You are on page 1of 11

Display,

Sensor actuators,
Input Processing Output
signals,
control

ADC Testing
Part 7 in a series of tutorials in
instrumentation and measurement

Thomas E. Linnenbrink, Jerome Blair, Sergio Rapuano, Pasquale Daponte,


Eulalia Balestrieri, Luca De Vito, Solomon Max, and Steven J. Tilden

nalog-to-digital converters (ADCs) are test- The following sections address general test setups,

A ed for several reasons. Manufacturers need


to measure the ADC’s performance so that
they can guarantee to their users what per-
formance to expect of the ADC (usually via a specifica-
tion sheet) and to assure the quality of the ADCs they
common test methods, and the assessment of uncertainty
in measurements. The same tests are also performed on
digital waveform recorders, but the test setups are differ-
ent and are described in IEEE Std 1057-1994. See “ADC
Versus Waveform Recorder—What’s the Difference?” for
produce. Users need to measure the ADC’s performance more information.
relative to its specification as well as for its intend-
ed use. Both static and dynamic performance Test Setups
parameters may be assessed. It is impor- A few general test setups can be used
tant for those measuring ADC perfor- to perform most of the ADC tests. We
mance to communicate their findings discuss several primary types below.
clearly. To that end, the IEEE
Instrumentation and Measurement Sine Wave Test Setup
Society’s Waveform Generation, Sine waves are commonly used in
Measurement, and Analysis ADC testing because appropriate
Committee (TC-10) developed the sine wave sources are readily avail-
IEEE Standard for Terminology and able and because it is relatively easy to
Test Methods for Analog-to-Digital establish the quality of sine wave (e.g.,
Converters (IEEE Std 1241-2000) [1]. This with a spectrum analyzer). The sine wave
standard presents a wide range of terms and test setup, shown in Figure 1, comprises a
test methods to serve as a common technical lan- sine wave generator providing the test signal and a
guage among users and manufacturers. clock generator providing the clock (or conversion) signal.
Since it would be prohibitively expensive for both If frequency synthesizers are used to generate the test and
manufacturers and users to perform all possible tests, clock signals, the synthesizers can often be phase locked to
only parameters critical to an application are generally maintain precise phase relationships between the signal and
assessed. Table 1 suggests parameters critical to common the sampling clock. Phase locking of synthesizers facilitates
applications. These suggestions are intended to provide testing and simplifies subsequent digital signal processing,
a starting point from which to select a set of parameters by preventing clock/signal walkthrough (beat patterns),
for a particular application. which may artificially increase or reduce measured spurious

April 2006 IEEE Instrumentation & Measurement Magazine 39


1094-6969/06/$20.00©2006IEEE
output. The type of circuitry used to capture the digital data clock or signal paths to reduce noise or harmonic distortion.
samples produced by the ADC is determined largely by the data For example, subharmonics in the clock path will degrade
rate. Slower ADCs may be interfaced directly to the computer. ADC performance, so the clock signal may require filtering
Faster ADCs often require a buffer memory to acquire data at to smooth edges that might otherwise feed through to the
the ADC sample rate and then download stored samples to the signal path. Also, low-pass or band-pass filters may be
computer at a slower rate. Even faster ADCs may require latch- required in the signal path to eliminate noise or other unde-
es or demultiplexers between the ADC and the buffer memory sirable signals (e.g., harmonics).
and perhaps data decimation. A logic analyzer might be used as
a buffer memory to capture data for some tests. Arbitrary Waveform Test Setup
Both the clock and the test signals must be suitable for the The arbitrary waveform test setup of Figure 2 can be used
test being performed. Filters may be required in either the for arbitrary test signals, such as ramps, chirps, and steps.

Table 1. Critical ADC parameters [1].


Typical Applications Critical ADC Parameters Performance Issues
Audio SINAD, THD Power consumption
Crosstalk and gain matching
Automatic control Monotonicity Transfer function
Short-term settling, long-term stability Crosstalk and gain matching
Temperature stability
Digital oscilloscope/ SINAD, ENOB SINAD for wide bandwidth amplitude
waveform recorder Bandwidth resolution/low thermal noise for
Out-of-range recovery repeatability
Word error rate Bit error rate
Geophysical THD, SINAD, long-term stability Millihertz response
Image processing DNL, INL, SINAD, ENOB DNL for sharp-edge detection
Out-of-range recovery High-resolution at switching rate
Full-scale step response Recovery for blooming
Radar and sonar SINAD, IMD, ENOB SINAD and IMD for clutter cancellation
SFDR and Doppler processing
Out-of-range recovery
Spectrum analysis SINAD,ENOB SINAD and SFDR for high linear
SFDR dynamic range measurements
Spread spectrum SINAD, IMD, ENOB IMD for quantization of small signals
communication SFDR, NPR in a strong interference environment
Noise-to-distortion ratio SFDR for spatial filtering
NPR for interchannel crosstalk
Telecommunication SINAD, NPR, SFDR, IMD Wide input bandwidth channel bank
Personal communications Bit error rate Interchannel crosstalk
Word error rate Compression/power consumption
Video DNL, SINAD, SFDR, DG, DP Differential gain and phase errors
Frequency response
Wideband digital receivers SFDR, IMD Linear dynamic range for detection of
SIGINT, ELINT, COMINT SINAD low-level signals in a strong
interference environment
Sampling frequency
Instrumentation INL, DNL, gain, offset Long term stability, traceability

Notes:
COMINT: communications intelligence SIGINT = signal intelligence
DNL: differential nonlinearity SINAD: signal-to-noise and distortion ratio
ENOB: effective number of bits THD: total harmonic distortion
ELINT: electronic intelligence IMD: intermodulation distortion
NPR: noise power ratio SFDR: spurious free dynamic range
INL: integral nonlinearity DP: differential phase error
DG: differential gain error

40 IEEE Instrumentation & Measurement Magazine April 2006


Display,
Sensor actuators,
Input Processing Output
signals,
control

In this setup, the test signal is generated digitally then Common Static Test Methods
converted to analog. You must quantify the performance Static parameters describe the ADCs fundamental perfor-
of the digital-to-analog converter (DAC) and filter to mance with essentially nonvarying or slowly varying input
assess (or remove) its impact on the measured perfor- signals. The ADC’s static performance is often regarded as a
mance of the ADC under test. baseline since the ADC’s performance tends to degrade as
the input signal’s frequency increases.
Step Waveform Test Setup
You can use the step waveform test setup, shown in
Figure 3, for testing with precision step signals that are
not digitally generated. Precision pulses and step signals
ADC Versus Waveform Recorder—
can be used to measure both time domain parameters
What’s the Difference?
here are two IEEE standards, IEEE Standard for
(such as impulse response, transition duration, over-
shoot, and settling time) and frequency domain parame-
T Terminology and Test Methods for Analog-to-
Digital Converters (IEEE Std 1241-2000) and IEEE
ters (such as frequency response amplitude and phase,
Standard for Digitizing Waveform Recorders (IEEE
bandwidth, and gain flatness). If the optional pulse repe-
Std 1057-1994), that cover very similar material. In
tition generator is phase locked to the sampling clock,
fact, all of the figures of merit introduced in the
then you can use equivalent time sampling and simplify
previous article (i.e., Part 6 in this series of tutorials
certain data analysis tasks. You must pay careful atten-
[3]) are essentially identical in the two standards,
tion to the phase linearity of any filters placed in the
and the test methods are the same at the highest
pulse or step signal path.
level. So why are there two standards? A digital
waveform recorder is a complete instrument, such
Feedback Loop Test Setup (Servo Test)
as a digital oscilloscope, that uses one or more
A widely used test method for determining transition lev-
ADCs as an important element, whereas an ADC is
els is based on a feedback loop. In this method, you apply
a single component. Most of the tests in this article
an input to the ADC, trigger the converter, and compare
are easier to perform on a waveform recorder than
the results of the conversion to a desired value. If the
on an ADC, because much of the test setup (e.g.,
ADC output is below the desired value, the input is raised
power supplies, buffer amplifiers, clocks, and
by a fixed amount. If the ADC output is equal to or above
memory) is built into the waveform recorder but
the desired value, the input is lowered by a fixed amount.
has to be supplied by the tester when testing an
You repeat this process until the ADC input has settled to
ADC. So the test setups in the waveform recorder
a stable average value that can be measured by the volt-
standard are generally less complex than those in
meter. After the loop has settled, the input value can
this article and in the ADC standard.
either be measured or, if the input source is well calibrat-
Information about both of these standards
ed, computed from its transfer function. Figure 4 illus-
and other activities of the IEEE Instrumentation
trates a block diagram for the Servo Test. In this diagram,
and Measurement Society’s TC-10 can be found
a DAC generates the feedback signal, but other implemen-
at http://grouper.ieee.org/groups/1057/.
tations are possible, including the classic analog one,
which is shown in Figure 5.

Clock Generator
(Frequency Filter Required
Generator) Optional

Sine Wave
ADC
Generator Latch/ Buffer
∑ Filter Under Computer
(Frequency Demux Memory
Test
Synthesizer)
Programable
Delay
Signal
Generator
(Frequency
Synthesizer)

Fig. 1. Sine wave test setup.

April 2006 IEEE Instrumentation & Measurement Magazine 41


An ADC has four parameters that describe the nature of that, at all points other than the center of the codes, the
the static transfer characteristics: gain, offset, integral nonlin- ideal ADC will not exactly describe the input signal. This
earity (INL), and differential nonlinearity (DNL). One of the deviation is called the quantization error.
simpler approaches to testing involves a single test process, ◗ The center of the codes cannot be directly measured;
ramp testing, which can be used to extract these parameters only the code transitions can be measured.
(with some caveats). The gain of an ADC is nominally defined as 1.0. It is cal-
Figure 6 describes the transfer characteristics of an culated from (1). The values corresponding to the ADC
ideal 3-bit ADC with a voltage input. There are several described in Figure 6 are given in brackets
properties that are demonstrated in the figure that are
worth noting. G × T[k] +V os + ε[k] = Q × (k − 1) + T1 , (1)
◗ The code edges, described by the array T[k], where
1 ≤ k ≤ 2N − 1, describe the transfer characteristic of an where T[k] is the input value corresponding to the transition
ADC. between codes k and k − 1; T1 is the ideal value corresponding
◗ T[1], the lower edge of the output code 0012 , is also the to T[1], [0.5]; V os is the output offset in units of the input quanti-
upper edge of the code 0002 . ty, nominally equal to zero, [0.0]; G is the gain, nominally equal
◗ The full scale range (FSR) of the ADC is, by convention, to unity, [1.0]; Q is the ideal width of a code bin, i.e., the FSR
defined as V max −V min , where V max is the center of a vir- divided by the total number of codes, [1.0]; and ε[k] is the resid-
tual code one greater than the maximum code that can ual error corresponding to the kth code transition.
be generated by the ADC, and V min is the center of the The values of G and V os , which cause ε[1] and ε[2N − 1] to
zero code. As an example, in the 3-bit ADC shown in be zero, are the terminal-based gain and offset. Once the val-
the figure, V max could be 8 V and V min could be 0 V. The ues of T[k] are known, the value of G can be computed as
FSR would then be 8 V, T[1] would be 0.5 V, and T[7]
would be 6.5 V. FSR N
(2 − 2)
◗ Figure 6(b) shows the difference between the output code 2N
G= . (2)
translated into input units and the input signal. Note T[2N − 1] − T[1]

Clock Generator Programable


Filter Required
(Frequency Generator) Delay
Optional

ADC
Control and Latch/ Buffer
DAC Filter Under Computer
Local Memory Demux Memory
Test

Fig. 2. Arbitrary waveform test setup.

Clock Generator
Filter
(Frequency Generator)

Pulse Repetition Required


(Frequency Synthesizer) Optional

ADC
Latch/ Buffer
Pulse Generator Filter Under Computer
Demux Memory
Test

Fig. 3. Step waveform test setup.

42 IEEE Instrumentation & Measurement Magazine April 2006


Display,
Sensor actuators,
Input Processing Output
signals,
control

The offset V OS can be computed from The values of the two end-point transitions can be com-
puted from the following expressions:
V OS = T1 − G · T[1]. (3)

N VP −VM
The values of G and V os , which cause 2k=1−1 ε[k]2 to be a T[1] = VM + Hc [0] (7)
S
minimum, are the independently based gain and offset. VP −VM
T[2N − 1] = VM + Hc [2N − 2]. (8)
The INL can be determined after the gain and offset have S
been determined by
The parameters A and C can then be computed from the end
ε[k]
INL[k] = . (4) point values with the following expressions:
FSR

INL can be expressed either in units of least significant


(T[2N − 1] − T[1])
bits (LSBs) or as a fraction of FSR [parts per million (ppm) A= (9)
(Hc [2N − 2] − Hc [0])
or percent].  
H[0] · (T[2N − 1] − T[1])
The DNL can be determined after the gain and offset C = T[1] − . (10)
(Hc [2 − 2] − Hc [0])
N
have been determined in units of LSBs by

 
T[k + 1] − T[k] The gain and the offset of the ADC under test based on the end-
DNL[k] = −1 . (5)
Q points can be computed once the values of T[k] are known.
The independently based gain and offsets are computed
Determining Transition Levels T[k]— with a least mean squares algorithm, which is described in
Ramp-Based Method IEEE Std 1241-2000.
Apply a ramp to the ADC under
test. The ramp should start at a
known voltage VM, which is Clock
less than V min . The ramp should k
Word
terminate at a voltage VP, which ADC
Reference
Comparator Code
is greater than V max . An example Under A B
kin
of such an input is shown in Test
A<B
Figure 7. Voltmeter
Triggers are applied to the
ADC while the slow ramp is mov- Trig 1
N1 N2
ing from its initial position to its
final position. The total number of
triggers needed should be enough
to trigger the ADC many times at SUB N1 ADD N2
DAC
each step. A count is kept of the
Adder
number of times each code of the
NDAC
ADC is observed. The resultant
histogram of the code hits is
Fig. 4. Feedback loop test setup.
stored in the array H[k], where
0 ≤ k ≤ 2N − 1. The array of the
threshold values can be computed from the histogram with Potential Problems—Ramp-Based Method
There are several concerns that may arise when the ramp-
T[k] = C + A · Hc [k − 1] for k = 1, 2, . . . , (2N − 1), (6) based method is used.
◗ The linearity of the applied ramp directly affects the
where A is a gain factor and C is an offset factor; accuracy of the INL data.
 j ◗ The precision of the starting level and the stopping level
Hc [ j] = H[i] , of the ramp (VM and VP) will directly influence the
i=0 gain and offset calculations.
where H[i], is the number of histogram samples received in ◗ The presence of ADC input noise will affect the accura-
code bin i; and cy of the measurement. Generally, the uncertainty of the
2
N
−1 evaluation of the code transition levels will vary as the
S= H[i] = Hc [2N − 1], square root of the number of conversions that occur at
i=0 each code. This topic is discussed in greater detail in
where S is the total number of samples taken. IEEE Std 1241-2000.

April 2006 IEEE Instrumentation & Measurement Magazine 43


Determining Transition Levels T[k]— The ADC is triggered at a rate near its maximum spec-
Servo-Based Method ified trigger frequency. After every conversion, you com-
A classic method for determining ADC threshold levels is pare the output to the reference code word. If the ADC
known as the servo method. A feedback system is set up to output is greater than the reference code, then the polari-
compare the ADC output code with a desired code. With ty of the integrator input current is reversed and the inte-
proper algorithms, the input to the ADC will average out to grator output decreases. You can measure the average
be the value that defines the code edge. A block diagram of value of the ADC input with a voltmeter; the measured
the classic implementation of the servo is shown in Figure 5. voltage corresponds to the lower transition level corre-
sponding to the reference code
word. Newer implementations
of the servo are described in
Clock IEEE Std 1241-2000.
Word Reference
ADC k Comparator Code
Other Methods
Under A B kin
Test of Determining
A>B Transition Levels T[k]
Voltmeter
c The code transition levels can be
I1 measured using several other
algorithms.
s1
◗ The ac histogram method. This is

useful when the ADC input is
s2 ac coupled. It is also useful in
I2 evaluating how the transition
levels change with the frequen-
Fig. 5. Block diagram of analog servo for determining ADC code transitions. cy of the applied input.
◗ The manual method. A voltage source is slowly varied
until a code transition is observed. The value of the
input is then recorded as the transition level. This is
3Q/2 generally a throwback to days when computers were
Q not as readily available as they are today.
111 ◗ The triangle histogram method. This method is a variation
T [2N−1]
on the ramp algorithm. It is useful if the ADC has some
filtering at its input.
Figures 8 and 9 are INL and DNL plots of a 9-bit ADC to
show some typical results.
100
Common Dynamic Test Methods
Dynamic tests establish the ADC’s performance with signals
Q/2 that change with time. The selection of which dynamic tests
to perform generally depends on the application (see Table
T [1] 1). The following subsections present some of the more com-
Input mon dynamic tests.
000 FSR
Vmin (Vmin + Vmax)/2 Vmax
Total Harmonic Distortion Estimation
(a)
You use the sine wave test setup to estimate the total har-
monic distortion (THD) of an ADC. Apply a test signal con-
sisting of a pure, large amplitude sine wave at frequency fi
Q/2 as the input to the ADC. This test, being based on discrete
Fourier transform (DFT) analysis of unwindowed sample
Input sets, requires knowing or controlling accurately the test fre-
quency fi so that it coincides exactly with a DFT basis fre-
−Q/2 quency. Moreover, the test and sample frequencies should
be selected such that all possible ADC states within the
(b) amplitude range of the test signal can be sampled.
For this analysis, you acquire K data records of M points
Fig. 6. The transfer characteristic of an ideal 3-bit ADC. each from the ADC under test at sample frequency fs. For

44 IEEE Instrumentation & Measurement Magazine April 2006


Display,
Sensor actuators,
Input Processing Output
signals,
control

each record of sine wave data, compute the DFT. You use 
|Xavm ( fi)|
the K sets of data to compute an averaged magnitude spec- SFDR = 20 log10   . (13)
max fsp , fh |Xavm ( fsp )|, |Xavm ( fh )|
trum of the DFT bins at each basis frequency fm.
At this point, identify the set of frequencies fh that cor-
respond to the chosen set of harmonics of the input test You must specify the amplitude and frequency of the input
frequency. For a test tone at frequency fi, the harmonics and the sample frequency for which SFDR measurement(s)
are aliased so that fh lies between zero and the sampling are made.
frequency fs. The input frequency and each harmonic map
into two different frequencies in the DFT, one correspond- Signal-to-Noise-and-
ing to positive frequencies and one to negative frequen- Distortion Ratio Estimation
cies. The choice of harmonic components included in the You again use the sine wave test setup to estimate the signal-
set should be done including the harmonics with a signifi- to-noise-and-distortion ratio (SINAD) of an ADC. Therefore,
cant portion of the distortion energy and excluding DFT apply a sine wave of specified frequency and amplitude to
bins whose energy content is dominated by random noise. the ADC input. A large signal (approaching full scale) is pre-
The tradeoff suggested in IEEE Std 1241-2000 to estimate ferred. Almost any error source in the sine wave input, other
THD includes in the set the lowest nine harmonics, second than gain accuracy and dc offset, can affect the test result. So,
through tenth, inclusive, of the input sine wave. Finally, we recommend that you use a sine wave source with good
compute the THD: short-term stability and that the sine wave input be highly


1 
(Xavm ( fh ))2
M h
THD = , (11) Vmax
VP
Arms

where Xavm ( fh ) is the averaged magnitude of the component


at the hth harmonic of the DFT of the ADC output data
record and M is the number of samples in the data record.


1  2  2
Arms = Xavm ( fi) + Xavm ( fs − fi) (12)
M
Vmin
VM
is the rms of the output fundamental component at the
signal test frequency, fi = Jfs/M. Triggers

Spurious-Free-Dynamic-Range Estimation
To estimate the spurious free dynamic range (SFDR) of an Fig. 7. Ramp waveform applied to ADC for evaluating transition levels.
ADC, the procedure for THD is
used, considering also a set of
spurious frequencies fsp. A spuri-
ous frequency is the frequency of Plot of INL of 9-bit ADC
0.8
a persistent spectral output com-
ponent that is neither the funda- 0.6
mental nor a harmonic distortion
component. You determine the set 0.4
by inspection of the ADC output
Error(LSB)

spectrum, so there are no direct 0.2


equations to determine it. For a
pure sine wave input, the SFDR is 0
0 64 128 192 256 320 384 448 512
the ratio of the amplitude of the
averaged DFT value at the funda- −0.2
mental frequency fi, to the ampli-
tude of the averaged DFT value of −0.4
the largest magnitude harmonic
−0.6
or spurious signal component Code
observed over the full Nyquist
band, max{Xavm ( fh ), Xavm ( fsp )} Fig. 8. Plot of 9-bit ADC INL results based on terminal value gain computation.

April 2006 IEEE Instrumentation & Measurement Magazine 45


Signal-to-Noise
Ratio Estimation
9-bit ADC DNL Plot The signal-to-noise ratio is called
0.2
SNR; it is called signal-to-nonhar-
0.1 monic ratio (SNHR) in [1], but the
0 name will be changed to SNR in the
0 128 255 384 512
next revision. To estimate the SNR,
−0.1
use the sine wave test setup. The test
−0.2
Error(LSB)

procedure is the same followed for


−0.3 the THD test. The SNR is obtained
−0.4 from the DFT by excising the signal
harmonic frequencies identified
−0.5
from the test signal spectrum.
−0.6 The rms signal, Arms (in LSB’s), is
−0.7 the same as reported in (12). After
the components bins corresponding
−0.8
Code to dc, the test frequencies fi and
( fs − fi), and the specified harmonic
Fig. 9. Plot of 9-bit ADC DNL results based on terminal value gain computation. frequencies are all set to zero, the rms
noise is found from the sum of all the
remaining Fourier components as:
filtered to remove distortion and random noise from the

input signal. M−1
1 
After applying a sine wave, a record of data is acquired. A rms noise =  Navm ( fm )2, (16)
sine wave function is fitted to the record by varying the phase, M m=0
amplitude, dc value, and (if needed) frequency of the fit func-
tion to minimize the sum of the squared difference between the where Navm is the averaged DFT spectrum after setting the
function and the data. IEEE Std 1241-2000 suggests two algo- bins to zero.
rithms for least squared fitting, one for known frequency solu- SNR is then computed by
tions, when the ratio between the sample frequency and the
input frequency is known and stable, and one for other cases. rms signal
SNR = . (17)
The SINAD is computed as rms noise

rms signal Effective Number of Bits Estimation


SINAD =
rms noise To estimate the effective number of bits (ENOB) of an ADC,

sine wave peak amplitude/ 2 the sine wave test setup is used. For an input sine wave of
=  , (14)
 specified frequency and amplitude, after correction for gain
1  M
 
 yn − yn
2
and offset, the ENOB is
M n=1
 
rms noise
where yn is the sample data set and yn
is the data set of the ENOB = N − log2
ideal rms quantization error
best sine wave fit.  
full scale range
SINAD can be determined equivalently from the fre- = log2 √ , (18)
Q/ 12
quency domain. You must follow the procedure for the THD
test. You can determine the rms input signal and the rms
noise and distortion from the DFT of data records. The rms where N is the number of digitized bits and Q is the ideal
signal, Arms (in LSBs), is the same reported in (2). code bin width, expressed in input units. ENOB generally
To determine the rms noise, set the dc and frequency depend on the amplitude and frequency of the applied sine
bins, fi and ( fs − fi), to zero. Then, the rms noise is found wave. You must specify the amplitude and frequency at
from the sum of all the remaining Fourier components as which the measurement is made.

 Bandwidth Estimation
M−1
1 
rms noise =  Eavm ( fm )2, (15) To estimate the bandwidth of an ADC, use the sine wave test
M m=0 setup and add an ac voltmeter. Use a large signal sine wave
(the signal must span at least 90% of the full-scale range of
where Eavm is the residual spectrum of Xavm . the ADC under test), unless the small-signal bandwidth is to

46 IEEE Instrumentation & Measurement Magazine April 2006


Display,
Sensor actuators,
Input Processing Output
signals,
control

be determined. When small-signal bandwidth is to be deter- high enough to make aliasing errors negligible. The fref is
mined, the peak-to-peak input amplitude used is less than chosen from the DFT bins. It must be one within the pass
one-tenth of full scale. band such that the dynamic gain is at or near the peak gain
The input sine wave source should produce sinusoids of of the pass band.
high spectral purity (i.e., harmonic distortion lower than You analyze the DFT bins to find the upper and, if appli-
that of the ADC under test) and should have stable output cable, the lower frequency samples closest to the reference
during the measurement time. The tested input frequencies frequency at which the gain is 3 dB below the reference gain.
should not be subharmonics of the ADC sampling rate The bandwidth is the difference between these upper and
since such frequencies can produce incorrect results in this lower −3 dB sample frequencies (or, if a lower −3-dB fre-
test. An input frequency at which the ADC’s dynamic gain quency does not exist, the bandwidth is simply the upper
is equal to or near its peak value in the pass band is select- −3-dB frequency value). To improve the bandwidth esti-
ed as a reference frequency. After connecting the sine gen- mate, interpolate between the frequency samples above and
erator to the ADC input, you set its frequency to the below −3 dB in amplitude to better estimate the actual −3-
reference frequency fref and acquire a sufficient number of dB frequency.
data records from the ADC output to determine the maxi- The disadvantages of this test are high noise at higher
mum peak-to-peak range of the signal using a three- frequencies and aliasing and first-differencing errors result-
parameter or four-parameter sine fit. The input amplitude ing from the frequency response estimation. The results of
measurement must be done with care if high accuracy is using the step response method are invalid in the presence
required. If the measured input amplitude parameter is the of slew rate induced errors.
rms amplitude, Arms , you must convert it to peak-to-peak

amplitude by multiplying it by 2 2. Divide the peak-to- Gain Error Estimation
peak ADC output amplitude by the measured peak-to- To estimate the gain error of an ADC, the sine wave based
peak input amplitude to determine the reference gain. If methods or the differentiated step response method can be
the chosen fref is zero, the reference gain is the static gain. used. Gain error, also known as gain flatness, is the differ-
Alternatively, to determine the static gain, you can use a ence between the dynamic gain, G(f), of the ADC at a
precision dc signal source to provide a constant input sig- given frequency and its gain at a specified reference fre-
nal. The dc gain is approximated by the constant output quency, divided by its gain at the reference frequency. The
signal level minus the measured static dc offset divided by dynamic gain of the ADC under test at a frequency f is the
the input dc level. magnitude of the frequency response at that frequency.
Once the reference gain is determined, you change the The reference frequency is chosen to be a frequency whose
input frequency to another value that is not a sampling-rate gain is at or near the peak gain of the ADC pass band; typ-
subharmonic. You then measure the maximum peak-to-peak ically, it is the same frequency as the one used in the band-
range of the recorded data and divide it by the input ampli- width test. For dc-coupled ADCs, the reference frequency
tude measured by the ac voltmeter to find the gain at this is typically dc ( f = 0).
frequency. You repeat as necessary to find the upper (and, if
it exists, lower) frequency, closest to the reference frequency, Aperture Delay Estimation
at which the gain is 3 dB below the reference gain. If no To estimate the aperture delay of an ADC, apply a ramp to
lower −3 dB frequency exists, the upper −3 dB frequency is the analog input and apply a clock signal to the clock
the bandwidth. If a lower −3 dB frequency exists, the differ- input of the ADC adapting the arbitrary signal test setup.
ence between the upper and lower −3 dB frequencies is the Instead of a ramp signal at the analog input, you can use a
bandwidth of the ADC. portion of another waveform (e.g., a sine wave) provided
This test uses sine wave inputs and can be done very that the slew rate of the waveform does not vary substan-
quickly if the reference frequency and the approximate tially over the aperture width of the ADC (e.g., a sine
limit frequencies are known. The disadvantage is the typ- wave of frequency less than half the analog bandwidth of
ically low accuracy of estimates of the analog input the ADC). The ramp signal slew rate should be as high as
amplitudes, which reduces the accuracy of the band- possible without exceeding the slew rate limit of the ADC
width result as well. or causing excessive dynamic errors. The ramp and the
IEEE Std 1241-2000 proposes an alternative bandwidth clock must be synchronized such that the ADC samples
test method, which is generally more useful for ADCs that the ramp at the center of the ADC full-scale range. Using a
contain analog bandwidth limiting circuitry before the time-interval meter or oscilloscope with sufficient resolu-
quantizer(s). You use the step signal test setup to deter- tion and accuracy, the time delay is measured from the
mine the ADC frequency response. It is desirable to have time instant the clock input crosses its threshold to the
as many samples in the record as possible, to increase the instant the analog input crosses the dc value correspond-
resolution with which the bandwidth can be resolved from ing to the center of the ADC full-scale range. For ADCs
the DFT of the derivative of the step response. The sam- with very high sample rate, where the aperture delay may
pling rate, or equivalent-time sampling rate, should be be similar in magnitude to the clock period, extra care

April 2006 IEEE Instrumentation & Measurement Magazine 47


must be taken to measure the correct clock edge. This can on that of a complete system can be critical. Therefore, it is
be done by repeating the measurement at various clock important to use commonly understood terms and test
frequencies. The aperture delay as a function of clock peri- methods to describe the ADC’s performance. Finally, it is
od should be roughly flat. important to understand the confidence with which the mea-
surement data are stated. (Tutorial 8 will go into detail about
Aperture Uncertainty measurement uncertainty.)
To measure the aperture uncertainty of an ADC, couple
the output of a stable signal generator to both the analog References
input and the clock input of the ADC, using appropriate [1] IEEE Standard for Terminology and Test Methods for Analog-to-
signal splitters, attenuators, dc blocks, frequency multipli- Digital Converters, IEEE Std 1241-2000.
ers/ dividers, and delay circuits, to ensure that the signal [2] International Vocabulary of Basic and General Terms in Metrology
amplitude, offset, and frequency at each input are appro- (VIM), ISO, 1993.
priate. When possible, no active component should be [3] S. Rapuano, P. Daponte, E. Balestrieri, L. DeVito, S.J. Tilden, S.
used for coupling, as any jitter in those components Max, and J. Blair, “ADC parameters and characteristics,” IEEE
would contribute to the overall measured aperture uncer- Instrum. Meas. Mag., vol. 8, no. 5, pp. 44–54, Dec. 2005.
tainty. The slew rate of the signal at the analog input port
should be as high as possible without exceeding the slew
rate limit of the ADC input or causing excessive dynamic Thomas E. Linnenbrink (toml@hittite.com) received a B.S.
errors. The delay of the path is adjusted from the signal degree in electrical engineering in 1967 from the Illinois
generator to the analog input port to be longer than the Institute of Technology and a M.S. degree in engineering sci-
delay to the clock input port by the amount of the aper- ence with emphasis in automatic control from the Rensselaer
ture delay, such that each active clock edge is sampling Polytechnic Institute in 1971. He cofounded Q-DOT, Inc. in
itself at its midpoint. The connection between the signal 1977. He is currently a business development manager for
generator and the analog input port are broken, and both Hittite Microwave, which acquired Q-DOT in 2005. He has
ends of the broken connection are terminated appropriate- been active in the IEEE Instrumentation and Measurement
ly to prevent reflections. Society’s Waveform Generation, Measurement, and Analysis
The aperture uncertainty is then given by Committee (TC-10) since its inception and has served as its
chair since 1997.

σA2 − σB2
σT = , (19) Jerome Blair received a B.S. degree in engineering physics in
slew rate
1966 and a Ph.D. in applied mathematics in 1970, both from
where σA2 is the measured noise variance with clock signal the University of California, Berkeley. He has worked since
applied to the analog input port, σB2 is the measured noise 1966 on the design and analysis of complex instrumentation
variance without the clock signal applied to the analog input systems and has been a Member of IEEE I&M TC-10 since
port, and slew rate is magnitude of the slope of the clock sig- 1989 and is an IEEE Fellow.
nal at the analog input port at the sampling instant.
Sergio Rapuano achieved the master’s degree cum laude
Measurement Uncertainty in electronic engineering from the University of Salerno in
The objective of a measurement is to determine the value of 1999. In 2002, he joined the Faculty of Engineering of the
the measurand that is the specific quantity subject to mea- University of Sannio as assistant professor of electric and
surement. In general, no measurement or test is perfect and electronic measurement. In 2003, he obtained the Ph.D.
the imperfections give rise to errors in the result. degree in computer science, telecommunications, and
Consequently, the result of a measurement is only an applied electromagnetism. He is an IEEE I&M TC-10
approximation to the value of the measurand and is only member and secretary of the IEEE TC-23 Working Group
complete when accompanied by a statement of the uncer- on e-tools for Education in Instrumentation and
tainty of that approximation. The International Vocabulary of Measurement.
Basic and General Terms in Metrology defines uncertainty as “a
parameter associated with the result of a measurement, that Pasquale Daponte obtained his master’s degree cum
characterizes the dispersion of the values that could reason- laude in electrical engineering in 1981 from the University
ably be attributed to the measurand” [2]. of Naples. He is a full professor of digital signal process-
Tutorial 8, to be published in an upcoming issue, will ing and measurement information at the University of
cover measurement uncertainty in detail. Sannio, Benevento. He is a Senior Member of the IEEE
I&M Society and a Member of IEEE I&M TC-10. He is
Conclusion coordinator of the IMEKO Working Group on ADC and
ADCs provide an important function in a wide range of DAC Metrology and deputy of the International Relations
modern equipment. The impact of the ADC’s performance for the University of Sannio.

48 IEEE Instrumentation & Measurement Magazine April 2006


Display,
Sensor actuators,
Input Processing Output
signals,
control

Eulalia Balestrieri achieved the master’s degree in 2003 in 1961, respectively. He has two patents involving mixed-sig-
software engineering from the University of Sannio. She nal technology. He is currently a staff scientist and LTX
joined the research activities carried out at the Laboratory of Fellow at LTX corporation where he was one of the founders
Signal Processing and Measurement Information of the of the company in 1976. At LTX, he developed many of the
University of Sannio, Benevento. In 2004, she began a Ph.D. test instruments used in ATE systems. He is a Senior
course in information technology. She is a member of the Member of the IEEE and a member of Eta Kappa Nu, Tau
IEEE I&M TC-10. Beta Pi, and Sigma Xi.

Luca De Vito achieved the master’s degree cum laude in Steven J. Tilden has more than 33 years electronics experi-
2001 in software engineering from the University of Sannio. ence, six in the U.S. Air Force and the last 27 in Tucson,
Then, he joined the research activities carried out at the Arizona, at Burr-Brown, which became Texas Instruments in
Laboratory of Signal Processing and Measurement 2000. He has worked in nearly every area from wafer fabrica-
Information of the University of Sannio, Benevento. He tion through final test in several engineering disciplines. He
achieved the Ph.D. in information engineering in 2005 from a Senior Member of the IEEE and a senior member technical
the University of Sannio. Currently, he is a software design- staff at Texas Instruments. He is chair of the IEEE TC-10
er at Telsey telecommunications. He is a member of the IEEE subcommittee that published ADC test methods and termi-
I&M TC-10. nology standard 1241, is chair of the IEEE TC-10 subcommit-
tee on digital-to-analog converters, and is a U.S. national
Solomon Max has worked in the mixed-signal field for delegate to the International Electrotechnical Commission
almost 50 years. He received his B.E.E. degree from City for semiconductor, integrated circuits, which are publishing
College of New York in 1957, and S.M. and E.E. degrees IEC-level ADC and DAC test and terminology standards
from the Massachusetts Institute of Technology in 1959 and through TC47/SC47A/WG4.

Call for Papers


2007 IEEE Sensors Applications Symposium
6-8 February 2007
San Diego, California, USA
www.sensorapps.org

The 2007 IEEE Sensors Applications Symposium (SAS-2007) provides a unique forum for sensor users and developers to meet
and exchange information about novel and emergent applications in smart sensors, biology, homeland security, system health
management, and related areas. Collaborate and network with scientists, engineers, developers and customers, in a balance of
formal technical presentations, workshops, and informal interface meetings—a unique feature of this conference.

Important Dates—Abstract submission deadline: 01 October 2006, Notification of acceptance: 01 December 2006, Final
manuscript submission deadline: 01 January 2007

For Additional Information


General Chair John Schmalzel Vice Chair Shreekanth Mandayam
+1 856 256 5332, j.schmalzel@ieee.org +1 856 256 5333, shreek@ieee.org

Visit the Sensors Applications Symposium Web site at: http://www.sensorapps.org

April 2006 IEEE Instrumentation & Measurement Magazine 49

You might also like