Stardot — scarybeasts’ SN76489 PCM playback threads (2025)

Two Stardot threads by Chris Evans / “scarybeasts” that build on his 2020 blog post:

1. t=30838 (April 2025) — bus-side behaviour analysis: discovers that holding /WE continuously low across consecutive writes works reliably, halving the per-sample bus dance and enabling 62.5 kHz sustained write rate (one byte every 16 µs).
2. t=31654 (September 2025) — applies the 62.5 kHz write technique to produce the first BBC PCM playback demos with comparable quality to MOD-style 4-channel mixed-sample audio: 12.5 kHz × 4-channel and 15 kHz × 3-channel mixed playback of Amiga-era tunes (Lotus Esprit Turbo Challenge, Chaos Engine, Escape from Planet of Robot Monsters).

These two threads together are the current canonical reference for high-quality PCM playback on BBC SN76489, superseding all earlier techniques (which were typically single-channel ≤ ~8 kHz).

Bibliographic

• t=30838 — “SN76489 sound chip bus-side behaviour”: started 2025-04-04 by scarybeasts. Documents the 62.5 kHz sustained-write discovery with sn_test.ssd / SNLEVEL test program. hoglet contributed FPGA verification.
• t=31654 — “BBC SN76489 sampled song playback demos”: started 2025-09-09 by scarybeasts. Posts demo .ssd files + YouTube links. Contributions from hoglet (BeebFPGA verification, “very hard to tell any difference” vs real Master), Matt Godbolt (jsbeeb fixes triggered by the demos), dominicbeesley (C20K FPGA implementation with user-controllable HPF).
• Source code: https://github.com/scarybeasts/misc/tree/master/beebmod/

Key technical claims

62.5 kHz sustained write rate via persistent /WE-low (t=30838)

“It is writing one new byte to the chip every 16us which is a rate of 62.5kHz.”

“There are a few references across the internet that discuss leaving WE active on a SN76489. Some say it’s reliable and some say it’s not. I think it is reliable if you are careful.”

hoglet’s verification:

“allow multiple successive writes to occur if the WR_N signal was held continuously low”

Mechanism: instead of the textbook latch-pulse cycle (drive bus, pulse /WE low, hold ≥8 µs, pulse /WE high, repeat), you set /WE low once and then just drop new bytes onto Port A at 16 µs intervals. The chip latches each on the rising edge of its internal sampling — provided the bus value is stable when the chip samples (the “9 µs alignment” mentioned in t=30838). Saves the latch-line addressable-latch dance per write, which is the bulk of the slow-bus cost.

Resulting playback rates (t=31654)

Demo	Rate	Channels
Lotus Esprit Turbo Challenge title song	12.5 kHz	4-channel mixed
Chaos Engine subsongs 4, 11	15 kHz	3-channel mixed
Escape from Planet of Robot Monsters	15 kHz	3-channel mixed

These are MOD-tracker-style mixed multi-channel samples — the SN76489 still only has one PCM channel (the volume DAC of one tone channel), so all multi-channel mixing happens in software on the 6502 before each sample byte is written. Inner loop runs at 62.5 kHz (= the bus rate) with a 4× or 5× decimation per output sample.

Sample pre-processing (t=31654, Lotus song write-up)

“The squishing of 8-bit into 4-bit values is done at runtime, again with lookup tables.”

Per-instrument adjustments:

• Slap bass: 2× gain for SNR
• Bass/snare drums: zero-level drift to match SN76489 output curve
• Guitar: magnitude-axis flip to preserve detail

Storage trade-offs (t=31654)

• Lotus consumes all available RAM — impractical from tape without strategic loading pauses.
• Alternative approaches noted: compress samples (suspected highly compressible), use generated waveforms (square/saw), short 1-2 KB looped samples (bass / organ / sax), play same sample at different offsets (TFMX-style reuse).

Hardware limits

• Built-in BBC speaker is the bottleneck. Low frequencies (Chaos Engine bassline) “mostly stripped by the speaker and the circuitry that immediately precedes it” — the LM324N 8 kHz low-pass + speaker normaliser (sn76489) butcher the bass.
• External amplifier (Monitor Audio Airstream A100 in hoglet’s case) needed to hear the full quality.
• BeebFPGA “very hard to tell any difference” vs real Master — establishes that the technique is hardware-honest, not an emulator artefact.

Emulator support (t=31654)

Recommended emulator order (as of 2025-09):

1. b-em — works out of the box
2. beebjit — needs HEAD from master
3. b2 — needs HEAD from wip/master

Other emulators produce poor audio. Matt Godbolt explicitly committed to fixing jsbeeb after seeing the demos.

Cross-references made in-thread

• Cinter (Amiga 4K demo scene synth) — https://github.com/askeksa/Cinter
• Sarah’s NES Castlevania player thread — viewtopic.php?t=18290
• pcmenc — already referenced in scarybeast-sn76489-sampled

What’s still missing from primary sources

• No cycle-by-cycle inner-loop kernel posted in either thread; the GitHub repo holds it.
• No formal SNR / THD measurements — quality is reported subjectively.
• No explicit “how to silence the other channels” recipe — implicit in the multi-channel mixer (other tone channels’ volumes are set to off; the noise channel is similarly off).

Filed into

• Created sampled-sound — synthesises the 2020 blog (carrier-modulation physics) + the two 2025 threads (62.5 kHz persistent-/WE trick + 4-channel software mixing achieving Amiga-MOD-equivalent quality) into one wiki technique page.
• Extended sn76489 — added the persistent-/WE write pattern as a high-performance alternative to the textbook latch-pulse dance.

See also

• scarybeast-sn76489-sampled — the 2020 blog laying out the physics.
• sampled-sound — the implementation pattern derived from these three sources.
• sn76489 — chip-level reference (clock, registers, write protocols).
• system-via — slow-bus dance (the persistent-/WE trick is a slow-bus optimisation).
• reference_emulator_accuracy — relevant emulator status as of 2025-09 per hoglet/Godbolt.

---

This wiki is curated by Claude following the LLM-Wiki methodology — a human curates source documents, the LLM compiles structured cross-linked markdown. Content may contain errors, omissions, or stale claims. For authoritative information refer to the original source documents in the bbc-documents GitHub archive.