platkus@en.ecn.purdue.edu (Shawn W Platkus) writes:

>I'm not sure what you are talking about.  If the GS only has eight output
>chanels, then it would seen to me that it could only play eight different
>sounds at a time.  So how can I digitize fifteen different sound and play
>them all back simultaneously if there are only eight output channels.

To summarize the following lecture:

There are 32 oscillators (16 generators) and 16 channels. Not all are used for
actual sound output.

Oscillators are "smart voices"
Generators are oscillator pairs that can generate extra effects with each other
Channels are actual independent output lines like left and right speaker

That said, let's start at the top.

The DOC is a coprocessor with its own dedicated 64K of RAM. All the sound
samples have to be put in this RAM before they can be played.

The DOC has 32 'oscillators' which are essentially smart DMA channels. Their
basic function is to sweep through areas of the DOC memory reading samples
and playing them. They can do so at variable speeds (automatically repeating
or skipping sample values as necessary), they can loop on a power of two
boundary, they can stop when they read a zero, they have independent volume
settings, and various other things that aid in reproducing complex instruments
without loading down the main CPU.

But when you come down to it, the DOC is capable of playing 32 sounds
simultaneously and independent of each other, provided that all the sample
data fits in the DOC RAM.

The oscillators are not all perfectly identical in operation. For the basic
sample playing and looping they are, but for some more complex functions
they must be paired. This is where the concept of 'generators' comes from --
the 16 generators ARE operationally identical and that is why software prefers
the generator concept. Both oscillators and generators are numbered from 0,
so oscillators 0 & 1 are generator 0, oscillators 2 & 3 are generator 1, and
so on.

Generator 15 (oscillators 30 & 31) is reserved for system use (one oscillator
is set to loop slowly at zero volume, to generate tempo; I forget if the
other is used by anything, it's probably used to play mono samples).

Most software use one generator per voice. Since 15 generators are left over,
spec'ing the GS as having "15 voice sound capability" is a fair statement.

The actual output that comes out the DOC is a 'time-domain multiplexed' sound
output and five digital bits. What happens is this: the DOC services each
oscillator in turn, and for each oscillator the current sample value is
multiplied by the oscillators' volume setting and a voltage proportional to
the product is output on the sound output. Four of the digital bits are set to
the 'channel number' setting for the oscillator and the last one simply changes
voltage from about 3 volts to about 0.5 (for you EE folks out there, this is
the negative edge of an output-valid strobe).

External hardware is responsible for splitting off the various channels (4 bits
means that there can be 16 of them) and outputting them seperately. The
motherboard hardware just ignores the channel setting and mixes all the sound
outputs into the speaker/earphone. The sound connector on the motherboard only
has room for 3 of the four bits, so expansion cards that plug into the sound
connector can only get 8 seperate output channels. Most stereo cards (AE's
sonic blaster, for example) only pay attention to the lowest bit, so even
numbered channels are left and odd numbered channels are right (or is it the
other way round? I forget).

No, it isn't simple, but it gives a lot of flexibility -- most of which is
largely untapped.

Todd Whitesel
toddpw @ tybalt.caltech.edu