I’ve been trying to figure out how to build an oscilloscope from e-waste (see files TV oscilloscope, VCR oscilloscope, Laser printer oscilloscope, and Disk oscilloscope). The difficulty is that a basic oscilloscope has frequency response up to 20MHz, which means that if you want to digitize the signal you need a sampling rate of at least 40 megasamples per second, and such fast analog-to-digital converters (ADCs) are hard to come by. To display it without digitizing it, you would need some kind of display device with response up to 20MHz, like an analog oscilloscope tube, which is challenging to construct and very rare to find in garbage streams, especially intact.
I finally have an answer that I am confident will work: use a laser to store the analog signal on a CCD long enough to digitize it with an easily-available low-sample-rate ADC. Discarded flatbed scanners contain linear CCDs, which are already connected to moderate-speed ADCs; discarded laser printers contain lasers connected to scanning mirrors. Point the laser at the CCD and modulate the laser brightness with the signal you want to measure as it passes across. Then you can digitize the signal from the CCD at a lower speed in the usual way.
Let’s consider a garden-variety 22ppm A4-size landscape-mode 600×600dpi printer laser assembly. A4 paper is 210 mm × 297 mm, so the paper moves at minimally 22x210 mm = 4.62 meters per minute, or 77 mm/s. 600 dpi is 23.6 “dots” per mm, so the printer must scan at a bit over 1800 scan lines per second; if it uses a 6-sided spinning mirror, that’s 300 revolutions per second, or 18000 rpm, which is a pretty high speed.
(And it seems like a pretty average speed. The Samsung M2020W costs AR$2139 = US$134 and is 20ppm with fake 1200dpi in portrait mode; the HP Wifi Pro M12 costs AR$2900 = US$181 and is 18ppm A4 with 600dpi in portrait mode; a Xerox Phaser 3020 costs AR$1900 = US$119 and is 20ppm A4 with 600dpi, but I can’t tell if that’s landscape or portrait.)
For a laser printer, the laser doesn’t need to be linear (it’s either on or off for each pixel) but it does to be modulated at a pretty high speed. Each of those 1800 scan lines is, hypothetically, 297 mm long, with 7000 pixels per scan line. (In portrait mode you need more scan lines, but they are shorter, for the same number of pixels per page and per second.) This means the laser needs to be turned on and off at a bit over 12 MHz to meet the 600 dpi spec, but possibly 12MHz is already past the low-pass 3dB point of the electronics hooked up to it. However, laser diodes that can be modulated at many tens of MHz are easy to come by, and can be tested using a garden-variety photodiode and existing known-good oscilloscope.
A 6-sided mirror sweeps the laser across some 120 degrees in each scan, which are linearized by some weird one-dimensional optics in the laser assembly. They are linearized over a much longer region than the CCD from a scanner, which is typically as small as possible to cut down on silicon wafer costs, but which has its own set of optics to focus an image from typically the width of A4 or A3 paper onto it. Although it limits bandwidth, these two complementary sets of optics could be left in place with a sheet of paper between them as a diffuser, but maybe you could get better performance (lower “dark current” due to stray light) if you could eliminate them and just scan the reflected laser directly across the CCD, maybe using a mirror and a second laser at a different angle to cut the sweep angle down to 30°, and compensate for the temporal nonlinearity across the CCD in software. This allows you to scan the laser across the CCD several times (such as four, in this example), which keeps the CCD’s spatial resolution from being the limiting factor on the bandwidth.
The number of points you can sample from the captured analog signal on the CCD presumably depends on the resolution of the original scanner, but even a shitty scanner is 300 dpi, which across a 210 mm A4 page gives you 2480 pixels. A normal scanner is four times that; a high-end scanner is eight times that. 2480 samples is several times what you need for a decent DSO.
I think there may be some “dead time” during which it is impossible to capture the signal, either because the scan angle is too large for the CCD or because the laser is hitting the corner of the mirror or because the CCD is going through some kind of reset process between frames or something. There is definitely “dead time” on the laser end of the process, at least in the printers I've disassembled, where the laser is directed elsewhere than through the optics onto the page for part of the scan, thus getting lost.
Any other kind of camera would also work for converting from spatially modulated light (integrated over some frame time) back into a signal temporally modulated slowly enough to be easy to digitize. It’s just that the CCD from a scanner is high-quality (typically more than 8 bits per pixel) and has a lot of pixels already in a line.
If no lasers are available, a possible alternative is to use an LED viewed through a spinning mirror and focusing optics, i.e. take a picture of the spinning mirror in which the LED is reflected with a camera. Many LEDs have relatively low junction capacitance and consequently can respond far more rapidly than they really need to for their intended use.
If no mirror is available either, it may be possible to do the job by spinning the camera (relative to the LED) instead, although this involves potentially tricky electrical connections to the rapidly spinning camera or LED. To avoid that problem too, you could nutate the camera or LED rather than rotating it, although that seems likely to be mechanically tricky and will also require trickier time base correction.