> > > btw.  has anyone done this before?
> > 
> > 	As it happens, the Hardware Reference Manual describes exactly how to 
> > do it.  By setting the mask in ADKCON, you can define the interaction of the 
> > channels for modulation of both period and volume.  Specifically:
>  
> On comp.sys.amiga.programmer or somesuch there has been a sort of war over this
> subject.
> 
> Basically, the proposed trick was to play one channel at maximum volume
> (playing the most significant bits of the sample) and at the same time the other
> channel at minumum volume (for the least significant bits). The actual sample
> played would be 64 * high_channel + low_channel.

Hmm..  the conversion from mu-law to linear is given by:

  segment = ~(bits 4-6)  (range 0 to 7)
  sign = ~(bit 7)        (range 0 to 1)
  offset = ~(bits 0-3)	 (range 0 to 15)

  linear = sgn(sign) * (2^segment * (offset + 33) - 33)
         = sgn(sign) * 2^segment * (offset + 33 - 33/2^segment)

seems that you can easily decompose this into two channels multiplied
together by:

  A = 2^segment                                   (range 1 to 128)
  B = sgn(sign) * (offset + 33 - (33/2^segment))  (range -48 to 48)

the resulting value of A modulated by B (A * B) is the linear
equivalent of the original mulaw data except for a small error
due to quantization of (33/2^segment).

Both values A and B can be placed in small lookup tables (8 bytes
and 32 bytes respectively).

Seems that mu-law can be implemented on the amiga with good quality
without much fuss.

> ___ Olaf 'Rhialto' Seibert    rhialto@mbfys.kun.nl     Ooey-Gooey-Fluffy-Barfie