Lately I’ve been writing a codec to communicate over a digital audio bus called a I2S. I2S actually stands for Inter-IC Sound and is meant to pass digital audio between ICs (duh) in large signal processing systems. Its actually fairly simple, but debugging the particular chip I’m sending data to is a pain. So I spent much of today tracking down random clicks and pops that have been showing up in my output.
As I was debuging, I would listen to the output post- Digital to Analog Converter (which happens to be what I’m sending data to), and had noticed it sounded a bit off. I shrugged it off at the time as the output was significantly quieter than the input I was comparing it to. After a certain point, however, I hit a wall and needed something to take my mind off the problem for a bit so I decided to measure the frequency response of the converter.
The device in question here is TI’s TLV320AIC23. Its an all-in-one Analog/Digital interface chip; that is, it contains A/D (Analog to Digital) converters, D/A (Digital to Analog) converters, headphone amps, and a microphone input. For this test, I was passing audio into one of the A/D converters, then immediately back out one of the D/A converters. Thus, the resulting frequecy response is that of an A/D and a D/A converter in series. Here is the measurement taken in Smaart 6.1.
The upper graph is phase response; the lower one is magnitude.
Now, in all fairness to TI, A/D converters are stupid hard to get flat. That ripple shown above is caused by the need for an extremely sharp roll-off in the upper end of the spectrum. That’s actually why, often times, higher sampling rates sound better (this is sampled at 48KHz, btw). Contrary to popular belief, it doesn’t have anything to do with sampling “more” of the waveform. Rather, its because with higher sampling rates, the resitrictions relax significantly on how sharp the rolloff needs to be at the top, allowing designers to end up with acceptable stop-band rejection (roll off) while minimizing pass-band ripple (see above).
That having been said, I’ve certainly heard much better converters and have no intention of using this particular converter in any final designs. However, for the time being I’m stuck being that the TLV320 is the only audio I/O built into the development kit I’m using.
After completing that, I thought it’d be interesting to find out what the response of my USB pre (M-Audio FastTrack Pro). It was a bit more difficult, but long story short, I recorded the output looped back to the input, dropped the wave files into Matlab and BAM!
Upper is Magnitude; Lower is phase.
Zoomed in on 1KHz to 22KHz.
The ripple response on this guy is much much better (almost +/- 0.05 dB). The only elephant in the room here is a big ole scoop from about 50Hz to about 200Hz. Uppon further examination (a larger FFT), the scoop only drops about -0.8 dB at its lowest; contrary to what this graph says. Still odd… Fortunately, I really only use this pre for measurement (similar to the Smaart plot above), which means as long as both channels are the same, the response of the device itself doesn’t actually matter.
Anyway, what’s the lesson to be learned today? If ever you should hear an audio engineer talking about how a certain preamp or converter sounds better than another, don’t just write it off as a placebo. Different pieces of gear do actually sound significantly different. It can be especially heated when you’re talking about the difference between analog and digital gear. I think the above plots say all that really needs to be said on that topic.
