How to Produce Music Tunes in MSX1

Describes how to efficiently play game sound track with MSX1
retro-computing
MSX
Author

Humberto C Marchezi

Published

March 9, 2024

Introduction

The sound generator contained in the MSX1 computer is called PSG (programmable sound generator) and it is composed by 3 tone generators + 1 white noise generator.

Programming PSG

PSG is controlled via its registers which serve as an interface to communicate with the sound generator.

Register Description Values
0 Least significant bits of channel A frequency 0~255
1 Most significant bits of channel A frequency 0~15
2 Least significant bits of channel B frequency 0~255
3 Most significant bits of channel B frequency 0~15
4 Least significant bits of channel C frequency 0~255
5 Most significant bits of channel C frequency 0~15
6 Noise generator frequency 0~31
7 Mixer setting 128~191
8 Volume of channel A 0~16
9 Volume of channel B 0~16
10 Volume of channel C 0~16
11 Least significant bits of envelope period 0~255
12 Most significant bits of envelope period 0~255
13 Envelope shape 0~15

Mixer Setting Register

Mixer is controlled by setting a byte value to register 7 where each bit can enable or disable tone generator or white noise generator per channel.

Bit Description Channel
7 PSG I/O ports (always 1) X
6 PSG I/O ports (always 0) X
5 Disabling noise C
4 Disabling noise B
3 Disabling noise A
2 Disabling tone C
1 Disabling tone B
0 Disabling tone A

PS: Bit 7 should be set to 1 and Bit 6 to 0 to avoid joystick malfunction

Setting Envelope Shape

Envelope shape is about the sound shape over time. Without an envelope shape, when any produced sound would sustain indefinitely while with envelope, the produced sound could be programmed to fade in or fade out or perhaps follow a repetitive envelope pattern.

Configuring envelope shape involves 2 parts:

  • select an envelope shape
  • set envelope frequency or length

Register 13 sets the envelope shape into one of the following shape options:

Value Shape
0~3 ＼＿＿＿＿＿＿＿
4~7 ／＿＿＿＿＿＿＿
8 ＼＼＼＼＼＼＼＼
9 ＼＿＿＿＿＿＿＿
10 ＼／＼／＼／＼／
11 ＼￣￣￣￣￣￣￣
12 ／／／／／／／／
13 ／￣￣￣￣￣￣￣
14 ／＼／＼／＼／＼
15 ／＿＿＿＿＿＿＿

Registers 12 and 11 store 8-bit values but they are combined to represent a 16-bit value for the frequency. Therefore the frequency value 0x123F is composed by setting 0x12 to register 12 and 0x3F to register 11.

PS: Note that the envelope will take effect on a channel only if the bit 4 in the channel’s volume register is set. In other words, the volume should be set to 16.

Music Tune with PSG Sound Generator

In each PSG channel, a tone frequency is defined by 12 bits stored in 2 registers of 1 byte where 4 bits are stored in most significative register while 8 bits is stored in the least significative register.

However for the purpose of music soundtrack, each frequency can to be mapped to a musical note as in the table below:

Octave
Note 1 2 3 4 5 6 7 8
C 0xD5D 0x6AF 0x357 0x1AC 0x0D6 0x06B 0x035 0x01B
C# 0xC9C 0x64E 0x327 0x194 0x0CA 0x065 0x032 0x019
D 0xBE7 0x5F4 0x2FA 0x17D 0x0BE 0x05F 0x030 0x018
D# 0xB3C 0x59E 0x2CF 0x168 0x0B4 0x05A 0x02D 0x016
E 0xA9B 0x54E 0x2A7 0x153 0x0AA 0x055 0x02A 0x015
F 0xA02 0x501 0x281 0x140 0x0A0 0x050 0x028 0x014
F# 0x973 0x4BA 0x25D 0x12E 0x097 0x04C 0x026 0x013
G 0x8EB 0x476 0x23B 0x11D 0x08F 0x047 0x024 0x012
G# 0x86B 0x436 0x21B 0x10D 0x087 0x043 0x022 0x011
A 0x7F2 0x3F9 0x1FD 0x0FE* 0x07F 0x040 0x020 0x010
A# 0x780 0x3C0 0x1E0 0x0F0 0x078 0x03C 0x01E 0x00F
B 0x714 0x38A 0x1C5 0x0E3 0x071 0x039 0x01C 0x00E

ref: https://www.msx.org/wiki/SOUND

Produce a Simple Sound Tone

In order to be able to listen to some tone from a MSX channel, the following is necessary:

  • Tone frequency - 12 bits value combined from the most significant and the least significant registers
  • Sound volume - values ranging from 0 to 15 or 16 (to enable sound envelope)
  • Mixer setting - enable or disable channel sound tone or white noise
  • Envelope shape - establishes sound envelope such as a short sound, continuous, fade-in, fade-out, etc.

Example below produces a tone for channel A in MSX BASIC for demonstration:

PSG register Value Comment
7 &B10111000 Disable noise, enable tone
13 &H01 Fade out envelope shape
12 &H13 Most signif envelope value
11 &H88 Least signif envelope value
1 &H00 Most signif tone freq value
0 &HFE Least signif tone freq value
8 &B00010000 Volume associate to envelope 
10 REM enable ch A tones with mixer 
20 SOUND 7, &B10111000
30 REM set short envelope period
40 SOUND 13, 1
50 SOUND 12, &H13: SOUND 11, &H88 
60 REM tone A4 for channel A
70 SOUND 1, &H00
80 SOUND 0, &HFE
85 REM associate volume to envelope
90 SOUND 8, 16

Producing White Noise with PSG

PSG should be programmed the following way, in order to produce white noise:

  • Set white noise frequency value to register 6 with values ranging 0 to 31 (5 least significative bits)
  • Set mixer to enable noise with register 7
  • Set volume and envelope shape to prevent indefinite noise
PSG register Value Comment
7 &B10110001 Enable noise for channel C
13 &H01 Fade out envelope shape
12 &H13 Most signif envelope value
11 &H88 Least signif envelope value
6 20 Most signif tone freq value
10 &B00010000 Volume associate to envelope 
10 REM enable noise for ch C
20 SOUND 7, &B10110001 
30 REM set envelope shape
40 SOUND 13, 1
50 SOUND 12, &H13: SOUND 11, &H88 
60 REM set white noise frequency
70 SOUND 6, 20
80 REM associate volume to envelope
90 SOUND 10, 16

PS: It is perfectly possible to enable tone and noise at the same time for a channel. Effectively noise can controlled as a 4th channel and several games use the white noise as a cymbal or drum effect in the game soundtrack.

Effectively Composing Soundtracks for Game

The size of background music depends on how elaborate the sound track is. Typically, for a decent soundtrack, 2 channels are used to build it.

In average, a soundtrack contains something about 128 notes (8 notes * 16 bars), therefore the byte size estimation is:

  • 8 notes * 16 bars * 2 bytes channel A * 2 bytes channel B = 512 bytes

However 512 bytes is a considerable size considering a multi-level game what opens opportunities to reduce soundtrack memory size.

Reducing Memory Space with Optimal Tune Range

1 byte per music note can be used instead of 2 bytes if a specific music frequency range is carefully chosen for the melody and bass channels.

Considering the frequency table of frequencies and music notes, the following music note ranges can be recommended:

description most significant fixed value starting note final note
bass channel 0x1 0xFD (A3) 0x0D (G#4)
melody channel 0x0 0xFE (A4) 0x0E (B8)

Using those frequency ranges above, the most significant register can be set to a fixed value while the least significant value can be variate within a single byte therefore saving precious MSX memory for the game.

The BASIC program below demonstrate this concept more clearly with a real example:

10 SOUND 1, &H01 'PSG channel A - bass
20 SOUND 3, &H00 'PSG channel B - melody

50 REM SOUNDTRACK BYTES
51 REM 1st BYTE is BASS - channel A
52 REM 2nd BYTE is MELODY - channel B
55 REM 64 BYTES FOR THIS MICRO SONG

60 DATA &HAC,&HD6,&HAC,&HAA,&HAC,&H8F,&HAC,&H7F
61 DATA &H68,&H78,&H68,&H78,&H68,&H78,&H68,&H78 
62 DATA &H40,&H7F,&H40,&H7F,&H40,&H7F,&H40,&H7F
63 DATA &HAC,&H8F,&HAC,&H8F,&HAC,&H8F,&HAC,&H8F
64 DATA &HAC,&HD6,&HAC,&HAA,&HAC,&H8F,&HAC,&H7F
65 DATA &H68,&H78,&H68,&H78,&H68,&H78,&H68,&H78 
66 DATA &H40,&H7F,&H40,&H7F,&H40,&H7F,&H40,&H7F
67 DATA &HAC,&H6B,&HAC,&H6B,&HAC,&H6B,&HAC,&H6B

100 PRINT "PLAY MUSIC"
110 SOUND 8,12: SOUND 9, 12 'set volume
120 FOR I = 1 to 32 'read tones
130 READ A: SOUND 0, A
140 READ B: SOUND 2, B
145 PRINT "SOUND 0, " + HEX$(A) + " SOUND 2, " + HEX$(B)
150 FOR TT = 1 to 90: NEXT TT 'delay workaround
160 NEXT I
170 SOUND 8,0: SOUND 9,0 'turn off vol

Add Duration to Music Tune

With a naive approach, each music note can be stored with frequency (2 bytes) and duration (1 byte), therefore occupying 3 bytes. However 3 bytes per music tone is too much space for the limited memory of MSX !

A - freq hi A - freq low A - duration B - freq hi B - freq low B - duration
0 D6 (C) 100 1 AC (C) 200
0 BE (D) 100 1 AC (C) 200
0 AA (E) 100 1 1D (G) 200
0 8F (G) 100 1 1D (G) 200

Option 1 - Use Optimal Tune Range with Duration

With optimal note range explained in previous section, the table becomes smaller.

This way music table can store bass tones and melody tones separately with 1 byte (instead of 2 bytes) and their respective duration with another byte.

Such as:

  • set channel B most significative tone to 0
  • set channel A most significative tone to 1
B - freq low B - duration A - freq low A - duration
D6 (C) 100 AC (C) 200
BE (D) 100 AC (C) 200
AA (E) 100 1D (G) 200
8F (G) 100 1D (G) 200

A program can use duration as a loop counter before moving on to the next music note.

/* least significative tune channel A */
unsigned char least_signif_tune_channel_a[]  = [ 0xAC, 0x68, 0x78, 0xAC, ... ];
/* channel A duration */
unsigned char duration_channel_a[] = [ 0x04, 0x04, 0x04, 0x04, ... ]; 

/* least significative tune channel B */
unsigned char least_signif_tune_channel_b[]  = [ 0xD6, 0xAA, 0x8F, 0x7F, ... ];
/* channel B duration */
unsigned char duration_channel_b[] = [ 0x01, 0x01, 0x01, 0x01, ... ]; 

unsigned char channel_a_counter = 0;
unsigned char channel_b_counter = 0;
unsigned char channel_a_tune_idx = 0;
unsigned char channel_b_tune_idx = 0;
unsigned char channel_a_size = ...;
unsigned char channel_b_size = ...;


while game_loop {
  channel_a_counter--;
  channel_b_counter--;

  if (channel_a_counter < 0) {
    channel_a_tune_idx++;
    /* Get duration for channel A */
    channel_a_counter = duration_channel_a[channel_a_tune_idx];
    /* Play tune in channel A */
    play_channel_a(0x01, least_signif_tune_channel_a);
  }

  if (channel_b_counter < 0) {
    channel_b_tune_idx++;
    /* Get duration for channel B */
    channel_b_counter = duration_channel_b[channel_b_tune_idx];
    /* Play tune in channel B */
    play_channel_b(0x00, least_signif_tune_channel_b);
  }

  if (channel_a_tune_idx > channel_a_size) {
    channel_a_tune_idx = 0;
    channel_a_counter = duration_channel_a[channel_a_tune_idx];
  }

  if (channel_b_tune_idx > channel_b_size) {
    channel_b_tune_idx = 0;
    channel_b_counter = duration_channel_b[channel_b_tune_idx];
  }

}

Option 2 - Use 1 Byte to Store Both Duration and Most Significative Tune

The PSG registers responsible for the most signiticative tune, make use of only 4 bits, therefore leaving other 4 bits available for other uses such as duration.

  • storage of most-significative register
bit 7 bit 6 bit 5 bit 4 bit 3 bit 2 bit 1 bit 0
X X X X tune tune tune tune

The idea here is to use 1 byte for the least significative tune information while optimising the use of the next byte to store: most significative tune and duration.

  • 1 byte to store least significative tune bits
bit 7 bit 6 bit 5 bit 4 bit 3 bit 2 bit 1 bit 0
tune tune tune tune tune tune tune tune
  • 1 byte to store both most significative tune bits + duration
bit 7 bit 6 bit 5 bit 4 bit 3 bit 2 bit 1 bit 0
duration duration duration duration tune tune tune tune

The C code below demonstrates how this could be implemented.

/* least significative tune channel A */
unsigned char least_signif_tune_channel_a[]  = [ 0xAC, 0x68, 0x78, 0xAC, ... ];
/* most significative tune channel A + duration */
/* PS: 0x41 => duration=0x4 and tune=0x1        */
unsigned char most_tune_duration_channel_a[] = [ 0x41, 0x41, 0x41, 0x41, ... ]; 

/* least significative tune channel B */
unsigned char least_signif_tune_channel_b[]  = [ 0xD6, 0xAA, 0x8F, 0x7F, ... ];
/* most significative tune channel B + duration */
/* PS: 0x10 => duration=0x1 and tune=0x0        */
unsigned char most_tune_duration_channel_b[] = [ 0x10, 0x10, 0x10, 0x10, ... ]; 

unsigned char channel_a_counter = 0;
unsigned char channel_b_counter = 0;
unsigned char channel_a_tune_idx = 0;
unsigned char channel_b_tune_idx = 0;
unsigned char channel_a_size = ...;
unsigned char channel_b_size = ...;


while game_loop {
  channel_a_counter--;
  channel_b_counter--;

  if (channel_a_counter < 0) {
    channel_a_tune_idx++;
    /* Divide by 16 to extract duration from byte */
    channel_a_counter = most_tune_duration_channel_a[channel_a_tune_idx] / 16;
    /* Get remaining from division by 16 to extract most significative tune info */
    play_channel_a(most_tune_duration_channel_a % 16, least_signif_tune_channel_a);
  }

  if (channel_b_counter < 0) {
    channel_b_tune_idx++;
    /* Divide by 16 to extract duration from byte */
    channel_b_counter = most_tune_duration_channel_b[channel_b_tune_idx] / 16;
    /* Get remaining from division by 16 to extract most significative tune info */
    play_channel_b(most_tune_duration_channel_b % 16, least_signif_tune_channel_b);
  }

  if (channel_a_tune_idx > channel_a_size) {
    channel_a_tune_idx = 0;
    channel_a_counter = most_tune_duration_channel_a[channel_a_tune_idx] / 16;
  }

  if (channel_b_tune_idx > channel_b_size) {
    channel_b_tune_idx = 0;
    channel_b_counter = most_tune_duration_channel_a[channel_b_tune_idx] / 16;
  }

}

Comparing Approaches

option pros cons
Optimal Tune Range + Duration * Simpler coding (best for beginners) * Limited octave use (from O4 to O8)
All Tunes + Duration in 2 Bytes * Allow all tune octaves to be explored * Code complexity and less performance

Conclusions

  • About PSG programming
    • Playing tunes and producing noise involves understanding how PSG registers work
    • Musical notes can be translated to tune frequency with PSG available channels with registers 0-5
    • White noise can also be programmed with respective frequency with register 6
    • PSG can be programmed to produce tune, noise or both depending on the mixer setting values with register 7
    • Different volume levels can be set to channels via registers 8-10
    • Different envelope sounds can be set for tunes or noise with registers 11-13
  • Game music tune
    • Music storage should be planned ahead to avoid soundtrack from taking too much memory from MSX1
    • Music notes can be expressed with 1 byte depending on the octave ranges being used
    • Music note durations and most significative frequency value can be both encoded in 1 byte
    • When choosing approach to play soundtrack, consider both pros and cons of both approaches