IEEE Solid-States Circuits Magazine - Winter 2022 - 16

SHOP TALK: WHAT YOU DIDN'T LEARN IN SCHOOL
Chris Mangelsdorf
The Other ADC Test
M
Get Your Head Out of
Your Spectrum!
Most converter folks I know are obsessed
with spectral performance, and rightly
so. After all, most analog-to-digital
converters (ADCs) these days work
as " waveform coders " ; their output is
intended for some frequency-domain
purpose. Of course, time-domain
" measurement " applications are still
out there, especially in the high-precision
realm, but the communications
boom has caused many to abandon
integral nonlinearity and differential
nonlinearity (DNL) for spurious free
dynamic range.
So, if the spectral performance
is what matters, why look elsewhere?
Here's the thing: the Fourier transform
might as well be a hash function. It's a
Digital Object Identifier 10.1109/MSSC.2021.3126985
Date of current version: 24 January 2022
20
-100
-80
-60
-40
-20
512
1,024
Frequency Bin
(a)
1,536
2,048
1,000
2,000
-2,000
-1,000
1,024
2,048
3,072
Original Time Sequence (Sample Number)
(b)
FIGURE 1: This is how most people see their ADC. Can you spot the errors? That's a rhetorical question. (a) Spectrum. (b) Time-domain data
used for the preceding FFT.
16 WINTER 2022
IEEE SOLID-STATE CIRCUITS MAGAZINE
4,096
one-way beast like a cryptoalgorithm. It
will tell you if you got things right, but
it will not tell you how to fix the problem
if you didn't. Some people may be
able to look at a fast Fourier transform
(FFT) and tell you what's broken, but
not most mortals. It all looks alike.
There are just too many ways to create
similar distortion and errors.
There's another problem: most
people ignore phase information. FFT
phase data are not even available in
some ADC software, even though they
are always present in the raw output of
the FFT algorithm. Again, this is not a
spec problem, but a diagnostic limitation.
The ADC user cares only about the
size of the spurs. But for anyone trying
to fix or improve the ADC design,
the phase of the distortion is a huge
clue. (Tip: The phase of the harmonics
relative to the fundamental versus
frequency is another monster clue.)
Take the simple example of amplitude
clipping: the phase of the second harmonic
indicates whether the clipping
is happening at the positive peak or
the negative peak. You'd want to know
that, right?
In what follows, we'll be looking at
ADC output data in some sort of time
sequence. You'll find it much easier to
infer what's wrong with the converter
or its driver. And the phase data that
you threw away when looking at the
FFT are automatically incorporated
in the result. So, keep the spectral
plots for production testing, but have
a look at the time domain when you
need to debug anything.
The Clouded Mind Sees Nothing
" But I looked at the time domain
data and couldn't see anything, " you
whine. Sure, that's because your timedomain
data looked like Figure 1(b).
Magnitude dBc
ADC Output Code

IEEE Solid-States Circuits Magazine - Winter 2022

Table of Contents for the Digital Edition of IEEE Solid-States Circuits Magazine - Winter 2022