,EEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, VOL. CAS-33, NO.

8, AUGUST 1986 775

Dynamic Testing and Diagnostics of

A/D Converters
M. VANDEN BOSSCHE, J. SCHOUKENS, AND J. RENNEBOOG

Ah.sfrrrcf -A method is derived to measure the integral and differential In a second step, it is possible to make a diagnostics to
nonlinearity of an ADC using a sinewave with unknown amplitude and identify which bits are responsible for the nonlinearity,
offset. The uncertainty of the measurement is also estimated. In a second
phase, the integral nonlinearity is analyzed, using Walsh Transforms, to
using the Walsh Transform. These diagnostics are related
identify the nonlinearity at the bit level of the ADC. to the working principle of the ADC. In this paper, Ana-
log-to-Digital converterswill be consideredwhich are based
I. INTRODUCTION upon binary-weighted bit voltages, currents, and the
successiveapproximation principle.
D URING THE PAST years, the importance of digital
signal processing has grown very rapidly. At the
same time, the performance of the ADC, which transforms
The knowledge of the analysis can be used to correct the
transfer characteristic of the ADC.
the analog signals to digital ones, has been improved. To
have an idea of its real performance, it is necessary to II. MEASURINGTHE INL ANDTHE DNL
measure its transfer characteristic.
Classically, this was done in a static way. However, a 11.1. Deriving the Transfer Function of the ADC
number of imperfections will not be detected by these The PDF of a sinewavey(t) = A-sin& + B is given by
tests. To improve the knowledge of the behaviour of an
ADC, a dynamic test is set up.
The most common dynamic tests used nowadays, are f(Y) =
histogram and beat frequency testing [l]. These methods
give a good qualitative idea of the ADC performance. The
During the experiment, a great number of samplepoints
method, presented in this paper, will give a quantitative
are taken of the sinusoid. The signal can be sampled at
evaluation of the ADC under test. This gives the possibil-
random (random sampling method) or at equidistant points
ity to correct the dynamic behaviour of the ADC.
(asynchronous sampling method). In the last case, the
The ADC is excited with a signal with a known prob-
sample frequency has to be chosen in such a way that the
ability density function (PDF). A great number of samples
ratio of the sampling frequency and the frequency of the
are taken and an estimate of the real PDF is made. By
sinusoid is a rational number given by the ratio of two
comparing the measuredPDF with the theoretical one, it is
prime numbers. After the completion of the experiment, a
possible to derive the differential and integral nonlinearity.
vector P can be defined in which the k th element is given
The main idea of this test has already been developed by
many authors [2]-[4]. In these articles, a triangular wave- by
form was used becauseits PDF is very simple. The funda- Pk = nk/N (2)
mental drawback of this choice is the distortion of the
waveform. However, a sinewave can be generated with a with nk being the number of samples on the k th level of
very low distortion, even at high frequencies. the ADC and N the total number of samples.
Starting from this idea, an analogous approach was The values pk can be considered as an estimate of the
developed as explained in [l] and [5]. The transfer char- probability to excite the kth level of the ADC. The prob-
acteristic iscalculated and the Integral Nonlinearity (INL) ability Qi to realize a measurementy < UB,, with UB, the
and Differential Nonlinearity (DNL) are derived. To use upperbound of the i th level, is
these results, it is necessary to have an idea about the
uncertainty on them. A simple statistical analysis will give Qi 7 P( y < UBi) = i pj measured
the solution. j=l
The concepts of INL and DNL are applicable on all
ADC’s. theoretical. (3)

Manuscript received February 24, 1986. This integral dependsupon the value of the amplitude A
The authors are with Vrije Universiteit Brussel, Dienst ELEC, B10550
Brussel, Belgium. and the offset B of the applied sinewave. In practical
IEEE Log Number 8608876. setups, it is not always possible to know these values. A

0098-4094/86/0800-0775$01.00 01986 IEEE

 176 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, VOL. CAS-33, NO. 8, AUGUST 1986

Transfer Function of 5 - Bit ADC (Simulation)

16.0

Real transfer function,

due to bit failure

0.0

0.0
-1.1 4.9 -0.7 -0.5 -0.3 -0.1 0.1 _ 0.3 0.5 0.7 0.9 1.1

Amplitude

Fig. 1. Transfer function of the simulated ADC.

and B can be eliminated using a linear transformation The integral nonlinearity is defined as the difference be-
y=Ay’+B-y’=(y-B)/A tween the measured transfer characteristic and the best
fitted straight line. This line is given by a linear regression
ubj = (Vi; - B)/A (Fig. 1). (4 analysis
measuredcharacteristic - regressioncharacteristic
INL = LSB. 0)
1,
The shape of the transfer characteristic is not changed 11.3. Study of the Uncertainty on the Measurements
by this transformation. Combining (4) and (3) the follow- The measurementsof the INL and DNL are a result of a
ing relation is derived: stochastic process analysis (the random sampling of a
ub, = - cos TQ;. (5) sinusoid). This implies, that even in a noiseless process,
With this relation, it is possible to make an estimate of there will be an uncertainty in these results. It is very
the scaled transfer function, using the results of the mea- important to have an idea of the uncertainty, to design the
surement, stored in P. Starting from this transfer function, experiments and to interpret the results.
the DNL and INL are calculated. In Appendix I, it is proven that the standard deviation
u“,, and the crosscorrelation (I,~ ub are given by
II.2. Determination of the DNL and INL
uuh,= /[r2Qi(l - Qi)sin2nQi]/N
The differential nonlinearity of the ith level is defined as
the ratio
UB, - UBipl %h,uh, =
/[Ir2Qi(l-Qj)sin7iQjsin7rQj]/N,
DNLi = -1
LR with Qj > Qi. (8)
in Least Significant Bit (LSB) with L, the referencelength From these results the following expressionsare derived:
of a level and with the transformed variables
uINL,, = %hi 11, in LSB
ubi - ubipl
DNL, = -1 LSB
1,
LR (Fig. 1).
uDNL, = \i[ dh , + uu”b,-, - 2u,2b,uh,-, 1 /I,* in LSB.’ (9)

withlr = 7 (6b) The ‘maximal uncertainty occurs for Qi = 0.5. The follow-
VANDEN BOSSCHE et cd.: A/D CONVERTERS 111

ing approximations can then be used: error estimation band gives an idea about the significance
of the measurementprocedure. It must be noticed that the
= 7r/(21,)4/4R in LSB
uINL,_
error band is calculated for a random sampling process.
uDNL,,,_= \Ilrlo.,/(@) in LSB
111.3. Fourier Transform - Diagnostics
=#i/l/fi if I,,, = I, 00) It is possible with Discrete Fourier Transform-tech-
with l,, the length of the level containing the value ub = 0 niques (DFT) to detect the repetition of the peaks in the
(Q = 0.5) The last result is found by substituting the values DNL and the square waves in the INL. The DFT is
Q, = 0.5 - AQi and Qj = 0.5+ AQj, with AQi and AQj -+K applied to the INL and DNL (Figs. 4 and 5).
1, in relations (8) and (9) and using a linear approximation Because the INL can be considered as an integrated
of relation (5) in Q = 0.5. form of the DNL, the l/j+function (integration effect,
These results are derived for the random sampling represented in the Fourier Transform) is reflected in the
method. If the asynchronous sampling method is used, it INL. The small constant peaks on Fig. 5 are due to small
can be proven that the uncertainties are smaller. In that aberrations of the calculated reference level. This aberra-
case the l/$v law becomes a C”l/N law. Indeed, the tion introduces also the small constant peaks on Fig. 3
sampling rate is chosen in such a way that an entire (DNL). Applying a DFT to this kind of signal, a new
number of periods of the sinusoid is measuredfor the first signal with constant peaks is created, as illustrated in Fig.
time after N sampling points. The number of samples nk 5. In actual experiments, the ADC is consideredas a black
on the k th level is proportional to N. The maximal varia- box with no additional information. Therefore, it is neces-
tion of nk is 1. Using relation (2) the maximal variation of sary to estimate the length of the referencelevel I, (equa-
pk = n,/N is given by jAp,l is l/N. From this result, it is tions (6a) and (6b)). This is accomplished by taking the
easily seen that the uncertainty on the DNL and INL is mean of the length ( ubi - ubiPl) of all the levels. The
proportional to l/N. reference level is not calculated as 1, = 2/,N, where 2 is
the full scale range of the normalized ADC. This is due to
III. DIAGNOSTICS ON ADC the practical realization of the experiment. It is almost not
possible to apply a sinewaveto the ADC, covering exactly
III. I. Introduction its full scale range without creating saturation effects in
Using the knowledge of the INL and DNL, it is possible the lowest and highest ADC-level. This saturation would
to run diagnostics on the INL to extract information at the cause deformation of the probability density function,
bit level. This technique can be used by the ADC hardware which must be avoided. For this reason,only a subrangeof
designer to qualify his prototype. The INL and the DNL the ADC is analyzed. This subrangecan be chosen as near
will give him the information about the acceptanceof his as possible to the full scale range. In this case, an aberra-
product. If improvement of the prototype is necessary,the tion of the reference level, only introduces a gain error on
diagnostics of the ADC will provide a straightforward the INL, in opposite to the DNL where small parasitic
feedback to the ADC designerby indicating which bit(s) is peaks appear. Becauseof the behavior of the INL, due to
(are) wrong. The diagnostic technique will give also the bit failure, the sine-cosine functions are not of the ap-
possibility of compressing the great amount of data from propriate class to analyze the DNL and INL. It becomes
high resolution ADC’s, offered by the INL and DNL difficult to separatethe effect of different bit failures when
(2 x 2N information items, N-bit ADC), into the essential they start to interfere or if bit-intermodulation does occur.
information on the bit level (N information items). The use of the word “bit-intermodulation” describes the
behavior of bits, influenced by the appearanceof a certain
111.2. Bit Failure, Manifesting in the INL and DNL other bit. This phenomenon will be mentioned later.
To develop a diagnostic method, the effect of bit failure
on the INL and DNL must be considered. The effect is III. 4. Walsh Transforms
illustrated by a simulation of a 5 bit SuccessiveApproxi- The effect of the bit failure on the INL, imposes as the
mation Register (SAR) ADC (fifth bit: Most Significant appropriate class of functions the use of the Walsh trans-
Bit, first bit: Least Significant Bit) with a full scale range forms. In this way, all the problems, coming along with
from - 1 to + 1 V (digitizing step: l/16 v). Enlarging the DFT, are avoided. The Walsh functions are adequate to
weight of the 4th bit (23) by 10 percent compared to the study step-like signals and are defined as follows:
ideal situation, a bit failure was introduced. This will result
in an INL of 0.8 LSB and a DNL of 0.8 LSB. The 01)
asynchronous sampling technique is used to simulate the
measurements.100 000 sampleswere taken into account to where
get small uncertainties on the INL and the DNL nppl, npp2; . -, no binary representation of n,
(a INL 3 uDNL< 0.08 LSB). Fig. 1 shows the transfer func- tp-1, tp-2,. ’ -7 to binary representation of t,
tion of this ADC. The INL (Fig. 2) and the DNL (Fig. 3) n the order of the Walsh function,
do reflect the failure of the ADC. It can be seen that the t the argument (often: time representa-
INL is an integrated form of the DNL. The 95-percent tion).
118 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, VOL. CAS-33, NO. 8, AUGUST 1986

Integral Non - Linearity (5 - Bit ADC)

p 1.0 r 1
I

0.6

0.4

0.2

0.0

-0.2
L Error band (random sampling)

-0.4 t

-0.6 -

-0.8 -

-1.0
0.0 8.0 16.0 24.0 32.0
Level

Fig. 2. The integral nonlinearity of the simulated ADC.

Differential Non - Linearity (5 - Bit ADC)

1.0

P
: 0.8

2
0 0.6

0.4

0.2

0.0

-0.2

-0.4

-0.6

-0.8

-1.0
0.0 8.0 16.0 24.0 32.0

Level

Fig. 3. The differential nonlinearity of the simulated ADC

XSCHE et al.: A/D CONVERTERS 119

Fourrier Transform of INL (5 - Bit ADC)

0.5

:
:
z
Failure effects on bit 4 (23) +
z
4 0.4
integration effect of DNL

(descendant peaks)

0.3

0.2

0.1

0.0
0.0 2io 4.0 6.0. 8.0 10.0 12.0 14.0 16.0
Harmonica1 Number (N-32)

Fig. 4. The Fourier transform of the INL of the simulated ADC.

Fourrier Transform of DNL (5 - lit ADC)

0.20
1

Failure effects on bit 4 (23)

0.14 -

0.12 -

0.1n -

0.08 -

0.06 -

11
0.04 -

0.02 .

0.00 1
0.0 2.0 4.0 8.0 10.0 12.0 14.0 16.0

Harmonica1 Number (N-32)

Fig. 5. The Fourier transform of the DNL of the simulated ADC.

780 ’ IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, VOL. CAS-33, NO. 8, AUGUST 1986

Yal(7,T)
I I I I I I I

Wal(6.T)
1 I I I I

Wal(5.T)
I I I

,
Wal(4.T)
u u

Wal(3.T)
u I

Wal(1.T)
I

Wal(O,T)
,
Time
A fev
---------- examples ---

Fig. 6. Example of the Walsh function O-7.

In Fig. 6, a few Walsh functions (sequenceordered) are in such an ADC, can be mathematically represented
illustrated as example.
The Walsh functions are an orthogonal complete set of n -bit ADC:
functions, that results in the Walsh transforms: bit pattern: pn-,pnP2 * * *p. with pi = 0 or 1
N-l v,, = pn-,w,-,2”-l + . . . + PoWo2?
xi= c X,WAL(n,i)
n-0 Ideal linear ADC:
w,-,= “. =w,
X, = ; &WAL(n, i). (12)
I-O = W: quantization step (Weighting Values)
Implementing the Fast Walsh Transform, the sequence + KefN = ( p,-12”-1+ . . . + po20). w. (13)
ordered algorithm was chosen. The sequenceorder, which
The reference voltage I&, is formed by switching in
stands for the number of transitions between - 1 and + 1,
can be compared to frequency in Fourier analysis, and to resistors of an R -2 R-bridge. Inaccuracies on the resistors
do cause errors on the weighting value W ( = on the
the effects of bit failure on the INL.
quantization step). The bit diagnostics consist of estimat-
ing the error SW, on the weighting value Wand the sign of
111.5. Characterization of Bit Failure the error (6%: positive or negative). Knowing this error
To characterize the bit failure, the general working contributes to the possible improvement of the perfor-
principle of the ADC has to be considered. This paper mance of the ADC during design.andproduction.
studies a Successive Approximation ADC with binary- Referring to the simulation, mentioned-earlier, enlarging
weighted voltages or currents. The reference voltage, used the weight of bit 4 by 10 percent means that SW, = O.lW.
VANDEN BOSSCHE et al.: A/D CONVERTERS 781

Walsh Transform of INL (5 - Bit ADC)

0.1

:
:
=:
8” 0.0
<

-0.1

4 Failure effect on bit 4 (Z3)

-0.2

-0.3

-0.4

-0.5
0.0 4.0 R.0 12.0 16.0 20.0 24.0 2R.0 32.0
Sequence WI.

Fig. 7. Walsh transform of the INL of the simulated ADC.

Suppose: error on bit i + 1 (2’) factor W

+w,=w+srq 6V 6IK2’
a,===--@. (17)
<ef = 6:ef N + pis&2’ (14) = F/,,fN + 6V.
The relative error, on the weighting value, derived from
The deviation of VrefN(= 6w2’) will influence the point the Walsh transform, can be calculated as follows:
of switching in bit i + 1, compared to the applied voltage.
If SW, is positive, the decision to switch in bit i + 1 will be SW, 2.ai
si = y- -100 = 2’ -100.
delayed, if SWi is negative,the decision will be accelerated.
Due to the deviation, the voltage, corresponding to the Until now, the sign of a, has not been considered. It
beginning of the levels involved, will be shifted over 6K.2’ gives information about the direction of deviation from the
= 6V. This results in an INL of &f/W, each time the bit is ideal ADC. The sign depends on the form. of the Walsh
evaluated: functions. Comparing the Walsh functions of Fig. 6 with
sv SW,” the INL of the j-bit ADC simulation (Fig. 2), where SW,
-=- LSB. (15) was positive, one concludes that the Walsh function
w w Wal(3,t) with a negative coefficient describes the INL
The relative error on the weighting value can be repre- completely. So, if the Walsh coefficient is negative, the
sented by error 6W on the weighting value W is positive and vice
SWi
versa.
- *loo. 06) This technique can be demonstrated by processing the
W information, obtained by the simulation. In Fig. 7, the
Applying the Walsh Transform to an INL with step Walsh Transform of the INL of the example in Section
6V/ W, will result in a value 6V/2W on sequencenumber III.1 is shown. Analyzing this Walsh Transform, results in
2N-’ - 1 (N-bit ADC). The factor 2 results from the Walsh the following conclusions:
function which toggles between - 1 and 1 with amplitude N=5 (5-Bit ADC)
1. If a failure occurs on bit i + 1, during the ADC-test, the
Consider Fig. 7 + SequenceNumber = 2N-i - 1 = 3
Walsh Transform will give a value ai on the sequence
number 2N-i - 1 (N-bit ADC), as stated before. -+i=3
Refering to (14) and (15), this value can also be calcu- + Bit failure on bit 4(23)
lated’ as a function of the error SW, and the bit-weighting + a3 = -0.39
782 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, VOL. CAS-33, NO. 8, AUGUST 1986

Integral Non - Linearity (10 - Bit ADC)

0.00 128.0 256.0 384.0 512.0 640.0 768.0 896.0 1024.0

Fig. 8. The integral nonlinearity of the tested ADC (10 bit).

Differential Non - Linearity (IO - Bit ADC)

0.10

2
ol
t3

2
cl

0.05

0.00

-0.05

-0.10
0.00 128.0 256.0 384.0 512.0 640.0 768.0 896.0 1024.0

Level

Fig. 9. The differential nonlinearity of the tested ADC (10 bit)

/ANDEN BOSSCHE et al.: A/D CONVERTERS 783

Walsh Transform of INL (10 - Bit AOC)

0.020

:
:
2
0.015
:
<

0.010

/-, bit 6 (Z5)

0.005
I
c--, btr 3 (2’)

0.000

-bit 2 (Z1)

-0.005

bit I (2’) W

-0.010 J
0.0 128.0 256.0 384.0 512.0 640.0 768.0 896.0 1024.0

Sequence Number

(4

Zoom on Walsh Transform of INL (10 - Bit ADC)

4 bit 6 (Z5)

I I . ..,.,I I
I n,,
I I

bit 5 C24)
\

-O.OlOL . . ’ . . ’ 1 . . ’ . . - ’ . . ’ . . . ’ - - ’ . . .
0.0 6.0 16.0 24.0 ¶2.0 40.04 48.0 56.0 64.0

Sequence Number

(b)
Fig. 10. (a) The Walsh transform of the INL of the tested ADC. (b) Zoom on Fig. 10(a).
784 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, VOL. CAS-33, NO. 8, AUGUST 1986

INL, = - 24 = 0.78 I t
N
compare with given value: 0.8 LSB
t
‘.
%i3= - 2a,/23..100 = 9.8 ’ Ni ij

(compare with lo-percent error on weighting value) I I .

’ N ’
Nj 0
These values correspond within the uncertainty to the Fig. 11. Study of the cross correlation of N, and N,.
given parameters in the simulation.
IAPPENDIX
III.6. Application on a IO-bit SAR-ADC. To estimate the uncertainty on the DNL and INL it is
The previous techniques have been applied to a real 10 necessaryto know the probability distribution of the cum-
bit SAR-ADC, preceded by a track-and-hold circuit and mulative probability Q. The variance and cross-correla-
used in a data acquisition channel. tion of ubi is derived using linear approximations.
Figs. 8 and 9 show the INL and DNL. The INL-char- To realize the value Q, it is necessary to have N,
acteristic shows a nonstochastical deviation, due to the measurements with a value < UB, and (N - N;) measure-
nonlinearity of the track-and-hold. This was confirmed by ments with a value > UB,. The distribution of Qj is a
a spectral analysis, made after the track-and-hold. binomial distribution, which can be very well approxi-
In Fig. 10(a), the resulting Walsh transform, applied to mated by a normal distribution [7]
the INL is illustrated. The influence of the bit failure is
easily seen.The bits, which are wrong, are indicated on the meas<UBi)N’(l-P(meas>UB,))N-N’
P(Q;) = CNN’P(
figure. To find these bits, the results of the Walsh Trans- =C$QF(l-Qi)“-” 09)
form of the INL are scannedby looking at the components
with sequencenumber 2N-’ -1. If only bit failure occurs, with Q; the estimated value of Q,.
as described in Section III.5 ( = single-bit failure), the The mean and standard deviation is given by [7]
Walsh Transform results in components with only se- PQ: = Qi, UQ;=~~. (20)
quence number 2N-i - 1 ( = odd number). Again looking
at Fig. 10(a), one seesimportant components with an even This means that Q, is an unbiased estimate of Qi. For this
number (sequencenumber 2 and 6). These are caused by reason no difference is made in notation for the measured
the non-linearities of the track-and-hold, which also re- and true value of Qj in the other parts of the text.
flects the parabolic evolution ( = even function) of the It’s also necessary to know the covariance between Qi
INL. This effect, however, will not disturb the odd compo- and Qj. Considering Fig. 11 the following relations can be
nents, describing single-bit failure. made:
Investigating bits 2, 3, and 5 on Fig. 10(a), another u&) = E[dq.dN,] (21)
effect is illustrated. The components, corresponding to
these bits, are accompanied by components with sequence with dNj, dNj the deviations of Ni, Nj to the expected
number 2N-i - 2. This effect can probably be explained by values E[ N,], E[ N,]
bit-intermodulation. The meaning of bit-intermodulation Nj=Ni+N,j+dNj=dNi+dN,j.
in SAR-ADC’s is that an error occurs on a lower bit
becausea higher bit is switched on. This phenomenon will Equation (21) becomes
be subject to further investigations building a fitting model 2 2
‘NY, = ‘it, + ‘N, N,, . (2-4
and extracting quantitive information.
On the other hand, we have the following relation:
N, + N, j + No = N = de + dNO = - dNij - dNi
IV. CONCLUSION
or
A method is presented to measure and analyze the
characteristics of an ADC. It is shown that it is possible to 2 =a; , +a; ,, +2u;, I 8,.
‘No
extract the information concerning the nonidealities from From this relation, u; ,N,, is derived and substituted in (22)
the bit-level. Using these techniques, the analysis can be
done on 2 levels. 4,T = [ 4, + Go - 4 /2
The first level ( = INL- and DNL-analysis) is used by
the designer of high-performance Digital Signal which can be further reduced, using
Processing-applications to evaluate the heart of his system,
u; = NQi(l - Qi)
the ADC.
While the designer will use the second level ( = Walsh u;;= NQa(l- Q,)
analysis + information extraction on bit level) to evaluate
and correct his ADC-hardware. u; ‘J= NQij(l - Qij)
VANDEN BOSSCHE et al.: A/D CONVERTERS 785

with M. Vanden Bossche was born in Reet, Belgium,

on August 9, 1961. He received the degree of
Q, = P(meas > ub,) engineer in 1984 from the University of Brussels.
He was scientific research worker of the
Qij=P(UBi<meas<UBj) =1-Qi I.W.O.N.L. and works presently for the National
Qj. Fund for Scientific Research (NFWO) as re-
Finally, (23) results in search assistant. The prime factors of his re-
search are situated in the field of Biomedical
Engineering (imaging techniques).
u,?,~= NQ,Qo = NQi(l - Qj)
and
u~,Q,= Qi (l- Qj)/N* (24
8
To derive the variance u,‘, the relation (5) ub, = - cos TQ,
is used. Differentiation gives
-dub, = r sin TQ, dQ,
and
J. Schoukens was born in Lennik, Belgium, on
u,‘h = E [ dub, dub,] = r2 sin*rQ& August 18, 1957. He received the degree of en-
gineer in 1980, the degree of Doctorate in ap-
plied Science in 1985 from the University of
= r2 sin21rQiQi(l - Q;)/N. Brussels (V.U.B.).
In the same way it is found that He is presently Senior Research Assistant of
the National Fund for Scientific Research
2
%b,ub, = E[dub,dub,] = vr2sin aQ, sin rQ,Qi (1 - Qj)/‘N* (N.F.W.O.). The prime factors of his research are
situated in the field’of parameter estimation for
linear and nonlinear systems.
ACKNOWLEDGMENT
The authors wish to thank Dr. R. Dearnly for critical
reading of the manuscript.
REFERENCES
[l] “Dynamic Performance Testing of A to D converters,” Hewlett
Packard product note 518OA-2.
[2] G. Pretzl, “A new approach for testing ultra fast A/D converters,”
in ESSCIRC ‘77, Ulm, pp. 142-145.
[3] -, “Dynamic testing of high speed aid converters,” IEEE J. J. Renneboog was born in Sint Genesius Rode
Solid-State Circuits, vol. X-13, pp. 368-371, 1978.
[4] 0. J. Downing and P. T. Johnson, “A method for the assessment of on March 29. 1939. He received the degree of
the performance of high speed analog/digital converters,” Electron. engineer in 1962, the degree of Doctorate in
Left., vol. 14, no. 8, pp. 238-240, Apr. 1978. Applied Science in 1967 from the University of
[5] J. Doember, H. Lee, and D. Hodges, “Full speed testing of A/D Brussels (U.L.B.).
converters,” IEEE J. Solid-State Circuits, vol. SC-19, pp. 820-827, He is presently Professor of Electrical Mea-
Dec. 1984. surements at the University of Brussels (V.U.B.)
[6] K. Beauchamp, Walsh Functions and Their Applications. London, and head of the Electrical Measurement Depart-
England: Academic, 1975. ment (ELEC).
[7] A. Papoulis, Probability Random Variables. New York: McGraw-
Hill, 1974.