SG-323

The reverberator is a large block of delay memory - 65536 words, one word every 1/32000 of a second - written through a permuted address sequence rather than as a plain ring buffer. That permutation is what makes the tap spacings irregular, and it is the reason the sound is a room rather than a comb filter.

Twenty-three taps read out of that memory: fifteen return to the input as feedback, and eight are summed into the two output channels. Where each tap sits is not a design choice made here - the addresses come from the original machine's EPROM tables, carried across verbatim. Changing program swaps the table, which is why the sixteen programs are genuinely different rooms.

The taps move. A modulation clock walks the read addresses around their nominal positions, and the amount it walks is driven by the input level, so playing harder changes the character of the tail and not only its loudness. That is the machine's own behaviour, and the Input Gain control sits in front of it for exactly that reason.

Before anything reaches the memory the signal passes a first-order high-pass and a first-order low-pass - the controls the hardware called Low and High Frequency Decay. They shorten the tail at one end or the other by never letting those frequencies into the memory in the first place, which is not the same as filtering the output.

Two switches reproduce the hardware's converters. Vintage rounds what is written into the memory to 16 bits, and Noise adds a low white-noise floor under it. Both are on by default because both were unavoidable on the original; turning them off is the one thing this can do that the machine could not.

The kernel is compiled restricted TypeScript, so the same source runs on the host, in the native plug-in and as a WASM reactor. It processes at a fixed 48 kHz internally and a host at another rate goes through the shell's rate adapter, because the delay memory is allocated in samples when the instance is created.