[anthell]

Lets begin with what people want to achieve - a flat response. I looked at a bunch of frequency graphs reponses to various headphones and IEM and it got me thinking. Those frequency response graph usually are based on a specific dB level and the general consensus seems to be good headphones generates flat graph. Looking at the graphs in headroom.com, they are generally produced as a response to a 30 dB flat tone. 
 
I assuming that my understanding is correct, those graphs are generated by listening to the output of the actual headphones / IEMs. 
 
Take for example earphone X with a generally flat response but a peak @ 7.5kHz (+10 dB @ 7.5 kHz), can i assume that this +10 isn't fixed but instead a variable ? its basically +33% of the original sound level? so that means @ 20 dB the increase in sound level would not be 10 dB but instead 6.6 dB? This is because my understanding of ear canal resonance and IEM / headphone resonance seems to suggest its highly unlikely if i supply a very quiet 1 dB sound it would increase it flat out by 10 dB.
 
Now lets assume a -10 dB EQ @ 7.5 kHz
Imagine a song with exactly the same dB level @ 2.5 kHz and 7.5 kHz ( 25 dB at max volume by this certain set up)
 
The IEM on its own would amplify it to:

     i. 25 dB @ 2.5 kHz
     ii. 33.3 dB @ 7.6 kHz --> 25 dB + 33% 
 
This is the second possibility with EQ:

     i. 25 dB @ 2.5 kHz
     ii. 20 dB @ 7.6 kHz --> (25 -10) dB + 33% 
 
None of the results are the wanted response of 25 dB @ 2.5 kHz and 7.5 kHz. The EQ would make the 7.5 kHz to be quieter than it should have been.
And we can tell by that example, if the sound had ben a 35 dB tone, the 7.5 kHz would still be quieter than the 2.5 kHz even after EQ (33.3 dB vs 35 dB). This is assuming my logic is correct that the EQ pre-process the sound level and the actual peaks on the IEM boost it post-process.
 
This is even more detrimental if let say the particular song suppose to generate a 10 dB tone. 
 
With EQ:

     i. 10 dB @ 2.5 kHz
     ii. 0 dB @ 7.5 kHz --> (10 -10) dB + 33%
 
The actual sound of the 7.5 kHz would be lost altogether. This is especially true with music recorded at high dynamic range instead of compressed dynamic ranged. A song can have a 40 dB level @ 2.5 kHz and 5 dB level @ 7.5 kHz , and the EQ would wipe out some parts of the song. What should have been the little quiet and fun details would instead be lost.
 
Please tell me which part of this article is wrong so i can have a better understanding of how to EQ my IEM which happen to be Shure 535 LTD-J. I have been messing around with EQ of multiple settings, with different songs, and i somehow still find it imperfect. Also, i generated my own frequency response graph not those found on the internet by actually listening to tone generator to calibrate it. 

 

[castleofargh]

how did you assume that the signature response was in %? because well... it's not.
 
also most measurements are done at 90 or 100DB. I would be very surprised to learn that headroom does it at 30DB like you suggest.
 
 
 EQ itself is something pretty straightforward, if you remove 5DB it should lower said frequency by 5DB. no % in the mix.
there can be a bunch of reasons why you don't get the results you're expecting, but it has probably to do with distortion, clipping, crossovers, physical limitations of the driver, or simply the fact that your ear is different from the measuring head that was used to make the graphs.

 

[cjl]

The thing is, dB is already based on a ratio. If something is 10dB louder than something else, that's the same ratio, regardless of whether you're talking about 0dB vs 10dB (oh, and 0dB isn't zero output, in case you didn't already know) or 110dB vs 120dB.

 

[mikeaj]

It's kind of difficult to go line-by-line understanding the first post because a couple key assumptions need to be addressed, so let's just go with some basics:
 

• A decibel is a logarithmic unit to express a ratio. 3 dB is twice as much, whether it's the difference between 20 and 23 dB or between 95.3 and 98.3 dB, for example. A decibel by itself is not really a unit of any physical quantity, just of relative amounts or ratios.
• Sometimes decibels are implicitly used to indicate sound pressure levels (loudness), referenced to 0 dB (level of 1) equal to 20 micropascals, the lower threshold of human hearing in extraordinarily quiet circumstances. If somebody cites a decibel amount for loudness, they're using this definition and assuming you understand what the reference level is. This is how a "60 dB (SPL; sound pressure level)" actually means some physical quantity. Zero and negative values on a dB scale are fine and just indicate lower levels, not anything being wiped out.
• A graph of headphone response could be produced at any loudness level, though it's more convenient to do it at louder volumes so measurement mics have a stronger and easier signal to deal with relative to the background noise or whatever else could be happening. When the response is graphed, the level chosen is arbitrary. The IEMs will produce the same relative loudnesses if the levels are all shifted up or down 10 dB or whatever.
• Songs aren't supposed to produce a certain loudness. That depends on the playback hardware and volume control, and normal listening levels would generally be a lot more than 30 dB.

 
 
There are a lot of details about how IEMs are measured, how the raw data is processed and graphed, how they sound for you, and so on that complicate things, but the actual adjustments should be straightforward in concept. If you think some frequency is 5 dB too loud relative to others, you can set  the EQ down 5 dB at that frequency and the result would be the level you want. That's it.
 
More or less you should be warned that equalizing by inverting the FR graph may not do exactly as you intend (though it's maybe a starting point). And also, equalizing by trying to match the loudness of all frequencies by listening is definitely not what you want (leads to a very colored response), so don't do that. This is because the ear naturally picks up some frequencies better than others. If you try a tone at each frequency at the same SPL, they will sound louder at some frequencies than others. Also, the relative perceived loudnesses depend on the volume you're playing them at—though the IEMs and electronics scale properly at different volume levels, the ears don't.
 
Best would be to use calibrated speakers and compare relative loudnesses coming from those, which of course requires being careful of room interactions. Then adjust an EQ to make the IEMs exhibit the same relative loudnesses at different frequencies to you, at least for the midrange and treble.

 

[anthell]

1. how did you assume that the signature response was in %? because well... it's not.
 
 - I assume this because otherwise let say there is this song that has completely flat graph beyond the 5k mark, it shouldn't increase the 7.5k mark by 10 since it starts at 0 ? and the +10 signature also take account of ear canal resonance?
 
2. If something is 10dB louder than something else, that's the same ratio, regardless of whether you're talking about 0dB vs 10dB (oh, and 0dB isn't zero output, in case you didn't already know) or 110dB vs 120dB.
 
- wait... so how does it work exactly? 0 dB isnt zero output?
 
basically.. how the graph in my IEM specifically looks like this...
 
 -around 5 dB on the bass
- around 1 db on the 700 Hz mark
 +around 5 dB on the 1.2k mark
 -around 5 db on the 6k mark
 +around 8 dB on the 7.7k mark
 
any tips how would i do this? people say i should not ever increase those negative peaks since it generates distortion but there are some zones that are negative some are positive...

 

[cjl]

 
- wait... so how does it work exactly? 0 dB isnt zero output?

Nope. 0dB just means that your signal is the same as the reference level. For sound pressure level, the reference level is defined based on the threshold of human hearing at 1kHz, equal to a pressure amplitude of 20uPa. Because the 0dBA level is basically defined to be the threshold of hearing, all decibel numbers when dealing with sound pressure level (SPL) will tend to be positive. However, in other cases, you might use something completely different as a reference level, which can result in negative values. Many home theater receivers show their power output in dB, but they usually reference full power when doing so, so a volume level of 0dB is full output. In that case, reasonable listening levels might be -30 to -15dB (on the amplifier) or so. You could even use the input signal level as a reference if you wanted (I'm not aware of any amps which do this, but it could be done). In that case, 0dB would mean the output signal is the same level as the input, with negative values referring to an attenuation of the signal, and positive values referring to an amplification.
 
Wikipedia has a fairly good article on decibels (http://en.wikipedia.org/wiki/Decibel) and sound pressure level (http://en.wikipedia.org/wiki/Sound_pressure), both of which contain a lot of detailed information on how all of this works.

 

[esldude]

  1. how did you assume that the signature response was in %? because well... it's not.
 
 - I assume this because otherwise let say there is this song that has completely flat graph beyond the 5k mark, it shouldn't increase the 7.5k mark by 10 since it starts at 0 ? and the +10 signature also take account of ear canal resonance?
 
2. If something is 10dB louder than something else, that's the same ratio, regardless of whether you're talking about 0dB vs 10dB (oh, and 0dB isn't zero output, in case you didn't already know) or 110dB vs 120dB.
 
- wait... so how does it work exactly? 0 dB isnt zero output?
 
basically.. how the graph in my IEM specifically looks like this...
 
 -around 5 dB on the bass
- around 1 db on the 700 Hz mark
 +around 5 dB on the 1.2k mark
 -around 5 db on the 6k mark
 +around 8 dB on the 7.7k mark
 
any tips how would i do this? people say i should not ever increase those negative peaks since it generates distortion but there are some zones that are negative some are positive...

cjl is rightly calling 0 db the basic threshold of hearing.  I think you are referring to frequency response plot of a headphone. 
 
Lets just say I am Bob the headphone maker.  I need a plot for my customers.  I pick 1000 hz for my reference frequency.  I decide to play that 1 khz tone at 90 db absolute sound pressure level relative to a defined 0 db.  For my graph my 1 khz 90 db tone gets marked as a reference which I plot at 0 db.  Let me say I play a tone at 5 khz and it is +7.5 db louder than the tone at 1 khz.  Will get graphed as +7.5 db while having an absolute SPL of 97.5 db.  Now let us say at 20 khz I get -5 db compared to the 1 khz tone.  It gets graphed as -5 db and will be absolute SPL of 85 db.  This is how a response graph of your phones has a 0 db that isn't 0 db in the sense of being at the threshold of hearing.
 
Now your example numbers would all look like you could boost the dips and cut the peaks assuming the boost doesn't overdrive the electronics.  Those numbers are relative to some arbitrary reference level and frequency which is well above true 0 db SPL. 

 

[castleofargh]

DB is used because it's convenient. so we make the best of it and try to use it for everything, dynamic, volume level, even to talk about digital values(when it's really just volts or bits at some point). as long as it can translate later in DB in the headphone we're good.
 
when we talk about a signal at 0DB we're usually talking about the loudest it will be on the record, not in the headphone. like music on a CD can go from 0DB to -96DB at the quietest, giving a dynamic or +96DB. that's simply because of how the DAC will handle the signal. 0DB is always the loudest signal a DAC can send, it doesn't matter how many volts that is, or how loud the sound will really be in the headphone. it's just a conventional value for the DAC meaning "I won't do higher that that". if you try to make a signal go higher, the DAC won't be able to do it so anything above zero will clip(get crushed at zero).
 
when you EQ to rise a frequency, there is always the risk this would happen. let's say the signal on the song get close to -0DB on some 40hz bass(some good strong rumbling bass), and only the IEM specs/defaults makes you hear it like it's -25DB (often because the BA driver fails to recreate such a big wave with the same amplitude). then setting your EQ to +25db in that situation would clip the frequency anytime the signal is between -25 and 0db. because you're telling your DAC to take a signal that was already close to the max(0db) and add 25db to it. something it cannot do so you end up with clipping all that signal to 0DB, making a flat head to all your loud 40khz sine waves in the music ^_^.
that's why one way or another you must make sure that the most boosted frequency on you EQ doesn't get past 0db. some EQ take care of it for you, others don't.
 
 
 
now back to measurement graphs, as I said they're usually done with the headphone sound at 90 or 100db. and the graph is often set for readability with the signal at 1khz as the zero value. it has no relation to the actual loudness. only the variations over frequencies have a meaning.
but from what you're writing, I suppose you don't know about the compensation curves. flat on a graph does not mean it will sound flat for you.
look around for "diffuse field compensation" it should explain a few things about those graphs.