Casio PV-1000/ASIC registers

The µPD65010G031 gate array has 10 total ports. They are fully decoded, available via Z80 I/O instructions at $F8-$FF.

There are eight write-only functions and two read-only functions.

Square 1 ($F8) > write

7  bit  0
..PP PPPP
  || ||||
  ++-++++--- Period for first square wave.

f_square = f_ASIC ÷ 1024 ÷ (63-P) ≈ 8739 Hz ÷ (63-P)

Where f_ASIC is the 17.897727MHz crystal divided by 2.

The output changes when the current count changes to 0. This means that:

• P=63 meaning period 0 is silent, and can leave the output in either direction
• P=62 then P=63 will instantly toggle the output

All three squares are running off of the same 17478Hz clock, and will change at the same moment as appropriate. This is equivalent to thinking the sound hardware has a sample rate of [crystal divided by 1024].

Square 2 ($F9) > write

Same as Square 1, but 3dB louder.

Square 3 ($FA) > write

Same as Square 1, but 6dB louder.

Sound control ($FB) > write

7  bit  0
.... ..ES
       ||
       |+--- 0: Normal
       |     1: XOR/ring modulation
       +---- 0: Mute
             1: Sound audible

Muting ANDs all above three channels with 0. Writes are still permitted, the period dividers still run.

XOR/ring modulation replaces:

• SquareF8.external ← SquareF8.internal XOR SquareF9.internal
• SquareF9.external ← SquareF9.internal XOR SquareFA
• SquareFA is unchanged

There are highpass filters on the square waves, but in normal operation they're inaudible (their corner frequencies are below the slowest period achievable). Unfortunately, when XOR modulation is enabled that's no longer true.

The three channels have first-order highpass filters at:

• SquareF8: 68kΩ · 47nF → 50Hz
• SquareF9: 47kΩ · 47nF → 72Hz
• SquareFA: 33kΩ · 47nF → 103Hz

Interrupt enable ($FC) > write

Existing games write 0, 1, or 3.

7  bit  0
.... ..BC
       ||
       |+--- 1: enable Vblank IRQ (1/vblank) 
       +---- 1: enable matrix scan IRQ (16/vblank)

Known that 0x1C bits have no effect.

Interrupt status? ($FC) < read

7  bit  0
ZAAA AABC
|||| ||||
|||| |||+-- Unclear. Counts 2 scanlines?
|||| ||+--- Usually 1?
|+++-++---- Always 0?
+---------- Always 1 (Probably pulled up by resistor next to Z80)

GPO ($FD) > write

Also acknowledges Vblank IRQ.

7  bit  0
.AAA BBBB
 ||| ||||
 ||| ++++-- 4 general purpose outputs, accessible via both controller ports
 +++------- Strongly implied to exist on the gate array, but not accessible externally

GPI ($FD) < read

Also acknowledges Matrix scan IRQ.

7  bit  0
ZAAA BBCC
|||| ||||
|||| ||++-- 2 general purpose inputs for reading the player 1 joystick
|||| ++---- 2 general purpose inputs for reading the player 2 joystick
|+++------- Always 0. (Strongly implied to exist on the gate array, but grounded there)
+---------- Always 1. (Probably pulled up by resistor next to Z80)

Tilemap and pattern RAM address ($FE) > write

7  bit  0
BBBB AA..
|||| ||
++++-++---- Address A15-A10 from where to fetch the tilemap. 
||||        Almost all software write $B8, meaning the gate array fetches tilemap from $B802 through $BAFD.
++++------- Address A15-A12 from where to fetch pattern RAM tiles (A11-A10 always 1)
            The common value of $B8 means that gate array fetches data for tiles $E0-$FF from $BC08-$BFFF

Display properties ($FF) > write

7  bit  0
AAAD FBGR
|||| ||||
|||| |+++-- Border color. (0=black, 1=red, 2=green, 3=yellow, 4=blue, &c).
|||| +----- 1: Display disable. (Screen solid border color, faster Z80)
|||+------- Pattern RAM disable 
|||          (0: Tiles $E0-$FF are fetched from address specified by port $FE
|||           1: Tiles $E0-$FF are fetched from address specified by port $FF)
+++-------- Address A15-A13 from where to fetch patterns
             (0: Patterns are fetched from $0008-$1BFF or -$1FFF
              1: starting at $2008
              2: starting at $4008 &c

"Fetch pattern" msbit and lsbit have been confirmed. Middle bit has not yet. Dig Dug uses both D and the lsbit of AAA.

Enabling the display in the middle of a scanline causes 8 to 16 pixels of noise, specific origin unknown.

Other sources (e.g. Enri) refer to the pattern memory as "PCG"