Headphone Measurements Explained - Total Harmonic Distortion Plus Noise Part 2
In Part 1 of this article, I explained what total harmonic distortion is and how it's measured. Here in this second part, we will take a look at various THD+noise plots and interpret them.
Maybe the most important thing to know about my THD+noise measurements is that they're the most likely plot to be erroneous. Because it's a very sensitive acoustic measurement and I don't have a laboratory grade isolation chamber, if a large truck goes down my street during a THD+noise measurement you'll probably see it in the plot as a small rise. Also, if water is running in my house or if the refrigerator kicks on, these might produce enough noise to increase the recorded THD+noise. Lastly, for about six months last year I was having some odd symptoms with my Audio Precision testerit took a while for me to even realize I had a problemand it does seem that it effected some of the THD+noise measurement. The AP tester is fixed now, and THD plots do seem more stable.
There are two spikes at 200Hz and 2kHz in THD in all plots, this is an artifact of the tester changing ranges at those frequencies. These can be ignored.
My point is that THD+noise plots should be taken with a big grain of salt. Here's the kind of thing you're looking for to determine if a feature of a THD+noise plot is real or not. In the top good plot above, you'll see that the average level of the 90dB plots (red and blue) is just above the 0.1% line and the 100dBspl plot (yellow and green) is just under the 0.1% line. This is probably near the noise limit of my system, so the headphones measured may actually be lower distortion than indicated.
On the yellow line of the top plot between 1500Hz and 2kHz you'll see a bump that doesn't exist in the green plot. My THD+noise plots are taken one channel at a time, is quite possible that this is just a truck going down my street. However, above 3kHz you can see both the yellow and green line becoming noisy and a little more elevated; because both channels are doing it and because the rest of the plot is quite low, I suspect there is a real rise in THD above 3kHz here. As a rule of thumb, 1% is considered the point at which distortion becomes a problem.
In the bottom suspect plot above, you can see that all four plots have near the same shape and relatively low levels of noise. Left and right channels are offset but of similar shape. I suspect there's something wrong with this plot due to the problem with the AP tester.
Bottom line here: Don't worry too much about small features on the THD+noise plots; it's the larger features that are most telling. Let's have a look at them.
Distortion Rise in Bass
As the frequency gets lower and lower headphones will often have a harder and harder time reproducing the long wave form faithfully. My current working theory is that in open headphones it happens because the driver goes below its primary resonance and begins to be unable to properly track the waveform or begins to hit its physical movement limits. In sealed and on-ear headphones it generally happens when you reach a limit of how much pressure can be contained in the chamber between the ear and the driver, and the waveform begins to soft-clip.
The plots above are three different dynamic, open, circumaural headphones. The red lines are positioned at the primary driver resonance, which is indicated by the peak in the impedance plots (center row of graphs). You can see in the frequency plots at the top of the image that this is the point at which the bass starts to roll off. You can also see that it is generally just above this frequency that the rise in THD+noise begins. Notice on all threeTHD+noise plots this is a straight line. This is very common, though not optimal, behavior for dynamic open headphones.
The other kind of bass distortion I see is a curved rise in THD, which I think comes from a limitation of how well the ear pad is holding pressure at low frequencies. It's often seen in on-ear type headphones, but it can be observed with most headphone types.
Both types of rising distortion in the bass are audible and generally impart a loose, wooly sound depending on the magnitude of distortion. Headphones without the bass distortion are likely to sound tight and punchy in comparison.
Another feature to be seen in the THD+noise plots are peaks both wide and narrow. These peaks arrise usually due to resonances or diaphragm modal break-up.
When interpreting THD+noise it's a good idea to look for related features on the impedance and frequency response plots for clues to the nature of the peak. The plot above is an Oppo PM-1. You'll notice it has elevated THD between 200Hz and 500Hz. You can also see that it has a couple of bumps in it's impedance in that area, as well as some wiggles in its frequency response there. What we're seeing is a pair of diaphragm resonances from the just the central part, and then the whole diaphragm surface. The magnitude of the distortion is just 1% at 100dBsplwhich is pretty looudso I would consider this a minor problem for these headphones.
The AKG K812 shown above is a good example of something being quite obviously wrong. We can see a clear rise in THD between 1kHz and 3kHz, and we can see some associated wiggles at those frequencies in the impedance plot. Generally speaking, the impedance plot should be smooth and relatively featureless. It's okay to have a fairly large hump from the primary driver resonance, and there are usually some small bumps in the 2-6kHz area from resonant spaces around and behind the driver, but the curve should look smooth and regular overall.
On the above AKG K812 impedance curve you can see that the bumps and wiggles starting at 1.5kHz are rather irregular shaped. This is not a good sign and coupled with the broad rise in distortion indicates some significant driver problem in this area, possibly modal break-up of the diaphragm. That the distortion only just surpasses 1%and not taking into consideration other plots for this headphoneis an indicator this problem may be modest.
On the other hand, here is a plot of the Beyerdynamic T5p and we can clearly see distortion exceeding 1% even at 90dB, and reaching 10% at 600Hz at 100dB. The associated impedance plot is also extremely irregular. This headphone would be expected to sound very hard through the mid-range.
Unlike the broad bump increase in the previously discussed headphones, the Skullcandy Grind, shown above, exhibits three distinct narrow peaks in the THD+noise response at about 1.6kHz, 2.1kHz, and 3.7kHz, and relatively small and regular related features on the impedance plot. This leads me to believe these are resonances in the space behind the driver rather than modal break-up, and even though the THD peaks exceed 1% at 100dB, this symptom is less damaging to the sound quality than the previous two headphones.
The last headphone we'll check out for a distortion peak is the Audeze EL-8 above. The THD+noise plot look really, really good...until you get to 4kHz where it starts to rise significantly. The impedance plot appears unchanged, however. These are planar magnetic headphones, which will often times not exhibit changes to impedance with resonance problems. But, if you look at the frequency response plot, you will see a very large dip in response centered at 7kHz that's likely due to a resonance of some type. The rise at the very end of the distortion plot is very likely related to the dip in frequency response.
THD+noise sweeps are done at both 90dBspl and 100dBspl. When the 100dBspl sweep is plotted, the increased volume improves the systems signal to noise ratio, and because THD+noise includes a noise component, the result can be a lower measured value at higher volumes...as long as the headphones aren't increasing in distortion at the higher volume.
Above is the THD+noise plot from an Audeze LCD-2. The entire 100dBspl plot (green and yellow) lies below the 90dBspl plot (red and blue). This means that even though the headphone is being played quite a bit louder there is very little increase in distortion otherwise it wouldn't have remained under the 90dBspl plot. This headphone has good power handling; it can be played loud without distorting.
The Ultrasone HFI-450, on the other hand, shows the 100dBspl plot significantly above the 90dBspl plot indicating it is increasing in distortion significantly as volume increases. This headphone has poor power handeling.
THD+noise Examples Summary
THD+noise plots are the measurement most prone to error in the InnerFidelity headphone measurement spreadsheets. Be careful not to over-interpret small features that may be cause by environmental noise during the measurement and not the headphones.
THD+noise plots are good at identifying three things:
- Rising distortion in the bass due to the driver running out of steam when it goes below its primary resonance, or due to leaky pads that can't retain the pressure needed for low frequencies.
- Distortion peaks due to resonances and diaphragm break-up. These features are usually accompanied by features in the impedance and frequency response curves that can help you determine what kind of problem is in evidence.
- Because THD+noise is measured at both 90dBspl and 100dBspl, it is possible to see if distortion is rapidly increasing with volume or not, which will indicate whether the headphone has good power handling or not. As a general rule: the more the 100dBspl plot is below the 90dBspl plot, the better power handling it has.