Manufacturers such as Coolaudio (whose new 1024-stage V3207 BBD is shown here), Xvive, and Belling are now manufacturing accurate replicas of classic analog chips of yore.
Would You Care to Try a Sample?
BBDs—or analog shift registers, as they are properly called (or sometimes charge-coupled devices)—were first developed in 1969 as integrated-circuit (IC) chips containing a large number of storage cells (tiny capacitors, actually) and switching transistors. The “register” part of a BBD chip is capable of holding a sampled voltage (i.e., the “analog” part) and passing (or “shifting”) those sampled voltages along in a first-in, first-out fashion. Like a film that creates the illusion of continuous motion using 24 frames per second, BBDs create the illusion of continuous sound by capturing and spitting out samples. The sampled voltages (or sounds) can seem to occur more slowly by setting the device to spit them out slower than they came in, or by simply holding onto them for a wee bit. Because the innards of these IC chips acquire and pass along the samples continuously, from input to output, in a manner similar to a line of firefighters passing along buckets of water from one to the other, they are referred to as “bucket brigade” devices. And just as with motion pictures, the more samples (or frames, in the case of a film) taken in within a set period of time, the smoother the apparent transition between samples.
When most people hear the word “sample” in a context such as this, they immediately think of digital samples. How are analog BBD samples different from digital samples? In a number of ways. First, they aren’t transformed or encoded, as digital signals must be. They are a literal snapshot of whatever the signal voltage is at the moment the snapshot is taken. One could argue that analog BBD devices have infinite resolution, because unlike with the analog-to-digital (and then digital-to-analog) conversion required in digital processors, there’s no rounding off of signal information to the nearest coded value. In fact, back in the early digital days when the best signal resolution one could hope for was 12-bit resolution, the analog sampling of a BBD circuit made for higher audio fidelity. With 24-bit digital encoding being common these days, the difference in audio quality between an encoded digital snapshot, and a literal analog snapshot, is largely unnoticeable to most players.
A second way in which analog samples are different from digital samples is that, unlike digital code’s ability to be retained in memory for as long as one wishes, analog samples have a limited lifespan. The capacitors storing them leak, so each sample has to get moved along to the output quickly enough. Otherwise there won’t be much left in the “bucket” to pass along. At the exact moment a sample is taken, it may have been 250 millivolts (mV), but eventually declines to 247 mV, then 243 mV, etc., if it’s not passed along right away. This signal loss is less of a problem when the delays are very short and the samples get passed along quickly, as in flanging. But when samples get passed along very slowly (for example, to achieve a long echo), they pass through all those leaky stages again and again for additional repeats. If you set an analog delay to its maximum delay setting, you can often hear a substantial loss in fidelity. Given this reality, it goes without saying that no matter how much you may like analog delays, the very idea of an analog looper pedal is a nonstarter.
Another thing to remember is that the sample that gets passed along is a snapshot of something that already happened. It may not have happened very long ago, but it still already happened. So, while BBDs can almost catch up to real time in some instances, in general they will always lag behind by some small amount of time. The various musical applications of BBD chips vary how much they lag, and how closely they nearly catch up to a nondelayed version of the signal.
The more Legos you use, the more realistic-looking your sculpture. Same goes for audio sampling.
Photo by Shawn Hammond
Year after year, a group of local enthusiasts shows off fantastic dioramas made entirely out of Lego bricks at the annual Maker Faire in my area. One of the things I’m always struck by is that the realism of the construction depends on how many bricks are used to make it—the more bricks there are, the more nuance there is in the final image. The same is true of audio sampling. Basic depiction of a sound at any given frequency can be accomplished with two samples: one representing the positive half-cycle of the sound wave, and one representing the negative half-cycle. But that turns the signal into a series of square waves, and does that really convey the complexities of the original, or are we building something with two bricks that really needs many more to look and sound like the real thing? A sound wave might have a gradual ascent and descent, so describing it as one step up and another step down just doesn’t cut it.
The general principle with digital audio sampling is that the sample rate should be three to four times the highest frequency to be reproduced, if not more. So, if a signal processor is intended to process and output a signal whose content goes as high as 5 kHz, it should not sample the audio signal fewer than 20,000 times per second (20 kHz). That will do a bare-bones job of accurately representing a 5 kHz signal, and a better job at representing lower frequencies. At a 20 kHz sampling rate, a single cycle of a 1 kHz signal would be portrayed with 20 samples—better than four, but still somewhat block-y and definitely not hi-fi. Think Commodore 64 graphics rather than 1080p.
Considering this, the amount of delay time one can expect to get out of a BBD depends on how wide a bandwidth you want to reproduce with it. If you want more frequency content to be included in the time-shifted signal, you’ll need to sample more of the signal and do it faster. But the trade-off is that, even though sampling faster yields better sound quality, it also yields less time from the BBD. Which brings us to….