Seconds to Microseconds

Choosing a Timebase Before the Signal Arrives

Nobody sets up a capture in seconds. Front panels are marked in µs/div, acquisition drivers want a sample interval, and the pulse you are chasing is quoted as a width or a frequency. Yet the arithmetic that ties them together — window equals record length divided by sample rate — comes out in seconds every time. Converting to microseconds is the small step between the number the maths produces and the number you dial into the instrument.

Conversion factor: 1 s = 1 000 000 µs, so multiply seconds by 10⁶. A one-million-point record taken at 100 MSa/s covers 1 000 000 ÷ 100 000 000 = 0.01 s of signal, which is 10 000 µs — 1 000 µs/div across ten divisions.

The Four Numbers That Decide a Capture

Ten divisions, one window

Almost every graticule is ten divisions wide, so the whole acquisition window is simply the time/div setting times ten. Read 50 µs/div and you are looking at 500 µs of signal, no more.

The sample interval is just the reciprocal

One divided by the sample rate gives the spacing between points: 1 GSa/s puts them 0.001 µs apart, 1 MSa/s exactly 1 µs apart. Everything the instrument knows about the signal happens at those instants.

Record length is the ceiling

Acquisition memory is finite, so rate and duration trade against each other. Ask for a long window at a high rate and the instrument quietly drops the rate rather than refusing the request.

Nyquist is a floor, not a target

Two samples per cycle is the theoretical minimum for recovering a frequency. For a waveform whose shape and edges you need to trust, practice asks for several times that.

Working From the Waveform Back to the Front Panel

Start from what you know about the signal — a frequency, a pulse width, a burst length — and work outwards to the settings that will actually catch it.

Enter the window you calculated

Divide record length by sample rate, then type that seconds value into the left field — 0.002, 0.0001, whatever the division gave you. The microsecond equivalent appears immediately, and spaces or a comma in the number are handled without complaint.

Divide by ten for the timebase knob

A 2 000 µs window is 200 µs/div. Round to the nearest step the instrument offers — 1-2-5 sequences are the norm — and let the window grow slightly rather than clipping the event you came for.

Flip the direction when the datasheet is already in µs

Pulse widths, propagation limits and conversion times arrive in microseconds. The swap button (↔) runs µs → s so the figure can go straight into a driver call that expects seconds.

Paste the digits into the acquisition script

Copy hands over the number with no unit attached, which is what a SCPI string or a configuration dictionary needs. Ctrl + C from inside the field works the same way.

Very small intervals change format: the sample interval of a 5 GSa/s acquisition is 0.0002 µs, which prints plainly — but ask for it in seconds and 0.000 000 000 2 drops below the threshold where the display switches to scientific notation, so it appears as 2.000000e-10. Same value, different dress.

Sample Rate, Point Spacing and the Window a Million Points Buys

Sample interval is one divided by the rate. The window columns fix the record length at one million points and divide, which is the calculation that decides whether an event fits on screen at all.

Sample rate	Sample interval (s)	Sample interval (µs)	Window at 1 Mpt (s)	Window at 1 Mpt (µs)
5 GSa/s	0.0000000002 s	0.0002 µs	0.0002 s	200 µs
1 GSa/s	0.000000001 s	0.001 µs	0.001 s	1 000 µs
500 MSa/s	0.000000002 s	0.002 µs	0.002 s	2 000 µs
100 MSa/s	0.00000001 s	0.01 µs	0.01 s	10 000 µs
10 MSa/s	0.0000001 s	0.1 µs	0.1 s	100 000 µs
1 MSa/s	0.000001 s	1 µs	1 s	1 000 000 µs
100 kSa/s	0.00001 s	10 µs	10 s	10 000 000 µs

Notice how brutally the top row is punished. Five billion samples a second sounds like the safe choice until you see that a million points of memory buys only 200 µs of history — not enough to hold a mains cycle, let alone a start-up transient. That trade is the reason deep memory costs money.

How the Page Fits a Capture Setup

Try three windows in one sitting

Both fields accept typing and update each other continuously, so you can compare a 200 µs, 2 000 µs and 10 000 µs capture without clearing anything between attempts.

Datasheet figures go the other direction

Swapping the pair sends microseconds back to seconds, which is the direction a component specification takes when it has to become an argument in an automation script.

Nanoseconds sit in the same dropdown

Both unit lists are searchable and cover the whole time range, so a point spacing that has slipped below a microsecond can be read in nanoseconds without leaving the page.

Clean numbers for SCPI strings

Copying gives the bare digits with no unit and no separators, ready to drop into a horizontal-scale command or a driver parameter that would reject anything else.

Questions From the Test Bench

What is the sample interval at 1 GSa/s, and how many points is that per microsecond?

One divided by a billion is 0.000 000 001 s, or 0.001 µs — so exactly a thousand points land in every microsecond of the record. Watch what the instrument does when you enable more channels, though: many mid-range models interleave their converters, so a headline 1 GSa/s can halve to 500 MSa/s the moment a second channel is switched on, doubling the spacing to 0.002 µs.

How many samples does a 10 ms capture window hold?

Multiply the window in seconds by the rate. Ten milliseconds is 0.01 s, or 10 000 µs, so at 100 MSa/s it needs one million points, at 1 GSa/s ten million, and at 10 MSa/s only a hundred thousand. If the memory available is smaller than that product, the acquisition will not run at the rate you asked for — the instrument keeps the window and lowers the rate.

Which time/div setting shows several cycles of a 5 kHz waveform?

One cycle at 5 kHz lasts 0.0002 s, which is 200 µs. At 100 µs/div the ten-division window spans 1 000 µs and shows five clean cycles — usually the sweet spot for judging shape and amplitude together. Drop to 50 µs/div and you get two and a half cycles with more edge detail; go out to 500 µs/div and twenty-five cycles pack the screen, better for spotting modulation or a dropout than for measuring anything.

Why does record length limit the window when the sample rate is high?

Because duration is memory divided by rate, and only one of those three is free. A million points at 5 GSa/s is 200 µs of signal; the same million points at 10 MSa/s stretches to 100 000 µs. Turn the timebase out past what memory supports and the instrument keeps the requested window by sampling more slowly, which is fine until the reduced rate starts under-sampling the edges you were trying to see. Segmented acquisition is the usual escape: capture only the intervals around each trigger and skip the dead time in between.

What does an under-sampled signal actually look like on screen?

Convincing, and wrong — that is what makes aliasing dangerous. A 6 kHz tone digitised at 10 kSa/s, a spacing of 100 µs, folds down and displays as a stable 4 kHz waveform with no hint that anything is amiss. Milder under-sampling shows up as understated peaks and edges that look slower than they are, because no sample happened to land on the real extreme. The cures are the same in both cases: raise the rate until several samples fall on the fastest feature, or band-limit the input before it reaches the converter.