🔋 Current, Power & the Art of Sensing

Every current sensor is a translation. You cannot read amperes directly off a needle or an analog-to-digital converter; you read a voltage that some physical effect makes proportional to the current, then you scale. A resistor turns current into a voltage drop. A magnetic field turns current into a Hall voltage. A coil turns a changing current into an induced voltage. Power is one more translation on top: multiply the right voltage by the right current at the right instant. The order of this lesson is that ladder, from the simplest translation to the subtlest, ending on the timing detail that wrecks more power measurements than any other.

By the end, you can

1. Explain how a shunt resistor plus a differential amplifier turns current into a measurable voltage, and size the shunt from a burden-voltage budget
2. Choose between a shunt, a Hall current probe, and a Rogowski coil given DC content, bandwidth, and whether the wire may be broken
3. Explain why a power analyzer reports real versus apparent power and why power factor falls out of the two
4. Justify de-skewing the voltage and current channels and estimate the power error a skew introduces

Intuition first

Picture three ways to gauge a river without getting wet. You can drop a known rock into the streambed and watch how high the water piles up behind it: the faster the flow, the bigger the bow wave. That is a shunt, a small, precisely known resistor placed in the current’s path, where the voltage that piles up across it is your reading. You can hold a compass near the bank and watch the needle deflect with the moving water’s magnetic pull, never touching the river at all: that is a Hall probe, sensing the magnetic field a current always carries with it. Or you can wrap a loose ribbon loosely around the channel and feel it tug only when the flow changes: that is a Rogowski coil, which responds to how fast the current is rising or falling.

Each method trades something. The rock is simple and reads steady flow perfectly, but it nudges the river a little and the bow wave is tiny, so you need good eyes (a clean amplifier). The compass never touches the water and reads steady and changing flow alike, but it is fussier and pricier. The ribbon never touches the water either and can feel violent surges no rock could survive, but it goes blind on perfectly steady flow. Pick the instrument that matches the river you actually have.

The shunt: current becomes a voltage drop

The shunt is the workhorse, and it is nothing but Ohm’s law run backward. You insert a resistor of very low but accurately known resistance $R_{sh}$ in series with the load. All the current you care about flows through it, and the voltage it develops is

$$V_{sh} = I \, R_{sh}.$$

Measure $V_{sh}$, divide by the known $R_{sh}$, and you have the current. Because $R_{sh}$ is tiny (milliohms or less), $V_{sh}$ is tiny too, often only tens of millivolts at full current. That small signal is the price of not disturbing the circuit, and it is why a raw shunt is almost never read directly. You read it with a differential amplifier that sits across the two shunt terminals, rejects whatever common-mode voltage the rail sits at, and lifts the difference up to a level the ADC can use. This is exactly the DIFF probe discipline from lesson 26: a shunt is a tiny differential signal riding on a possibly large common-mode rail.

Shunts come rated by a current and a drop at that current, and the convention is to pick a full-scale drop of 50, 75, or 100 mV. A 500 A, 75 mV shunt, for instance, is just a resistor of $75\,\text{mV} / 500\,\text{A} = 150\ \mu\Omega$. The voltage you allow the shunt to develop is called the burden voltage, and it is the heart of the design trade. A bigger $R_{sh}$ gives a bigger, easier-to-read signal but steals more rail and dissipates more heat ($P = I^2 R_{sh}$); a smaller $R_{sh}$ barely touches the rail but buries the signal in amplifier noise and offset. You size the shunt by deciding how many millivolts you can afford to lose.

There is also a thermal ceiling. Shunt alloys like manganin are stable only over a limited temperature, and continuous operation forces a derating (66% of the rated current is a common rule) so the metal never anneals and shifts its resistance. Run a shunt hot and the very number you trusted, $R_{sh}$, drifts out from under you.

The Hall probe: sensing without breaking the wire

Sometimes you cannot break the wire. The motor is running, the cable is fat, the bus bar is bolted down, or you simply want to clamp on and read without rewiring anything. Here the Hall current probe earns its keep. Every current is wrapped in a magnetic field, and a Hall-effect sensor placed in that field produces a voltage proportional to the field, hence to the current. Clamp the jaws of the probe around a single conductor and you read its current with no electrical contact at all.

The defining virtue is that a Hall probe senses both AC and DC. The magnetic field of a steady current is just as real as that of a changing one, so unlike a plain transformer clamp, a Hall probe reads a battery’s quiet DC draw and a motor’s switching ripple in the same instrument. Hand-held Hall units routinely resolve down to a couple of hundred milliamps, and better ones reach a milliamp. The cost is more circuitry, more drift, and a price tag higher than a humble resistor. Shunts are cheaper and often more accurate (tenths of a percent), but they cannot give you galvanic isolation from a high-voltage line the way a clamp does. Choose the clamp when the wire must stay intact or when isolation matters; choose the shunt when accuracy and cost rule.

The Rogowski coil: built for fast di/dt

The third instrument is the most specialized and, in a switching robot, often the most revealing. A Rogowski coil is a helix of wire with no iron core wrapped around the conductor. Because it has no core to saturate, it stays beautifully linear even at enormous currents, and because it has very low inductance, it can follow currents that change in nanoseconds. It is the probe you reach for when $di/dt$ is brutal: motor phase switching, welding, pulsed power.

But the coil does not give you current directly. The voltage it induces is proportional to the rate of change of the current, not the current itself:

$$v(t) = -\frac{A N \mu_0}{l}\,\frac{dI(t)}{dt},$$

where $A$ is the area of one small loop, $N$ the number of turns, and $l$ the length of the winding. To recover the current you must integrate that signal, which a Rogowski system does with a dedicated integrator circuit (analog or digital). That integrator is lossy on purpose, which buys stability but costs you the bottom of the spectrum: a Rogowski coil cannot read a constant DC current at all. It samples a changing field, so a steady current produces no output. Steady DC is exactly the river the ribbon goes blind on.

So the three line up cleanly. The shunt reads DC and AC, breaks the wire, and is cheap and accurate. The Hall probe reads DC and AC, leaves the wire intact, and costs more. The Rogowski coil leaves the wire intact and laughs at fast transients, but it is deaf to DC. Knowing the current’s frequency content, and whether you may cut the wire, picks the tool for you.

From current to power: real, apparent, and the skew that lies

Knowing voltage and current separately is not the same as knowing power. A power analyzer is an instrument that takes a voltage channel and a current channel together and reports the numbers that actually matter for a power budget: real power, apparent power, efficiency, and power factor. The definitions are worth holding in your head.

Real power is the average of the instantaneous product of voltage and current over a full cycle, the energy genuinely delivered to the load per second:

$$P = \frac{1}{T}\int_0^T v(t)\,i(t)\,dt.$$

Apparent power is just the product of the RMS voltage and RMS current, $S = V_{rms}\,I_{rms}$, the power that looks like it is flowing if you ignore timing. When voltage and current are out of phase, or distorted, the real power is less than the apparent power, and their ratio is the power factor:

$$\text{PF} = \frac{P}{S} = \frac{P}{V_{rms} I_{rms}}.$$

A power factor of 1 means every volt-amp does useful work; a low power factor means current is sloshing in and out without delivering net energy, loading your wiring and your supply for nothing. The analyzer computes all of this by sampling $v(t)$ and $i(t)$ fast and multiplying them point by point, which is why the timing of those two channels is everything.

Here is the trap. Your voltage probe and your current probe almost never have the same delay. A Hall clamp’s electronics, or a Rogowski integrator, can lag a plain voltage lead by tens or hundreds of nanoseconds. If the two channels arrive at the analyzer misaligned in time, the instantaneous product $v(t)\,i(t)$ is formed from mismatched moments, and the integral comes out wrong. This is channel skew, and you must de-skew the channels (apply a compensating delay to the faster one) before you trust a single power or power-factor reading. Skew does almost nothing to the RMS values, so apparent power looks fine, which is exactly why a skewed setup fools you: $S$ is right, $P$ is wrong, and the power factor you compute is a fiction.

The robot hand: per-actuator current is a KPI

Bring all of this back to the dexterous hand. Each finger has an actuator, and the current that actuator pulls is one of the most informative numbers in the whole system. It tells you, in real time, how hard the finger is pushing, whether it has stalled against an object, whether a tendon is binding, and how much of your power budget that finger is spending right now. Per-actuator current is a key performance indicator, and it deserves a dedicated sense path, not an afterthought.

The standard answer on a control board is a low-side shunt plus a differential amplifier, one per actuator. The shunt sits in the return leg of each motor driver, so the common-mode voltage at the amplifier stays near ground and the design is gentle. The diff amp lifts the few-millivolt drop into a clean signal the microcontroller’s ADC can sample at the motor’s control rate. It is cheap enough to fit one channel per finger, it reads DC (the steady torque-holding current matters as much as the transients), and it ties straight into the firmware that decides when a grasp is firm enough. The Hall clamp and the Rogowski coil are bench instruments you reach for to characterize the hand; the low-side shunt is the sensor you build into it.

   +V motor rail
        |
     [ MOTOR / ACTUATOR ]
        |
     [ DRIVER FET ]
        |
        +------+-----> to diff-amp (+)
        |      |
      [ R_sh ] |   V_sh = I_motor * R_sh   (a few mV)
        |      |
        +------+-----> to diff-amp (-)
        |
       GND          low-side: common-mode ~ 0 V, sees actuator current

?A 100 mΩ low-side shunt sits in a finger actuator's return leg. The differential amplifier across it reads 45 mV. What is the actuator current, and why is the shunt in the return leg?

• A 450 mA; the return leg keeps the amplifier's common-mode voltage near ground, which is easier to handle
• B 4.5 A; the return leg gives galvanic isolation from the motor rail
• C 0.45 mA; low-side placement lets the shunt see leakage current that high-side misses
• D 45 A; the return leg makes the shunt read AC only, which is what a motor needs

?On a motor-drive power measurement the analyzer shows a sensible apparent power but a wildly wrong real power and power factor. The most likely cause is:

• A The shunt resistance has drifted with temperature
• B Channel skew between the voltage and current probes was not de-skewed
• C The current probe is a Rogowski coil and cannot read DC
• D The power factor is genuinely near unity and the analyzer is broken

Lab: build a per-actuator current sense and read its power

On the bench, sense one actuator’s current with a low-side shunt and a diff amp and sanity-check it against a clamp. First pick the shunt from a burden-voltage budget: decide the most millivolts you can lose at full motor current, divide by that current, and choose the nearest standard $R_{sh}$. Wire it in the return leg, put the diff amp across it (mind the common-mode just as in lesson 26), and scale the output to amps. Cross-check by clamping a Hall probe around the same motor lead, remembering to clamp a single conductor so the out-and-back currents do not cancel. Then move to power: feed the rail voltage and the shunt current into a power analyzer, and de-skew the channels first. Watch how real power and power factor settle into sensible numbers only after the skew is removed, while apparent power barely moves. Log real power, apparent power, and power factor for an idle finger, a moving finger, and a stalled finger, and you have a power fingerprint of the actuator.

↘ Why the Rogowski coil is deaf to DC, and what 'self-integrating' buys you

The Rogowski coil’s blindness to steady current is not a flaw to engineer away; it is baked into Faraday’s law. The coil’s output is $v(t) = -\frac{A N \mu_0}{l}\,\frac{dI}{dt}$, so a constant $I$ has $\frac{dI}{dt} = 0$ and produces exactly zero volts. To get current back you integrate, and the recovered signal is

$$V_\text{out} = \int v\,dt = -\frac{A N \mu_0}{l}\,I(t) + C_\text{integration}.$$

Real instruments use a lossy integrator whose time constant is much smaller than the period of the lowest frequency of interest. The loss bleeds away offset voltages and forces the integration constant to zero, which keeps the output from wandering, but it also sets a low-frequency floor below which the coil simply stops responding. Practical Rogowski systems can reach down to around 1 Hz, yet never to true DC.

There is an elegant special case. If the coil’s $L/R$ time constant is much longer than the current pulse you are measuring, the coil becomes self-integrating: shorting its ends (or closing them through a small resistor) makes the coil’s own current, and the voltage across that resistor, proportional to the measured current without any external integrator at all. This is how some very-fast pulsed-power diagnostics work, exploiting the coil’s inductance as the integrator. The coil’s inductance itself follows $L = \mu_0 N^2\!\left(R - \sqrt{R^2 - r^2}\right)$ for a toroid of major radius $R$ and minor radius $r$, and at high frequencies that inductance starts to roll the output off, which sets the coil’s upper bandwidth. The same coreless geometry that makes it linear and fast against huge currents is what denies it DC: no core means no way to hold a steady flux, and a steady flux is exactly what a constant current would need to register.

The authoritative facts pin the division of labor cleanly: a Hall probe handles AC+DC without breaking the circuit, a shunt plus diff amp turns current into a small voltage you read with a DIFF probe, and the Rogowski coil owns high $di/dt$ but surrenders DC. The grounding material adds the welding, arc-furnace, and lightning-research uses that make the coil’s nanosecond response and saturation-proof linearity worth its integrator.

Grounded in Wikipedia: “Current clamp”, “Rogowski coil”, “Shunt (electrical)” (CC BY-SA).

Key takeaways

• ▸Current is always inferred from a voltage: a shunt drop, a Hall voltage, or an induced coil voltage. The art is picking the translation.
• ▸A shunt plus differential amplifier is the cheap, accurate workhorse; size it from a burden-voltage budget and watch its thermal derating.
• ▸A Hall probe reads AC+DC without breaking the wire; a Rogowski coil handles brutal di/dt but is deaf to DC because it senses the rate of change.
• ▸A power analyzer reports real vs apparent power, and power factor is their ratio.
• ▸De-skew the voltage and current channels before trusting any power reading: skew leaves apparent power alone but corrupts real power and power factor.
• ▸On the robot hand, per-actuator current is a KPI, sensed with a built-in low-side shunt plus diff amp, one per finger.

Practice 1 warm-up

You have a 50 mΩ shunt in a motor’s return leg and your differential amplifier reads 30 mV across it. What current is the motor drawing? If the same shunt must read up to 3 A at full scale, what burden voltage will it develop there, and is that a comfortable signal level for an amplifier?

↘ Show worked solution

Current now: $I = V_{sh}/R_{sh} = 30\ \text{mV} / 50\ \text{m}\Omega = 0.6$ A.

At full scale: $V_{sh} = I R_{sh} = 3\ \text{A} \times 50\ \text{m}\Omega = 150$ mV. That is a healthy, easy-to-read signal (well above amplifier offset and noise, and only 150 mV of burden lost from the rail). The shunt also dissipates $P = I^2 R_{sh} = 3^2 \times 0.05 = 0.45$ W at full scale, so it needs to be rated for that power.

Practice 2 core

For each task, pick the best of (shunt + diff amp, Hall current probe, Rogowski coil) and say why in one line: (a) reading a battery’s steady 8 mA standby draw to verify a sleep spec, (b) clamping onto an unbroken motor phase wire to catch the current spike during a hard switching transient, (c) building a permanent per-finger current sense onto a robot hand control board.

↘ Show worked solution

(a) Shunt + diff amp (or a sensitive Hall probe). The draw is steady DC, so the Rogowski coil is out (no DC). A small shunt with a good diff amp resolves milliamps cheaply and accurately; a Hall probe also works if you cannot break the wire.

(b) Rogowski coil. You cannot break the wire, and a fast switching spike is high $di/dt$ with violent peaks. The coreless coil follows nanosecond changes and will not saturate on the surge, which is precisely its niche.

(c) Shunt + diff amp, low-side. It is cheap enough for one channel per finger, reads DC (steady holding torque matters), is built into the board rather than clamped on, and keeps common-mode near ground in the return leg.

Practice 3 stretch

A power analyzer measures a switching converter. With the current channel skewed by 200 ns relative to the voltage channel, it reports apparent power $S = 12.0$ VA and real power $P = 9.0$ W, giving a power factor of 0.75. After you de-skew the channels, $S$ is unchanged at 12.0 VA but $P$ rises to 10.8 W. Explain qualitatively why $S$ did not move while $P$ did, and compute the corrected power factor and the percentage error the skew introduced into the real-power reading.

↘ Show worked solution

Apparent power $S = V_{rms} I_{rms}$ depends only on the magnitudes of the two waveforms, not on their relative timing. Sliding one channel a fixed delay does not change either RMS value, so $S$ stays at 12.0 VA. Real power $P = \frac{1}{T}\int v(t)\,i(t)\,dt$ depends on the alignment of $v$ and $i$ instant by instant. A 200 ns skew multiplies mismatched moments together, which (here) understates the true overlap and drags $P$ down.

Corrected power factor: $\text{PF} = P/S = 10.8 / 12.0 = 0.90$.

Skew-induced error in real power: the skewed reading was 9.0 W versus the true 10.8 W, an error of $(9.0 - 10.8)/10.8 = -1.8/10.8 \approx -0.167$, about a 17% under-read. The lesson: apparent power looked perfectly believable the whole time, which is exactly why an un-de-skewed measurement is dangerous.