Power Supply Rejection
The CS5012A/14/16’s power supply rejection
performance is enhanced by the on-chip self-cali-
bration and an "auto-zero" process. Drifts in
power supply voltages at frequencies less than
the calibration rate have negligible effect on the
CS5012A/14/16’s accuracy. This is because the
CS5012A/14/16 adjust their offset to within a
small fraction of an LSB during calibration.
Above the calibration frequency the excellent
power supply rejection of the internal amplifiers
is augmented by an auto-zero process. Any
offsets are stored on the capacitor array and are
effectively subtracted once conversion is initiated.
Figure 13 shows power supply rejection of the
CS5012A/14/16 in the bipolar mode with the
analog input grounded and a 300 mVp-p ripple
applied to each supply. Power supply rejection
improves by 6 dB in the unipolar mode.
The plot in Figure 13 shows worst-case rejection
for all combinations of conversion rates and input
conditions in the bipolar mode.
CS5012A/14/16 PERFORMANCE
Differential Nonlinearity
One source of nonlinearity in A/D converters is
bit weight errors. These errors arise from the de-
viation of bits from their ideal binary-weighted
ratios, and lead to nonideal widths for each code.
If DNL errors are large, and code widths shrink
to zero, it is possible for one or more codes to be
entirely missing. The CS5012A/14/16 calibrate
all bits in the capacitor array to a small fraction
of an LSB resulting in nearly ideal DNL. Histo-
gram plots of typical DNL of the CS5012A/14/16
can be seen in Figures 14, 16, 17. Figure 15 il-
lustrates the DNL of the CS5012 for comparison
with the CS5012A (Figure 14).
A histogram test is a statistical method of deriv-
ing an A/D converter’s differential nonlinearity. A
ramp is input to the A/D and a large number of
samples are taken to insure a high confidence
level in the test’s result. The number of occur-
rences for each code is monitored and stored. A
perfect A/D converter would have all codes of
equal size and therefore equal numbers of occur-
rences. In the histogram test a code with the
average number of occurrences will be consid-
ered ideal (DNL = 0). A code with more or less
occurrences than average will appear as a DNL
of greater or less than zero LSB. A missing code
has zero occurrences, and will appear as a DNL
of -1 LSB.
Integral Nonlinearity
Integral Nonlinearity (INL; also termed Relative
Accuracy or just Nonlinearity) is defined as the
deviation of the transfer function from an ideal
straight line. Bows in the transfer curve generate
harmonic distortion. The worst-case condition of
bit-weight errors (DNL) has traditionally also de-
fined the point of maximum INL.
Bit-weight errors have a drastic effect on a con-
verter’s ac performance. They can be analyzed as
step functions superimposed on the input signal.
Power Supply Ripple Frequency
1 kHz
10 kHz
100 kHz
1 MHz
P
o
w
e
rS
u
p
ly
R
e
je
c
tio
n
(
d
B
)
90
80
70
60
50
40
30
20
Figure 13. Power Supply Rejection
CS5012A, CS5014, CS5016
DS14F6
23