Headphone Measurements Explained - Frequency Response Part Two
In Part One of this article I introduced a bunch of concepts that resulted in a target headphone response curve (as measured at the ear drum) that looks something like the black plot above. (The green plot is what speakers, equalized to be flat in the room, look like at the ear drum.) To some extent, it's necessary to memorize certain aspects of the curve above in order to be able to mentally compare it to the raw frequency response plots when you look at headphone measurements. The first thing to do is establish a 0dB reference level, which should be the lowest point, or the average level, of the mid-range (~200hz-1kHz) on the raw frequency response plots. These are the characteristics you should look for:
- +4dB rise in the bass, starting no higher than 200Hz and at full level by 60Hz.
- From 200Hz to just over 1kHz, you should have a gentle 3dB rise.
- The main peak of the plot should be at 3kHz +/-500Hz ideally. If the peak starts getting too high in frequency the sound can get quite piercing in nature.
- Although the curve can get fairly noisy above 3kHz, it should generally follow the profile shown and be roughly back to 0dB at 10kHz. And a less noisy frequency response curve is generally better sounding.
- Above 20kHz the plot should continue to fall off to about -13dB below baseline.
The above curve is right out of Sean Olive's paper "Listener Preferences for In-Room Loudspeaker and Headphone Target Responses" and my notes above are for that curve. My personal sense is that it's close but not quite right. So here are a few variables to consider:
- I think the bass may be high by a small amount, I'd rather see 3dB in the bass. I do agree that the bass emphasis should all be below 200Hz, any higher at all, and you start to hear the lower mid-range start to thicken up unnaturally.
- The 3dB rise from 200Hz to 1.2kHz is so rare that I really can't say much about whether it's right or not. This rise mainly comes from the head and torso interaction that is bypassed by headphones, so I would think it's something that should be there on headphones, but it might be very difficult to achieve.
- I think +12@3kHz might be a a few dB high. I'd say you want this peak between 10dB-12dB, but not higher lest there be daggers in your ears.
- Although headphone measurements may be noisy and have lots of peaks and valleys above 3kHz, on average it is good to see the level back to between 0dB and 3dB at 10kHz. It is very common to have a peak in response near 10kHz due to an ear canal resonance; this peak being higher than the average level can be okay if nearby frequencies remain near the target curve level on average.
- It is very difficult to get a real sense of where the levels are above 10kHz since measurements are so noisy and plagued with resonance artifacts, but my sense is the Harman response curve shows too much roll-off. I think that while levels above 10kHz should continue to fall, the drop should be more like -10dB.
Okay, lets take a look at headphones that approach the above curve, and some that don't, and what that means when relating measurements to the listening experience.
Later, will also look at some of the characteristics of ear pads that can be seen in the frequency response curves; characteristics that are unique to in-ear monitors; and a few other odds 'n ends that can be seen in the frequency response plots of headphone measurements.
Historical Headphones Relative to the Target Response Curve
The point of this section is not only going to be identifying headphones that approach the target response, but also noting how they deviate from the curve and what character it might give the headphones apart from sounding neutral. Remember, we're going to be talking primarily about the lower gray curves that are the raw frequency response measurements from the dummy head with microphones at the ear drum position. (The upper blue and red curves are compensated.)
Lets start with a few headphones that have historically been considered "good" sounding.
Sennheiser HD 580
For this plot, we'll call the -25dB line the 0dB reference. You can see that the peak at 3.3kHz and +12dB over baseline is just about right when compared to the target response curve at the top of the page. The fall above 3kHz is somewhat too steep with the peak at 10kHz just reaching the 0dB line, and the average falling significantly below. This would cause the HD 580 to sound somewhat laid back, and, of course, this headphone is famous for having very good, but "veiled" sound.
You'll also note the bass is not accentuated but is, in fact, rolled-off in the low bass. Fifteen years ago headphone enthusiasts really didn't complain of the HD 580 not having enough bass; that kind of response in headphones was the norm. We were used to it.
Another old favorite is the AKG K701, which generally considered a brighter headphone than the HD 580 but still fairly neutral when it was first introduced. Here you can see there is no well defined peak at 3kHz and that area only rises about 8dB above baseline. More importantly, above 3kHz it continues at +8dB to 5kHz before beginning to descend to baseline around 10kHz. There's quite a bit more treble energy in the K701 than the HD580 above 3kHz relative to each other.
You'll also notice the broad mid-range hump is centered at about 300Hz, while it's centered at 120Hz on the HD 580that's more than an octave higher in music, it's quite a bit of difference in the overall center of emphasis. This makes the the bass sound just a bit warmer on the 580 than the overall cooler sounding K701. The point here is that the K701 and HD 580 clearly sound different while there measurements are not that dramatically different. Subtle interpretation is important to get things correct.
Beyerdynamic DT 880
We might as well have a look at the Beyerdynamic DT 880 as it joins with the first two as the triumvirate kings of the headphone hill for quite a while. Here you can see that the DT 880 has quite similar response to the HD 580 to 3.5kHz. Above 3.5kHz the DT 800 remains high and the 10kHz peak is at +10dB relative to baseline, with an average level at about +3dB at 10kHz. Energy above 10kHz is about 5dB higher than either the HD 580 or K710. The DT 880 was a bit brighter sounding than the K701.
Another old school headphone that had a strong following for its good sound is the Denon AH-D5000...and its siblings D2000 and D7000, which were similar. While it doesn't have the step up in the bass of the target response curve it at least does rise about 5dB from baseline, which gave this headphone a sense of heft and body the the first three lacked, and which was broadly appreciated by enthusiasts.
Unfortunately, while the 3.5kHz peak is in a good place 12dB above baseline, it subsequently looses little energy and remains 5dB above baseline to about 13kHz before dropping. This made these headphones somewhat too hot in the treble for some listeners (me included), and was particularly noticeable on the D2000.
Sennheiser HD 800
Introduced in 2009, and remaining still one of the finest reference headphones on the planet, the Sennheiser HD 800 shows some improvements over the HD580/DT880/K701, but also some troubling features. The bass response doesn't roll off as fast as the first three, which gives the HD 800 a stronger punch down low. The rise to 3kHz is good and the subsequent response has fairly good shape without excessive peaks and valleys. However, everything above 3kHz is also about 3dB too high as well, making these a headphone that can be too bright for comfortable listening.
Then about five years ago we started hearing planar-magnetic cans and people were immediately drawn to their powerful and tight bass response. The graph above shows the LCD-3 with a wonderfully flat bass extension. Even though it's below the target response, it's much flater and better extended in to the lowest notes than anything that preceded it. People love planar magnetic headphones for their well extended bass response. I have to say though, after having spent some time now with headphones that more closely conform to the base boost of the target response, I do feel that it's more pleasing, and a more subjectively accurate representation of neutral.
The plot above is from one of the earlier LCD-3s that were considered somewhat lush in the mids and polite in the treble. In the plot above we do see a slight rise from 200Hz to 800Hz, which gives the sense of strength and character to the vocal harmonics. But the subsequent dip before the peak at 3.5kHz puts the LCD-3 a little off the pace in the presence regionthe sound of spit on the lips or saliva on the reed of a sax is going to sound a little laid back. Also, above 3kHz it falls too quickly and, other then the level of the 10kHz peak, is about 5dB below the target curve. All this might make for a headphone that is too polite in sum, but the treble above 10KHz actually rises above the target and somewhat makes up for the slightly too dark treble otherwise.
Two things to note here. First, this is where the value of measurements end, and only experience can fill in the blanks. Yes, the LCD-3 is a bit too low in the low and mid treble (though the peak at 3kHz is the right height), and a bit too hot in the top octave above 10kHz, but it's not too far off. The question is: does that sound bad or good? Because it can go either way. And second, we're only looking at the FR measurements and not able to analyze all the other plots. There are further hints about sound quality gained by looking at transient and distortion characteristics, but there again there are limits to how much information about the quality of the listening experience that can be gleaned through measurements.
Last old school headphone that needs to be viewed would be a Stax electrostatic. Many would say the SR-007 shown above is the best of the lot.
I'm not certain, but I would assume the drop in the bass is due to a pad resonance. Electrostatics have a reputation of having poor punch in the bass, but I think that's a bit of an overstatement. I think there has been some problems with pads and how they fit historically, and I do think that larger electrostatic panel speakers may have had poor bass impact, but from what I've heard and can tell from measurements, electrostatic headphone have fairly good bass response; certainly as good or better on average than open dynamic headphones.
Above 200Hz the curve does have a nice rise to 1kHz, but subsequent rise to 3.5kHz is a bit of a roller coaster and on the low side of optimal at 8dB over baseline. The fall after 3.5kHz starts off about right but at about 8kHz begins to gather about 5dB of energy and keeps that emphasis for the remaining treble. Like the LCD-3, Laid back in the lower treble ranges, hot in the top treble ranges, but overall close. It's just how it all comes together in you ears that will determine wither it's good or bad for you.
Current Headphones Close to the Target Curve
The next group of plots will be from current headphones that I've measured and appear to come close to the target response curve. They all sound fairly good, in my opinion. That's not to say there arent good sounding headphones that farther off the target curve, or headphones that may be close to the curve but sound poor, I'm just saying the headphones on the target curve seem to have a pretty good probability of sounding good.
NAD VISO HP50
The first headphone that I've measured that had a raw response very close to that of the target response curve is the NAD VISO HP50. Paul Barton designed this headphone with a very clear picture of the acoustics and studies surrounding the development of target response curves in general. He went straight to the drawing board with his own target response curve using the trademarked term "RoomFeel." The result is a headphone with a very pleasing and fairly neutral sound, and a measured performance quite close to the Harman target response.
In the plot above, you can see that though the bass has a very typical broad hump characteristic of dynamic headphones, there is an overall rise of about 5dB over baseline. The transition upward into the bass starts at about 400Hz, which is a bit high in frequency and will typically cause a bit of thickness to appear in the low mids. From 500Hz up to 3.5kHz we see a beautiful profile that very closely fits the target response and indeed the HP50 does a very nice job of rendering vocals with a natural timbre. Though there's a dip at 6-7kHz and a small peak at 10kHz, the overall profil of the upper treble on the plot matches the target response well.
It's probably a good time to note that a dip somewhere between 5-8kHz in response might actually be preferred to the Harman target response in the region. Work by Philips on the X2 headphone suggested to them that a notch in this region is subjectively preferred. Too much energy in this area can be quite abrasive to the ears. I don't know this for sure, but if you see a dip in the 5-8kHz region it might be a good thing.
Another thing to start noticing as we look at cans that have good matching with the Harman curve is the shape of the compensated curve. As I mentioned in part one of this article, I use the "Independent of Direction" compensation provided by Head Acoustics, which is quite similar to the "Diffuse Field" response. One thing you'll notice is that if a headphone closely matches the Harman curve, it results in a relatively featureless sloping line that gradually drops of faster as you go higher in frequency. The one exception is the peak at 10kHz, which can be overemphasized by the compensated curve.
Focal Spirit Professional
While the Focal Spirit Pro plot below 3kHz is a little smoother than the HP50, it doesn't do quite as good a job of producing the long linear rise from the mids to the 3kHz peak. It's more liquid and coherent sounding bass through mids, but it's also a tad polite in the presence region sounding a bit more laid back than the more correct to my ears HP50.
The valley at 6-8kHz, peak at 10kHz, and subsequent peaks and valleys are not optimal, but after having looked at hundreds of headphone frequency response plots I've found that this feature is surprisingly common and often fairly benign. I'm going to break my rule about remaining with frequency response plots only, and show the 300Hz square wave for the FSP and HD600 (which has a similar look to its FR plot). You can see that rather than a clean initial step there is some ringing. I don't know for certain, but I think this may be a fairly natural result of ear canal resonances due to the way some headphones acoustically interact with the ear, and it might be that the auditory perception system knows how to ignore it. In any case, when you see the a frequency response plot with three big peaks at 3kHz, 10kHz, and 15kHz, and the triple ringing front end of the 300Hz square wave, it's a good bet that while the treble may not have superb transient response it will probably sound okay. This measured feature seems to look worse than it sounds.
At first glance the Shure SRH1540 looks quite a bit like the target response curve. The bass boost is nicely placed with a start to the boost at 180Hz, but at +8db over baseline at 50Hz is a tad excessive. The peak at 3.5kHz is of proper height, but lacks the long slope upwards from 200Hz, causing it to loose some presence and richness of vocal overtones. The decent of the treble above 3kHz has fairly good shape, but should probably fall of a little faster and be down another 3-5dB at 20kHz.
In short, it's got a bit too much bass and high treble, and lacks a bit in the midsa mild "U" shaped response, and that's exactly the way it sounds. One thing worth mentioning here is that while a "U" shaped response may be too exciting for some (me included) at solid listening levels, it tends to sound great at lower volumes as it delivers a bit more bass and treble, just like Fletcher and Munson prescribe. So, if you listen a lot at low levels (as I often do) this is a terrific headphone.
All the headphones above are around-the-ear, sealed headphones, which seem to be the most likely type of can to get close to the target responseespecially in the bassbut there are a few headphones of other types that get close.
Philips X2, circumaural, open
Open, dynamic driver headphones have a much tougher time getting the bass extension of sealed dynamic headphone, and you can see that the X2 does fall off at about 6dB/octave below primary driver resonance at about 60Hz. However, unlike the three open headphones at the top of this article, the mid-range is not a broad hump but rather is flat with a hump only below 200Hz. This is a fairly good approximation of the target response for an open headphone.
The mid-treble peak has about the right level, but centered at 4.5kHz is a bit high. However, this headphone was designed with a lot of feedback from trained listeners, and it was determined that a notch at 6-8kHz makes for better listening. This notch could very well have required a bit more energy both before and after to help compensate for the loss of treble energy in the notch. The result is a nice, warm, friendly sounding headphone, but I think all the tuning to get the curve into the desired shape has caused the response to lack a little smoothness and, as a result, sounds a bit grainy to me.
V-Moda XS, on-ear, sealed
The XS has a somewhat excessive bass response that bleeds too far up into the mids giving them a bit of thickness to the sound. But the fact that the rise to 3kHz starts below 1kHz, and the fall above 3kHz (though starting off a bit fast) has good proportion makes for a really lovely sounding headphone.
V-Sonic GR07 Bass Edition, dynamic driver, in-ear
This is a headphone I heard only briefly after measuring it and seeing such good measurements, so I really can't comment in depth. ljokerl can however, and in his GR07 review on TheHeadphoneList.com he gave the GR07BE a 9.1/10 in sound quality.
One thing I will point out however is the peak at 5kHz being equal in level to the one at 3kHz, which might skew these headphones into sounding a tad bright. ljokerl's comment at the end of the review has him pointing to "Mildly sibilant on some tracks" as one of the GR07BE's cons.
Here are a few more IEMs that measure relatively close to the target response. I'll just post them without comment.
Headphones a Little Farther Off the Mark
The above headphones are about all there are that are very close to the target response curve. Let's look at some fairly good sounding headphones that are a little farther off the curve and see if their failings match their deviations from the target.
A big round of applause for Beats getting this damn close to the target response, the new Beats Solo2 is a good sounding headphoneWAY better than the first Solo. However, you can see that the bass boost goes all the way up to 500Hz, which causes the low mids to sound quite thick. And while the peak at 3.5kHz and subsequent drop to 10kHz look fairly good, the drop between 10kHz and 20kHz is far too excessive, causing the headphones to lack a sense of spaciousness.
AKG K267 Tiesto
The K267 has a switch on the earpieces that changes the internal acoustical tuning to adjust the bass to three different settings. In the "Club" setting it does a fairly good, if a bit rough, job of mimicking the target response, especially in the long rise to the 3.5kHz peak from ~300Hz. It has a fairly wide notch above 3.5kHz, which may sound okay, but it does seem a bit too wide to me. Above about 8kHz the curve looks good.
Musical Fidelity MF100
When I first saw my measurement system draw out the plots for the Musical Fidelity MF100 I was giddy with excitement, it looked like it was very close to the target response. But as you look at it a little more closely, you'll see the 3.5kHz peak is about +3dB too high, and the subsequent run up too 10kHz remains far too high in level. In listening this headphone sounded mostly good but somewhat piercing to me, and caused me to put more credence on Philip's idea that taming the 6-8kHz area might indeed be a good idea. The MF100 is a bit hot there and suffers dearly for it. You may also want to take a peak at the full measurements to see how this treble boost made for a very significant leading edge spike. So, close, but no cigar.
Audio Technica ATH-M50x
Though pretty rough looking due to some pad bounce artifacts between 80Hz-300Hz, the M50x response is quite close to the target curve from 20hZ to 4kHz, especially between 300Hz and 3kHz, and this is indeed a good sounding headphone. Above 4kHz is a notch probably too large, but response is in place above 10kHz. Overall, the tonality of these headphones is excellent, but, as evidenced by the herky-jerky response, they aren't particularly refined or liquid sounding.
Okay, that's probably enough on the target response curve, let's turn the page and look at some of the other things that can be seen in the raw frequency response measurements.