52 CHAPTER 2 Performance Characteristics of Sensors and Actuators

Noise and errors are additive since the outputs are in series. Because the inputs are
in parallel thermally, and because the thermocouples are essentially identical, the errors
are also identical (or nearly so) and therefore the output error is n times the error (or
noise) of a single thermocouple.

2.2.6 Hysteresis, Nonlinearity, and Saturation

Hysteresis (literally lag) is the deviation of the sensor’s output at any given point when
approached from two different directions (see Figure 2.17). Specifically, this means
that the output at a given value of stimulus when it increases and when it decreases is
different. For example, if temperature is measured, at a rated temperature of 50! C, the
output might be 4.95 V when the temperature increases, but 5.05 V when the tem-
perature decreases. This is an error of "0.5% (for an OFS of 10 V in this idealized
example). The sources of hysteresis are either mechanical (friction, slack in moving
members), electrical (such as due to magnetic hysteresis in ferromagnetic materials), or
due to circuit elements with inherent hysteresis. Hysteresis is also present in actuators
and, in the case of motion, is more common than in sensors. There it may manifest itself
as positioning errors. Also, hysteresis may be introduced artificially for specific
purposes.
Nonlinearity may be either a property of a sensor (see, e.g., Figure 2.1) or an error
due to deviation of a device’s ideal, linear transfer function. A nonlinear transfer func-
tion is a property of the device and, as such, is neither good nor bad. One simply has to
design with it or around it. However, a nonlinearity error is a quantity that influences the
accuracy of the device. It must be known to the designer, must be taken into account,
and possibly minimized. If the transfer function is nonlinear, the maximum deviation
from linearity across the span is stated as the nonlinearity of the device. However, this
measure of linearity is not always possible or desirable. Therefore there are various valid
ways of defining the nonlinearity of a sensor or actuator. If the transfer function is close
to linear, an approximate line may be drawn and used as the reference linear function.
Sometimes this is done simply by connecting the end points (range values) of the
transfer function (line 1 in Figure 2.18). Another method is to draw a least squares line
through the actual curve (line 2 in Figure 2.18), usually by first selecting a reasonable
number of points on the curve and then, given the selected (or measured) pairs of input
and output values (xi, yi), draw the line y ¼ a þ bx by calculating the slope a and the

FIGURE 2.17 ¢ V
Hysteresis in a 10 V
sensor. Increase in
stimulus
Ideal transfer
function Hysteresis

Decrease in
stimulus

Stimulus
0 Max
 2.2 Input and Output Characteristics 53

Output FIGURE 2.18 ¢

Linear
Transfer function:
using end points approximations of
Transfer function: (1) nonlinear transfer
using least squares functions.
(3)
(2)

Transfer function:
using tangent to a point Stimulus

axis intercept b as shown in detail in Appendix A. The linear best fit to the data (see
Equation (A.15)) is
! "! " ! "! " ! "! "
P
n P
n P
n P
n P
n P
n P
n
n xi yi % xi yi yi %xi2 xi xi yi
i¼1 i¼1
i¼1 i¼1 i¼1 i¼1 i¼1
a¼ ! n "2 , b¼ ! n "2 : (2.19)
P
n
2
P P
n
2
P
n xi % xi n xi % xi
i¼1 i¼1 i¼1 i¼1

This provides a ‘‘best’’ fit (in the least squares sense) through the points of the transfer
function and can be used for purposes of measuring the nonlinearity of the actual sensor
or actuator. If this method is used, nonlinearity is the maximum deviation from
this line.
There are many variations on both of these methods. In some cases a sensor is only
expected to operate in a small section of its span. In this case, either method may be
applied for that portion of the span. Another method that is sometimes employed is to
take a midpoint in this reduced range and draw a tangent to the transfer function through
the selected point and use this tangent as the ‘‘linear’’ transfer function for purposes of
defining nonlinearity (line 3 in Figure 2.18). Needless to say, each method results in
different values for nonlinearity and, while these are valid, it is important for the user to
know the exact method used.
It should also be noted that in spite of the foregoing, nonlinearity is not neces-
sarily a ‘‘bad’’ thing or something that needs correction. In fact, there are instances in
which a nonlinear response is superior to a linear response and there are sensors and
actuators that are intentionally and carefully designed to be nonlinear. An example is
the common potentiometer used as a volume control, especially in audio systems.
Notwithstanding the fact that most current volume control systems tend to be digital
and many are linear, our hearing is not linear—in fact, it is logarithmic. This allows
the ear to respond to minute pressure changes (as low as 10%5 Pa) as well as high
pressures (high power sound)—as high as 60 Pa. Normally the range is given between
0 and 130 dB. To accommodate this natural response, potentiometers for volume
control were also designed as logarithmic to conform with our ears’ response. Even
some digital potentiometers are logarithmic. The following example discusses these
issues a bit more, but the important point is that this nonlinear response has been
designed on purpose to fit a particular need and in this case a nonlinear response is
superior.