NwAvGuy's Heaphone Amp Measurement Recommendations

Editor's Note: NwAvGuy is an audio engineer who showed up recently in the headphone hobby scene. He made a really big splash with measurements and criticisms of commercial products and DIY designs on his blog, and with his "politically incorrect" postings on some of the headphone forums. If the quality of investigative journalism is measured in ones ability to create a stir, NwAvGuy should win a Pulitzer.

I must have a thick skin: I found his posts interesting ... a little too long maybe, but quite interesting. Let's put all that aside for the moment, however, that's not what we're here for today.

I've been struggling to find just the right set of measurements to implement for testing headphone amps at InnerFidelity, and thought it might be cool to get some advice from NwAvGuy. He obviously has the chops, so I asked him if he would like to write his recommendations for my test routines in the form of an article to introduce me, and InnerFidelity readers, to the various sets of measurements he would recommend. I did place some constraints on the article: it needed to be around 2000 words, and I had room for only 7 graphs and maybe a dozen or so single point measurements.

He's done that ... and even more. The following is a very good and nicely concise introduction to the important headphone amp measurements. It also includes links to his blog to provide a more in-depth understanding of the information. All told it's likely in excess of 10,000 words. A mighty large chunk of work, written largely at my request to help headphone hobbyists understand the numbers, all completely voluntary and uncompensated. I'm grateful, and hope InnerFidelity readers will set aside the drama for a moment and just absorb some good info.

STIG’s COUSIN: Top Gear is one of the BBC’s biggest shows. They evaluate expensive cars in entertaining ways then turn them over to their anonymous racing driver, known only as The Stig, to find out how fast they’ll lap their track. Some say NwAvGuy is Stig’s geeky cousin. Instead of testing cars, as an Electrical Engineer, I design and test audio gear. I want to thank Tyll for requesting my thoughts on measuring electronics. I'm all for improving measurements

WHY MEASUREMENTS? Here are some key reasons:

  • Compatibility - Measurements can predict if the UltraCans play loud enough with the UberAmp and more.
  • Comparisons - Measurements allow comparing gear in fair and useful ways that are otherwise impossible.
  • Objective - Measurements are unbiased and not affected by preferences, cost, aesthetics, etc.
  • Many More - See: Music vs Sine Waves

BACKGROUND: Audio measurements are specified in the following units:

  • dB SPL - Acoustic sound pressure
  • dBr - Relative to some (hopefully) specified reference level
  • dBFS - Relative to Full Scale (the maximum) digital signal
  • dBv - Relative to 1 Vrms
  • dBu - Relative to 0.775 Vrms
  • Volt - A measure of signal level usually given in Vrms and you can convert between volts and dBu.
  • Watt - A measure of energy usually in mW (0.001 Watt) for headphones
  • Hz - A measure of frequency

NECESSARY NINE: The following nine measurements shine a spotlight on what matters most. An amp clearing all the hurdles should work well with your headphones and not get in the way of the music.

1 - OUTPUT IMPEDANCE: Guideline: 2 Ohms or less.This is perhaps the most common reason different headphone sources sound different and it's often not disclosed or measured:

  • Frequency Response - Higher output impedances can create audible peaks and dips in the response.
  • Bass Damping - Higher output impedance can reduce bass damping resulting in audibly less controlled bass.
  • Guideline - With the exception of a few rare headphones designed for higher impedances, divide the headphone impedance by 8 to get the maximum recommended output impedance. For 16 ohm headphones it's 2 Ohms. For more see: Output Impedance
  • Measurement - It's measured by comparing the 100hz output at 100 mV with no load to the voltage with a low impedance load and cranking the math: Zout = (Rload * (Vnoload - Vload)) / Vload

2 - OUTPUT POWER: Guideline: Ten times the headphone's 90dB SPL voltage. Tyll's headphone measurements make it easier to calculate necessary amplifier power. Headphones cover a huge range of sensitivities and impedances creating a wide range of power requirements. See More Power for a complete explanation but here are the basics:

  • Interactions – Sources produce wildly different amounts of power depending on what headphones are used. All sources have a maximum output voltage that limits their power into higher impedances, some have limited current into lower impedances, and many have significant output impedance. All three can interact with the headphones in complex ways.
  • Impedance – The headphone impedance determines the relative amounts of voltage and current required. Typically low impedance loads need more current and high impedance loads need more voltage. For math geeks, Power = (Voltage^2) / Impedance while Voltage = Current * Impedance.
  • How Much Is Enough? – Most consider peaks of 110dB SPL sufficiently loud. Tyll measured the 300 ohm HD600 at 0.23V for 90dB so they need ten times more, or 2.3V, for 110dB. The power is (2.3 * 2.3) / 300 = 18 mW. This works for any headphone Tyll measured.
  • Measurement - The maximum output level at 1Khz and 1% THD should ideally be specified in both voltage (Vrms) and power (mW) at several impedances. THD+N vs Output Voltage or Power can be graphed with a line for each load impedance.

3 - NOISE: Guideline: -100dBu/7uV. Many headphone amps have audible noise especially with sensitive headphones. For all the details, see Noise Explained, but here are the essentials:

  • Guideline - Tests show 85dB below the maximum listening level will usually be inaudible. With sensitive IEMs that means noise below -100dBu or 7 uV. For a rule of thumb, divide Tyll's 90dB headphone sensitivity number by 2000 to find the maximum noise voltage.
  • Absolute vs Relative - Noise can be measured in absolute terms as a voltage but it's most often specified as Signal to Noise Ratio (S/N or SNR) where the noise is compared to a (hopefully) specified reference level.
  • Upstream/Downstream – Noise upstream of the volume control is mostly relative to the music--changing the volume doesn't change the SNR much. But downstream noise is fixed (absolute) and always present even when listening at low volumes. Some amps are noisier at half volume than full volume.
  • Headphone Sensitivity – The more sensitive the headphones the louder the downstream noise.
  • Reference Level – A noise measurement in just “dB” is nearly worthless without a reference. If you see such a number it’s probably referenced to the maximum output of the device. This makes the spec more impressive but is more accurately the dynamic range. For example -100 dBu (reference 0.775V) becomes -120dB referenced to a full output of 7.8 Vrms.
  • Weighting - Noise is measured as the sum of all noise at audio frequencies and can be un-weighted (the raw number), or “A-Weighted” adjusting for the sensitivity of human hearing typically improving SNR by several dB. You can’t fairly compare A-Weighted to un-weighted measurements.
  • Measurement – Noise should be measured at the worst case volume setting and, if applicable, the highest and lowest gain settings with inputs shorted. The A-Weighted results should be listed in microvolts and dBu. A graph showing the noise spectrum from 20hz - 20Khz is useful to distinguish hum or digital noise from broadband hiss. If the same reference is used, this graph can be compared to distortion graphs. Use an online calculator to convert between volts, dBv and dBu.


LFF's picture

Very nice article! Very nice!

Extremely informative and to the point. Great job NwAvGuy!

Reticuli's picture

I knew it! NWAVGUY is a member of Daft Punk!

Limp's picture

Absolutely stellar.
Two of the driving forces of awesomeness in the hobby getting together.

khaos's picture

Clear, concise, with links for a more in-depth study. Thanks!

sgrossklass's picture

Nice article overall, just a minor goof when it comes to phase response: Within 1° between 10 Hz and 10 kHz? Methinks there's a couple of zeros missing there. 10° and 100 Hz would make a lot more sense.

Personally, I would also like to see a 1 kHz distortion spectrum. Not a terribly big fan of THD+N either, it just doesn't correlate too well with hearing impressions. (Hence the former.) Maybe it doesn't have to be the fancy GedLee metric (which is kinda hard to evaluate in practice), but some kind of weighting would definitely be advantageous (maybe D. E. L. Shorter's from 1950).

sgrossklass's picture

Oh, and measuring nominal power at 1% THD is reasonably silly in something claiming to be hi-fi these days. In many concepts this will be at the onset of clipping already (beyond the THD "knee"), which means they'll sound audibly bad, while oldschool concepts with little feedback may still be acceptable - with places swapped at a little less power. If you want to go all DIN 45500, at least keep in mind that this norm called for power to be sustained for 30 minutes or so.

SpaceTimeMorph's picture

... if the THD graphs vs. output power are given you have the option of taking the cusp proceeding the upturn in THD as the max power output. For headphone stuff, THD @ 1% is the standard (as silly as it is to do so), so quoting that number still provides a good comparison point among different equipment.

sgrossklass's picture

This topic just sprang to mind as one of the little things that can ruin your day. Few amps would have enough DC offset to upset linearity in very sensitive cans, but it should definitely be checked (in all gain stages at normal volume settings, plus a defined load of like 100..300 ohms for AC-coupled outputs so as not to be fooled by coupling cap leakage currents). If a headphone barely needs 10 mVrms to output 90 dB SPL with noteworthy amounts of distortion (and there are a few among those measured, mostly IEMs), 10 mV of offset is anything but negligible. Some insensitive 600 ohmers wouldn't care one bit, of course. For rating DC offset, I'd use 10-dB steps: <3 mV (good), <10 mV (OK), <30 mV (meh). I hope you have a good, well-calibrated multimeter...

Power-on/off noises are a closely related area. While truly dangerous voltage spikes have been the exception, they can nonetheless be quite annoying in practical use.

NwAvGuy's picture

...for all the encouragement!

@sgrossklass, you are correct about the phase but just one zero is missing. Thanks for catching that and I let Tyll know. It should read 100hz - 10Khz for 1 degree of error. Even AC coupled amps (with direct coupled outputs) can manage < 1 degree at 100hz. The idea is to keep the phase error low in the range where spatial information is most important. But it's not an exact science as there is considerable debate over how audible phase error is. But I would consider 10 degrees at 100hz fairly marginal. Such an amp would have significant frequency response error in the deep bass.

1% THD is indeed the benchmark standard for maximum power and it is the onset of clipping in most amps. As SpaceTimeMorph mentioned, what's most important is to apply the same standard across the board so amps can be fairly compared. By graphing THD+N vs output you also get to see the distortion at lower levels so if you want to know the max output at 0.05% it's there. Setting the standard much lower than 1% would cause a lot of single-ended and tube amps to fail the test completely (they often have distortion approaching 1% at any output level) and require using a different standard making comparison difficult. Manufactures and other reviewers also typically use 1%.

I agree it's worth checking DC offset with a DMM and at least subjectively checking for turn on/off transients. Neither really impacts the sound quality of an amp and Tyll wanted to keep the results page to the essentials. If there's room, the DC offset could be added but it's really more of a pass/fail sort of measurement. I would agree anything over 30 mV deserves a red flag.

SullivanG's picture

Is it worth mentioning the polarity? (absolute phase) I am sensitive to this, for low frequency waveforms. Thanks for the great write-up.

13mh13's picture

It seems the Guy's been bad-mouthin' Tyll and the so-called "subjective" community over on his Blog. He -- and his small gang of supporters -- don't raise any worthy claims -- just keep repeating/rehashing the same ol' line that a select group of narrow-minded "objectivists" have been pushing for years. If they REALLY had meritorious, science-supported claims, the hi-fi world may give them more than 2sec air-time.
On a more interesting note, just saw this on Head-Fi re a possible IP rip-off by NwAvGuy. If true, then "may the best man win."

Colin Shaw's picture

A couple of comments on phase and output impedance, motivated by a change from measuring specs on some designs I have been working on to mostly just listening to them and trying to understand the reason for what I perceive to be good qualities and improvements.


The output impedance of the amplifier is definitely an concern.  Let's talk in terms of damping factor, or the relative value of load impedance inclusive of mechanical properties to the output impedence of the amplifier.  In my experience the specifics of it are quite dependent on the type of load, motivated in great part by work with open baffle loadspeakers.  A very light, high sensitivity driver can sound fantastic for natural tone music if it is used in a system with a relatively low damping factor, the reason being that there is more expression of resonance.  This, however, does not seem to be a general maxim, but rather a fairly special case.  Also, depending on the type of amplifier (Zen type amplifiers as a prototypical case here), a difficult region for reproduction, particularly with higher impedence headphones is where the driver is near resonance and the impedence is much higher than nominal.  I suppose the point of my comment is that while there may be some rules of thumb, it is not a one size fits all issue, and it depends on the load, the amplifier and the way that the two of them work with what you are listening to.  


On the issue of phase I have a complex comment.  The ability of an amplifier to accurately track the signal has more to do with bandwidth than is generally appreciated, in my opinion.  Our ears are amazing devices that have been honed over many generations of selection to allow us to escape wild cats and whatnot, and have an amazing ability to discriminate phase for that purpose.  Moverover, they are well suited to doing so in a frequency dependent manner.  What I have found is that amplifiers with enormous bandwidth tend to sound (to me) more accurate and more enjoyable.  Much of the reason for my concern stems from considering jitter in digital sources, where a very small temporal error results in diminished accuracy.  I decided to try to apply this concept to amplifiers and see what happens, and the result (to me) is that with high bandwidth (some of the amps are above 1MHz), the higher audible frequencies become much more detailed and pure sounding.  There is an issue of whether or not you can hear the frequency, and there is an issue of whether or not you can tell that the phasing of the signal aligns in a way that sounds most sincere and true to your perception of the source.  Any amplifier has some limitation on upper frequency response, but by pushing it out there as far as you can, the phase limitations in the audible portion are minimized.  I can't claim to have any expert knowledge or rule of thumb ideas on this issue, but I can claim that I tend to prefer the higher bandwidth amplifiers, and I believe this has a great deal to do with accurate phasing.  

Tyll Hertsens's picture

Thanks for the comment. Makes me think I need to try to make my frequency response plot go up much higher than 20kHz. I'll work on that.

MGbert's picture

I'm an engineer, but not an electrical engineer.  I can run formulas, though, so I'm wondering if you could help me out since I'm not sure how to interpret some of these measurements.

I just got a Parasound Zdac and my headphones are a vintage pair of 1980 AKG K240 Sextett's.  The Zdac makes them sound great, but the volume needs to be cranked up near max (9 out of 10) for the volume to hit about 90 dB.  Your measurement sheet has the AKG, at 90 dB, as requiring 893 mV, 1.27 mW, and 630 ohms at 1K Hz.  Doing the math, that equates to a current of 1.42 milli-amps*.  Which of these parameters define how the Zdac will behave with other headphones?  If I were to get a pair of AKG Q701's, will the fact that they would need 5.29 milliamps to get to 90 dB (based on 318 mV and 1.68 mW at 60 ohms, also from your measurement sheet) mean the Zdac will never get them to 90 dB based on available current and or wattage, or does the fact that the Q701's need less voltage at 90 dB than the K240's mean the Q701's will do fine?  Thanks in advance.


* Math being Amperage = SQRT(Wattage/Ohms) at 90 dB