When equalizing a headphone, I often wonder: at which point (in Hz) is it ideal to “anchor” the target line on the frequency response graph?
Depending on the headphone and the chosen target curve, a different anchor point might make more sense. Placing it too high or too low could unintentionally lead to over- or under-boosting/cutting. I’ve seen some people use a fixed reference, like 500 Hz or 1 kHz, which seems useful for comparing frequency responses, but I’m not sure if that’s really headphone-specific enough for EQ purposes.
It also seems to me that there’s surprisingly little information available on this topic, even though it might have a significant effect on EQ results.
I’m curious: does anyone know of a reliable (ideally scientific) way to determine or calculate the ideal anchor or crossover point? Or is this something that mostly comes down to trial and error?
So you want to determine the ideal frequency to have the target and measurement match in SPL when EQing a measurement to that target? When I’m doing this I just eyeball it to where there is the most amount of overlap between the two lines aka the least amount of EQ needed to match the target. Either that or I use 500hz. AFAIK it doesn’t matter as long as you stick to the 200hz-1khz range. Match the mids, basically.
Agree with AudioTool. Align the curves at the frequency or range where you want the EQ curve to be flat. And adjust your preamp/pregain so your PEQ filters don’t clip.
I generally use 633 Hz with my GEQs though. With a preamp.
There are a couple tips on adjusting preamps in EAPO here btw…
Thanks so much for your helpful responses so far. I’m curious if anyone knows of a mathematical or scientific method for this—could it even be calculated? I find the idea of determining the exact “starting point” of an EQ quite intriguing.
Also, in the case of a “cut-only” EQ setting, what would be the best way to choose the anchor point?
I’m not aware of a mathematical way to calculate this. In theory it should not matter, but I would still avoid using frequencies from areas of the FR curve that are not flat.
All EQ is “cut-only” if you apply the required amount of negative pregain and you should always be doing that or you will get distortion.
If there is a scientific method, it would be psychoacoustic rather than abstract mathematics. Mathematics are human tools for understanding the universe, but understanding human biology comes first. Then, the two can often be paired up.
The most functional part of human hearing lies in the male-to-female vocal range. Hearing is tuned to match human voices (e.g., mothers with babies; speech for communication; danger calls and warnings; screams for help and from pain; hunting coordination; etc.).
If you’re using parametric filters, it may be a bit more difficult to design an effective EQ curve using just cuts. AudioTool is correct though that a negative preamp/pregain allows you to use both boosts and cuts, while avoiding clipping.
It’s somewhat easier to do cuts-only with a GEQ. Even if you design the GEQ with both boosts and cuts, EAPO can automatically shift the entire GEQ curve an appropriate amount in dBs along the Y-axis to align the highest points in the curve at the clipping point of 0 dBFS. (This is possible with a GEQ, because it’s just one filter, rather than several. And it can be done with one of the tools on the righthand side of the GEQ filter panel in EAPO’s Configuration Editor.)
It may be possible to do this mathematically with a set of PEQ filters as well. But it’s likely to be more involved. And a preamp filter (with negative gain) will accomplish the same thing much more simply and painlessly.
If you really want to use a cuts only approach to construct your EQ curve though, then you’d probably want to normalize the target and headphone response curves at the point where your headphone would need the greatest boost in its response to match your target. Everything else should be downhill from there, so to speak.
I can demonstrate how to find that frequency graphically, if you want. It would be fairly easy to show something like this in one of the Squiglink interfaces, if @listener would give his consent to post a few graphs for this purpose. I can also show it in EAPO’s Configuration Editor though. Just not quite as well. I cannot guarantee though that this is the best appraoch to use with Squiglink’s automatic EQ features, because I haven’t really had the chance to play with them yet.
There are a variety of different ways to normalize or align two curves btw. I use 633 Hz most of the time, because it’s in the center of the 20 Hz to 20 kHz range of most FR graphs. With 5 octaves above, and 5 octaves below.
Frequency response curves are just a series of points on a 2D graph though, so there are various ways they can be normalized, aligned, arranged, or compared. They can be normalized at a single frequency value. Or over a given range or bandwidth, including the entire bandwidths of the curves. I believe Squiglink gives you the option of using either a single frequency band, or some average of the curve’s entire bandwidth (from 20 Hz to 20 kHz).
Oratory1990’s old graphing tool also had the option of using a specific range of frequencies (from 500 to 2000 Hz, for example), which was also sometimes handy. Especially if you want to exclude some of the whackiness at the two ends of the curves from the normalization process.
Oratory1990, Crinacle, Super* Review, or the gal who designed the Squiglink interface might know about the specific maths used for something like this. You can also ask Google AI (or your AI of choice) how to normalize two curves on a graph for comparison, and it will likely give you several different methods. Min-Max, Z-score, Divide by Maximum, and Normalize by Area appear to be some of the standard methods. Whether one of those would work the way you want though, I don’t know.
I cannot guarantee though that this is the best appraoch to use with Squiglink’s automatic EQ features, because I haven’t really had the chance to play with them yet.