How much does frequency response actually matter?

So this is a topic I've been sitting on for a while beacuse I've been trying hard not to be too snarky and, at the end of the day, I am genuinely ignorant. Unlike a very many folks on the forum and elsewhere I have no real engineering knowledge and very often get principles backwards or things outright wrong (e.g. "The Studio Jr 300B sounded *significantly* better out of XLR out so it *must* be a balanced topology amp... whoops, nope!).

Now and again I feel that ignorance can be an asset in a sense, mainly in how I pepper folks who actually know things with often silly questions that, hopefully, might generate meaningful and useful discourse down the line. At this point in time I think I'd have hoped that I'd know more concrete things about how amplifier or DAC topologies work or what sort of fine nuance actually goes into developing the analogue output stages on converters, but nah I'm as much an idiot as I've ever been and with an increasingly worsening memory.

I've got vague ideas of how pairing certain DACs and amps and transducers work, but is the ENTIRE reason that the Sennheisers love those big boi tube amps just because they can provide GOBS of voltage? Is all of what can determine how much a headphone can scale just a matter of power handling and impedance matching, accounting for those massive low-frequency humps with dynamic drivers? How about how much sense of SLAM an amp can give a transducer like that one oldish amp I got to borrow that made my HD600s kick like Focals, is it all just a matter of the OI of that amp being REALLY high so the Senns were severely under-dampened? Is headstage really just an artefact from an upper midrange dip? What about image delineation?

Why do different amps make as much of a difference as they do with planars then, which measure flat as a typical anime protagonist's character arc? Is it just that planar amps can run way more current without melting? And for that matter how much do those little jaggedy ripples you often see in unsmoothened FR traces for lots of planar headphones actually matter? Could those be indicative of how many planars tend to exhibit an undetailed, hazy/hashy/splashy/plasticky timbre up top?

One thing that comes immediately to mind in this conversation is how @GoldenOne recently posted a video with headphonesdotcom where he managed to validate via ABX that yeah, DACs (or more pedantically, DAC *filters*) can be appreciably, audibly different:



Okay, shocking news to the forum DACs and even DAC filters can sound different. Wow. Contestable as that statement is, I think we can all agree with Golden's conclusion that it's all just a matter of FR then? But wait, basically any modern DAC that isn't busted to hell measures like the Maldives (i.e. flat as a pancake) with maybe the occasional design exhibiting disgusting harmonic distortion spikes when measuring SINAD at... -90dBFS while pushing out 2VRMS. Ah, significantly below my noise floor even when listening at louder levels. Odd.

Well it turns out that all the DAC filter differences can be easily ascribed to ultrasonic activity! Heck, lots of IC manufacturers have in recent years seem to have taken to attempting to address grass above 20kHz, so hey that could support the hypothesis that everything to do with DACs from presentation of stage to refinement of treble to sense of macrodynamic impact to detail retrieval to timbral differences to background clarity and even trailing decays can ALL be ascribed to FR (presuming distortion levels are well below threshold of perception of course)!

And let's also bring the HATS vs flat plate coupler conversation back around because sure, some headphones will more deliberately elicit interactions with pinnae which will drastically alter measured FR relative to how others work. FR derived from flat plate measurements are useless then, right, and only ought be used as a supplementary point of reference and *only* ever be used to compare to other measurements on the same rig? They ought have no real crossover with what's perceived because everything from the acoustic impedance encountered by the transducer to the shaping of mid and high treble response (anyone still remember Jude vs Tyll re:the MDR-Z1R's treble?) on such a non-human-representative rig will be *incomparable* to what we actually perceive! What does HR(Transfer Function) stand for again?

Okay I'm turning the sarcasm off now.

To touch on distortion briefly (because I'm well and truly aware that I'm rambling at this point nearly) it ought be a criminal offence on here to not refer to @atomicbob 's great and painstaking contribution to the conversation about how SBAF's prior preferred method of visualising distortion as plotting D2, D3, and D4 across frequency, while leagues above how lots of folks just refer to SINAD/upside-down THD+N, still didn't fully capture how we might realistically perceive distortion because music and such are NOT steady-state signals. Distortion is also a part of the conversation of perceived sound, but one that many assert *must* be a solved problem owing to how high-feedback designs push things down to inaudible levels with NO deleterious effect to sound whatsoever (... sorry sarcasm off for real this time).

Also, the R70x Redux thread reiterates how much @purr1n 's work with burst responses is equally important, if even more contentious because it deviates even more from what's commonly accepted as conventional wisdom among many well-educated audio engineers that time-domain elements such as CSD are useless for headphones as it's all resolved in FR anyway. I read it on Reddit so it HAS to be true! (sarcasm, sorry)

Sincerely, I dislike EQing headphones. It can definitely help in many instances but whether it's because there's inexplicably a difference in sound when absolute polarity is flipped (that's ANOTHER contentious thing to lots of well-educated folks it seems) or placebo and stubbornness (I'm not above admitting I'm prone to both), it just sounds/feels/is wrong to me. Also, it just seems to me that folks who say to EQ everything are putting the hay before the cart-- I've always seen FR as being a symptom of physical realities, but many seem to act as if it were the other way around and you could just use DSP to try and alter how a physical thing sounds in reality, constrained at least a bit by distortion and such.

But no those old EQs were all based on measurements from older, inferior couplers, the 5128 presents different measurements versus what the older GRAS stuff did so THAT must be the truth! It's significantly more accurate a model than anything that precedes it (and it's $41,000+!)




... Hmm, wait weren't some folks complaining about the treble smoothness on the HD660S2? That actually looks rather good, if a bit muted!







... Ah. Right.


So, thoughts on FR?

 

I’ve been thinking about this a lot, too, lately, especially since some headphonescom stuff. Like the video you posted — I can easily hear the difference between HQPlayer filters (at least between some of them, some are very similar, too), but I can’t hear above 20k.

And, even stranger to me, that headfi thread where Resolve makes clear that there’s nothing but FR, and every feature like resolution or stage can just be EQed in if you know which regions to adjust.

That sounds completely wrong to my ears, but I don’t know enough about the subject to be able to prove him wrong or even be sure I’m right.

I hope knowledgeable people will chime in here and explain it all to us.

 

One stupid(?) question I have is how distortion would be represented in FR. In sweep tests it isn't obviously as you're playing/measuring one exact frequency at a time.. but what if you did sweeps overlaid with a constant 20hz tone, 21hz, 22... and every combination of tones possible? This is obviously impossible without an extreme degree of automation and still would need to be simplified significantly to not take an infinite amount of time to test; would such testing better represent a headphone in use than one sweep or single frequency distortion test?

This line of query is assuming FR morphs due to contributions of a mix of harmonic/intermodulation distortion, resonances, & comb filtering as different frequencies are played simultaneously & that this interaction is different with different headphones(it may also extend to electronics.)

A simple example of what I'm saying is that if you ran a sweep with a constant 50hz tone you'd see higher volumes as you hit 100, 200, etc tones due to the harmonics of 50hz than with a pure sweep. Bumps in the FR correlating to other tones present. Right? This seems more like what would happen during real use.

I'm rather technically illiterate on these topics too but if there's any validity to this real-use apples-to-apples testing is impossible as the only situation one product will behave predictability to another is in artificial testing scenarios with a singular frequency playing at a time. This would certainly indicate the limited use of FR graphs for determining performance beyond tendency(dark/bright) & specific pain points(Beyer spike.)

 

No clue honestly, I do think that there'd be merits to presenting tonality via RTA as opposed to frequency sweeps (which I remember was also done by others before) but I'm actually rather unsure of how common practise that is for whatever reason; maybe takes too long to get a high-enough resolution capture that environmental factors might end up more easily polluting the data?

 

I think a great part of it is just that people go for the whitepapers without having had the necessary experience to translate sensible things into what might be perceptible.

@purr1n it seems that there's a growing contingent of folks in the recording space who use impulse responses to try and mic-match things, e.g. make an AKG C414 sound like a Neumann U67 (cuz hey at least they're both condensers so ought be similar tech, no? Easy!). One guy sent four recordings in: these were of a TLM102, a TLM103, a TLM102 EQed to sound like a TLM103, and a TLM103 EQed to sound like a TLM102.

The hilarious thing is I got them all wrong, i.e. without a baseline I had no idea which sample was of the dry 102 and which was the dry 103. The fun thing is that I did not confuse any of the dry samples from EQed ones, so even within pretty similar designs there's something different about the sound of EQ. I know the topic of linear phase vs minimum phase EQ is beaten to death now and it's possible they could have just used a better program and completely fooled me, but there's just something to making a diaphragm act differently than it is naturally predisposed to that makes it sound WEIRD.

None of that even approaches how headphones are wholly different things, often outputting more complex signals than most mics record all the while interacting with relativelt large and complex structures e.g. pinnae and tufts of hair along the sides of head. Chaotic it seems.

Also, distortion is a MASSIVE part of the picture too but what about time domain elements that so many well-read folks seem to dismiss as meaningless in headphone playback?

 

Eh you put the ideas out into the world and now it's up to others who might actually have the time and motivation to see where this goes. I do plan on investing in *some* source measurement gear when I have my headphone rig in a place where I can say that I'm close enough to DONE chasing better and better because it's all super interesting stuff, but I do hope those with more resources at their disposal might beat me to the punch.

Also, love that you brought the older visualisation for burst cycles back here, I still find those significantly more intuitive than the multicoloured attack and decay things (even though the latter bunch are honestly more informative and look cooler). I think that distortion is understood and accepted enough in the current space that it's more transient response and time domain elements that need proper investigation by the erudites.

The decays of e.g. the ESP950 goes a LONG way towards describing how I feel those estats sound so weirdly bouncy! Could you be bothered to maybe do similar for the JAR600 vs a stock one? Curious how the subjective differences there'd align (or not, who knows!)

 

Oh hell yeah. I know nothing about software really-- I love reading the threads on here but real talk you, ultrabike, scblock, and loads of others just post things that I skim, try to comprehend, smile and nod at, and ignore in favour of how they might have an audible effect on things, which I do understand. There are enough memes on the internet about how many systems nowadays are running on legacy code that's impossible to fix pending a complete rework from the ground up cuz of how the foundations were written by some dude in the 70s or 80s and absolutely no one else knows how the Rube Goldberg machine works.



Yeah, I know this statement is gonna age me but it does seem obvious that folks nowadays just want information spoonfed to them. Even taking a few minutes or maybe a couple of days to really learn how to read different graphs is more effort than can be warranted when trying to integrate insightful objective measures into the "which headphone?" discourse.

And hey with how things are nowdays it's REALLY hard to blame folks who want to chase the money. There are a few people on planet who have way too damned much of it, but orders of magnitude more who you can't blame for wanting more. Real curiosity doesn't always prove useful I guess.

 

So we have acoustic measurement devices to measure the transducer during a burst response... but can we also measure the actual output of the testing device? Do they generate a clean start/stop of the burst or is there some garble there too? If so, how much of the garble is affected by the load?

 

Sewing a few things together...

1) If the influence of ultrasonics on things other than FR down in the audible range turns out to be important - as suggested by apparent audible differences in DAC filters - then in practical terms the measured FRs of transducers are never going to tell you anyway, because the responses of measurement mics don't extend high enough.

2) Connected with this, if we use FFTs of frequency measurements to analyze time-domain responses and want to see how those vary with frequency, we also lose high-frequency information: mathematically, the top 50% of the samples are lost during the transformation.

3) There seems to be evidence that humans' temporal perception of sounds is really good. It's worth directly quoting parts of MN Kunchur's response to a criticisms of his work that were referenced in a recent thread here:

So parameters purporting to characterize perceptions of temporal phenomena but which are obtained from frequency-based measurements aren't going to be precise enough anyway.

3) Distortion - yep; see my sig. None of this is new.

Having said all that though, I do think the effect of FR on our perceptions of some sonic characteristics deserves a bit more credit than maybe we're inclined to give it here. Staging, for one, can be very dependent on troughs and peaks in the upper mids through treble. These days Sennheiser seems to be very deliberately using a right-shifted 'BBC dip' in some of its products (certain recent IEMs, and the HD 620S - or HD 620W, as it really should have been named) to try to achieve better perceived separation.

 

Thinking it's case of certain transducer types presenting a more challenging load to amplification ergo the resultant sluggishness in coming to stable performance? That's actually a really good point given how different headphones do appear to stress amplification a lot more than others, I'm thinking that maybe sticking with one transducer type ought help at least eliminate that from consideration, not to mention yeah checking the source (somehow?) for any feedback from the transducer. Thinking that could end up bollixing the signal in itself though, sort of like observer effect, slapping something in the signal path.

I do agree with this much cuz of how e.g. some Sennheisers and LOTS of the HiFiMAN headphones do have those pronounced upper-mid troughs and all seem to throw a large stage out. Not quite sure how this'd necessarily tie in with imaging performance though, which I do view as being adjacent but not the selfsame thing as staging-- delineation on some things that stage diffusely can be pretty great or nah. I'm looking kinda specifically at the Campfire IEMs for this where both the Andromeda 2020 and Solaris have relatively large stages for the form factor; the Sols are pretty pinpoint in localisation whereas things are just generally diffuse on the Andros out of the same gear. Granted though, there ARE FR discrepancies between them so I could be eating crow on that.

Rather curious still at the mechanisms behind how high frequency ultrasonics might have influence on perceived slam and stage, among many other things. I know that you do need a fair bit of high frequency information present to make a drum sound *snappy* and aggressive, it's never just more bass, but that's all within audible bandwidth I think so not sure how much of that'd apply to here.

Myself the moron asking questions again but I'm aware that many opine IR can be directly correlated to other changes in FR; given that, is it possible to directly translate pre-and-post-ringing in DAC filters to FR?
https://www.superbestaudiofriends.o...d-post-effin-ringing-and-shit-like-that.2627/

 

I can just barely still hear so 17kHz so yep I'm with you there, but I get the feeling that at the very least an argument can be made that ultrasonic sound is still "information" that can saturate a system e.g. a transducer and sort of distract from what else that component could be doing.



Screenshot of what a snare drum hit from one of those high-res Chesky Records things (Dave's True Story - Misery, I actually like this album haha) looks like in visualisation here. I wonder how much smoother this'd be if it were a regular redbook recording and whether that'd influence transducer motion any?

n.b. this is not a measurement but one of the stock foobar2000 oscilloscope visualisers so it ought be what's being shoved into my DAC.

 

I looked up the thread: https://www.head-fi.org/threads/frequency-response-at-the-ear-drum.972580/

I respect Resolve as one of the best YouTube reviewers I know, and I think that he’s really a nice guy, but that stuff is strange.

Disclaimer: I only read the first few pages of the thread, so maybe there’s something further on that puts that into perspective.