Test 18-bit ADCs with an ultrapure sine-wave oscillator
With careful design, this circuit challenges test equipment's ability to verify its performance.
Jim Williams and Guy Hoover, Linear Technology -- EDN, August 11, 2011
At A Glance
|
The ability to faithfully digitize a sine wave is a
sensitive test of high-resolution-ADC fidelity.
This test requires a sine-wave generator with
residual distortion products approaching 1 ppm
(part per million). It also requires a computer-based
ADC-output monitor to read and display
the converter’s output spectral components.
Performing this testing at reasonable cost and
complexity requires the construction of its
elements and performance verification before its use. A low-distortion
oscillator drives the ADC through an amplifier (Figure
1). The ADC’s output interface formats the converter output,
which communicates with the computer. The computer executes
spectral-analysis software and displays the resulting data.
Oscillator circuitryThe system’s oscillator is the most difficult-to-design part of the circuit. The
oscillator must have transcendentally
low levels of impurity to meaningfully
test 18-bit ADCs. You must then verify
these impurity characteristics by independent
means.
Start with a design based on the work of Winfield Hill, director of the electronics-engineering laboratory at the Rowland Institute at Harvard University. You can then adapt this design for a 2-kHz Wien-bridge design (Figure 2). Using all of the amplifiers in inverting mode eliminates CMRR (common-mode-rejection-ratio) errors from the signal path.
Low-distortion amplifiers A1 and A2 are the active components of this oscillator. The JFET of the original design would introduce conductivity-modulation errors, so you can replace it with an LED-driven CdS (cadmium-sulfide) photocell isolator. You then combine the output of A2 with a filtered dc offset at the input to A3. The capacitor in A3’s feedback network limits the bandwidth of the amplifier. The output of this 2.6- kHz filter drives the input amplifier of the ADC under test.
|
Click here to read more articles by Jim Williams. |
Current-summing resistors can be used to balance the dc
value against a voltage reference that the Linear Technology LT1029 IC creates. The current-summing
resistors feed the AGC single-supply amplifier, A7. This amplifier
drives Q1, which sets the LED current. The LED current
closes a gain-control loop because it ultimately varies the CdS
cell’s resistance, stabilizing the oscillator’s output amplitude.
By deriving the gain-control feedback from the circuit’s
output, you maintain the output amplitude, despite the attenuating,
bandlimiting response of A3 and the output filter. This
topology also places demands on the loop-closure dynamics of
amplifier A7. A3’s bandlimiting, the output filter, A6’s lag, and
the ripple-reduction components that attach to Q1’s base combine to generate a significant amount of phase delay. You can
accommodate this delay with a 1-μF dominant pole at A7, along
with a zero-value RC (resistor/capacitor), to achieve stable loop
compensation. This approach replaces closely tuned high-order
output filters with simple RC roll-off responses, minimizing
distortion and maintaining constant output amplitude.
It is essential that you eliminate oscillator-related signal
components from the LED bias to maintain low distortion.
Any such residue modulates the oscillator’s amplitude, introducing
impure frequency components. The bandlimited AGC
signal path is well-filtered.The heavy RC time constant in Q1’s base provides a final,
steep roll-off response. Q1’s emitter current shows approximately
1 nA of oscillator-related ripple from a 10-mA total—less than
0.1 ppm (Figure 3). The oscillator needs only one 100Ω trim
to achieve its performance. This adjustment is set in accordance
with the notes in Figure 2 and centers the AGC’s capture range.
Oscillator distortion
Verifying oscillator distortion necessitates sophisticated measurement
techniques. You will encounter limitations if you
attempt to measure distortion with a conventional distortion
analyzer, even a high-grade type. An oscilloscope can be used
to indicate distortion residuals at the analyzer’s output (Figure
4).The amplifier’s floor faintly outlines noise and uncertainty
on any signal activity that relates to the oscillator.The Hewlett-Packard HP-339A analyzer specifies a minimum measurable distortion of 18 ppm. The figure shows the instrument indicating 9 ppm, which is beyond the unit’s specification and, hence, highly suspicious. Measuring distortion at or near the limits of your equipment yields pronounced uncertainties. Distortion measurements at or near equipment limits are full of unpleasant surprises (Reference 1).
Specialized analyzers with low uncertainty floors are
needed to meaningfully measure oscillator distortion. The
Audio Precision 2722 analyzer has a maximum
2.5-ppm THD+N (total harmonic distortion plus noise)
and a typical THD+N of 1.5 ppm. This instrument measures
the oscillator’s THD in three tests and finds THD
figures of −110, −105, and −112 dB at 3, 5.8, and 2.4
ppm, respectively (Figure 5). These measurements provide
confidence in applying the oscillator to ADC-fidelity
characterization.ADC testing
When you test ADCs, you route the oscillator’s output to the ADC through its input amplifier. The test measures distortion products produced by a combination of the ADC and the ADC’s input amplifier. You then examine the ADC’s output with a computer, which quantitatively indicates spectral-error components (Figure 6).
You can download the code to take measurements and obtain input-amplifier, ADC, computer-data-acquisition, and clock boards from the Linear Technology Web site. Appropriate parts include an oscillator; the Linear Technology LT6350 amplifier; the LTC1279 ADC; the DC718 interface card; and any stable, low-phase-noise, 3.3V clock capable of driving 50Ω.
The computer display includes
time-domain information showing
the biased sine wave centered in the
converter’s operating range. It also
displays detailed tabular readings and
a Fourier transform indicating spectral-error components. The amplifier/ADC combination under test produces
second harmonic distortion of −111
dB, which is approximately 2.8 ppm.
The higher-frequency harmonics are
well below this level, indicating that
the ADC and its input amplifier are
operating properly and within specifications.
Harmonic cancellation
may occur between the oscillator and
amplifier/ADC combo, mandating
that you test several amplifier/ADC
samples to enhance your confidence
in the measurement.|
Reference |
|
Talkback
-
It would be very much appreciated and save many reader's time if we could download the LTspiceIV model for this wonderful circuit.
Mark Clemmons - 2011-23-8 19:44:02 PDT -
Testing ADCs in the frequency domain is not a new concept. Back in 1986 we were incorporating the latest generation of 16-bit 125kHz ADCs in the front end of our whole-body magnetic resonance imaging systems. Improvements in our antenna designs meant that the noise floor of the demodulated signal was suddenly 10dB lower than it had been and our images were now full of artifacts. Understanding that our images were just 2-D FFTs, any patterns in the image ("stars and stripes") could be attibuted to other periodic signals caused by harmonic distortion due to jitter and nonlinearities. At my insistance we began a project to qualify our ADCs based on their frequency domain performance. I had a diploma student as my sidekick and his thesis was about how we designed, built and qualified the system. The test results led to us changing vendors to an upstart out of California and a huge improvement in image quality.
Jeff The FFT Guy - 2011-23-8 15:12:47 PDT -
I'm a huge admirer of Jim Williams work! His designs always transcended the obvious to reach greater performance.
In this app, I wonder why a 3 op-amp solution was employed for level control instead of an LTC1966 rms-dsc converter?
richard becker - 2011-17-8 19:15:01 PDT -
In the second from last paragraph it says
"You can download the code to take measurements and etc." I can not find this ?
L H Sparrow - 2011-17-8 07:29:53 PDT -
And if it needs to be better still, consider using a passive filter on the output. Difficult but not impossible to make an air-core L that's big enough (air-core so as to be linear), at least for not-terribly-unreasonable values of C. It would be fun.
And to measure it: notch out the signal (with a notch filter made with your linear passives), then gain-up the residual with a low-noise amp. Notch doesn't have to be perfect, just known depth. More fun.
Steve Cahill - 2011-17-8 00:28:29 PDT



















Jim Williams was a staff scientist
at Linear Technology
Corp, where he specialized
in analog-circuit and instrumentation
design. He served
in similar capacities at National
Semiconductor, Arthur D Little, and
the Instrumentation Laboratory at the
Massachusetts Institute of Technology. He
was a former student at Wayne State University
(Detroit) and enjoyed sports cars,
art, collecting antique scientific instruments,
sculpture, and restoring old Tektronix
oscilloscopes. A long-time EDN contributor,
Williams died at age 63 in June
2011 after a stroke.
Guy Hoover is an applications
engineer at Linear
Technology Corp in the
mixed-signal-products group
supporting SAR (successive-approximation-register)
ADCs. He has a bachelor’s degree in electronics-engineering technology from DeVry
Institute of Technology (San Francisco,
CA). Hoover has written several application
notes and articles. His hobbies include
watching classic sci-fi movies and collecting
high-tech gadgets.



