[sdiy] Measuring Audio Sine Purity

[Robin Whittle]

Hi Jerry,

Thanks for pointing to this page:

> http://dogstar.dantimax.dk/tubestuf/thdconv.htm

I think the approach of using a high quality analogue to digital
converter system - such as those which handle 8 channels and connect to
an RME sound card via ADAT plastic optical fibre - is a good one.

These 8 channel ADC-DAC units typically have 20 or 24 bit sigma-delta
converters which are highly linear and have good phase and frequency
response.  Digitizing a nearly full-scale sine-wave with small harmonic
distortion products is easy for these devices.  (If they were to have
audible problems, I expect they would be with extremely complex signals,
rather than sine-waves with mild distortion.)  Even if the output is
reduced to 16 bits, there is a vast dynamic range which will show up any
distortion components, even if they are far below what anyone could
perceive.

In 2009 I was investigating microphone distortion at high sound
pressures.  I generated a 1kHz sine signal in an audio editing program,
perhaps an old Windows program Cool Edit Pro 1.0 (Syntrillium Software)
from 1997), recorded this to a WAV file and recorded this on a DAT tape.
 The DAT player (Sony TCD-D8) uses a novel 18 bit Burr Brown (now TI)
PCM69AU DAC:

  http://www.datasheetcatalog.org/datasheet/texasinstruments/pcm69.pdf

I hadn't looked at this DAC in detail before.  It is a combination of
conventional R-2R network DAC and a sigma-delta DAC.  This is driven by
an 8 times oversampling digital filter (SM5840B).  The DAT recorder
produced a suitably low noise and low distortion 1kHz signal which I fed
to a stereo amp.  (I had an inverted version on the right channel and
fed the two channels into a stereo amp so I could drive the loudspeaker
in bridge mode and get twice the power.)

The speaker, a massive soft-dome midrange ATC SM75-150S, drove the
microphone which was enclosed in a small cavity.  So the speaker was
driving into this sealed cavity, rather than radiating into space.  This
created very high sound levels, which I somehow roughly calibrated to be
115, 120, 125, 130, 135 and 140dbA by turning the volume control.

The electret mic and my test pre-amp drove one channel of an 8 channel
Fostex VC-8 converter.  This drove an RME sound card (which doesn't
alter the bits) via optical fibre, so there was complete electrical
isolation from the electrically noisy PC.

The VC-8 is probably regarded as a primitive device now that everyone is
into 96 and 192kHz sampling with so-called 24 bit converters.  It runs
at 44.1 and 48kHz and uses four 20 bit AKM AK4522 dual ADC, dual DAC
chips.  The ADC is a 64x oversampling sigma-delta / delta-sigma (these
mean the same thing) device from AKM in Japan, who worked with Crystal
Semiconductor in the USA.  Crystal's design and AKM's manufacturing
pioneered sigma-delta ADCs and then DACs.  These Crystal/AKM ADCs in the
late 80s were hugely significant in providing low noise, flat frequency
response, low distortion, low phase "distortion" ADCs which finally
provided high quality conversion, replacing conventional successive
approximation R2R network ADCs and the difficulties these normally
incurred with oversampling filters, distortion and phase problems in the
oversampling filters.  Crystal Semiconductor was acquired by Cirrus in
1991.  Cirrus still make some CSxxxx chips which I think date from
Crystal/AKM days.  AKM still make the same or a similar range of
devices, so maybe it is AKM making them for Cirrus:

  http://www.cirrus.com/en/products/a-d_converters.html
  http://www.akm.com/prod-adc.asp

After modifying a noisy internal powersupply inside the VC-8, I measured
the ADC performance as being about 2 to 3 bits better than the optimal
dither noise which would be required to linearize the steps of a perfect
noise-free 16 bit ADC.  So in my view, these ADCs are effectively 18 or
19 bits, which is very good.  Just because other ADCs put out 24 bits
doesn't mean their noise floor is necessarily 24 bits, though they may
be somewhat better than this 18 or 19 bits.

This is a long way of saying that although the DAC and ADC in this setup
are both of rather old designs, from the 1990s, I think their
performance was more than adequate to the task of handling a 1kHz sine
wave and looking for harmonics at 2kHz, 3kHz etc.

I recorded the ADC signal as a 16 bit wave file at 48kHz because I think
Cool Edit Pro couldn't handle 24 bit WAV files.  Then I used Cool Edit
Pro's spectrum analysis function on selected parts of the recording.
Even with a relatively low FFT size of 1024, the noise level in each of
the 1024 frequency bins was quite low - about -115dB.  This was plenty
low enough since I had the 1kHz signal at about -13dB and the strongest
of the harmonics, at 2kHz, was at -60dB.  Here is a screenshot:

  http://www.firstpr.com.au/temp/sdiy/distortion-analysis.png

I could easily see that my distortion products at 140dBA were quite low,
such as 47dB below the main signal.  This would be high for an
amplifier, but I thought it was quite low for a microphone being driven
at this sound level.  This is the sum of multiple types of distortion:

  In the DAT recorder's DAC - very low, I think I checked it.

  In the hi-fi amplifier - likewise.

  In the loudspeaker - there's no way of testing it independently, but
  this is a humongous mid-range driver operating in the middle of its
  frequency range.

  In the microphone and my pre-amp.  I guess most of the measured
  distortion was from here, but some of it would have been from the
  speaker.  There's no way I could separate the two, without the use
  of another microphone, or perhaps by using the same microphone via
  an acoustically attenuated version of the same drive signal, so
  the microphone was operating at a lower level.  If this showed lower
  distortion, then this would indicate that the higher level of
  distortion at 140dBA was due to the microphone itself operating at
  this high sound level.

I think I tested the distortion of the DAT, hi-fi amplifier and ACD in
series and found it to be very low.  So the distortion I found was
almost entirely the sum of that in the speaker, the mic and the mic pre-amp.

  - Robin