InnerFidelity Headphone Amp Measurement Program Update July 2012
The headphone amp measurement routine consists basically of two things: (1) the main program to run the various tests and (2) the tests themselves. The individual tests (Frequency Response, THD+Noise Vs Frequency, SMPTE IMD, etc.) are configured with a set of panels on a computer that runs the Audio Precision tester. Then a program is written that calls the various tests in sequence, prompts the operator to do various things like change loads or switch cables, and pokes all the data into an Excel spreadsheet as it is gathered. So far, I've been working on the tests, and Brian (NA Blur on Head-Fi.org) has been working on the main program to execute the tests, and the spreadsheet that accumulates and displays the data.
Over the last couple of weeks, we've both been working on our respective tasks. Brian has managed to set up the software on a computer in his home; has started looking at the headphone measurement test to see how the programming language works; and has started to configure the Excel spreadsheet template in which the data is stored. There is a mode in the software that allows it to simulate being hooked up to the Audio Precision tester, and that will allow Brian to do the basic software development at his home. I can't express how grateful I am to Brian for voluntarily taking on this task (well, he'll get some headphones out of it) and helping me get this ball rolling. I can manage the programming, but I'm sure he'll be able to do it much quicker than I. Feel free to thank Brian in the comments, or NA Blur on Head-Fi if you bump into him there.
My task over the last couple of weeks has been to run and tweak the tests I've been slowly configuring on NwAvGuy's O2 headphone amp, and compare my measurements with his measurements with the hope that they match fairly well. If you've not heard of the O2 amplifier, it is NwAvGuy's attempt to design a very inexpensive amp that performs very well and can be built by DIYers. My experience with this amp so far is that it's very clean sounding, and does indeed measure very well. You can find out about it here.
The tests I'm configuring are designed to match up with NwAvGuy's recommendations in his InnerFidelity post here. In that article he recommends producing a datasheet that looks like this mock-up:
Let's have a look at the O2 measurements I've done in the last fortnight, and how well they match up with NwAvGuy's measurements. Graphs with black backgrounds are NwAvGuy's and mine have white backgrounds:
This is essentially a measure of Gain vs. Frequency. Ideally, a headphone amplifier should have the same gain at all frequencies producing a "flat" frequency response plot. Apart from the fact that NwAvGuy's measurement gear can go down to 5Hz and my AP can only go down to 10Hz, you can see that the two measurements are quite similar.
This is a plot of how much energy is at each frequency without any input signal. What you're seeing in this plot is spectral density of noise in the amp. It is in this plot that you'll typically observe AC power hum with peaks at the AC fundamental at 60Hz and some harmonics at 120Hz and 180Hz.
As you can see, my measurements have some additional artifact peaks between 50Hz and 600Hz, and about a 4dB higher noise floor. It's not bad, but I'd like to do better. Unfortunately, these low-level problems usually amount to things like grounding and cable lengths and shielding. Usually not too much fun to hunt down. That's life, I guess.
Total Harmonic Distortion + Noise vs. Frequency
This plot measures the amount of distortion vs. frequency with various loads on the output. As the load resistance gets lower and lower, the amp has a harder and harder time swinging the amount of current needed. Amps have more difficulty at the frequency extremes here, which is why the curve is "U" shaped; the drop-off in THD+noise above 7kHz is actually a measurement artifact as the harmonic distortions being measured begin to exceed the measurement system's bandwidth setting. For example, at 7kHz the third harmonic is 21kHz, and because the system bandwidth is set to 20kHz, the harmonics begin to be attenuated when the fundamental tone rises above 7kHz.
These two plots match up pretty nicely! In fact, my measurements show slightly less THD+noise, which on first blush might indicate my system has less noise in its measurements. I will try to spend some time talking with NwAvGuy about these data--his help will likely be invaluable in interpreting the measurements in order to troubleshoot potential problems.
Total Harmonic Distortion + Noise vs. Output Level
This plot shows the amount of THD+noise from a 1kHz signal as output voltage rises. The steep slope at the extreme left of the plot occurs because as the output gets lower and lower the ratio of noise to output signal gets higher and higher. In other words, at the low end of output voltage, the measurement is dominated by the amplifier's noise.
As soon as the output signal gets strong enough the THD+noise measurement becomes dominated by the harmonic distortion and flattens out. As the output voltage rises at some point, the amplifier begins to "clip" and the distortion begins to rise dramatically. By placing various resistive loads across the amp and running this test, we can begin to see how good the amp is at delivering clean power to headphones of differing impedance. By calculating the power being delivered at the "knee" of the curves (the point to the right at which each curve begins to rise dramatically) we can get a look at the output power limit of the amp into different loads.
- With 600 Ohm and the knee at 6.8 Vrms indicates 77 mWatt.
- With 150 Ohm and the knee at 6.6 Vrms indicates 290 mWatt.
- With 32 Ohm and the knee at 3.8 Vrms indicates 451 mWatt.
- With 16 Ohm and the knee at 2.1 Vrms indicates 275 mWatt.
Bottom line here is that both sets of plots are remarkably similar...and that makes me quite happy!
CCIF Intermodulation Distortion
Intermodulation distortion comes from two signals non-linearly mixing, which produces sum and difference signals. In the case of the CCIF IMD (CCIF is a French standards setting organization: International Telephone Consultative Committee), the two tones used are at 19kHz and 20kHz. When these two tones are mixed together, you will get a difference tone at 1kHz. This difference tone will also intermodulate with the two probe tones to produce distortion peaks with 1kHz spacing around the probe tones.
NwAvGuy's measurement gear goes higher in frequency than mine, and my measurement is set to show down to -140dBr (his can go that low, but for whatever reason it was set to -130 for his measurement), otherwise the two plots are quite similar. Differences could be chalked up to very slight differences in system noise and grounding. Again, I'm going to have to chat with him to fully asses my measurements and system performance.
SMPTE Intermodulation Distortion
This is another standardized method for testing IMD, this time from the SMPTE (Society of Motion Picture and Television Engineers). In this case, the probe tones are at 60Hz and 7kHz, with the 7kHz signal at 1/4 the power of the 60 Hz tone.
As with the CCIF test the plots are similar enough to make me happy, but contain enough differences that I want to get a little education from NwAvGuy on the results.
Woot! We're making progress here! The above plots show much progress has been made in getting a headphone amplifier characterization capability in place for InnerFidelity. There's still some room to tweak grounding and cables to see if the measurements can be improved--and that's the name of the game with this type of endeavor--but it seems to me that the measurements are so close that the differences might simply be chalked up to different measurement systems in different environments. Nonetheless, my next step is getting together with NwAvGuy and working on getting any little gremlins out of the measurement system and methods.
I gave Brian the data from these measurements and he will be working on a mock-up of the spreadsheet that produces the graphical datasheets for the amps. And he'll also be working on the program to automate the tests. I suspect we'll see some mock-up datasheets sometime next month.
Damn glad to see progress here!