Frequency response

back to Learn
back to How to interpret graphs

post separation

Frequency response deviations, as well as phase response, are so called ‘linear distortions’. Phase response is not covered in this topic.

Linear distortion does not produce new (unwanted) signals, contrary to non-linear distortion. When a headphone, speaker, amplifier or DAC has little to no linear distortion the frequency- (and phase-) response is flat.
All frequencies are there in equal amplitude and in the same phase. When this is the case AND non-linear distortions as well as the time-domain issues not being there the reproduction of the original recorded signal is perfectly reproduced.
This does NOT mean the reproduced signal sounds like a real-life recording. In fact a recording may sound quite poor but can also sound extremely nice.
How good a particular recording sounds, under these circumstances, depends completely on the recording itself. ALL aspects of that recording and those are MANY.

Deviations in the frequency response (linear distortion) largely determines the tonal character of a headphone.

Certain linear distortions in the frequency response are often preferred over no linear distortion (flat frequency response) by most listeners. This can depend on factors like:

  • taste/preference of a specific sound signature/color
  • music genre (also a form of taste)
  • listening level (loudness)
  • circumstances/surrounding
  • mood

Think of preferences like:

  • boosted bass
  • rolled-off treble
  • scooped out mids
  • bass light
  • elevated treble

Some poorer recordings may sound somewhat better on headphones with a specific coloration such as ones with elevated bass, ‘warmer’ or ‘colder’ sounding headphones or headphones with a dip between 1kHz and 5kHz (takes the ‘edge’ of recordings).
Headphones like that, however, may sound ‘exaggerated’ in certain aspects or ‘veiled’ with well made recordings though.
This thus is a very good reason to own 2 or more headphones with a different tonality so you can pick one based on your taste.
A one headphone does all perfect solution does NOT exist unless all recordings were perfect and all tastes were the same.

One should chase their OWN personal preferences/sound signature and not those of others. What others might find ‘perfect’ may sound as an abomination to others/you.

That being said, my personal preference lies in ‘flat’ with a few dB (sub)bass boost to compensate for the lacking tactile ‘feeling’ of bass. Believe it or not but low frequencies are also picked up by nerves in the skin which ADDS to the low frequencies we hear with the ears. This ‘lack’ of sensory input can be more or less compensated by adding some extra bass. At least for me this helps.
For me otherwise ‘flat‘ (+ that small bass boost) sounds as perfect as can be expected with well made recordings and my personal taste in music. Therefore this is what I chase and hope to find. The plot below thus sounds good to me with well made recordings. When poorer quality sound is heard then I realize it sounds poorer because the recording is poorer.
Sometimes I then reach for a different headphone or skip the song/album.
When a headphone would exist with a flat frequency response as shown below in the plots on this website it has slightly elevated bass in reality.

graph-bass correctionNo headphone nor speaker will ever have a response as shown above. Amplifiers and DAC’s could well have a response as shown above.

All frequencies should be reproduced equally loud when the same amplitude signal is applied. This is something that is easily achieved in the electrical domain by amplifiers, and music sources. At least all good equipment should at least have a range between 30Hz and 18kHz without a deviation relative to a signal of 1kHz smaller than 0.5dB. Differences of 0.1dB are almost impossible to detect. An amplitude step of 0.2dB is detectable by most. A gradual sloping frequency range that drops of a maximum of 0.5dB at the extremes of the range (20Hz and 20kHz) is considered to be indistinguishable by most people. A difference of 3dB is already clearly noticeable and 6dB is a halving of the amplitude. For the human hearing a 10dB difference is considered a halving (or doubling) of the perceived loudness of a signal. Strangely enough a 10dB doubling is not perceived the same as a 10dB decrease in amplitude due to the way our hearing works.

Sonic characterization

Loudspeakers (the better ones) can be reasonably ‘flat’ from 50Hz to 20kHz within +/- 3dB. Top of The Line speakers may even go from 30Hz to 30kHz within +/-1dB when measured in an anechoic room. It is impossible to reach those values is a normal (listening room) though. For headphones it is even harder, if not impossible, to create a flat frequency response over the entire frequency range for many physical reasons.

In reviews a certain terminology is often used when we speak off bass, mids and treble e.t.c. All of these ‘regions’ have a specific band in the audible range. The graph below shows where these sonic descriptions are in the frequency range. This will help to understand how the terminology is used to describe if a certain ‘range’ is lifted (peaks) or is lowered (dips) with respect to the average signal.


25Hz is about the lowest frequency that can be heard. It manifests itself as the deep rumble that can be heard in the movies theater. 32Hz is about the lowest bass note in most music. Only some organs can go lower. Still we can ‘perceive’ even lower frequencies just not as a ‘tone’.
18KHz is about the maximum most people can perceive as a single tone, some younger people might be able to detect up to 20kHz. For most adults about 16kHz is the maximum. This is not perceived as a tone but rather as a ‘pressure’. Some research has shown related harmonics above 20kHz can be perceived by human hearing when single test tones are used.

frequency bands

I am not aware of any scientific sound research that has shown humans have this ability when listening to music which is more complex and the brain might take some ‘shortcuts’. Plenty of anecdotal reports can be found though, but these are only based on subjective findings.

THIS interactive chart shows what frequencies are present in various instruments.

Even though specifications of headphones always state optimistic figures like ’20Hz to 20kHz‘or even ‘5Hz to 35kHz’ these numbers are often meaningless and should be ignored unless it also states the ‘cut-off’ points.
Those are the points at the extremes of the frequency range where the output signal has dropped by a certain amount of dB’s. Those stating +/- 3dB are the most realistic even though 3dB is already more than barely noticeable. Most headphones, however, are specified at -10dB and those that do not specify these cut-off points usually quote their numbers at -10dB, or -20dB to create more impressive numbers. 10dB is already a halving in perceived loudness and 20dB yet another halving.

No headphone is perfectly flat from 20Hz to 20kHz. The best ones out there may vary ‘only’ +/- 3dB. Most of the headphones around vary MUCH more and can have peaks and dips of over 10dB, even 20dB is not uncommon while the better ones generally keep the variations within +/-6dB. This is the main reason why all headphones sound so different from each other, because of these relatively large differences in frequency response. It is also why the frequency plot reveals the most when it comes to the sonic character.

One of the ‘flattest’ headphones around is the Beyerdynamic DT150 with DT100 pads: 20Hz to 20kHz +/- 3dB (not counting the dip at 3kHz)

FR DT150 with DT100 velours pads

Since human hearing is pretty sensitive to human voices a specific part of the frequency range should be pretty ‘flat’ to sound accurate, that is to have a horizontal and not wobbly response in that area. If the lower part of that spectrum is lifted (even just a few dB) with respect to the higher frequencies the voices sound ‘warm‘ (full bodied) sounding. Deep male voices sound that way. If the lower part of that spectrum is lower in amplitude than the highest part of that frequency range we say the voices sound ‘cold‘ (thin sounding). female voices are generally associated with this. When voices sound neutral that specific part of the frequency range is horizontal (flat).

voice colorNOTE: the tilted lines are shown exaggerated and less steep sloping plots (less dB difference) will also show this effect. The plot above merely shows the general idea and in reality the slope may differ or look quite different. It is the general idea of a sloping curve in that specific area that counts.

Of course there is also the possibility a headphone is tilted upwards towards the lower and higher part of this band (100Hz to 6kHz) in which a headphone is found to be sounding ‘sucked out in the mids’. The DT990 below is a nice example. Bass is slightly but not terribly elevated as well as the treble. Relatively the mids between 500Hz and 3kHz are lower in amplitude. As the frequency response between 100Hz and 2kHz is sloping downwards the sound is ‘warm’. The part between 3kHz and 6kHz is higher in level again makes it still have good clarity (clarity resides between 2kHz and 6Hz) while still having a ‘warm’ signature.

dt990-600 nieuwe meting

Also the opposite can be there, tilted downwards in the lower part (300Hz) and lower again in the right part (3kHz) in which a headphone is found to be sounding ‘mid centric’ (a bit ‘cuppy’ sounding). An example below.. a fake MSR7.msr7-fixed

In short … NO headphone (nor speaker) is perfectly ‘flat’. The majority of them is even FAR from flat. Therefore ALL headphones add a certain ‘color’ to the sound.

All headphones (even the most expensive ones) thus have a kind of curved (more or less wobbly) frequency response with peaks and dips that may be present at certain point(s) in the frequency range. As long as these peaks and valleys (dips) are gradually sloping and not too big in amplitude (+/- 3dB) they are hard to detect as such. Sharp peaks AND dips are usually the result of a resonance. A resonance is a small part of the frequency range where the driver/cup likes to vibrate by itself.

Sharp and or high peaks (and sometimes a resonance that manifests itself as a dip) is detrimental to the sound quality.

A rather weird phenomenon is that human hearing is less susceptible to (gradual) dips even if they are substantial, yet peaks in the frequency range are easily heard and perceived as coloration of the sound.

graph sharp peak

The plot below shows where the typical sonic coloration of headphones is located. When the peaks are not too close together in the frequency plot that is. When for instance the part between 4kHZ and 15kHz is raised opposite the part between 200Hz and 4kHz most people will find the headphone to sound ‘detailed‘. It really isn’t in reality, it’s just bright and initially perceived as very detailed but often will be perceived as ‘sharp’ once accustomed to it. ‘Details’ are found in the same region as ‘piercing’ is. A gradual downwards sloping FR from 10kHz is usually described as ‘smooth’ sounding is there aren’t too many sharp peaks and dips. The side effect is that the finer details aren’t as present as when it doesn’t slope downwards in that region.


In short….
are not easily noticed and mostly do not harm the sound in a very negative way.
(of over 3dB) are easily heard and define the sonic character, more often than not in a negative way.
Dips in the 200Hz region for instance will give the sense of a ‘dis-attacted’ bass but generally sounds clear with big bass. A peak, however, will give it a boomy/muddy sound. 
A dip in the 2 to 3kHz region will give the headphone a laid-back sound somewhat lacking ‘bite’ and is generally find pleasant where a peak in that same region gives the mids a sharp and ‘shouty’ edge to music.

the importance of scales

When looking at (frequency) plots the first thing to watch for is the used scales.
The horizontal frequency scale should go from at least 20Hz to 20kHz.
The vertical dB scale (and its divisions) varies considerably between different websites and headphones can appear to be very ‘wobbly’ to ‘very flat’ because of this.
When comparing graphs take a good look at the scales and divisions. An example is shown below.
The upper plot has a 5dB/division scale and looks a bit stretched out vertically where the bottom plot is the EXACT same plot except for a different vertical (dB) scale. The bottom plot has 20dB/division.
When looking to buy a headphone I am willing to bet the bottom plot will generate more sales.

It’s also the main reason why lots of people say these plots are meaningless. Graphs, however, are anything BUT meaningless when you know HOW to interpret them and they do tell the biggest part of the story.
Only a frequency plot doesn’t tell the whole story but one needs to take a few different types of plot into account to get the bigger picture of how it behaves.

The scale of the plots I use for frequency and distortion is not chosen randomly.

FR DT880

The chart above bottoms out at around 35dB SPL. Others may have plots down to 0dB or even below it. 0dB SPL is just a value and does not represent 0 sound pressure.
The reason I chose to use 35dB as the bottom value is not because of one of standards prescribe this but based on experiments I conducted a long time ago.
I found that when listening to music with average levels of 80 to 90dB SPL (which is quite loud) and then attenuating the signal I was unable to hear anything about 60dB below it.

So when testing at 90dB the audible floor would be around 30-35dB SPL. Only when my ears got used to the silence when playing nothing for quite a while I could still hear something when played that soft. The dynamic range of the hearing thus isn’t as big as the limits that are found in labs.

As my mic pre-amp combo starts to clip around 110dB the max. SPL level is set just below 110 dB as well. Headphones peaking above 110dB with 90dB @ 1kHz are unlistenable bright/sharp/nasty sounding anyway.

As we can easily hear a few dB difference but do not want to clutter the graphs with lines the vertical scale is usually shown in 5dB divisions unless noted otherwise.

post separation

back to Learn
back to How to interpret graphs