I mentioned the CCD oscilloscope idea from Dercuano on ##electronics, and Stipa pointed out that the whole laser mirror thing was far more complex than needed. You can just use a VGA monitor!
In more detail, the problem to be solved is that of building an oscilloscope out of discarded junk, for ghettobotical purposes. With an oscilloscope, designing, debugging, and characterizing electronics becomes enormously easier; some old vacuum-tube oscilloscopes had a rolloff (-3dB point) of 10MHz, but the standard basic analog oscilloscope for decades was 20MHz. So expedients like wiring your signal source to your sound card through a resistor and some limiter diodes are grossly insufficient: your sound card is 20kHz, three orders of magnitude crappier.
Once you can bring the problem into the digital domain, everything else becomes easy, because you can do it as slowly as you like. But how can you digitize a 20MHz signal if you don’t have access to the market? 50Msps ADCs are quite scarce in the municipal waste stream.
Analog oscilloscopes achieved these bandwidths by using a cathode-ray tube; the beam’s Y deflection was what you observed on the screen. So you might think that you could salvage an old TV and use its CRT, but TV CRT beams are magnetically deflected, while oscilloscope CRT beams are electrostatically deflected, allowing deflection frequencies two or three orders of magnitude higher at reasonable voltages. TV CRTs are not going to be useful for that; they’re designed for horizontal scanning with a 15.734-kHz sawtooth wave (for NTSC; PAL and SECAM vary slightly.)
However, the beam intensity is modulated at higher frequencies, up to about 5 MHz in the case of NTSC, PAL, or SECAM TV; more promisingly still, tens of MHz in the case of computer monitors, whose VGA connectors still accept analog waveforms for red, green, and blue. A monitor running at 1600×1200 at 85 Hz, which was high-end in the 1990s but quite likely junk today if it’s a CRT, is drawing 163 million pixels a second, so it can “sample” signals up to 80 MHz or so. A lower-end 60Hz 1024×768 monitor is only drawing 47 million pixels per second, but that’s still enough for more than the 20 MHz for a basic oscilloscope. Pixel brightness is not linear in signal voltage, being transformed by the “gamma curve”, but it’s not outrageously nonlinear. And an RGB VGA monitor, like nearly all of them, will “sample” three channels at once, which is very respectable indeed.
You also need electronics to generate sync pulses, of course.
So how do you get the signal off the screen? With a camera, of course. Many-megapixel cameras are now common on cellphones, and they are fast enough, high enough resolution, and clean enough (with good signal-to-noise ratios) that you can use them capture individual pixels from an individual video frame off a monitor.
(My previous proposal used a laser diode and two spinning mirrors to scan the laser beam across a reflective screen — certainly doable with discarded materials, but substantially more difficult.)
Once you have that data, it’s a Simple Matter of Programming to find the individual scan lines, compensate for lighting and viewing-angle variability, undo the gamma-curve transformation, look for triggers in the signal stream, and draw a waveform with the resulting data.
There are some limitations of this method. All data is lost during the horizontal blanking interval and vertical blanking interval, unless you are driving three separate monitors out of phase to avoid this. (Two monitors is not enough to eliminate these dropouts because the VBI of each monitor will inevitably contain many HBIs of the other.) There are cases where this matters: where you’re watching for a single-shot event and you really need a lot of data on both sides of it, for example. Calibration may drift, since slight brightness changes don’t affect the normal use of monitors; occasionally displaying a calibration frame or two instead of the input signal may help with this.
Aside from its use as electronics test equipment, this approach should work well for software-defined radio, in which case it is probably possible to use analog-domain downconversion and frequency filtering to smear the signals of interest around in the time domain enough that the HBI and VBI are less problematic.