Skip to content

86Box with Aureal Vortex 2 (SQ2500) emulation 0.5

Latest

Choose a tag to compare

@z2k-retro z2k-retro released this 17 Nov 15:30
650de2f

Aureal Vortex 2 for 86Box, build 0.5

This build adds an emulated Aureal Vortex 2 (SQ2500) sound card with hardware A3D 2.0 to 86Box. It runs Aureal's own Windows 98 drivers.

It is a reverse-engineering project. Aureal never documented the chip; the emulation was built by tracing and measuring a real SQ2500 and is checked against it. The method, the register-level reference and all the tools are in [aureal-vortex2-tools](https://github.com/z2k-retro/aureal-vortex2-tools).

More testers are wanted. If you have A3D games, and especially if you also have a real Vortex 2 to compare with, please try it and report what you hear. The [user guide](https://github.com/z2k-retro/aureal-vortex2-tools/blob/main/docs/using-86box.md) covers setup and what to include in a report.

What's new since 0.4

The theme of this build: parts of the chip are now measured digitally, straight off a real card, instead of through recordings of its analog output. With the card idle and no driver running, the host machine can load the chip's buffers by hand, route them through one of its blocks, and read the block's output back as numbers. The sample-rate converter and the wavetable voices' output levels were measured this way, and both are now modelled exactly as read.

You may hear these

  • Half-Life's ordinary sounds are about 4 dB louder against its 3D sounds. The wavetable voices Half-Life plays its non-3D sounds on (gunfire, footsteps, voices) go out through the chip's effect send buses. Measured on the card, a voice comes out 4 dB hotter there than 0.4 assumed. The overall balance of the mix in Half-Life should now be closer to a real card.
  • Plain DirectSound sounds the driver puts on wavetable voices are about 2 dB quieter, for the same reason seen from the other side: at the driver's centre pan the card plays them 2 dB below an ordinary stream.
  • Sounds that overflow onto wavetable voices in busy 3D scenes are about 2 dB quieter going into the A3D engine.

Accuracy work you probably won't hear

  • Sample-rate converter: now bit-exact. All 213,018 of its coefficients (26 taps at each of its 16,384 phases) were read off the card. Every value is backed by at least two readings that agree; none is fitted or smoothed. The output is rounded exactly as the chip rounds it, and a sum that overflows wraps and then saturates as on the card. Any 16-bit stream through a converter now comes out the same as on a real SQ2500.
  • Converter speed: the converter now works in integers on a compact table. All 16 converters take about 2% of one CPU core.
  • Wavetable voices:
    • Their pitch is now decoded exactly, correcting 0.4's small errors of up to about 100 ppm.
    • Pan, both effect sends and both gain bytes were measured digitally and match the model to the last unit. The pan acts on both output pairs, as the chip does.
    • The rarely used gain states are now modelled as measured.
  • Mixer: its gain law was confirmed digitally (0.5, 0.75, 1, 1.5, 2 at codes 0x00–0x10), as 0.4 already had it.

Build

  • Windows builds failed for a day after an MSYS2 OpenSSL update; fixed.

Checked against real hardware

  • Half-Life on a Pentium III 700 with a real SQ2500, headphones and speakers, with CD music. CD music sits in the right place against game sound at the default CD audio input level (−12 dB). The speaker modes' level and midrange are within about 0.5 dB of the card. Some bass and treble differences remain open.
  • Half-Life's tram ride on an Athlon XP with a real SQ2500, headphone mode: the long-term spectrum is within about 1 dB of the card from 160 Hz to 10 kHz (as in 0.4).
  • Bench measurements read digitally from the card: every converter and wavetable output level above, reproduced exactly by the model.

Known issues

  • Turn off "trace" in the card's settings for normal use. It writes every register access to disk, and at Half-Life's rate of access that is enough to cause pops and skips even on very fast machines. "Write diagnostic log" is light, but is also only needed for reports.
  • Only the converter is bit-exact. The rest of the chip is modelled in floating point.
  • Half-Life only plays CD music if the disc is in the drive when the game starts. That is the game's behaviour, not the emulation's. Mount the disc image first.
  • Slave Zero's sound can drop out after a while and not return until Windows restarts. The cause is not yet known.
  • Sound Blaster emulation can freeze the machine at boot if another PCI device is on IRQ 5. Disable "Vortex AU8830 Sound Blaster Pro Emulation" in Device Manager, or move the other device.
  • Aureal's A3D splash screen can crash PowerVR games at start-up; turn the splash screen off in the registry (see the user guide).
  • 86Box outputs stereo, so in four-speaker mode the rear pair is dropped or folded into the front.
  • Not emulated: recording from microphone or line in, the game port and MPU-401, real-mode DOS, S/PDIF, the wavetable filter's resonance (the driver never uses it).
  • Only driver 4.06.2048 under Windows 98 has been tested.

What to listen for

Reports on these would help most:

  • The balance of Half-Life's ordinary sounds against its 3D sounds, compared with a real card. This build changes it the most.
  • How loud 3D sound is in Satellite speakers and Monitors mode, compared with a real card.
  • Echoes and reverb in games with A3D reflections.
  • Any sound that cuts out, sticks or crackles, with "trace" off. Please include vortex2.log.

Credits

Hardware, testing and direction: Z2k Retro. Technical assistance: ZanQuance, whose method of reading the converter by pumping its buffers by hand made this build's measurements possible. Chip model, tools and documentation written with substantial help from Claude, Anthropic's AI model. Register names from the Linux ALSA au88x0 driver.

This project is not affiliated with Aureal or its successors.