Four Ways to Make a Sound

Code status: illustrative code. Adapt the listing to your program and check it on your target machine. This page links no assembly or execution record for the listing itself.

Overview

The tune sequencer introduces a reusable idea: store events and advance through them. That does not make the drivers interchangeable. Their timing, voice state and sound-generation work depend on the target. We can compare where each machine does the work, which resources a cue needs and what continues after our routine returns.

Four approaches place the work in different parts of the machine:

• Spectrum 16K/48K — CPU-driven beeper. The CPU changes the speaker output through the ULA. Repeating those changes produces a tone; the spacing of the edges determines its waveform and pitch. The 128K also has an AY sound chip.
• NES — specialised channels. Two pulse channels, triangle, noise and DMC sample playback each have different controls. We arrange sounds around those differences.
• C64 — oscillators and envelopes. Each of the SID’s three voices has waveform selection and an ADSR envelope. Selected voices can feed a shared analogue filter; synchronisation and ring modulation connect particular oscillators.
• Amiga — sample playback. Four audio channels stream signed 8-bit sample data from Chip RAM using DMA. We can supply recorded sounds or draw our own waveforms, within memory and playback-rate limits.

The comparison

Hardware-generated audio can continue while the CPU does other work. That does not make the whole music player free: someone still advances notes, applies effects and decides which sound owns each resource.

Starting a sound, four ways

The comparison above and the completion question below can be explored without reading assembly. Open a fragment when we know that machine’s registers and instructions; we do not need to learn all four syntaxes together.

These are illustrative fragments, not complete programs or four versions of the same note. Startup, duration, stopping and resource ownership belong to the surrounding program. Read each fragment’s assumptions before adapting it.

    ld   b,50          ; 50 complete high/low pairs
loop:
    ld   a,$10
    out  ($fe),a       ; speaker bit high, black border, MIC bit clear
    call delay         ; supplied delay routine must preserve B
    xor  a
    out  ($fe),a       ; speaker bit low, same border and MIC bits
    call delay
    djnz loop

    lda #$01
    sta $4015          ; enable pulse 1; disable other channels
    lda #%10111111
    sta $4000          ; 50% duty, length halt, constant volume 15
    lda #$08
    sta $4001          ; sweep disabled, negate set to avoid upward overflow mute
    lda #$fd
    sta $4002          ; timer low byte
    lda #$00
    sta $4003          ; timer high bits 0, load length, restart pulse phase

    lda #$00
    sta $d404          ; gate off before this trigger
    sta $d417          ; no voices routed through the shared filter
    sta $d402          ; pulse-width low byte
    lda #$08
    sta $d403          ; pulse width $800: half of the oscillator cycle
    lda #$0f
    sta $d418          ; master volume 15, filter modes off
    lda #$67
    sta $d400
    lda #$11
    sta $d401          ; frequency $1167: approximately middle C on PAL
    lda #$09
    sta $d405          ; attack 0, decay 9
    lda #$f9
    sta $d406          ; sustain 15, release 9
    lda #$41
    sta $d404          ; pulse waveform ($40) + gate on ($01)

    move.l  #sample,AUD0LC(a5)
    move.w  #sample_words,AUD0LEN(a5) ; length in 16-bit words
    move.w  #428,AUD0PER(a5)          ; spacing between output samples
    move.w  #64,AUD0VOL(a5)           ; full channel volume
    move.w  #$8201,DMACON(a5)         ; set master DMA and channel 0 DMA

Choosing a sound that communicates

For a collected item, we might use a short beeper tone, a bright NES pulse, a SID note with a quick attack, or a small Amiga sample. Those can communicate the same success without having identical waveforms.

Compare a sharp attack with a slower one. Does the cue still feel connected to the collection? Compare a short tail with a long one. Do repeated collections remain distinct, or do they obscure one another? Keep the visible collection feedback too, so the information does not depend on hearing the sound.

Our creative choices follow the available controls. On triangle we cannot simply turn down an amplitude envelope. On the beeper, delaying waveform work changes the sound itself. On Paula, playing a sample faster changes both its pitch and the time to traverse its data. Understanding those differences helps us choose an effect that works in the game.

When a shared operation stops fitting

Consider an operation called play_cue that takes a cue number. In our first memory game it returns after the cue finishes. The caller then restores the panel and checks input. That is a useful, simple promise.

Before tracing, write down two separate moments: when the call returns and when the cue finishes. They coincide in the waiting version.

Now we want another implementation that starts hardware playback and returns while it continues. Predict what the unchanged caller does: does the panel stay active until the sound ends? When can it read input again?

A cue identifier need not promise an identical instrument. SID envelopes, NES channel-specific controls and Amiga sample data are not interchangeable parameters. Keep those choices visible in each implementation. Compare whether the cue communicates the same event, then account separately for its sound, storage and execution cost.

Hardware references

• Sinclair, ZX Spectrum BASIC Programming, chapter 23: port $FE, speaker, MIC and border bits.
• MOS Technology, 6581 SID datasheet, register descriptions: waveform bits, ADSR and filter routing.
• NESdev’s hardware research: APU channels, pulse, sweep and DMC.
• Commodore, Amiga Hardware Reference Manual, Audio Hardware: waveform data, length, volume, period and DMA.

Related

Patterns: Tune Sequencer — event data and playback updates, with target-specific contracts · foundational audio: Sound Beep (Spectrum) · A Square Wave (NES) · SID Note Trigger (C64) · Playing a Sample (Amiga)

Vault: Paula | APU | The SID Chip | ULA | AY-3-8912

Reuse an effect with an explicit lifetime

Decide who stops the sound as well as who starts it. A blocking tone owns the CPU until it returns. A serviced effect needs its mute routine called at the agreed rate. A queued MOS effect depends on firmware service and queue policy. These entries include native working and omitted-trigger recordings.

ZX80 and ZX81 have no stock programmable speaker. Their input entries state that boundary. The PET example requires a CB2 sound connection or the selected emulated output. Other language and expansion routes remain separate work.

Eight service calls do not automatically mean eight complete elapsed frames: the first decrement may happen in the triggering update. A sound queue, effect priority and music channel sharing are useful next components; a register writer does not supply them.