Cirrus Logic AN31 - Manual (Page 27)

a single +5 V supply, use two 100 ohm resistors
connected in series between +5 V and ground.
Then connect the input of the digitizer circuit to
the 2.5 V node of the resistor pair. While a load
cell simulator may be used in many
circumstances, this can be a source of some
problems. Some simulators exhibit 1/f noise
which can adversely affect the data output from
a high resolution digitizer. And some simulators
may not work well with the circuits which use
ac-excitation. This is because some simulators
use switches which rectify the ac excitation
signal; therefore the actual signal to be measured
is corrupted. This can result in greater noise than
expected as well as a dc offset error.
The biggest difficulty in evaluating the noise
performance of a circuit is that some means of
getting the data out of the digitizer and into a
computer must be designed into the circuit. For
the CS5504/5/6/7/8/9 devices this can be
accomplished by making the SCLK, SDATA
and DRDY signals available on a header. The
CDBCAPTURE system from Crystal has a
standard 10 pin (two rows of 5 pins) stake
header which can interface to the
CS5504/5/6/7/8/9 products and capture data from
these converters. Alternatively, a designer may
include some other type of interface in his
system to port data to a PC-compatible computer
via the serial or parallel port.
Once an interface is available, it is a matter
collecting enough conversion words to perform
meaningful statistical analysis on the data. The
CDBCAPTURE system enables the user to
capture data from the CS5504/5/6/7/8/9 and to
produce noise histograms. The CS5516 and
CS5520 are not supported with the
CDBCAPTURE system, but the CDB5516 or
CDB5520 evaluation boards can be configured
to collect data from these chips. Once data has
been collected into a file on a computer,
spreadsheets such as Quattro, Lotus, or Excel
can be used to analyze the data using a
frequency distribution function and statistical
functions. The data should also be plotted as
shown in Figure 18 to give the user an indication
that the data actually follows a Gaussian
(Normal) distribution. Thermal noise will have a
"bell-shaped" histogram. If the data words
represent thermal noise, one standard deviation
is equivalent to the rms noise; while 99.9% of all
the data should fall within
±
3.3 standard
deviations of the mean. Therefore the peak-to-peak
noise is approximately 6.6 times the rms noise.
When performing statistical analysis on a
digitizer’s output, at least 500 to 1000 conversion
words should be included to lower statistical
uncertainty to an acceptable level.
Once the rms noise is known (by calculating the
standard deviation of the data set), averaging can
be used to improve system resolution if it has
been confirmed that the noise follows a Gaussian
distribution. Data may not follow a Gaussian
distribution because it includes interference due
to dc-dc converters or to clock coupling which is
picked-up by the sensitive analog circuitry. In
this case averaging output words may be
deceptive. Averaging will reduce the
peak-to-peak noise but the mean can be
adversely affected by the interference which is
included with the signal.
One additional noise test is to measure noise
over the entire input span of the converter. If
noise increases with higher signal amplitudes, it
suggests the voltage reference input to the
converter is excessively noisy.
Conclusion
The circuits in this application note were
designed, constructed, and tested with the intent
of illustrating a wide variety of bridge digitizer
solutions. The circuits demonstrate various
power supply arrangements and various levels of
measurement resolution; all with the intent of
helping designers understand the flexibility of
the A/D converters which have been used.
Bridge Transducer Digitizer Circuits
AN31REV3 27