110 IEEE TRANSACTIONS ON INSTRUMENTATION ANDMEASUREMENT, VOL. 54, NO. 1, FEBRUARY 2005
Uncertainty of ADC Random Noise Estimates
Obtained With the IEEE 1057 Standard Test
Francisco André Corrêa Alegria and António Manuel da Cruz Serra
Abstract—In this paper, the uncertainty of the estimated stan-
dard deviation of the random noise present in analog-to-digital
converters (ADCs), using the test suggested in the IEEE 1057-94
Standard, is thoroughly analyzed. Expressions for the test result
uncertainty and for determination of the test parameters are de-
rived.
Index Terms—Analog-to-digital converter (ADC) testing,
random noise, uncertainty.
I. INTRODUCTION
T
HE TRADITIONAL test method used to estimate
analog-to-digital converters (ADCs) random noise is the
one suggested in the IEEE Standard for Digitizing Waveform
Recorders [1]. For low random noise standard deviation, a
triangular stimulus signal is applied to the ADC input, and two
sets of samples are acquired synchronously ( and ). The
output codes obtained are then subtracted to eliminate possible
ADC fixed errors, and the mean square error is calculated using
mse (1)
This error is then used to estimate the standard deviation of the
random noise present in the ADC, , expressed in least signif-
icant bit units (LSB), through
mse mse
(2)
Actually, all the random additive noise present in the test setup
is estimated. Special care should be taken to minimize the noise
not due to the ADC itself, namely, the one introduced by the
function generator and by improper connection cables.
The knowledge of this kind of noise is important, for instance,
in selection of the proper ADC for a given application. It is also
important, when performing the standard static test [1] and the
standard histogram test [1]–[7], for the determination of the un-
certainty of the ADC transition voltages and code bin width es-
timates.
The estimation of the noise standard deviation obtained using
(2) is biased, as referred in [1] and demonstrated in [8]. The
Manuscript received July 9, 2003; revised June 5, 2004. This work was
supported by the Portuguese National Research Project entitled “New
error correction techniques for digital measurement instruments,” reference
POCTI/ESE/46995/2002.
The authors are with the Instituto Superior Técnico, Technical Uni-
versity of Lisbon, 1049-001 Lisboa, Portugal (e-mail: falegria@lx.it.pt,
acserra@ist.utl.pt).
Digital Object Identifier 10.1109/TIM.2004.840226
value of the estimation error is very small, lower than 0.022
LSB, if a triangular amplitude greater than five ideal code bin
widths is used [8].
The uncertainty of the noise standard deviation has, however,
never been studied, as far as the authors are aware. We use the
study presented in [8] and take it a step forward in order to an-
alyze the probability distribution of the estimate of the random
noise standard deviation and to obtain an expression for its un-
certainty. An expression for the minimum number of samples
required in order to limit the estimates uncertainty to a chosen
level is also proposed.
In Sections II–VII, we will analyze the probability distribu-
tion of several random variables, starting with the ADC output
codes, continuing with the square error, the mean square error,
and finally, the estimated random noise standard deviation.
Note that all voltage variables are normalized (divided) by the
ideal ADC code bin width .
II. ADC OUTPUT CODES
As seen in [8], the output code of a sample has a discrete
distribution with probability given by [8, eq. (11)]
(3)
where is the ADC number of bits, is the normalized tran-
sition voltage , is the normalized input voltage
and
(4)
The conditional mean, 2nd moment and variance of the output
codes are, by definition [9]
and (5)
It was seen in [8] that the ADC output codes standard deviation,
for small random noise standard deviation and when twice the
amplitude of the triangular wave amplitude is an integer, is given
by
(6)
0018-9456/$20.00 © 2005 IEEE