Cirrus Logic AN330 REV2 - Manual (Page 5)

AN330REV2 5
AN330
deviation was calculated with a result equal to 1.8 codes. The amount of voltage represented by one code of the
converter would be the full scale span of the converter (8 Vpp) divided by 2
20
codes, or 8 / 1048576 = 7.629 μV per
code. Note that one standard deviation of a Gaussian noise distribution is equivalent to the rms noise of the distri-
bution. Therefore, with the computed standard deviation of the 256 samples equal to 1.8 codes, the rms noise of
the distribution would be 1.8 times 7.623
μV = 13.7 μV. This agrees with the calculated estimate of 14 μVrms.
The noise level in the output codes of the converter can be improved by using the microcontroller to perform aver-
aging. If 20 samples at 107 Sps are averaged together, the averaging produces a sinc filter with an output rate of
107 / 20 = 5.35 Sps. The noise bandwidth of the Sinc
1
filter would have a bandwidth of one half the output rate, or
2.67 Hz. The noise floor across this frequency was 3104 nV /
√
Hz, therefore the noise in this bandwidth would be
about
√
2.67 times 3110 nV /
√
Hz = 5.07 μVrms. Peak-to-peak (±3.3 standard deviations) noise would be
30.4 microvolts. This would yield about 2.04 / 30.4
μV = 67,035 noise-free counts on the 5 mV signal from the load
cell.
Some comments about testing the circuit are appropriate. The circuit is amplifying very small signals. If the averaged
result above is to achieve 67,035 noise-free counts, this would indicate that one noise-free code, referred to Input,
is equivalent to 5 mV / 67,035 codes, or 75.5 nV per noise-free count. This is at a level that air currents in the room
can create changes in the parasitic thermocouple connections in the circuit. Therefore, the circuit board should be
covered up with a towel or put into a box and allowed to stabilize for a time before measurements are taken.
The load cell itself can be the source of unexpected spectral content in the signal. This can occur if the load cell
experiences vibration. Once while performing tests with a load cell, the level of noise did not match what was ex-
pected. The cause of the descrepancy was traced to the fact that the load cell was vibrating continuously. The sys-
tem was being tested on the fourth floor of an office building that had an underground parking garage. Due to
automobiles moving around in the garage, the building was always subject to minute vibrations that were being
sensed by the load cell. The testing had to be moved to a measurement environment on the ground floor in a differ-
ent building (without an underground parking garage) to achieve the level of performance that was expected. An-
other recommendation to help eliminate spurious output from the load cell itself while taking noise performance
measurements is to remove the load cell from its normal suspension system and to lay it on a padded surface in
such a way that it is orthogonal to the direction of the normal applied force.
When collecting output data from the ADC it is always beneficial if one can actually see the shape of the distribution
of the output codes. A method used to do this is described in Appendix A. Using Excel
®
to Produce a Noise Histo-
gram Plot.
The CS3001 + CS5513 circuit uses only one op amp with the ADC. It is, therefore, a lower-cost solution, but this
single-amplifier circuit has some disadvantages. First, this circuit is recommended only when the load cell is in very
close proximity to the op amp. If the load cell is some distance from the op amp, this introduces the possibility of the
circuit picking up interference in the wiring length from the load cell directly into the sensitive op amp inputs. Second,
the temperature coefficient of the two resistors that set the gain of the circuit (the load cell resistors in parallel as the
input R, and the feedback resistor around the op amp) can have different temperature coefficients. This may result
in gain drift with temperature changes.