Before a self-calibration the scope asks you to disconnect every channel. It is tempting to think that a probe clipped to a circuit that is switched off is good enough: there is no signal. We nearly did exactly that, and the scope showed why it is not.
CH1's probe was still clipped to a pin of an XBee radio whose board was not powered. Nothing was sending. Yet CH1 showed about 0.47 V peak to peak at about 61 Hz. That is mains hum: the 60 Hz of the building's wiring, picked up by an input that nothing holds at a fixed voltage. At 100 mV/div it is almost five divisions tall.
A pin of a circuit that is switched off is floating: nothing drives it, so it has a very high impedance (it resists current strongly, so a tiny current makes a large voltage). The scope's input is high impedance too: 1 MΩ, or 10 MΩ through a 10× probe. Electric fields from the mains wiring and from your body push a few nanoamps through such a node, and that is enough for hundreds of millivolts. The probe's ground lead adds its own pickup: tip, lead and circuit make a loop, and a changing magnetic field drives a voltage around any loop. The full explanation, with numbers.
A powered circuit holds its pins at a low impedance, and the hum shrinks to almost nothing. A probe unplugged from the scope is no longer part of the input at all. A probe clipped to a dead circuit is the worst of both: a long antenna on a sensitive input.
Each input has small errors of its own: an offset (it reads a little above or below 0 V with nothing on it) and a gain error (it reads a little more or less than the true voltage). Both change with temperature and differ at each vertical scale. A self-calibration measures them with the inputs empty and stores corrections that the scope applies from then on. On the DHO800 and DHO900, NovaOS measures each channel's zero and gain at every scale; on the MHO900 it also measures the correction for its two interleaved converters (two converters sampling in turn, which must match).
The measurement assumes the input is empty. Anything on it, hum included, is measured as if it were the input's own error and stored as a correction. 0.47 V of hum is five divisions at 100 mV/div and dozens at the most sensitive scales: a calibration taken with it would be worse than none.
Rigol's own firmware only tells you to disconnect the channels. NovaOS also looks: while the question is open it reads each input's live trace, and names any that shows a signal: "Remove the probe from CH1." Press Start and it waits ("Waiting…"; "It starts by itself once every input is free."), and once you unplug the probe it starts on its own. If a signal appears while it runs, it stops at once, keeps nothing it measured (the calibration in force stays), says "Self-calibration waiting" and starts again by itself when the input is free. After ten minutes of waiting it gives up and names the channel.
Continue anyway is for an input that only looks connected: one that has never been calibrated can show a small offset or noise of its own. Use it only when nothing at all is plugged into that input; the scope then calibrates it without judging it. While it waits, Calibrate the others leaves a channel out, for a bench where one input must stay wired.
On an MHO900 with no calibration stored, NovaOS offers it once the scope has run 30 minutes: "No self-calibration stored for CH1-CH4. The scope has warmed up: run it now?" with Start and Not now. NovaOS keeps its corrections in its own storage: the factory calibration Rigol stored on the scope is never changed.
Yes: unplug them from the scope's input connectors. A probe left clipped to a switched-off circuit picks up mains hum, and the calibration would store it as an error of the input.
Almost certainly not. A floating probe picks up 50 or 60 Hz mains hum. Measure its frequency: 50 or 60 Hz (or close) means hum. Why it happens.
On NovaOS about a minute on a DHO800 or DHO900, and about three on an MHO900. Other scopes differ; their manual gives the time.
On NovaOS, no: it stores its own corrections and runs them on top of the factory calibration, which it never writes. Other firmware may differ; check the manual.
When the scope is warm and the room temperature has changed a lot since the last one, after a firmware update that asks for it, or when an empty input no longer reads zero. Not every day.
Mains hum on a floating input: why a probe on nothing shows a wave, why a 10× probe shows more of it, and how to tell hum from a real signal.
Why the scope waits for a triggerAuto, Normal and Single; slope and level; and why a level that looks right still catches nothing on an idle line.