This is only a preview of the April 2025 issue of Practical Electronics. You can view 0 of the 80 pages in the full issue. Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
|
Precision Electronics part four
Precision
Electronics
Part 4: Signal Switching
In this fourth article in this series, we will look at how to extend the current
measurement range of the circuit we’ve been working on so far. To achieve that, we’ll
have to switch between two or more shunt resistors.
By Andrew Levido
I
n the previous article in this series, we developed our current-sense circuit (Fig.1)
to the point where we could measure a 0–1A current in the high-side of
a hypothetical power supply with a
worst-case 25°C precision of around
±0.03%. Over the temperature range
of 0–50°C, this error rose to just under
±0.2%.
That was the analog error only;
it did not include any errors introduced by the analog-to-digital converter (ADC), which we will go into
in a future article.
To achieve this level of precision,
we were planning to apply a fixed gain
calibration and a dynamic zero offset
calibration in software, using the two
switches shown in Fig.1. This level of
precision would allow us to meaningfully measure current from 1A down
to a few tens of milliamps, since our
resolution is limited to ±2mA.
To achieve the microamp or better
current sensitivity that we desire, we
determined that we needed to switch
in different shunt resistors to provide
a series of current ranges.
So far, we have been using a 0.1W
shunt resistor for the 1A range, which
develops 100mV across it at full scale.
This requires a differential-mode gain
of about 25 to get our signal to a nominal 2.5V level for the ADC.
Assuming our power supply has
some voltage headroom, there is noth-
ing stopping us from increasing the
shunt resistance by an order of magnitude, so it drops 1V at full scale. We
can then decrease the gain to a nominal value of 2.5.
The power dissipation in the shunt
resistor will increase accordingly, but
any offset errors we see on the input
side, including those that change with
temperature, will be smaller in relation to the full-scale signal. That will
improve the overall precision of the
circuit. This will be important as the
complexity – and therefore sources of
uncertainty – of the circuit increases.
Table 1 (below) shows the ranges we
could potentially implement, the current resolution we could expect, and
the shunt resistors we would need for
each one.
This table assumes we can maintain
the ±0.2% error we have achieved so
far. It suggests we should be able to
realise our sub-microamp resolution
ambitions if we can maintain a similar level of precision as we did with
our previous efforts.
Before we get into the details of how
we will switch the shunt resistors in
and out, and the impact that will have
on precision, we should look at the
options available for signal switching.
There are basically only two options:
1. We can use a mechanical switch
such as a signal relay if we want to
control it with a microcontroller.
2. Alternatively, we can use some
form of electronic analog switch, which
will most likely be based on field-effect
transistors (FETs).
Signal relays
Signal relays are similar to power
relays, but their design is optimised
for low on-resistance and high linearity instead of power handling. They are
usually rated for currents of 2A or less
and for switching voltages under 50V.
These aren’t hard and fast definitions;
there is plenty of grey area between the
top end of signal relays and the bottom
end of power relays.
Relays have the advantage of excellent on-resistance linearity with applied voltage and temperature. They
have a very high off-resistance (essentially infinite) and virtually zero leakage since the switching path is electrically isolated.
Typical initial on-resistances for
signal relays range from about 10mW
to 200mW. The word “initial” is important here – the on-resistance of
signal relays generally increases with
the number of operations, as shown on
the right side of Fig.2. This is an extract from a data sheet for Panasonic
TQ-series relays, although all brands
behave in more or less the same way.
It’s also worth noting that the operating and release voltages, shown on
the left, also worsen slightly with time
Table 1 – current ranges using a fixed 0.2% error
Fig.1: this is
the circuit we
designed last
time. It is capable
of a measurement
resolution of
a couple of
milliamps; to
measure lower
currents, we need
to switch ranges
somehow.
68
Current range
Resolution (±0.2%)
RS (gain ≈ 2.5)
1.00A
±2.0mA
1.00W
100mA
±200µA
10.0W
10.0mA
±20µA
100W
1.00mA
±2.0µA
1.00kW
100µA
±200nA
10.0kW
10.0µA
±20nA
100kW
1.00µA
±2nA
1.00MW
Practical Electronics | April | 2025
Signal switching
Fig.2: relays make great signal switches, but we should be aware that their contact
resistance and operating voltages deteriorate with the number of operations.
as the relay’s mechanical parts wear.
The failure rate data for the TQ relays
suggests that 1% will have failed after
3.5 million operations and 10% after
about 10 million operations when
switching 5V at 1mA into a resistive
load. That is a lot of operations, so it
probably will not be of concern to the
designer, but relays do have a limited life.
Panasonic deserves a lot of credit for
publishing very comprehensive data
for their relays. Not all manufacturers
are this up-front in their data sheets.
Relays are not always good for very
high-frequency applications, since their
stray inductance and capacitance can
be relatively high. Specialised highfrequency relays are available if you
need them.
For precision circuits, we often use
reed relays, which can have very low
stray capacitance (0.5pF) and are available with internal electromagnetic
screens which can help minimise induced noise or be used as a “guard”
electrode when measuring minuscule
currents. A reed relay is essentially a
reed switch that’s actuated by an electromagnet.
On the downside, relays are usually somewhat bulky and expensive, so
designers tend to use them only when
their unique characteristics are absolutely necessary. Instead, they generally
use more compact and cheaper analog
switches where they can (which offer
the added benefit of an almost indefinite lifespan).
Analog switches
Analog switches are typically built
from Mosfets since their drain-source
resistance is controllable via gate voltage and the channel can conduct current in either direction. Because a Mosfet’s channel resistance is non-linear
with applied voltage, most analog
switches use back-to-back N-channel
and P-channel Mosfets.
The parallel on-resistance of the two
devices is more linear than either one
alone, as illustrated in Fig.3. The Mosfet
substrates are connected to the analog
power rails to maximise linearity.
By the way, if you are familiar with
using discrete Mosfets as high-power
switches, you may be puzzled by the
comment that they can conduct current in either direction. That’s because
Fig.3: most analog switches use
parallel N-channel and P-channel
Mosfets to minimise the effect
of the non-linear channel onresistance of Mosfets.
power Mosfets usually have an unavoidable ‘body diode’ in parallel with
the channel in one direction, meaning
they can only really switch current in
one direction by themselves.
When fabricating multiple Mosfets
on a single substrate as in a CMOS
integrated circuit, the body diode is
still there, but it is possible to choose
where one end of that diode connects.
Depending on what potential it is connected to, that body diode may never
conduct under normal conditions, so
it can effectively be ignored.
Thus, Mosfets in ICs (as well as the
fairly unusual four-terminal discrete
signal types that expose the bulk connection separately) can operate bi-
directionally, similarly to JFETs.
The NMOS+PMOS architecture is
used in switches such as those in the
industry-standard DG41x series. Fig.4
shows the simplified circuit of one
channel, extracted from the data sheet.
As well as the back-to-back switching
Mosfets, you can see a level shifter,
which allows the control voltage (VIN)
and logic supply (VL and GND) to be
anywhere within the V+ to V– analog
supply range.
Fig.4: this simplified
diagram of one
switch from a
DG41x series analog
switch shows the
parallel N-Channel
and P-channel
Mosfets. The level
shifter allows the
control signal and
logic supply to be
anywhere within the
analog voltage range.
Practical Electronics | April | 2025
69
Precision Electronics part four
Fig.5: the onresistance
characteristic of
this DG41x series
analog switch shows
the non-linearity
and temperature
dependence of the
on-resistance.
The DG41x series switch on-
resistance characteristic with ±5V
rails is shown in Fig.5. The nominal
on-resistance is anywhere between 10W
and 20W, depending on temperature,
and varies about 30% as the signal
voltage changes.
The imprecision associated with
analog switches can best be understood by looking at the on- and offstate equivalent circuits in Fig.6. In
the on state (left), the on-resistance
Ron appears in series with the source
resistance Rsource to produce a voltage
divider with the load resistance Rload.
As we have seen, Ron is non-linear
and temperature-dependant, so the
voltage error due to this divider will
be uncertain. For this reason, we usually try to keep the load resistance as
high as possible with respect to the
sum of Rsource + Ron.
In the on state, a leakage current
Id(on) will produce a DC error voltage
proportional to Rload in parallel with
Rsource + Ron. This can be minimised
by keeping the source impedance as
low as possible.
The channel capacitance Cd(on) will
appear in parallel with Cload and form
an RC low-pass filter with Rsource + Ron
– another reason to keep Rsource low if
you can.
In the case of the DG41x family of
switches, Ron can be up to 35W, Id(on)
can be up to ±15nA and Cd(on) is typically 35pF.
In the off state (shown in Fig.6),
the leakage current I s(off) will produce a DC voltage across Rsource, and
ID(off) will produce a voltage across
the load impedance, Rload. The latter
can be more difficult to manage, since
we generally want to use a high load
impedance for reasons described
above. The DG41x switches have
Fig.6: these equivalent circuits show the leakage currents and internal
capacitances present in analog switches in the on and off states.
70
off-state leakage currents (ID(off) and
Is(off)) of up to ±15nA, and CD(off) can
be up to 9pF.
Charge injection is another concern
with analog switches, especially those
with a low Ron value. Achieving low
Ron requires physically large Mosfets,
which have higher levels of gate capacitance. Whenever the gate of the
Mosfet switches, this gate capacitance
is charged or discharged via the drain
and source. This means a charge is injected into the signal path when the
devices switch.
The resulting voltage disturbance is
a factor of the switch output and load
capacitance, as shown in Fig.7. The
charge is injected via Cq and appears
as a voltage spike or dip at the output,
as CD(ON) in parallel with Cload charge
or discharge.
Each DG41x switch has a charge injection of 5pC. If the external load capacitance were 50pF, this would result
in a voltage spike or dip of 59mV every
time the switch changes state. This
could very well create a significant
‘pop’ when switching audio signals –
something to be aware of.
Of course, the input signal to this
type of analog switch must stay within
the power rails. For switches with backto-back complimentary Mosfets, the
signal voltage can extend all the way
to both rails.
There are some newer analog
switches with very good Ron linearity. These appear to use a single N-
Channel Mosfet with a very flat Ron
characteristic.
Fig.8 shows the on-r esistance
characteristic for one channel of the
TMUX821x series of analog switches
from Texas Instruments (TI). The onresistance is very flat all the way from
Fig.7: charge injection can cause
voltage transients in the signal path
when an analog switch is opened or
closed. The effect is usually worse in
low-Rds(on) switches, where the gate
capacitance (Cq) is higher.
Practical Electronics | April | 2025
Signal switching
Fig.9: this interesting
class of optically coupled
analog switches may
be suitable for some
applications. They can
switch a few hundred
milliamps and provide
good isolation between the
control signal & switch.
Fig.10: this circuit ensures the current-carrying
switches (S1a, S2b and S3c) are not in the
measurement path. That’s helpful since the voltage drop across them is unpredictable. The shunt voltage sensing
switches (S1b, S2b and S3b) carry no appreciable current, so the voltage drop across them will be minimal.
the negative supply up to a few volts
short of the positive supply.
With the ±15V supplies shown here,
the upper limit on signal voltage is
around 10V to 12V, depending on
how much non-linearity you can put
up with.
Before we leave this discussion of
analog switches altogether, I want to
mention one more type that I have
found useful in certain applications:
optically coupled Mosfet switches,
such as that shown in Fig.9.
These are a bit of a hybrid between
relays, analog switches and opto-
couplers. They use inverse series
Mosfets (for polarity independence),
which are switched optically via an
internal LED. A typical example, the
AQY282GS, is rated for switching up
to 60V (AC or DC) at 0.8A. It has a maximum on-resistance of 0.8W at 25°C,
rising to twice that at 85°C.
The manufacturer does not provide
any linearity data, but we can assume
it will not be great.
They do have good input–output isolation (1000MW and 1.5pF), but up to
1µA of leakage between the output terminals when off. These devices are not
Fig.11: this circuit configuration was used to obtain the results described. Not
shown are the DIP switches used to control the analog switches & relay coils.
super-fast – the switch-on time can be
up to 5ms and switch-off up to 0.5ms.
They are driven exactly like you would
drive an optocoupler.
Updating our design
So, armed with all this knowledge,
how do we go about designing our
multi-range current sensing circuit?
Whatever type of switch we use to
select the shunt resistors, it will add
a material and unpredictable voltage drop.
We therefore can’t just put the
Fig.8: the onresistance
characteristic of
the TMUX821x is
remarkably flat
for signal voltages
from the negative
rail up to a couple
of volts short of the
positive rail. This
suggests a single
Mosfet is being
used. Note how
the Rds(on) is still
highly temperaturedependent.
Practical Electronics | April | 2025
switching element in series with the
shunt and measure the voltage across
them both. Instead, we need to use the
topology shown in Fig.10.
One of the “a” switches (S1a, S2a or
S3a) is closed to select one of the shunt
resistors, depending on the chosen
current range. The corresponding “b”
switch is also closed, connecting the
relevant shunt resistor to the instrumentation amplifier’s inverting input.
Since this input has a very high
impedance, very little current flows
through the “b” switch, so its on-
resistance and non-linearity are largely
irrelevant. The voltage drop across the
active “a” switch, where appreciable
current does flow, is not in the measurement path, so it does not impact
the reading.
As a bonus, we get the zero-
calibration state for free. If we close
any “b” switch that does not have its
corresponding “a” switch closed, we
effectively short the inamp’s inputs
together via that shunt resistor, which
will have close to zero voltage across
it.
I decided to build a version of this
circuit with 1A, 10mA and 100μA
full-scale ranges. In a real application, you would probably implement
71
Precision Electronics part four
Fig.12: this graph,
copied from the
manufacturer’s
data sheet, shows
the various
leakage currents
in the TMUX821x
series of analog
switches. As you
would expect, they
increase rapidly
with temperature.
a range for each decade, but I wanted
to keep things manageable for my experiments. I chose to use relays for
S1a and S2a (the 1A and 10mA range
respectively), although an analog
switch could certainly be used for
the latter range.
The 100µA range (S3a) and the three
“b” switches used analog switches. This
meant I could get away with just one
quad analog switch package.
The key parts on the test board are
shown in Fig.11. A 3.3V logic power
supply and the dip switches driving
the relays and analog switch control
lines are not shown. I used a 1% tolerance 3W resistor for R1, since highprecision power resistors are super
expensive.
I did, however, select a resistor with
the best tempco (±20ppm/°C) that I
could afford, since we can’t trim out
the temperature drift as easily as we
can trim out the absolute resistance
error. It is easier (and cheaper) to get
high-precision 100W and 10kW resistors, so I chose devices with 0.1% tolerance and 10ppm/°C tempcos.
The relays I used were 3.3V coil 1A
relays from Fujitsu’s SY series that I
happened to have on hand. The primary concern with selecting the analog
switch was to get a unit with a sufficient voltage rating, since the supply
voltages would be +24V and -5V, giving
a total supply span of 29V. DG41x-series switches are limited to a supply
voltage span of 12V.
Figs.13 & 14: the voltage error due to analog switch leakage is calculated by
substituting the on and off equivalent circuits. As discussed in the text, the
600pA source can be
ignored but the other
two will cause an error.
This diagram shows the
100µA range where the
error is worse than the
others. The simplified
version is shown at
right; it summarises the
sources of error.
72
The TMUX821x range is good to
±50V, which is more than enough. The
TMUX8212 includes four independent normally open switches, which
is perfect.
From Fig.8, we can see that the analog
switch on-resistance is under 5W at
room temperature, with about ±1W
change over the 0°C to 50°C range we
are designing for. Fig.12 shows the
leakage currents. At 50°C, the worst
case for our design, Id(on) is ±10pA or
less, while Id(off) and IS(off) are each less
than ±300pA.
Those figures are for ±36V supplies,
so with our lower supply voltages,
the values we experience are likely
to be lower. However, in the absence
of more detailed data, we have little
choice other than to use those figures.
I used the cheaper of the two instrumentation amplifiers that we tested
last time, the INA821, but this time
with the gain set to about 2.5. Like
last time, the op amp is powered from
+24V and -5V rails.
Error budget
The easiest way to manage the error
budget for a circuit with several configurations like this one is to calculate
a separate budget for each range. The
process is exactly the same as for the
examples we created in previous instalments, except for the errors introduced by the analog switches.
We can distil the impact of the analog
switches down to a single voltage
error by substituting them with their
equivalent circuits, as shown at the
top of Fig.13.
Here, the circuit is shown with the
100µA range active (with the two analog
switches closed and both relays open).
Fig.14 shows the same configuration with the leakage current sources
consolidated. The 1W and 100W resistors disappear, since they are in series
with current sources, which themselves
have very high (theoretically infinite)
source resistances. This simplification
leaves us with three potential sources
of leakage-induced voltage error.
The 600pA current feeding into the
power rail on the source side of the shunt
resistor can be ignored, since this current
must flow either back into the regulator
(where it does not matter), or through
the shunt to the load (where it will be
measured as part of the load current).
The 10pA source on the load side
of the shunt will cause an error since
this current can flow into the load
Practical Electronics | April | 2025
Signal switching
Table 2: 100μA range
At Nominal 25°C
Error
Nominal Value
Shunt Resistor: ERA-6ARB103V (±0.1%, 100ppm/˚C)
10kW
Abs. Error
Rel. Error
Additional error over
0-50°C (Nominal ±25°C)
Abs. Error
0.10%
Rel. Error
0.025%
Input voltage error due to shunt
1V
1mV
0.10%
0.25mV
0.025%
Input voltage error due to switch leakage
0V
6.2μV
0%
0mV
0%
Input voltage error due to bias (Ios ±0.5nA, ±20pA/˚C)
0V
5μV
0%
5μV
0%
InAmp: INA821 (Vos ±35µv, 5µV/˚C)
0V
35μV
InAmp Input Voltage error total (Sum of Lines 2-5)
0V
1mV
0.10%
0.380mV
0.038%
InAmp Gain Resistor Rg: ERA-6ARB333V (±0.1%, 10ppm/˚C)
33kW
33W
0.10%
8.3W
0.025%
125μV
InAmp Gain Error (0.015% ±35ppm/˚C)
0.02%
0.088%
InAmp Gain (Line 7 × Line 8)
2.5
0.0029
0.12%
0.0028
0.113%
Vout DM (Line 6 × Line 9)
0V
5.5mV
0.22%
3.8mV
0.151%
Vout CM (20V, 100db, ±1.5db over 0-50˚C)
0V
200μV
Vout (Line 10 + Line 11)
0V
5.7mV
Table 3: 10mA range
37.7μV
0.23%
At Nominal 25°C
Abs. Error
Rel. Error
3.8mV
0.152%
Additional error over
0-50°C (Nominal ±25°C)
Error
Nominal Value
Abs. Error
Shunt Resistor: ERA-6ARB101V (±0.1%, 10ppm/˚C)
100W
Input voltage error due to shunt
1V
1mV
0.10%
0.25mV
0.025%
0.10%
Rel. Error
0.025%
Input voltage error due to switch leakage
0V
95nV
0%
0nV
0%
Input voltage error due to bias (Ios ±0.5nA, ±20pA/˚C)
0V
50nV
0%
50nV
0%
InAmp: INA821 (Vos ±35µv, 5µV/˚C)
0V
35μV
InAmp Input Voltage error total (Sum of Lines 2-5)
0V
1mV
0.10%
0.3751mV
InAmp Gain Resistor Rg: ERA-6ARB333V (±0.1%, 10ppm/˚C)
33kW
33W
0.10%
8.3W
125μV
InAmp Gain Error (0.015% ±35ppm/˚C)
0.02%
0.025%
0.088%
InAmp Gain (Line 7 × Line 8)
2.5
0.0029
0.12%
0.0028
0.113%
Vout DM (Line 6 × Line 9)
0V
5.5mV
0.22%
3.8mV
0.150%
Vout CM (20V, 100db, ±1.5db over 0-50˚C)
0V
200μV
37.7μV
0.038%
Vout (Line 10 + Line 11)
0V
5.7mV
3.8mV
0.152%
Table 4: 1A range
0.23%
At Nominal 25°C
Abs. Error
Error
Nominal Value
Shunt Resistor: VMP-1R00-1.0-U (±0.1%, 20ppm/˚C)
1W
Input voltage error due to shunt
1V
10mV
1%
0.5mV
0.05%
Input voltage error due to switch leakage
0V
4.5nV
0%
0nV
0%
Input voltage error due to bias (Ios ±0.5nA, ±20pA/˚C)
0V
500nV
0%
500nV
0%
InAmp: INA821 (Vos ±35µv, 5µV/˚C)
0V
35μV
InAmp Input Voltage error total (Sum of Lines 2-5)
0V
10mV
1%
0.625mV
0.063%
InAmp Gain Resistor Rg: ERA-6ARB333V (±0.1%, 10ppm/˚C)
33kW
33W
0.10%
8.3W
0.025%
Abs. Error
InAmp Gain (Line 7 × Line 8)
2.5
0.0029
0.12%
0.0028
0.113%
Vout DM (Line 6 × Line 9)
0V
28mV
1.12%
4.4mV
0.175%
Vout CM (20V, 100db, ±1.5db over 0-50˚C)
0V
200μV
Vout (Line 10 + Line 11)
0V
28.2mV
1%
InAmp Gain Error (0.015% ±35ppm/˚C)
Practical Electronics | April | 2025
Rel. Error
Additional error over
0-50°C (Nominal ±25°C)
Rel. Error
0.05%
125μV
0.02%
0.088%
37.7μV
1.13%
4.4mV
0.177%
73
Precision Electronics part four
without being measured. This is the
equivalent of under-reading the load
current by 10pA, so it will result in a
voltage error of up to 100nV (10pA ×
10kW) at the op amp input.
The 610pA leakage will similarly
cause a voltage error, but this time the
error will be seen across the series combination of the shunt resistance and
the switch on-resistance. This error
will be 6.1µV (610pA × [10kW + 6W]).
The total voltage error introduced by
the switches will therefore be ±6.2µV,
which you can see in line 3 of the error
budget table for the 100µA range.
This is a meaningful amount compared with the instrumentation amplifier’s ±35µV input offset voltage.
Given the relatively high shunt resistance, we also have to account for the
impact of the instrumentation amp’s
input bias currents. The difference between these currents (the input offset
current) will cause an additional voltage error across the source resistance.
The INA821’s data shows the maximum input offset current is ±0.5nA at
25°C, with a tempco (estimated from
the graphs) of ±20pA/°C.
This will result in a voltage error
of ±5.0uV at 25°C with an additional ±5.0µV over the 0°C to 50°C operating range. This error, shown on
line 4 of the error budget, is also
similar in magnitude to the input
offset voltage.
Other ranges
As you might expect from the above
calculations, the error voltages will be
lower for the other ranges where the
shunt resistances are lower.
I went through the same exercise
for these ranges and came up with
error voltages due to switch leakage
of 4.6nV for the 1A range and 95nV
for the 10mA range, plus input offset
current errors of 500pV and 50nV,
respectively.
These are included in the relevant
error budget tables (Tables 2-4), but
are frankly so small as to be irrelevant
given the instrumentation amplifier’s
±35µV offset voltage.
The rest of the error budget tables are
calculated as we did the last time. The
upshot is a worst-case untrimmed 25°C
error of ±1.13% for the 1A range and
±0.23% for the 10mA
and 100µA ranges.
The big difference
td
is due to the 1% tolerance of the 1W shunt
compared to the 0.1%
tolerance of the other
All of our stock is RoHS compliant and CE
two.
approved. Visit our well stocked shop for
Over the operating
all of your requirements or order on-line.
temperature range,
the 1A range has an
We can help and advise with your enquiry,
additional ±0.18%
from design to construction.
error, with an extra
±0.15% for the 10mA
and 100µA ranges.
Recall that the cir3D Printing • Cable • CCTV • Connectors • Components •
cuit in the previous
Enclosures • Fans • Fuses • Hardware • Lamps • LED’s •
article had a worstLeads • Loudspeakers • Panel Meters • PCB Production •
Power Supplies • Relays • Resistors • Semiconductors •
case untrimmed 25°C
Soldering Irons • Switches • Test Equipment • Transformers
error of 0.65% with
and so much more…
±0.28% additional
Monday to Friday 08:30 - 17.00, Saturday 08:30 - 15:30
error over temperature.
This circuit is
better (except on the
1A range, where the
shunt tolerance range
has doubled) because
we have used tighter
Station Road
tolerance resistors
Cullercoats
and have reduced
North Shields
the instrumentation
Tyne & Wear
NE30 4PQ
amplifier gain by a
Tel: 0191 2514363 sales<at>esr.co.uk www.esr.co.uk
factor of 10.
ESR Electronic Components L
74
Testing
As usual, I built the circuit and carefully measured its performance. The
results are shown in the tables opposite (Tables 5-7). Again, we achieved
much better performance than the
worst-case calculations would suggest. The measured untrimmed errors
were ±0.5%, ±0.06% and ±0.18% for
the 1A, 10mA and 100µA ranges, respectively.
To calculate the trimmed error results, I used a gain correction based
on the line of best fit, but just used
the measured zero-current output
value as the offset, mimicking the
dynamic offset correction process.
The trimmed errors were ±0.036%,
±0.054% and ±0.031% for the three
ranges – very similar to the values we
achieved previously.
The errors over the operating temperature range are around ±0.11%, assuming the offset calibration eliminates
the offset component of the input-side
temperature drift error. It would be
around ±0.18% otherwise.
We can probably say that, across
all ranges, our circuit achieves better
than ±0.06% error at 25°C and ±0.25%
over the operating temperature range.
This is on par with the performance
we saw last time, and means we have
more-or-less met the expectations we
set in Table 1 for these ranges.
As a paper exercise, I calculated
the error budget for a possible 1µA
full-scale range, assuming a 1MW
0.1% ±10ppm shunt. The worst-case
untrimmed error at 25°C is ±0.35%,
and the total error over the temperature range would be within ±0.6%,
which is pretty good. With trimming,
we could probably assume a current
resolution in the order of ±5nA. This
is about as low as I would go with
this circuit.
Once we get down to measuring
such small currents, things become
very challenging.
A next obvious step will be to look
into the analog-to-digital conversion
process, to complete our theoretical
PSU current-sensing design.
However, in all of our work so far,
we have entirely ignored one important source of uncertainty and error:
noise. This is an interesting but complex topic that we need to know about
before moving on. So we will cover it
in next month’s article, before moving
onto the digital side of measurement
PE
systems (ie, DACs & ADCs).
Practical Electronics | April | 2025
Signal switching
Measured Data
Untrimmed Error
Trimmed Error
I (mA)
Vout (mV)
Absolute (mV)
Relative
Absolute (mV)
Relative
0.0000
-3.480
-3.48
-0.14%
0.00
0.000%
9.0707
223.366
-3.13
-0.12%
0.38
-0.001%
20.1549
500.500
-2.76
-0.11%
0.78
-0.007%
29.1271
723.712
-3.58
-0.14%
-0.02
0.006%
38.4297
955.808
-3.77
-0.15%
-0.18
0.016%
49.9203
1243.030
-3.46
-0.13%
0.16
0.018%
58.7674
1464.160
-3.24
-0.13%
0.41
-0.021%
72.2879
1801.780
-3.23
-0.13%
0.47
0.031%
80.1932
1998.150
-4.25
-0.17%
-0.53
0.000%
86.6674
2160.290
-3.77
-0.15%
-0.03
0.000%
95.1638
2373.240
-2.97
-0.12%
0.79
0.000%
www.poscope.com/epe
Table 5 – 100μA range (Vcm = 20V).
Measured Data
Untrimmed Error
Trimmed Error
I (mA)
Vout (mV)
Absolute (mV)
Relative
Absolute (mV)
Relative
0.00000
-0.773
-0.77
-0.03%
0.00
0.000%
0.98420
244.922
-0.83
-0.03%
-0.06
-0.002%
1.98602
495.514
-0.39
-0.02%
0.38
0.015%
2.93840
733.371
-0.34
-0.01%
0.43
0.017%
4.18878
1045.728
-0.20
-0.01%
0.56
0.022%
4.98283
1244.140
-0.06
0.00%
0.70
0.027%
5.85370
1461.660
0.01
0.00%
0.76
0.030%
7.11774
1775.860
-1.42
-0.06%
-0.67
-0.026%
7.99387
1996.360
0.31
0.01%
1.06
0.041%
8.68506
2169.050
0.42
0.02%
1.16
0.045%
9.53879
2382.450
0.64
0.03%
1.39
0.054%
10.64341
2658.070
0.44
0.02%
1.18
0.046%
Table 6 – 10mA range (Vcm = 20V).
Measured Data
Untrimmed Error
Trimmed Error
I (mA)
Vout (mV)
Absolute (mV)
Relative
Absolute (mV)
Relative
0.000
0.055
0.05
0.00%
0.00
0.000%
100.303
251.400
0.95
0.04%
-0.27
-0.011%
199.851
500.786
1.76
0.07%
-0.61
-0.024%
300.618
754.046
3.41
0.13%
-0.14
-0.005%
400.330
1003.724
4.11
0.16%
-0.59
-0.023%
500.944
1255.870
5.03
0.20%
-0.85
-0.033%
601.552
1508.490
6.43
0.25%
-0.61
-0.024%
701.079
1758.470
7.90
0.31%
-0.31
-0.012%
800.656
2008.760
9.55
0.37%
0.18
0.007%
901.122
2261.360
11.29
0.44%
0.75
0.029%
1000.709
2511.350
12.61
0.49%
0.92
0.036%
Table 7 – 1A range (Vcm = 20V).
Practical Electronics | April | 2025
- USB
- Ethernet
- Web server
- Modbus
- CNC (Mach3/4)
- IO
- PWM
- Encoders
- LCD
- Analog inputs
- Compact PLC
- up to 256
- up to 32
microsteps
microsteps
- 50 V / 6 A
- 30 V / 2.5 A
- USB configuration
- Isolated
PoScope Mega1+
PoScope Mega50
- up to 50MS/s
- resolution up to 12bit
- Lowest power consumption
- Smallest and lightest
- 7 in 1: Oscilloscope, FFT, X/Y,
Recorder, Logic Analyzer, Protocol
decoder, Signal generator
75
|