Power electronics and energy
Voltage and current on one time base: the channel skew nobody warns you about
On a multiplexed front end the converter visits voltage, then current, then back again. Whether the microseconds in between matter depends entirely on the frequency content of what you are measuring — and that is a question you can answer before you buy anything.
Channel skew and when it corrupts a power measurement, choosing between shunt, Hall, CT and Rogowski, why DC offsets integrate into phantom capacity, and what sweep rate does to a PV curve.
Almost everything interesting in power electronics reduces to the same measurement: a voltage and a current that have to be compared with each other. Efficiency, power factor, switching loss, charge and discharge curves — each one is a statement about the product of two signals, which means each one depends on those two signals having been captured at the same instant.
Where the skew comes from, and how big it is
Every 62xx and 6xxx module on this site uses one converter behind a multiplexer. Conversions happen one at a time, so two channels that appear in the same scan are separated by the reciprocal of the total conversion rate. Add channels and each channel is revisited less often — but the gap between neighbours inside a scan does not change.
| Total rate | Adjacent-channel gap | Phase error at 50 Hz | Phase error at 10 kHz | Phase error at 100 kHz |
|---|---|---|---|---|
| 400 kS/s | 2.5 µs | 0.045° | 9° | 90° |
| 250 kS/s | 4 µs | 0.072° | 14.4° | 144° |
| 100 kS/s | 10 µs | 0.18° | 36° | — |
| 1.25 MS/s | 0.8 µs | 0.014° | 2.9° | 28.8° |
Swipe the table sideways to see all columns
The 100 kHz column is deliberately included to show where the argument breaks down. Once the skew approaches a tenth of a period, the voltage and current records are describing different parts of the same waveform, and no post-processing recovers the pairing. Above roughly 100 kHz of signal content — which is to say, the edge of any modern switching device — a multiplexed card is not the right instrument and the answer is a scope with a differential probe.
When simultaneous sampling stops being optional
The USB-4431 and USB-4432 carry a converter per channel, so every input is sampled at once at its full rate — no skew, by construction. The USB-4432 is the more useful of the two for power work, because its ±40 V range on every channel and its per-channel choice of AC or DC coupling make it suitable for a DC bus and an AC ripple measurement on the same instrument.
That is the honest boundary of the range. Below roughly 10 kHz of signal content, a multiplexed card with the arithmetic in the table above is the right and cheaper answer. Above it, simultaneous sampling is not a refinement — it is the difference between a measurement and a coincidence.
Current sensing has no single right answer
Nobody makes a sensor that is simultaneously accurate at DC, fast to the megahertz, non-invasive and cheap. Every current sensor on the market is a different compromise, and choosing wrongly tends to produce a smooth, plausible curve that is simply wrong.
| Sensor | DC | Bandwidth | Isolated | The catch |
|---|---|---|---|---|
| Shunt resistor | Yes | Very high | No | Its own inductance rounds off fast edges, and it sits in the power loop; needs Kelvin connection and a differential input |
| Hall effect | Yes | Moderate | Usually | Offset drifts with temperature, and there is a residual offset even at zero current |
| Current transformer | No | Moderate | Yes | AC only, and the low-frequency response depends on burden and core |
| Rogowski coil | No | High | Yes | AC only and needs an integrator; the light weight and lack of core saturation that make it good for fast pulses are exactly why it reads nothing at DC |
Swipe the table sideways to see all columns
The classic version of this mistake is measuring a battery discharge curve with a Rogowski coil. The instrument is working perfectly. It is reading the true AC component of the current, which is essentially zero, while the DC component that constitutes the entire discharge curve never reaches the integrator. The result is a flat line that gets blamed on the cell.
Why a DC offset turns into phantom capacity
Charge is current integrated over time, so any constant error in the current measurement accumulates without limit. This is the one place where a small offset is not small.
Work it through with a 1 mΩ shunt. A 100 µV of residual offset on that shunt reads as 100 mA of current that is not flowing. Left running for ten hours, that integrates to one amp-hour of capacity that was never in the cell. On a 50 Ah pack that is 2% of the nameplate number, produced entirely by the measurement chain — and it looks exactly like self-discharge or a weak cell.
- Use a Kelvin, four-wire connection to the shunt so the sense pair carries no load current and the measured voltage is the shunt's and not the wiring's.
- Zero the channel with no current flowing, and repeat it — a drifting zero is the failure mode, not the absolute value.
- Prefer the narrowest input range that fits the shunt's full-scale drop. A 50 mV full-scale signal on a ±10 V range is throwing away two orders of magnitude of resolution.
- Log the offset channel alongside the data. An offset that is visible is an offset you can subtract with confidence.
PV I-V curves: the sweep is part of the measurement
Tracing an I-V curve looks like the easiest task in this article and is not. A module or string has capacitance, and that capacitance has to charge and discharge as the operating point moves. Sweep too quickly and the measured current includes the capacitor's current rather than only the cell's. Sweep slowly and irradiance and cell temperature drift underneath you, so the curve you record is a curve no single operating condition ever produced.
Sweep direction matters too. Going from open circuit down to short circuit and going the other way do not give the same answer on a high-capacitance module — one direction tends to overestimate the maximum power point, the other to underestimate it, and the truth sits between them. Common practice is to record both directions and treat the discrepancy as your uncertainty rather than picking the flattering one.
Harmonics: measurable, but check what is actually limited
Harmonic current emission is a solved measurement problem with a well-defined instrument standard behind it — IEC 61000-4-7 specifies how the spectral components are grouped and measured, and the limits themselves are in IEC 61000-3-2 up to 16 A per phase and IEC 61000-3-12 from 16 A to 75 A. The bandwidth this needs is modest. The 40th harmonic of 50 Hz is 2 kHz, and even a 16-channel multiplexed card running at 15 kS/s per channel resolves it with an order of magnitude to spare.
The thing worth knowing is what the limits do not cover. Harmonic limits stop at the 40th order. Between roughly 2.5 kHz and 150 kHz — the supraharmonic band, which is exactly where a modern switching converter puts a great deal of its energy — there is no harmonised emission limit. An inverter can measure clean against IEC 61000-3-2 and still emit substantially at its switching frequency and its multiples. If that matters to your application, it has to be measured deliberately; it will not appear in a compliance report as a pass or a fail, because there is nothing to pass.
Picking a part for a power bench
| The job | Part | Why |
|---|---|---|
| Line-frequency efficiency, power factor, energy metering | USB-6212 | 16 inputs at 400 kS/s and 32 digital lines is more than line-frequency work needs, and the arithmetic above shows the skew is negligible there |
| Driving a programmable load or a gate while measuring | USB-6229 | Four 16-bit outputs updating simultaneously at 833 kS/s — the simultaneous part is what makes it usable as a multi-phase drive |
| Multi-string PV monitoring, or many current and voltage channels at once | USB-6225 | 80 analog inputs on one module, which keeps the wiring centralised and the channel cost low |
| Faster switching content, up to the low hundreds of kilohertz | USB-6259 | 32 inputs at 1.25 MS/s with 48 digital lines and four fast analog outputs |
| Anything where the voltage-current phase relationship is the measurement | USB-4432 | A converter per channel, ±40 V with per-channel AC or DC coupling — no skew, and a DC bus and its ripple on one instrument |
Swipe the table sideways to see all columns
Errors that survive review because the curve looks fine
- Measuring a battery discharge with an AC-only current sensor, and diagnosing the cell from a flat line.
- Comparing a voltage channel and a current channel on a multiplexed card at switching frequency without accounting for the skew between them.
- Ignoring shunt insertion inductance, then reading the rounded edge as the device's real switching behaviour.
- Leaving a coulomb-counting channel on a wide input range, so that a small offset becomes a large capacity error overnight.
- Treating an IEC 61000-3-2 pass as evidence that the converter is electrically quiet.
Every figure in this note is either arithmetic from published resolution and sample-rate specifications, or a value stated by the standard it cites. Where a number depends on a particular module, confirm it against that module datasheet before design freeze.