Headphone Amp Measurements ... Almost There!

InnerFidelity's second anniversary is coming up April 1st, and I'm bound and determined to get the amp measurement program up and running by then. Two years is just too long; the time is now.

I was spending a lot of time listening for the differences in headphone cables (not easy to hear) and wanted to try to do something while I'm listening. Turns out, programming is one of those things that doesn't like to share time with anything else in your head, so I had to abandon the listening and just carry on programming the tester. The good news is that I've made great progress, and the basic unbalanced solid-state amp test is working.

My basic philosophy of the test routine was to run an amp through its paces, but not to try to really stress the amp during testing. For example, I had initially wanted to test amps with a 16 Ohm load, but found that too many amps weren't very happy with this low impedance load. I want the test routine to be as universal as possible so that it would be easy to compare one amp to another, so I opted to make the test a bit less stressful on the amps so it could be used as broadly as possible. In the end, after looking at a half dozen amps or so, I think the test as it stands does a good job of identifying differences between amps, and will identify problematic amps fairly readily.

The Test Procedure
The basic test runs like this:

  1. The amp is connected to the ins and outs of the tester with the 150 Ohm load on the amp output. The tester outputs 0 dBu and the headphone amp volume is set to unity gain. Gain setting on the amp will be low for two gain setting amps, and medium for three gain setting amps.
  2. Shorting plugs are inserted into amps inputs and the noise spectra is measured.
  3. Tester outputs are reinserted into the amp inputs, and frequency response, phase response, SIMPTE IMD spectra, CCIF IMD spectra, and cross talk are measured.
  4. Single point measurements are made with 150 Ohm load with 0 dBu output for output impedance and channel balance.
  5. Volume is turned all the way up, and measurements are made for gain and noise at all amp gain settings. Gain setting is returned to low for two gain setting amps, and medium for three gain setting amps. Volume is reset to unity gain.
  6. THD+noise Vs. Frequency is plotted.
  7. Volume is turned all the way up. (This is the tricky bit.) The tester is put into THD test mode and is commanded to regulate the input voltage until the output THD is 1%. Then that input voltage is put into the maximum sweep voltage setting. The tester then plots amp output voltage vs. THD+noise up to 1%. (This will change for tube amps as the output distortion is much higher and it will be informative to see the distortion vs output voltage to a higher level. I'll determine the actual THD% setting when I start measuring tube amps.)
  8. With volume still at full, input is regulated until 1%THD is reached, then output voltage is measured and output wattage is calculated for single point datapoints.
  9. Input is set to 0 dBu and volume is reset for unity gain. Single point measures for THD+noise @ 1kHz, SMPTE IMD, and CCIF IMD are recorded.
  10. The load is changed to 32 Ohm and steps 6 through 9 are repeated. Load is again changed to 600 Ohms and steps 6 through 9 are repeated.

Here's the current look of the datasheet.

Click here to download datasheet .pdf of initial amps measured.

What's Next
The current task is to run a bunch of unbalanced solid-state amps through the ringer to see if everything runs smoothly and to tweak any hick-ups. Currently I've measured the: Meyer Audio Corda Rock; Lake People G109; Apex HiFi Butte; Burson HA160D; HeadRoom Desktop, and HeadRoom Max. A .pdf of all the measurements can be downloaded here. Once it looks like everything is working, I'll start a Headphone Amp Datasheet download page and will post links to amp datasheets there, and will start commenting on amps measured in monthly updates similar to the way I do with headphones.

Once it seems everything is working well with the unbalanced solid-state test, I'll start the process of converting the current test to tests appropriate for balanced and tube headphone amps.

Once that's done, I'll start working on using the digital outputs of the Audio Precision tester so that I can measure headphone amps with DACS that have optical or coax digital inputs. USB amps will be the last to get tests as the AP does't have USB output capability. I can however create digital test tones that can be stored on the computer and called by the tester, but it's a more complex process with a bit of a learning curve so that will take a while yet.

Phew! I'm I glad to have made it to this point, it's taken far too long.

KikassAssassin's picture

I'm looking forward to seeing how some of the more popular amps measure once you start getting a bunch of them done. It should be pretty interesting!

Impulse's picture

How'd you settle on 32, 150, and 600 ohm? Are you thinking your test will be run over far more desktop amps than portables?

I would think something like 16, 32, and 300 might cover a wider range of usage cases out there, given the number of IEM users these days... Tho perhaps it wouldn't yield as much of a complete data set.

Just curious, personally 32/150/600's fine by me since I mostly just use my IEM off portable players and whatnot.

KikassAssassin's picture

Yeah, it seems to me that if a lot of amps have trouble with 16 Ohm loads that's all the more reason to test at 16 Ohms, to sort out which amps are usable with IEMs and which aren't. An amp with a high output impedance isn't necessarily a bad amp as long as you're using high impedance headphones, it just means you don't want to use it with low impedance headphones, and that's useful information to have so know what you're getting into when you buy an amp.

If the output impedance is listed in the graph somewhere, that would at least give us a general idea as long as everyone knows about the 8x rule (the rule that says you want your headphone's Ohm rating to be at least 8x the amp's output impedance) but it would still be nice to see the measurements.

Tyll Hertsens's picture

You guys are making me reconsider. Once I run a few more amps through the ringer, I'll go back and try the 16 ohm load on them.  The problem is that some amps act so weird on that low load that I can't get the system to regulate the THD to 1% and the test fails.  It takes about 45 minutes to complete the test, and if it fails midstream it's a pretty big time sink. Anyway, that's the big problem, if the test can't reliably run with some amps then I'd have to make some exeptions to the test, and that test wouldn't be the same as the rest of the tests.

One possibility is to not run a complete test sequence on the 16 Ohm load, but do a single point test with 16 Ohm to record THD+noise at 0dBu out.  Hmm, maybe that would work.

Also, yes, output impedance is measured and available in the single point measurements.

Another thing worth mentioning is that it is possible that I might have to do a different test with portable headphone amps as they may not be able to deal with the somewhat higher input voltages that desktop amps are capable of. If so, I certainly will explore the 16 Ohm load. 

Limp's picture

Very good.
I was starting to wonder how relevant these measurements would be for me as I search for a new IEM amplifier. One particular worry was the 150Ω IMD plots. The undemanding load aside, why do you visualize it with one coninuous line, instead of several one-dimensional dicrete 'spikes'. I mean, there won't be any CCIF IMD products at say 10.1kHz, the they will all be at points like 1, 2, 18 and 21kHz, i.e discrete frequencies. Not a major concern, but it would make it easier to read the graphs with precision.

Another, less serious, puzzlement is the window you've chosen for the THD vs. Voltage plot. The stupidly inefficient HE6 needs 8.4Vrms to reach 115dB, so why should we care about outputs all the way up to 25V?


Your work is highly appreciated. We only bust your bones like this to make sure your effort comes to the most use possible :)

Tyll Hertsens's picture

I'll have to ponder your first paragraph a bit, not quite sure what you're driving at.

The reason the window is so wide is that the Lake People amp had to go all the way up to that voltage to reach 1% distortion. Um...I know you don't really need that much to drive headphones, but that's where the numbers lead me. Not really quite sure what to do about it.

Feel free to bust my bones all you want. cheeky

Limp's picture

As I understand it the intermodulation distortion products of two pure tones will also be pure tones. As it is visualized in your graphs it looks more like tight tone clusters, you know, Burj Kalifa/The Shard vs. a flag pole.

No biggie, it just at times is a bit hard to see what is a side band, and what is part of the test tone 'structure'. Christ, I'm just making things worse, aren't I.

1% distortion at 25Vrms? Gosh.

ultrabike's picture

It may be that with music recordings that preserve quite a bit of dynamic range, the Vrms requirements could be well south of 25V, but peak voltage might be a different story (specially with the likes of an HE6.)

NA BLur's picture

I think Tyll is trying to keep it simple for now because amps can have a ton of features making testing difficult.  To look at it another way trying to test a headphone through all of the possible amps on the market is a bit ridiculous so the 'standard' Tyll is moving toward makes sense.

I agree that 16 Ohms makes more sense due to the fact that even Tyll mentions some amps have trouble there.  Sure the basic measurements will show how each amp compares in an ideal cause ( i.e. unity gain, nominal impedance, etc ) but many of us want to know where the amp has downfalls.  Perhaps something like a best case and worst case could be overlapped with the current set of graphs?

Nice work so far.

khaos's picture

I think measuring at 16 ohms may be a good thing even of many amps don't like it. Or should say that the idea is to tell the reader:

"This amp has difficulty with current hungry low impedance loads, don't buy it if you have such headphones".

It's not necessary to get a 16 ohms measurement, but if would be nice to know which headphones an amp isn't electrically suited for.

Stefraki's picture

Would be interesting to see frequency response with a couple of headphones with impedance that varies quite a lot to see how well the amp deals with it.

This headfi thread, for instance, has some graphs which show how the HM801 and Clip + deal with the Shure SE530 and its quite revealing - http://www.head-fi.org/t/650523/if-it-graphs-bad-then-it-is-bad-yes-or-no ... I guess it sort of brings up a debate about which headphone to use and so on, and there's no easy answer to that, but its maybe one to think about for the future.

As it stands these look like a useful set of tests. I can't think of a headphone under 32ohms that is not designed for portable use, so maybe use a different set of loads for portable and desktop amps?

Maybe 16ohm, 32ohm, 80ohm and 300ohm for portable amps when you start measuring them?

Tyll Hertsens's picture

I think it may be possible to do some math using the impedance curve of the headphones and the output impedance of the amp to calculate a combined FR.  Sometime (WAY down the road probably) I might be able to have an on-line tool in which you could pick a headphone and an amp and then show the combined FR curve. Don't hold your breath.

I don't think it's a good idea to make the measurement though as it will only be good for one headphone, and it would require a very time consuming step in the test proceedure where I'd have to recable everything for a headphone measurement. The step of inserting the IEMs in the ear canals (or putting the headphones on the head) would create some variability in the test. 

Willakan's picture

This is looking great! Can we have the noise figures expressed slightly more accessibly though :D?

Tyll Hertsens's picture

What would you like to see?

Willakan's picture

dBV would be fine, or perhaps dBu for consistency with the units used on the graphs.

Tyll Hertsens's picture

Yeah, I struggled a bit with what units to use.... for the whole thing, actually.

The thing is that the noise is measure at max volume for each gain. dB metrics are, generally, use to compare audio levels. So, one way of looking at noise would be how many dB it is below some reference level.  Maybe I shouuld measure it relative to 0 dBu out. But that's exactly how the noise spectra plot is measured, so there's that for one reference.  Really, I did the noise at max gain single point measurement that way because that's the way NwAvGuy suggested.  I'll think about it some.

sue4's picture

Hello, Tyll! In this modern age w/ so much options of modern OS, is there any particular consideration to use Windows XP? This is just a silly question since I know there will be no different results from the test measurement based on different OSes. cheeky

I remember that you once mentioned about Dirac HD Player. Could you please give any thought of yours about it? Measurement, maybe? I found similar App: JAYS Curves, any opinion?

Tyll Hertsens's picture

...I'd be curious as well. It's actually quite a pickle.  The Audio Precision SYS2522 is a decade (or More) old now, and back in the day the common interface for scientific instrumentation was the GPIB (general purpose interface bus) connection. USB wasn't as common back then.  So the connection between the tester and computer requires an interface card installed in the computer.  Unfortunately, it's the old ISA standard, so I need to use a computer with an ISA motherboard, which means an old machine, and an old OS.  I've also got a backup computer that I keep synchronised with all the files on the main machine in case the computer fries. 

Hifihedgehog's picture

And what a technological pickle that is, given the fact that the last relatively reputable PC manufacturer who built motherboards with ISA was ASROCK about five years ago. But did you ever get wind of this manufacturer, ADEK, and their offerings? Maybe it is what you are after: http://www.adek.com/ATX-motherboards.html

Tyll Hertsens's picture

I'll look into that!

miceblue's picture

...is the volume level set on the maximum level? I'm just getting into the DAC and amp stuff, but from this website it sounds like you might lose resolution if your source volume isn't on the maximum volume for XP, hence why NwAvGuy built the ODAC with a 24-bit USB interface (from what I understand).


Tyll Hertsens's picture

The computer is just used to control the tester, which is what's generating the test signals. So no worries there.

warpdrive's picture

This impedance is common for many IEMs so it would be nice to do a compatiability test at least. My new Momentums are rated at 18ohm. If I drop big bucks on a HP amp, I want something versatile.

mikeaj's picture

Any amp actually show any significantly different behavior at 600 ohms compared to 150 ohms?  (aside from maybe if they have huge output impedance and your 150 ohms is not a resistor but some kind of headphone with a wonky impedance plot)

If you have three impedance settings, I recommend 16 ohms, 50 ohms, and 150 ohms.  It's not about trying to expose some headphone amps, but just find out what they're good for.  16 ohms represents some IEMs.  50 ohms represents some IEMs and headphones and then also very importantly—the class of relatively insensitive planar magnetics that are the rage these days, that actually need the high power outputs.  What amps are good for these?  Then 150 ohms covers everything high impedance, where a lot of amps can stretch out their legs.  If an amp can do 150 ohms, it can do 600 ohms too.

Then again, it could be argued that at ~16 ohms, a level of about 0 dBu is higher than you'd ever use for those IEMs.

Or replace 150 ohms with 300 ohms, if 150 ohms is too low for some amps.  Anyway, you can kind of extrapolate ~32 ohms from 16 and 50, at least with some educated guess.

Good luck sorting this all out.  I don't envy your task.

NA BLur's picture

The issue with the 16 Ohm load and loads that drastically depend on frequency is that it creates myriad tests that Tyll will not have time to conduct.  Even the 16 Ohm load creates havoc.  Take the Ultra Desktop Amp from HeadRoom for example.  It has 3 gain settings one of which works quite well for low impedance loads.  Tyll wants to standardize his tests for 3 impedance mode amps to the middle impedance.  Now if the 16 Ohm load causes a problem as seen on the graphs does that mean that the amp cannot work driving the load?  Of course not, just drop the switch down to the low gain mode and the problem is solved.

In an attempt to create a standard as many have in the past amps are typically measured at unity gain.  Why this is the case the book I am reading does not provide, but it makes sense because it is the least load on the op-amps, but unity gain in my mind is taking the "amp" out of the amplifier.

It will be nice to see if what Tyll measures with various amps coincides with what we all hear and have experienced.  If this is the case at unity gain and nominal impedance values then perhaps the need for 16 Ohm loads and gain setings other than unity will be unecessary.  Not to say they will not give additional information, but if the standard tests say the amp is poorly designed then there is no reason to conduct further testing.

Amps are just tricky buggers to test.  Keep up the great work, Tyll.

Limp's picture

I like the sound of that last part. Weed out the duds and focus on differentiating the good ones, a bit like how he does it with headphones. Only problem is that making an amplifier that isn't terrible is rather easy, even I can do that, so Tyll's work load might not get that much reduced.

And besides, how does one decide if a tube amp is a dud? I mean, from a solid state point of view pretty much every tube amp is a dud.

ultrabike's picture

Would it be possible (and practical) to make an exception for the 16 Ohm load case, and measure all amps in question using the lowest gain mode (only for the 16 Ohm load case)?

silver_85's picture

the "1% THD mW" data are wrong?

Apex Butte: 0.074 mW @32Ohm??? Maybe 0.074 W, or 74 mW...

Tyll Hertsens's picture

I'll check into that! yes

mikeaj's picture

I love how everyone is just bombarding Tyll with opinions and suggestions, me included.  Poor guy!

On second glance, I don't think the THD+N vs. output voltage graph is as appropriate as it might be.  I think voltage could be on a log scale.  After all, a whole lot of the "action" happens between 0.1V and 1V.  The difference between 0.1V and 1V is the same as the difference between 1V and 10V:  a 20 dB difference in volume.  And it would make plotting the axes easier (maybe? I hope) as that would avoid the ugly 5.1V, 10.1V, etc. issue.

Also, are the axes set up in the proper places?  I mean, for some IEMs you'd normally use under 0.1V a lot of the time, and that region isn't even shown on the graph.  Sure, it's mostly noise down there, but isn't that a significant finding if you want to know if you can use IEMs with the amp?  For some tube amps and some others, topping out at 1% THD on the graph might not show as much of a range as they're really capable and intended for.  (Still quoting 1% THD max power levels at the bottom, is okay)  Maybe going up to 10% THD would make more sense?