Headphone Amp Measurements Done...Almost (Aaaaarghh!)

Today, April 1st, is InnerFidelity's second birthday. Not going to make a big deal out of it because what I really wanted to do is get this darn amp measurement routine done. So, about five days ago I just ignored all else and started cranking away at code. Oh boy, that first day was murder. I was getting all sorts of weird readings; I though I was going crazy. Turns out it was a bad solder joint in one of my cables...made by me, argh!

I'm not going to waste a bunch of time like that again, and I remembered Ted Paisley of The CablePro had offered to hook me up with some cables for my tester when we chatted at RMAF last year. I gave him a call---with much panic in my voice---and he did indeed get me set up with two Freedom UP-OCC Mini to RCA cables, and two sets of Freedom UP-OCC RCA to RCA interconnects. Not only did I need them overnight, but they had to be built as well. The box arrived on my doorstep the next morning; the cables are sweet and work dandy; Ted, you're a cable stud! Thanks man! (FWIW, both HeadRoom and TTVJ Audio carry CablePro cables. There's a reason for that: Ted and crew do very nice work.)

New cables in hand, measurements now seemed to be moving along nicely, so I began modifying the previous test. The biggest changes since last go around was that I changed the loads I was using, and the new test now uses 150, 32, and 16 Ohm loads. I've gotten rid of the THD vs Frequency plot because it didn't seem to indicate much, and added a 500Hz tone spectrum for folks who just want to see the harmonic series. Let's have a look at one of the new data sheets.


Measurements taken of the HeadAmp Pico.

I've shown the HeadAmp Pico in the plot above because it measures so darn well; better, in my opinion, than the JDS Labs O2, which is no slouch. I love the Pico...I wonder how much that has to do with its outstanding measurements. In total, I measured 23 unbalanced solid-state amps of a variety of types. You can download the headphone amp measurements .PDF here.

All the graphs on the sheet are measured with the 32 Ohm load except for the THD+noise Vs. Output Vrms which has plots for all three loads. The output from the AP tester was set to 0dBu (0.7745Vrms), with the exception of the 'Noise Spectrum' plot where shorting plugs are inserted into the input of the amp. I used the lowest gain setting on the amp that allowed me to get to a unity gain setting with the 32 Ohm load. Single point measurements of gain and noise were measured with the 32 Ohm load and the volume control at maximum.

Some of the amps measured very well (like the Pico and O2) and some did not. The problem is figuring out if an amp was measuring poorly because it's just poorly designed, or because of some problem with my test equipment or procedure. Sadly, I think there's a bit of both going on. Let's step through some of the amps in the .pdf (Headphone Amps .PDF here), I'll explain some things about the measurement interpretation as we go along.

ALO International
Frequency response is nice and flat, but something went sideways with the phase response. I'll remeasure this amp sometime soon, but the thing that really caught my eye on this sheet is all the steps in the THD Vs Output Volts plot at the top right. There are a couple amps with strange features on this plot. The tech support at Audio Precision is second to none, I gave them a call this morning to suss out what was up with the 'steps' on the THD vs Volts plot. It turns out this may be just one of those facts of life that we'll have to live with. This is a very sensitive measurement, and when conditions are just right, small calibration differences in the various ranges the the tester goes through can cause these steps to appear. The engineer at AP said there may be some things I can do to make them go away or minimize them (mainly putting attenuators on the output of the AP's generator), but those work arounds can't be automated. So, I think we might be stuck with these features at times.

The four plots in the middle of the page are somewhat similar. All are amplitude vs. frequency plots. For the 'Noise Spectrum" plot, I insert a shorting plug into the input and simply look at the noise spectra of the output. You'll notice the top of this plot is -50dBu and the bottom -160 dBu; this allows us to see as much information as possible in this chart. Mostly what you will see in this spectra is noise from 60Hz AC sources getting into the amp. This may be from poor regulation, power transformer radiation, and/or radiated noise from fluorescent lights. You can tell it's from the AC because of the frequencies involved: 60, 120, 180, 240, 300, and 360Hz, etc in multiples of 60Hz. This plot also allows you to observe the general noise level of the amplifier, in this case the baseline for noise is about -135dB below 0dBu.

The '500Hz Spectrum' plot is simply the spectra of a 500Hz tone. You'll see in the ALO International plot three small peaks to the right of the main 500Hz tone, these are the harmonics of the 50Hz tone at 1kHz, 1.5kHz, and 2kHz. The sum of these peaks would be the 'Harmonic Distortion' of the amp, and the baseline of the curve is the noise level. What you're seeing here is a visual representation of the THD+noise of the amp. To the left of the 500Hz peak you'll see some features. If you look carefully you can see that these are the same 60Hz noise spikes seen in the 'Noise Spectrum' plot.

The next two plots are the 'Intermodulation Distortion' (IMD) graphs. In both, two signals are present, and the idea is to see how these two signals mix and create sum and difference signals---which is their intermodulation distortion products. In the SIMPTE IMD plot the two signals are at 60Hz and 7kHz. It's a little hard to pick out the intermodulation products on this amp (because they're small and are hiding in the 60Hz noise peaks), so let's move on to the next page in the .pdf.

Alpha Design Labs Cruise
This is a fairly quiet battery powered amp, so the noise plot doesn't have any 60Hz appearing. But as you can see, in the SIMPTE IMD plot you can see numerous bumps to the right of each primary tone (at 60Hz and 7kHz). The CCIF IMD plot has two peaks on either side of 20kHz spaced 1kHz apart (19.5kHz and 20.5kHz). What this will give is a signal at 1kHz (called the difference signal) and one at 40kHz (called the sum signal), and in serious cases, peaks every 1kHz from 1kHz up to the two probe tones.

Now, here's where I have a problem: It looks like this amp is measuring poorly. Not only is there a lot of signals generated in the IMD plots, but you can also see a strong series of ongoing peaks in the '500Hz Harmonic Spectrum' plot. This type of result will often come from 'clipping' of a sine wave signal. After talking with the engineers at AP this morning, I think a 0dBu signal may be to hot for some of the portable amps. Most portable device line outputs evidently run in the few hundred millivolt region and the 773mV of the 0dBu signal may simply be overdriving the input stage of the amplifier. This could easily account for the strong harmonic series seen in the Cruise. The engineers at AP suggested I test the amps at -10dBu, which, according to them, is a common test level for consumer audio gear. The downside of testing at this level is that noise in some measurements will become more dominant, and, of course, I'd have to do all these measurements over again. (I've pretty much resigned myself to that.)

At the bottom left of the page is the 'Crosstalk' plot. Here one channel is swept through the frequency range, while the other channel gets no signal. The tester listens to the un-driven channel for any signal that's bleeding over from the driven channel. The rising straight line of the Cruise is most common as the capacitive and inductive coupling that's often the source of crosstalk gets more and more efficient as frequency increases. The most common place for crosstalk to occur is in the volume control.

Apex HiFi Butte
Let's see if we can start quickly going through amps for a good snap-shot view of performance.

The Butte is flat from 10Hz to 40kHz. THD+noise vs. Output V shows an amp with ample headroom. (Vertical parts of curves at right hand edges are 3V@16Ohm, 6V@32Ohm, and 7V@150Ohm.) The Butte has fairly low distortion (below .01% above 0.3Vrms output). It has a moderate amount of AC noise as seen in the 'Noise Spectrum' plot, and little harmonic and IM distortion as seen in the other three frequency spectra. This can also be seen in the THD vs. Output V plot: If the left hand side of the curve is going downward and is a straight line, then it is mostly a measure of noise, not distortion. Crosstalk is about average ending at about -50dB@20kHz. Output impedance of 1.8 Ohms is acceptable.

Apex HiFi Glacier
Dead flat in both frequency and phase. THD Vs. Output Volts shows good headroom for a portable; curves are mostly noise limited, but upper third or so of curves bend upward showing increasing distortion products. The amp is fairly low in noise at between -130 and -140dBu. 500Hz harmonics show a moderate amount of THD, but if you look closely you'll see the even order harmonics are slightly more emphasized than the odd, likely leading to the comfy sound of this amp. SIMPTE and CCIF IMD shows mild IMD distortion. Crosstalk in this case is a flat line to 400Hz. I'll need to verify this with some engineers, but I believe a flat crosstalk line indicates that most crosstalk is being directly coupled through the power supply. 2.3 Ohm output impedance is acceptable.

Beyerdynamic A1
Flat in frequency and phase to 30kHz. Good headroom, and mostly noise dominated THD. Moderate AC noise and -130dB noise baseline. Very little harmonic distortion in 500Hz spectra. Very good IMD results with little showing but AC noise. Good crosstalk numbers, right channel worse than left. 99 Ohm output impedance!!! Not good at all.

CEntrance DACmini
Flat in frequency and phase. Good headroom, but has some significant distortion issues. Very little AC noise, and this is not a battery powered unit. Pretty strong harmonic series on 500Hz plot, looks like borderline clipping, though this being a home unit should be able to deal with the 0dBu levels. Fairly strong SIMPTE IMD products especially in the mid-frequencies. Noise level in the bass increases significantly with CCIF IMD signal. Crosstalk mostly power supply related. 1.4 Ohm output impedance is groovy.

I'm going to let you look over the rest, but I will draw your attention to the HeadAmp Pico and the LDS Labs O2. These are the two best measuring amps of the group. Both are battery powered. Generally, the O2 is lower noise, but the Pico is lower distortion.

Dag nubit! I really wanted to be done with this today, but I think it's going to take looking at these tests using a -10dBu standard to tell whether it should be changed or not. A buddy who I sent these to this morning also pointed out that the first line in the single point measurements is likely in Watts, not milliWatts. So, a bit more to do, I'll try to redo the measurements at the lower voltage this month so we can move forward one way or the other.

I have to say it's also very helpful to get comments from folks, sometimes I use them and sometimes not, but I always read them. Please feel free to one last time lobby for your desires on this sheet. It'll be frozen pretty hard in place next time through.

pingu_turbo's picture

I just wanted to say Happy Birthday to innerfidelity.com and what a great site it is too. Great work Tyll! I have thouroughly enjoyed this website for some while now, I only joined the other day too. Congratulations!

Cami's picture

I totally agree, Tyll, you have put up a great job with innerfidelity. I check in more than once a week and I always find good stuff to read, and more as I browse older posts. The site has never stood still and there's always new features like these Amp measurements and new collaborators. My appreciation to you and Innerfidelity Tyll, keep it up!

Now to a couple of questions :)

Is there a chance that you could get the Violectric V100/ V200 to the test bench for a comparison with Lake People's G109 (John Grandberg mentioned he would be willing to lend you his V200)?
I would also love to see the Leckerton UHA-6S MKII in comparison to Headamp's Pico and JDS Lab's O2, and also more expensive DAC/Amps like Anedio's D2, Benchmark Media's DAC 2 HGC or Sennheiser's new HDVD 800, etc., in cmparison to Grace design's M903.

I feel kind of ashamed for just asking more of you right after congratulating you, lol, but I guess that's how we audio freakss are, obsessive, impatient and demanding, lol.

Have a good one!

John Grandberg's picture

I can't speak for Tyll and his timeline, but I imagine once everything is ironed out and somewhat "easy" to do, he will want to build a large database much like the headphone measurement list. At that point, I can send him my V200, Anedio D2, Resonessence Labs Invicta, UHA6S mkII, and a bunch more stuff. 

johnjen's picture

Thanks for the extra effort and time spent.  This is most appreciated and imho, long overdue.  And I don't mean to single out you specifically, but the whole industry.  The dearth of available info has long bugged me.

And since you did ask (even though these are probably already on your 'to-do' list) I thought I'd mention an observation or 2.

Add some text to the 2 IMD graphs saying what the test frequencies being used are.

Add more resolution on the x axis of the graphs (a 5 between the 1 and 10's etc.) would probably be sufficient as a guess.

But the mere fact that this info is now becoming available is a landmark in and of itself.


Keep up the forward momentum… :thumb


Dano91's picture

Thx a lot for your effort!

One thing, I think that "THD+noise vs. Output V" chart should be labeled with 16, 32, 150 ohms as in "summary table" and not with 32, 150, 600 ohms.

Also, the max output power in spec table seems to be way off? Even if they are in W and not mW, they still seems to be too low for all amps. (E.g. for JDS O2 at 16 ohms it reads 87.6mW and 63.1mW for 32 ohms. But according to nwavguy's blog it should be ~337mW @16 ohms and 612mW @32 ohms.

Tyll Hertsens's picture

...you are correct, those changes will be made.

ultrabike's picture

Tyll, you might have used W = sqrt(V)/(1k*R) instead of W = (V^2)/R for the "mW @ 1% THD" entries.


TheManko's picture

From what I understand a high output impedance affects the frequency response on whatever headphones you plug in, so is the 99 ohm impedance there to tame the infamous Beyerdynamic treble spikes? I assume most headphones would sound pretty terrible with it, but if it's designed especially for Beyerdynamic headphones it might make some sense.

I look forward to more tests! Seems like it's been a massive timesink, so I'm glad you're seeing light at the end of the tunnel. I'm interested in measurements of the SPL Auditor or Phonitor and Violectric V200. I own both the Auditor and V200 and the Auditor is clearly superior, apart from the high output impedance making it unusable with some headphones like the Momentum. The difference in clarity and soundstage between them makes the V200 sound mid-fi when I use the HD 800 and LCD-2. Since this is what I'm hearing, I'm curious of how the difference would look in measurements. I'm not chomping at the bits for measurments, so I'd be ok if it took a couple of years for them to show up in the database.

mikeaj's picture

There shouldn't be too much impact on the frequency response if the headphones are very resistive (flat impedance plot... thankfully InnerFidelity's got that covered too) and also if the impedance is high in general.  Like for the planar magnetics, you wouldn't really see any difference other than the lower volume from losing part of the signal across the output impedance.

That's not too far from the 1996 IEC 61938 spec of 120 ohms being standard.  You know, several years before the iPod and such.  It's been said that maybe Beyerdynamic designed some of their headphones around this higher output impedance, where they get less electrical damping.

But such a difference shouldn't really dampen a treble spike, as far as I know.  Check the impedance plots.

John Grandberg's picture

I hear it the exact opposite way, with V200 being preferable by far over the SPL models. It would be interesting at some point in the future to have both of them measured and see if we can correlate what we hear to what we see. It might not offer anything as conclusive as the headphone plots but then again maybe it will.

Cami's picture

and all I can say is I clearly go with the V200. I also believe the specs published by the manufacturers are quite sufficient to tell who's got the advantage. There's also a price quality ratio which is clearly in favor of the V200, and even the V100.

TheManko's picture

When I got the Auditor it cost about as much as the V200 does, but here in Europe it has dropped in price over time, so it's now cheaper than the V100. At thomann.de the Auditor costs 599€, the V100 625€ and the V200 795€. This makes the Auditor pretty reasonably priced.

What I like about the Auditor sound is how it feels pushed back away from your head compared to the V200. Does the soundstage show up in measurments? I did volume matched comparisons between the two at one time and the V200 bass seemed a bit slow.

Anyway, it'll be fun to see the list of measured amps fill up over time so we can see what they're "really" like.

itsastickup's picture

"What I like about the Auditor sound is how it feels pushed back away from your head compared to the V200"

I've written a related comment, above.

What do you mean by pushed back? In the audience vs in the band, or perhaps more out there and less in the head as is so common to IEMs and unangled headphones.

What do you mean by 'slow' violectric bass? According to the specs, its bass should be tight, hard and dynamic.


TheManko's picture

With the Auditor all the sound played through it sounds as if it’s further away from your head than with the V200. Most reviews I’ve seen on Head-Fi talk about how the Auditor has an almost artificial airy quality to it. This makes every “layer” of the audio seem more separated from each other. The effect is particularly easy to hear with the HD 800, but it’s there with all headphones I've tried, like the LCD-2.

A couple years ago I was playing around with changing op amps on sound cards, and with some combinations I got a similar effect where the sound got an airy quality to it. It wasn’t nearly as good as the one the Auditor has, but I’m guessing the way the Auditor sounds is thanks to the 120 volt op amps SPL make themselves for the Auditor and all their mixing boards.

What I meant by slow was that I noticed that the bass with the V200 seemed to hang around in the air for a bit longer than with the Auditor. In direct comparison this made the V200 sound slower, as if it was smearing over details slightly.

attilahun's picture

One of the most popular headphone amp manufacturers, Ray Samuels, seems like an obvious brand to include.

His amps are a popular choice among both portable and desktop systems. 

Looks like you're off to a great start on a challenging journey, appreciate all the hard work and look forward to following your progress. 

anaxilus's picture

Hah!  I always thought the Micro was Headroom's best amp.

Let me know if you want me ship over the UHA6S, unless John still has one.

ProTofik's picture

You just created measurements that will be used by lot of people from all around the world for the next couple of years.


bugmenot's picture

Any chance you'll make the raw data available? I'd like to make it easier to browse/compare/etcetera.

ultrabike's picture

Just wanted to wish Innerfidelity a Happy B-Day!

Tyll Hertsens's picture

Thanks mate, that made me smile.

miceblue's picture
  1. Is the offset between the left/right frequency responses of the O2 really a channel imbalance, or is it to show the difference between the channels' responses?
  2. Is there any particular reason why the frequency responses extend to 100 kHz?
itsastickup's picture

I would be interested in some exploration of the meaning of crosstalk. There's an example on headfi of crosstalk at -40db that seems to make zero difference to the audio, in terms of stereo left/right/imaging etc. However, my clip+ which according to Nwavguy measures at -50db crosstalk, has distinctly less stereo than my laptop or my ODAC/O2 as also reported in that headfi thread; to the extent that some of my music comes alive on the ODAC/O2 compared to the clip+, although rockboxing and increasing stereo width works well to compensate (a marvellous revelation to me of the power of DSP). The O2 measures at -65db if I remember correctly, which doesn't seem like much of a difference.

-50db crosstalk as 'average' is either insignificant or significant depending on who you talk to.

Someone mentions the lack of soundstage of the Violectric v200, which is also greatly surprising for two reasons. It's crosstalk is excellent, according to their own numbers, and they have a rep for top-grade kit. 

mikeaj's picture

I just took a more careful look at the Pico and O2 measurements, because those were pointed out, and...

what's going on?!

The THD+N graphs show voltage outputs of about 1.8 V / 2.2 V / 2.7 V for 32 / 150 / 600 ohms.  The table at the bottom-right has listings for 15 ohms, 32 ohms, and 150 ohms, with approximately the same 1.8 V / 2.2 V / 2.7 V.  At 32 ohms, 2.22 V and 0.0466 mW are listed as the points for 1% THD+N, but 2.22 V should be 154 mW at 32 ohms.  It's not just off by a factor of 3, but by more than that.  Weird.

Furthermore, HeadAmp lists output power as 0.7 W (700 mW) for 32 ohms, but also output voltage as 2.84 V.  What?  Not only do these figures not agree with what you got, but they don't agree with each other.  You only get 0.25 W from 2.84 V at 32 ohms.  Even if they're counting 0.25 W per channel, that's still not 0.7 W.  And these figures are all greater than the 154 mW implied by the graph.  The measured gains don't really match what's listed either.

For the O2, there are similar inconsistencies with the graph and the table.  But let's assume that the graph is really for 16 ohms, 32 ohms, and 150 ohms.  The table's power figures are way off as before, but voltage until 1% THD (2 V, 4.1 V, 7.2 V) seem slightly lower than what NwAvGuy got for 1% THD (2.25 V, 4.5 V, 7.3 V) with 15 ohms, 33 ohms, and 150 ohms respectively.  You'd expect lower voltage at 15 ohms than at 16 ohms.

Hm.  I wonder if maybe the +/- 12 V rails are falling out of regulation.  I think JDSLabs ships the WAU12-200 AC adapter?  You might be drawing over 200 mA in some parts of the cycle when benching those sine waves into 16 or 32 ohms.  (a higher-output adapter such as WAU16-400 delivering 16 VAC at up to 400 mA may be better if maxing out the power draw)  But if that's really what's going on, then I guess those are the correct results for JDSLabs' O2.


A belated Happy Birthday from me too!  And sorry for the ultra-long comment here.

castleofargh's picture

i didn't test them for long, but to me the glacier was idd flat if not a little cold (shy on bass) when the pico slim or the fiio were pretty warm and happy to show they had bass (in db if not in quality).

the freq response is pretty much telling the glacier is the only one standing its bass till the end. what happened?


it's a mess but plz pretty plz don't give up on this. you've been doing huge things latelly.

toads's picture

Big fan...keep up the great work...!

I'm a bit disappointed with the output impedance number for this amp.... 

A couple of Q though:

-was your measurement for the SE/unbalanced output..?

-does L, M, H gain settings effect this measurement...?

-any idea if the balanced out has a different impedance...?

thanks !

elmura's picture

Big effort firstly.

Coming from both a product comparison & technology background, the number 1 rule is to have meaningful and consistent test criteria.

Rather than setting all amps to max volume, they should all be set to produce a set relevant output voltage. This may require higher gain on some units.

The input voltage is only moderately high. I suggest a 0.5V rms input for portables, and 2Vrms for desktops which reflects their real-world source levels.

Also, not many amps will be happy driving a 32-ohm load. Ideally, such amps should be tested at higher impedances to reflect their purpose. If an amp is designed to drive high impedance phones, testing it at 32ohms is unfair.

Preferably, test all amps with a 32-ohm load and 0.5Vrms input, and say 0.7Vrms output. Then repeat test with a 300-ohm load, 2Vrms input and say 3Vrms output.

This would reflect real-world usage scenarios - low impedance, low source, low output for portables driving IEMs and low impedance headphones. Then the other test for high impedance, high voltage phones.

Charles Hansen's picture

Hello Tyll,

Charlie Hansen from Ayre Acoustics here. The glitches you see on the THD + N measurements are an artifact of the Audio Precision. When it changes ranges (generator frequency and/or amplitude measurement), it takes a few tens of milliseconds for the reading to settle down. For some reason, the default setting on the factory setup test is not long enough for things to settle down.

So every time that the machine changes ranges, typically at 200 Hz and 2000 Hz, but depending on how you have the machine set up, you will see a "step" in the THD + N level. This is JUST the machine settling down.

The problem is that the older machines (pre-x58x series) have THE WORST instruction manuals in the world. A degreed engineer has to read it four times to even understand what they are talking about. But once you find out what to change, it is trivial.

For example one of our machines is an old DOS machine, just like JA's original. (We also have the latest, greatest 2722 and an easy-as-pie to use x585.) If you are using the DOS software, just hit "Page Down" once or twice and there will be a value for "Settling Time", which determines how long after changing ranges that it actually measures the value. If you just double this ONE setting, it should solve your problem. Don't forget to save the "Procedure" also, preferably under a different name. The tech support guys should be able to walk you through this in five minutes or less.

Have fun!

Charlie Hansen, Ayre Acoustics, Inc.

miceblue's picture

You stated in your article: "I've shown the HeadAmp Pico in the plot above because it measures so darn well; better, in my opinion, than the JDS Labs O2, which is no slouch."

So are you using subjective opinions to interpret objective data, which defeats the purpose of comparing things objectively? I'm a bit confused.

The Pianist's picture

Why don't you use dScopeIII?

Poimandres's picture

Very new to graphs, is there a good website to read up on them? I recently ordered a pico dac/amp and since have seen the graphs you posted and was wondering what would cause the db difference in the freq response graph between the left and right channels? I realize it is only around .4db and shouldn't be audible however I am curious.