A 48 kS/s module reading eight channels is really eight channels at 6 kS/s

Why aggregate sample rate, input range and ADC resolution decide what a teaching lab can actually measure — with the arithmetic for per-channel rate, LSB size, aliasing and floating sources.

A teaching lab buys thirty identical modules, which means the wiring diagram has to survive being redrawn by a first-year student and the specification has to survive being read quickly. Both of those push in the same direction: fewer things that can silently go wrong. This article is the short list of things that do.

The sample rate is shared, not per channel

Every low-cost and multifunction module on this site has one analog-to-digital converter behind a multiplexer. The converter is fast; the multiplexer takes turns. So the headline rate is the total number of conversions per second across all enabled channels, and the rate each channel actually sees is that number divided by however many channels are switched on.

Swipe the table sideways to see all columns

Nyquist says you need more than two samples per cycle of the fastest component you care about. In practice, sampling at five to ten times the highest frequency is what makes a waveform look like a waveform rather than a set of disconnected points. Eight channels at 6 kS/s therefore gives a genuinely useful bandwidth of roughly 600 Hz to 1.2 kHz — fine for temperature, load cells, photodiodes, RC transients and audio-band signals, and not fine for anything switching fast.

Why a floating source reads garbage single-ended

A thermocouple, a battery, a function generator output and a photodiode are all floating sources: neither terminal is connected to earth. A single-ended input measures one wire against the module ground, which means the floating source has no defined reference and will drift, rail, or sit at whatever the leakage currents decide. The fix is either to give it a reference or to stop measuring it against ground.

• Differential input — the module measures the difference between two wires, so the source never needs a ground reference. Every analog module in the range offers a differential mode; on the eight-channel parts that pairs the inputs up, giving four differential channels.
• Bias resistors — two equal resistors from each input to ground, typically 10 kΩ to 100 kΩ, give a floating source a defined DC level while loading it negligibly.
• Twisted pair — run the signal and its return as a twisted pair so the loop area stays small and magnetic pickup cancels in the pair.

The USB-6008 and USB-6009 add a second reason to prefer differential mode: their widest ranges, ±20 V and ±4 V, exist only when inputs are paired differentially. In single-ended mode the largest range is ±10 V. On a student bench, that extra headroom is worth having — it is the difference between a mis-wired 12 V supply being a teaching moment and being a dead channel.

Resolution is not accuracy

Students reliably quote the least significant bit as if it were the error bar. It is not. The LSB is just how finely the converter divides the range; the actual error comes from the voltage reference, the amplifier gain and offset, and how all of those drift with temperature. They are usually one to two orders of magnitude apart.

Swipe the table sideways to see all columns

The ±0.2 V row is the one worth teaching explicitly. The multifunction tier offers ±10 V, ±5 V, ±1 V and ±0.2 V, and choosing the narrowest range that still contains your signal is the cheapest gain stage available — it costs nothing and buys 50 times the resolution of the ±10 V setting. It is also the point at which students discover that a small signal on a wide range is mostly measuring noise.

Aliasing: the demonstration worth doing every year

Aliasing is the one measurement error that produces a completely convincing wrong answer. A signal above half the sample rate does not disappear — it folds down and appears as a lower frequency that was never there. The demonstration takes ten minutes and it sticks: sample a clean sine at a deliberately low rate and watch the display show a slow, stable, entirely fictitious waveform.

1. Feed a 1 kHz sine into one channel. Sample at 100 kS/s and confirm the waveform looks correct.
2. Drop the sample rate to 10 kS/s and note that the waveform is still recognisable, with five samples per cycle.
3. Drop it to 6 kS/s — just above Nyquist — and watch the amplitude and shape start to lie.
4. Drop it to 1.1 kS/s and the display shows a stable sine of about 100 Hz instead. Nothing in the display reveals that this is wrong.

The lesson that follows is not "always sample fast". It is that the anti-alias filter, not the sample rate, is what makes a digitised spectrum trustworthy — which is why the IEPE instruments in the high-performance tier track their anti-alias filter to the sample rate automatically, and why the general-purpose tiers rely on you to keep the input bandwidth below Nyquist by other means.

Ground loops, and the two grounds on every student bench

A laptop on a USB cable and a bench supply on a mains lead share an earth through two different paths. Any current flowing in that loop appears as a voltage in series with the signal. At DC it is a constant offset; at 50 Hz it is hum; at switching frequencies it is spikes. The symptom students report is "noise", so the first diagnostic question is whether the noise survives disconnecting the bench supply from the circuit.

• Use one ground reference point and star everything back to it, rather than letting the breadboard ground rail wander.
• Prefer differential inputs whenever the source can float — it removes the ground loop from the measurement instead of trying to filter it out.
• Keep signal wires away from the supply leads and from anything with a switching converter in it.

Pulses, speed and frequency belong on the counter

Anything that arrives as a train of pulses — a slotted wheel, a tachometer, a reed switch, a flow meter — is much better counted than sampled. Every entry-level analog module carries one 32-bit counter, and the USB-6501 carries one alongside its 24 digital lines. A counter counts edges in hardware, so it does not consume sample rate, does not alias, and does not care that the pulse train is far faster than any waveform you could digitise.

This is also where the teaching sequence has a natural fault line. A module with a counter and 24 digital lines and no analog input at all — the USB-6501 — is the right purchase for a digital-logic or microprocessor course, and the wrong one for an instrumentation course. Buying a mixed bench of both types across a lab is usually cheaper than buying one type that does neither well.

What this looks like across a lab

Swipe the table sideways to see all columns

The mistakes worth putting on the wall

• Quoting the card's total sample rate as if each channel got it. Divide by the number of channels you enabled.
• Reading a floating source single-ended and calling the result noise. Use differential or add bias resistors.
• Quoting the LSB as the accuracy. Check the absolute accuracy specification instead.
• Leaving the range at ±10 V because it always works. The narrowest range that fits is free resolution.
• Trusting a waveform without asking what the sample rate was. Aliased data looks completely plausible.
• Counting pulses by sampling them. Use the counter.

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.