Interestingly, analog reverbs and some other kinds of time-delay effects (echo, flanger, chorus) were made using so-called bucket-brigade devices which are indeed analog in the amplitude domain, but not in the time domain, where they are very much digital...
True. One caveat though. Before someone rushes to purchase bagfuls of these chips on Aliexpress/Ebay/Amazon etc, it should be noted that the sellers are perfectly aware of those chips popularity among guitar effects builders, therefore they're among the most faked items over there. If you're extremely lucky you get the real thing, which doesn't happen often, or if you're moderately lucky you get a smaller version or a rebadged clone (resulting in shorter delays or louder noise, or both). If you're totally out of luck you get common cheap random chips relabeled as BBD delay chips. This happens very often with pricey chips when shopping online on those platforms.
However, the BBD sound can be almost easily emulated in digital. On the "wet" return line of a digital delay put a N dB low Q high pass filter before say 200 Hz, a N dB low Q low pass filter above 6-8 KHz, add some gentle clipping and some noise, compress overall wet dynamics and feed the return back to the delay input. Then adjust delays, feedback and filters amount to make it sound like a real one. The trick is simulating the progressive way in which the sound when going through a BBD device will lose in fidelity while getting more distorted and noisy, so we filter and add those artifacts inside the feedback path.
Fun fact about the origin of flanger, it was originally made with tape machines and got the name because you'd do it by running two tape machines side by side, and press your finger against the metal "flanges" of one to make it slow down relative to the other.
> The finished music track is recorded simultaneously to two matching tape machines, then replayed with both decks in sync. The playback-head output from the two recorders is mixed to a third recorder. The engineer slows down one recorder by lightly pressing a finger on the flange (rim) of one of the playout reels. The "drainpipe" or subtle "swoosh" 'flange flango' effect "sweeps" in one direction, and the playback of that recorder remains slightly behind the other when the finger is removed. By pressing a finger on the flange of the other deck, the effect sweeps back in the other direction as the decks progress towards being in sync.
The story goes John Lennon coined the name. Ken Townsend at Abbey Road had worked out a way to mimic vocal double tracking without actually recording the singer twice using this technique. They called it Artificial Double Tracking, or ADT. But Lennon would just call it "Ken's Flanger" and the term stuck.
I think it would be more correct to say they are discrete in the time domain rather than digital. This is just semantic nitpicking though since digital usually implies discrete time. I never heard about continuous time digital electronics, it might be interesting to think about though it feels like an oxymoron.
Would it be continuous time if not clocked, e.g. switches or 555 timer feeding into logic gates, and levels just switch when they switch. Of course no logic transitions are actually instant...
The clock is what defines the sampling rate of the delay, so a 555 is just a simpler version of what they already use. The true audio is always moving continuously. When the BBD receives a clock signal, it samples the incoming audio, and to reconstruct that audio as accurately as possible you have to play it back at that same rate it was sampled in. Very similar to how digital (PCM) audio works, except using charged capacitors instead of bits in a digital memory.
If you don't clock it at all, you'll just be sending out the constant voltage equal to whatever your first capacitor has charged, which results in no audible sound. You can adjust your clock speed and have it play back at speeds slower or faster than the incoming rate, with the pitch changing too, just like slowing a record or tape machine.
The output isn't really discrete though, only the internals, and even then like you say, even logic transitions aren't actually instant (but at a rate so much faster than audio we ignore it). The BBDs use filters on both the input and output that remove the very high-end of the audio range (a lowpass filter).
These filters will be made to match or be lower than the current clock rate [0]. One before going into the BBD removes any frequencies too high to be stored accurately (an anti-aliasing filter), and then one after the BBD removes any high frequencies which were created by the near-instant shapes of the capacitors discharging (a reconstruction filter). These means your output audio is "continuous", but also has zero frequency content above your filter cutoff (okay there's a teeny amount of information, nothing in reality is perfect).
To create the delay effect you take a long line of them (usually 512-4096) and the length of your delay is equal to the number of "buckets" times your clock rate. If you had a near infinite amount of buckets and ran them so fast there was no loss in fidelity from the sampling rate, you'd basically have a tape delay. The fun thing about BBDs though is that they don't always pass on their charge to the next bucket perfectly. So longer delay lines mean introducing more errors accumulating as your audio passes through the delay line. A near-infinite BBD would turn into audio mush. Though you can still get a very high number of buckets and a very high clock rate for near digital fidelity, but at that point it's getting more expensive than just using a digital effect, and you're ditching the unique charm of a nice effect.
Here are some more detailed explanations of a bbd [1] [2] The posts by Richard Crowley in [2] are particularly cool to the HN audience I'd hope.
> Curiously enough, BBD were a pre-cursor to EEPROM, DRAM, and FLASH RAM. Microscopic capacitors are used to store the data. In the case of DRAM, the capacitors aren't terribly good, and they must be "reminded" (refreshed) on a regular basis (many times per second). Very similar to BBD.
[0] okay technically 1/2 the clock rate because of nyquist-shannon, I just mean it will be linked to the sampling rate.
