Merge pull request #361 from ludovic111/feat/dualsense-vibration

DualSense vibration: shaped pulses, HD haptics on USB (opt-in)
This commit is contained in:
Ludovic Marie
2026-10-15 22:30:54 +02:00
committed by GitHub
7 changed files with 295 additions and 3 deletions
+45 -1
View File
@@ -248,7 +248,7 @@ sensitivity, slower while scoping with a spyglass.
Vibration (on by default, off in the options): taking damage (stronger with
the damage), a block broken, an explosion nearby (scaled by distance), a totem
of undying. Short, low-intensity pulses, as on Bedrock. A DualSense gets them
through its own output report (below), on every OS. Other pads go through
shaped, through its own output report (below), on every OS. Other pads go through
gilrs force feedback: Linux (evdev force feedback, including Steam Input's
virtual pad) and Windows support it; macOS gilrs has no force feedback, so it
is silently skipped there.
@@ -393,6 +393,49 @@ documents (none of it measured on Horizon's pad yet):
get a finer, faster grit, harder ones a slower, heavier chug, never above
5 of 8, since the attack trigger is held for seconds.
### Vibration shapes and HD haptics
Each event plays a short shape rather than one flat block: a hit opens at
full strength for 50 ms and dies away in two steps, an explosion rolls on for
half a second, a totem swells, a broken block is a 35 ms tick. Over HID they
drive the motors' emulation (the compatible-vibration bytes). Measured on the
pad, that emulation has inertia: after a full-strength step it takes about
half a second to settle, so short shapes blur.
The DualSense has no motors, but two voice-coil actuators, and on USB it is
also a four-channel 48 kHz audio device whose channels 3 and 4 drive them
(1 and 2 are its speaker and headset). Played there as waveforms (an 80 Hz
tone for the heavy part, 200 Hz for the light one, 4 ms ramps between steps,
channels 1 and 2 silent), the same shapes come out crisp. Per OS:
| | USB | Bluetooth |
|---|---|---|
| macOS | the pad lists as a 4-channel CoreAudio output ("DualSense Wireless Controller", seen on the Mac mini by dualsense-1); cpal opens it by name | no audio device |
| Linux | snd-usb-audio exposes the card "Controller" with 4 channels; cpal opens it through ALSA (`plughw:CARD=Controller`), which fails while PipeWire or PulseAudio holds the card, as desktops do: then the motors' emulation stays | no audio device |
| Windows | a WASAPI endpoint named "Wireless Controller" (or "DualSense Wireless Controller"), stereo until its speaker setup is set to Quadraphonic in the Sound control panel (`mmsys.cpl`, Configure); until then Horizon finds no 4-channel output and keeps the motors | no audio device |
The audio haptics play only while the pad is out of its rumble emulation:
any report with the rumble flags (flags 0, bits 0 and 1) silences them until
a report without them (SDL: "Leaving emulated rumble bits off will restore
audio haptics"; nondebug/dualsense and
[dse-haptics](https://github.com/iranran666/dse-haptics) arm them with an
all-zero report 0x02). While HD haptics are open the motors stay at zero, so
every report Horizon sends, starting with the one at opening, keeps them
armed; that first report also clears the attenuation a PS5 may have left.
The frame layout `[0, 0, left, right]` at 48 kHz is the one
[KytyPS5 #843](https://github.com/KytyPS5/KytyPS5/pull/843) plays on
Windows.
Over Bluetooth the haptics travel as 8-bit 3 kHz samples in vendor output
report 0x32, reverse-engineered by
[SAxense](https://github.com/egormanga/SAxense) and since used by DSX 3.2
and [dualsense-bt-haptics](https://github.com/awalol/dualsense-bt-haptics);
it needs Bluetooth output, which only Linux gets (see above), and isn't
done: there the motors' emulation is used. HD haptics are opt-in for now
(`HORIZON_DUALSENSE_HAPTICS=1`) until they are confirmed on a real pad;
games started for tests (`HORIZON_NO_AUDIO`) never open them, and the pad's
audio device is only ever opened by name, never the system's default output.
Byte layout from public references, read for facts; no code is copied from
them (the kernel and dualsensectl are GPL):
@@ -404,6 +447,7 @@ them (the kernel and dualsensectl are GPL):
| Nielk1's [TriggerEffectGenerator](https://gist.github.com/Nielk1/6d54cc2c00d2201ccb8c2720ad7538db/f73a299fdc52f43be84d39e62734192ad82bb393) (MIT) | rev. `f73a299` | the trigger modes (official 0x05, 0x21, 0x25, 0x26) and their zone packing |
| [pydualsense](https://github.com/flok/pydualsense/blob/01445f5ed6ba/pydualsense/pydualsense.py) (MIT) | `01445f5` | sets the attenuation flag with 0 on every report |
| [nondebug/dualsense](https://github.com/nondebug/dualsense) | `b87450e` | an all-zero report 0x02 leaves rumble emulation for the audio haptics |
| [KytyPS5 #843](https://github.com/KytyPS5/KytyPS5/pull/843), [dse-haptics](https://github.com/iranran666/dse-haptics), [SAxense](https://github.com/egormanga/SAxense) | read 2026-10-15 | the 4-channel layout, the Windows and Linux device names and setup, the Bluetooth haptics report |
Verified on hardware (USB, dualsense-1, Mac mini): the motors with the V2
flags (the IMU shook, and settled after a stop that then still sent the