and only amplify the difference voltage, but in the real
world this doesn’t happen because of imperfect com-
mon mode rejection ratio (usually given in dB). The
simplest one op-amp, unity-gain differential ampli-
fier circuit can only suppress common mode voltage
by 66 dB if 0.1% resistors are used, which means 300
V in common mode will result in 150 mV of error
at the output! Of course, there are other differential
amplifier circuits with higher intrinsic CMRR (e.g. the
instrumentation amplifier) but they cost more money
(and board area) and still don’t isolate the signal, so
the “cheap and cheerful” shunt might not be either of
those things!
It’s not all bad news for shunts, however. They tend
to be very robust and can usually withstand enormous
overloads (the signal conditioning circuit might “hit
the stops” - that is, run out of supply voltage, how-
ever). Also, if a zero temperature coefficient metal is
used (e.g. Manganin), then stability over time and
temperature will be unparalleled. Finally, shunts have
an almost unlimited dynamic range - limited only by
the signal-conditioning circuit, really. When you need
to accurately measure over a span of several decades
(for example, to keep track of traction battery state of
charge) then only a shunt will do.
Conversely, if you can tolerate poorer absolute ac-
curacy and/or drift over time/temperature, but really
need isolation, then a magnetic current sensor will
likely prove superior. The first of these we’ll cover is the
Hall effect type, which relies on an effect discovered by
Edwin Hall all the way back in 1879. Basically, when-
ever an external magnetic field is applied perpendicu-
lar to the flow of current in a conductor, it will bendtance in the way. For a concrete, yet realistic, example,
if a shunt has a mere 1 nH of stray inductance, then
it will generate a short spike at every transition of 1 V
if 20 A is switched in 20 ns (V = L [dI / dt]). Given
that many PWM controller ICs have an overcurrent
trip at 1 V, it’s clear that spikes like this can cause all
sorts of trouble. The usual solutions are to integrate
the spikes away with an RC filter that has the same
time constant as that of the shunt (that is, R C = L /
R), or else ignore, or blank out, the leading edge of the
current sense signal (or both). An RC integrator slows
down the transient response of the shunt, however,
while leading-edge blanking disables overcurrent
protection for the time it is in effect (and also enforces
a minimum switch on time in current mode control).
Increasing the resistance of the shunt while somehow
maintaining the same inductance will help, but at the
expense of generating more heat (and therefore lower
efficiency) from increased I^2 R losses.
Two other potential issues with shunts are interre-
lated: the signal they produce is obviously not isolated,
which means that they are not intrinsically safe; and
also, they are exposed to any “common mode” voltage,
which can produce huge errors in the signal condition-
ing circuit. Common mode voltage is basically that
which is the same at two different nodes in a circuit, as
referred to circuit ground. A classic example of where
this is a problem is in the case of monitoring the phase
current from an inverter. In this position the current
sensor is in between the inverter supplying a chopped
voltage and the motor load it drives, so neither end is
grounded, and since a shunt will have a negligibly low
resistance, practically the same magnitude of chopped
voltage will be present at both ends of it; that is, in
common mode. Reading a 100 mV to 1 V signal from
the shunt in the presence of what is effectively the
traction battery voltage requires a differential ampli-
fier to (theoretically) ignore the common mode voltage
30
Shunts tend to be very
robust and can usually
withstand enormous
overloads.
The Hall effect occurs in all
conductors, but the voltage
produced is downright
minuscule, even in special
semiconductors made
for the job, so follow-on
amplification is de rigueur.
THE TECH