On the Measurement and Audibility of Headphone Break-in
A couple of days before I left for CanJam at the Rocky Mountian Audio Fest, I stuck a brand new pair of AKG Q701 on my Head Acoustics measurement head, positioned it properly in the measurement chamber, and closed the door. Then I waited a couple of days for the ear pads to settle in. Before heading out the door for the 700 mile drive to Denver, I pushed the button to start the last break-in test.
For the next 300 hours or so, the headphones played pink noise at 90dBSPL for an hour, then rested silently for 10 minutes to let it cool down, and then underwent a battery of tests including: frequency response; THD+noise vs. frequency; intermodulation distortion spectra, and impulse response. The system glitched a couple of times and had to be restarted, so the test was really done in three roughly 100 hour tests; there was a one day break between the first two, and a couple of hours break between the middle and last data sets. (For some dumb reason, Excel would choke at about 110 hours. Argh!)
A total of about 1.7 million individual measurements were made over the 330 hours.
After the test was completed, and the spreadsheet of the three chunks of data assembled onto one spreadsheet, I sent it off to a couple of folks for a look-see. Arnaud was a real trooper and did a bunch of number crunching. It took him numerous tries to get it to do his bidding, but after persistent effort, he managed to produce some movies of the data changing over time. Let's have a look ....
Here's Arnaud's video of the frequency response over the 330 hours of testing. The first bit of the video toggles between the beginning and the end FR, and then the video goes through all 330 hours of measurement.
Here's what I see in this data:
- At about 2kHz I see a dip that kind of breathes back and forth. We'll see this action in a more pronounced way in the cumulative spectra decay plots. I believe what we're seeing here is the day/night temperature cycle. We did see some similar artifacts in previous experiments.
- I see the data jump around sometimes. I'm not sure where this is coming from, but I think Arnaud might have been calculating FR from impulse response data. Because the system sets the level for the measurement using an maximum length sequence signal, which sounds much like pink noise, it might be that the signal level of this test is not set reliably at the same number due to the noisiness of the signal.
- If you assume the "breathing" and the glitches are not from break-in, there's not much left to point at as artifact that might point in that direction. If break-in happens and is measurable, it doesn't seem to be visible in the frequency response data.
I've offset the right channel down 5dB for clarity. I breathed a sigh of both relief and angst looking at these data. It's fairly obvious that:
- The system is performing quite reliably, as the start and finish plots of both channels overlay each other almost perfectly over roughly 12 days of measurements. I'm very happy to see the gear looking so stable.
- There is virtually no change in FR data over the 330 hours. If break-in exists, it's not affecting basic frequency response.
THD+Noise vs. Frequency
Having measured THD+noise Vs. Frequency for this test, I plotted the data over time.
This test is very sensitive to sound in the measurement environment. If a large truck happens to be driving down the street outside my home during the test, it could very easily cause a blip in the measurement. Here we can see some random blips here and there, but no observable change that trends over time. Bottom line: I see nothing here to indicate break-in.
I guess we're going to have to look elsewhere. Fortunately, Arnaud also made movies of cumulative spectral decay. Let's have a look ...