Shepard tones (also called Risset tones, after J-C. Risset, the INA/GRM composer who used this quite a bit) are sort of weird. It sounds like the tone is continuously going up or down, but that's not what's up at all. The spectrograph above gives a few clues, tho...
You can see a continuous pattern of up-swept pitches that overlap. But you'll notice that you can't actually see the origin of the pitches, and that the amplitude reduces the higher the pitch goes. That's part of the trickery. The other part is purely psychoacoustic, and relies on how we perceive sonic foregrounds via amplitude.
OK...let's build one of these things. Take several VCOs, for starters. Set each one to the same low pitch, and mix them all together. Send this to a bandpass VCF so we can limit the passband we're listening to...this eliminates the obvious "bottom limit" and "top limit" pitches.
Now, here's the voodoo part: each VCO also receives a modulation signal from...oh, let's say we'll use several AD EGs. Long linear attack, pretty much no decay. The attack has to keep going while the signal is within the VCF's passband, but it starts moving the VCO's pitch up before the signal enters the passband, and keeps going until it's gone past the top limit. So now, we've got several VCOs and several individual ADs sending them pitch-rise modulation. Then the key to this is a trigger sequencer. Each EG gets triggered in turn; if you have six VCO/EG combos, use six steps. As each EG receives a trigger pulse, it starts to raise the VCOs pitch through the passband, then out the top where the decay kicks the pitch back down. Technically, you could stop there, as that right there gives you the continuous sweep illusion. But if you want to really make this perfect, you then use the EG triggers to trigger a second EG and then a VCA on each VCO, so that the VCA is only open for the time needed for the sweep, eliminating the possibility of the bottom pitch drone from bleeding through or the very rapid downsweep of each VCO being reset to the initial pitch.
So what WE hear now is that continuously-rising tone, with tones entering at the bottom, vanishing at the top, and yet it doesn't move...or so we think!