I have
Of course its always a hassle to build things from basic components..but if its not a critical system, and the original poster didn't know much about ADC's, its a great way to learn. Theres many 'leaps of the imagination' there, but hey, might be of some use. Something that may be easier than even an SAR, but still interesting (no offence to the comparators idea) would be the brute force approach....a simple 4-bit binary counter, and a comparator.
Assuming you can't get a 4-bit ADC (I've never seen one tbh) building one isn't that daunting...
For 4 bits, it'll only take 16 clks, and with a CMOS/TLL counter we're talking MHz. Of course, for high resolutions the counting becomes a lot longer, but 16 clks (4-bit) isn't that bad. With your analogue input (0-5V?) into the comparator, your counter's 4 outputs to a 4 bit R-2R ladder which goes into the other comparator input (I'm sure many know how an SAR ADC works). The comparator output latches a flipflop which stops the counter, and also acts as a 'done' line, and you just read the counters outputs for your digital value. This FF can be reset to start another conversion.
Actually, I'm gonna try this as a quick experiment, can't say its 'fun', but its definately interesting, seeing as the input range (assuming we use the LM393 and of course an opamp buffer after the R2R) can be pretty big, and completely variable. So theres a 'Vref'. Also with the speed it could run at on its own clock, 1Mhz sampling frequency shouldn't be a problem. Well, a lot of hassle, but a stripboard, 2"x1" 4-bit ADC might come in handy one day (even though I'm surrounded by PIC's with ADC's coming out their ears
)
just a few thoughts,
Blueteeth
ps. super easy option? Use a standard 8-bit ADC, and only use the 4MSB's, discard the lower nibble. Not ideal, but you'll have 16 'levels' of voltage to measure. (you'll need to correct the input voltage level).