit.

What I’m saying is that provided all else equal, if you could measure the FR at your own eardrum relative to your own HRTF, that’s likely to be the acoustic cause of our experiences, and that there’s no extra layer or dimension of ‘technicalities’ over and above this that’s conferred by driver story.

Importantly note that the FR at your eardrum would look different from a typical FR graph and you’d be able to exhaustively correlate it with what you hear, where as currently you cannot, and there are gaps between what you measure on a rig and what you hear. That’s sort of the whole point of this discussion.

But is that all that matters? Also no. As said earlier, when it comes to preference, we have a good understanding of what people like for wideband balance, not so much for fine-grained features. Maybe people like a certain coloration and this amounts to ‘technical’ qualities to them, or maybe people prefer the opposite. Even if you had a perfect picture of the FR at your eardrum relative to your own HRTF, that still wouldn’t guarantee matching it would sound best to you.

1 Like

Do you know of some of the scientific studies that’s being done to bridge that gap? Or maybe even stuff that’s been done? I know Apple has outer ear scanning for their Atmos playback for example. There’s the Greisinger ear drum silicone probe for measurements, are there any products or diys like that out today for reviewers like you to perhaps use so the discussions could move further forward? And are you going to try to make molds using different materials from your mri scans and put that in a measurement rig to see if it gets closer to getting your HRTF correct?

Same question as above. This theory (yours I assume?) sounds plausible, it would be interesting to read more about research that’s been done (specifically related to music since lots have been done using tones and noises) or is ongoing if you have any links.

I’m not aware of any specific listening experiments being done on this type of thing. From what I can tell the only public research on all of this is still just Dr. Sean Olive doing it for fun. So the industry needs to do more there in some respect.

As to the other stuff, yeah we’ve been basing these ideas off of blocked canal data from the HUTUBS database (Fabian Brinkmann et al, from Technische Universität Berlin). It’s referenced in one of my recent videos on the subject and you may have already seen that. The thing is, that was just one example. There are other examples that show even more variation, but I wanted to use the best case scenario.

I see. I was hoping maybe you’d know headphones designers or experts who are doing ear drum level sound capture. I thought maybe someone like Oratory would be doing that since he talks about the theory of how everything is just Fr at the eardrums and he’s literally working for a company that designs MEMS?

That sounds interesting, are there any links so I could do some extra reading on these topics?

I don’t have this at the moment, but the references are in the description of this video. To be clear, this is our conjecture on the subject, as Blaine regularly reminds me. But much of it comes from the work they’ve done.

1 Like

What are MEMS?

Thanks

https://audioxpress.com/article/the-impact-of-mems-speakers-in-audio

Solid state speaker driver technology

1 Like

Still playing catch up, and deleted one of my posts from yesterday (on phase cancellation) because it contained some wrong thinking and info on how phase works in EQ.

@MichaelJames, I think both you and Luke had similar questions about some comments by Rob Watts. I haven’t heard his presentation. But there is loss whenever you change the volume of audio content in the digital domain, because the data has to be resampled. This is true in digital EQ as well, which changes the volume of selected frequencies in your audio content.

In my experience, the loss will be more noticeable at lower bit depths, of 16 bits or below. To keep the audible effects of this loss to a minimum, I suggest using a higher bit depth on your audio device. How high you can go depends on your hardware’s capabilities, and how much latency is acceptable. My gear only goes up to 24 bits. So I haven’t tested higher depths than that to see if I can detect any audible improvements.

I used the standard CD spec of 16 bits, 44.1 kHz on my computer for quite awhile, until I realized that the YouTube video clips I was listening to were actually 24/48. So all of the content was being inadvertentlly downsampled to 16/44.1. When I switched my audio device to 24/48, I noticed a subtle, but immediate improvement in the resolution and details I was hearing in the content.

This was before YouTube and other streaming platforms decided to start experimenting with volume stabilization though, and I have no idea what effect those features are having on the audio quality, if any. If I see them though, I usually disable them.

If you can keep the audio content at its original volume, bit depth, and sample rate up to the point of digital-to-analog conversion, that is usually best for preserving all the original details in the recording. But it means you can’t use a digital EQ to adjust the FR for better playback in your headphones.

IMPORTANT FOOTNOTE: When you change the bit depth and sample rate on your audio device, you may also need to reconfigure or reinstall your EQ software to work properly with it. This has been true for me with EAPO. The sample rate of your audio device determines the maximum frequency you can adjust in EQ applications like EAPO.

1 Like

In my opinion, there can also be a loss of some sound quality due to some of the phase characteristics of EQs. And I think it can be audible in some circumstances.

The loss from phase changes in EQ can be mitigated by using headphones that are close to your target FR, and by using smaller, broader band FR adjustments to tweak them. Or by using some of the other methods of FR coloration that I’ve mentioned previously, that either don’t involve EQ or lessen the need for it.

This is probably something I need to do a better job of practicing in my own EQ curves.

By “slows the impulse response”, I’m trying to parallel the language often used when talking about time-domain tests. Here in particular, though, I’m just talking about the slowed rise and fall time on the short pulse. While the short pulse is perfectly vertical (given the sampling rate), once lowpass filtered, it takes a bit of time to reach the peak – “slowing” the response.

Again, “slowing” is “removing the high frequency components that allow for a rapid rise time to the shape of the impulse”. It’s the HF bits that allow for it to be “fast”.

So just for a ‘sanity check’.

Are we saying that most of everything or possibly everything which matters is Freq Response in headphones?

1 Like

Indeed. Provided all else is equal or sufficiently inaudible, like harmonic distortion. And there will be other non-acoustic factors as well that contribute to the psychoacoustic effects.

We keep an open mind to there being other things, like acoustic Z, but at the very least FR at the eardrum is likely to be the thing that matters as far as the acoustic cause of our experiences… and the variable of non-HRTF related HpTF variation is what explains the gaps between measured frequency response and our experiences, including providing reasonable answers to the question of “what is technicalities?”.

Thank you for this reply!

OK; this makes sense to me, and, FWIW, I would agree. Whatever features there are in the driver material or build that are relevant to producing sound waves, the results should show up in the frequency response, the aggregation of waves characterized by frequency and amplitude. Now whether or not we can measure those results at a high-enough resolution, let alone depict them, I don’t know; I suspect we can’t at present, but my skepticism isn’t that important as it isn’t very well-informed.

Very interesting discussion!

1 Like

And I want to emphasize this, that we’re not saying driver design is irrelevant here, since ultimately its parameters contribute to the FR at the eardrum, as is the rest of the headphone’s acoustic design. Is it more modal? is it more damped? Is it well extended? These parameters are all relevant and measurable, and they do play a role in how these products ultimately perform.

It’s just that there’s no extra metric here of ‘technicalities’ that isn’t being measured. The limitation lies in measurement conditions and analysis, particularly that most people doing measurements only do so on one head and one set of ears - specifically not yours - and that we don’t have a strong sense of what people like for fine-grained features.

It’d be reasonable to suspect that a good fit with one’s HRTF is desirable… and indeed it is for me. But I can’t know that for anyone else.