Cirrus Logic AN31 - Manual (Page 26)

rate of 1/second. Averaging 50 words reduces
the noise by
√
50 , or by a factor of 7.07. Since
the standard deviation, or rms value of the noise
illustrated in Figure 18 is 1.07 LSB, the rms
output noise in the post-filtered samples will be
1.07/7.07 = 0.151 LSB rms. You can use the
rule of thumb that peak to peak noise is
approximately 6 to 6.6 times greater than the
rms value to predict the peak-to-peak noise in
the post-processed output words. This results in
a peak- to-peak noise in the post-filtered output
words of less than
±
1 LSB for greater than
99.9% of the post-filtered output words.
To illustrate the dc stability and noise of the
post-filtered output words over time, the 1 Hz
post-filtered output words were collected for a
period of one hour. Note that for this test the
input to the bridge amplifier was removed from
the load cell and tied to ground through two 350
ohm resistors. This eliminates the load cell’s
sensitivity to vibration when studying the
digitizer input noise characteristics.
Figure 19 illustrates the peak-to-peak noise of
the digitizer over a one hour period. The plot
indicates that the drift and noise are less than
±
1 LSB for more than 99.9% of the output
samples over the hour long period. This is
superb performance and illustrates the benefit of
synchronous detection. Figure 18 and Figure 19
indicate that the 50 Hz output data from the
converter can averaged to yield a 1 Hz update
rate which is stable to 1 count in
±
524,000 when
the converter is set up for bipolar mode. The
CS5520 includes a DAC and a ratiometric offset
register which can be used to offset the span in a
negative direction by 500,000 counts. This
allows the weigh scale to have 24,000 counts of
underrange to accommodate any zero drift or
creep in the load cell. The measurement span for
the 18 mV load cell output would be over
524,287 counts, but above this would be another
500,000 counts which would allow the digitizer
to accurately measure overranged weights. In
this configuration the digitizer can accurately
digitize an overrange signal, even up to 195% of
full scale.
The GSE 4444 platform has mechanical stops
which activate at approximately 120% of
capacity, so the synchronous detection weigher
will yield a noise-free 19-bit measurement, with
a 20% overrange capacity. If the digitizer was
used with a tension-compression load cell such
as the BLH Electronics model LPT, the digitizer
would yield better than
±
500,000 noise-free
counts.
Digitizer Noise And Averaging
As illustrated in the previous example circuit, it
is good practice to evaluate the performance of a
prototype digitizer. While many measures of
performance should be investigated (linearity,
stability over temperature, etc.), one of the
primary factors which limits measurement
resolution is noise in the digitizer circuit itself.
Investigating the noise performance of the
digitizer should begin in the design phase.
Analysis should yield an estimate of the amount
of noise in the circuit. This discussion will not
focus on the analysis but will instead be limited
to evaluating the noise in the digitizer circuit.
One simple method of evaluating digitizer noise
is to "ground" the input and collect enough
samples to evaluate the noise statistically.
"Grounding" the input involves connecting the
signal + and signal - leads of the digitizer input
amplifier to a quiet node which has a voltage
equivalent to the common mode output of the
bridge to be measured. In a system with load cell
excitation of +5 V and -5 V the inputs can be
tied to ground. If the load cell is excited with a
single supply ( for example, +5 V or +10 V), a
quiet source with a common mode voltage
compatible with the input of the amplifier should
be generated. For example, if the circuit runs on
Bridge Transducer Digitizer Circuits
26 AN31REV3