[ripps818]

Summary

Adds haptics that use the Steam Controller 2's actuators directly instead of motor-style rumble:

• DualSense mode: the puck presents a real pad's 4-channel USB audio function and plays the game's haptic channels on the controller. By default it streams them as PCM to the grip actuators, as a real DualSense does.
• Switch Pro mode: renders the console's HD rumble. Both bands play on the grips, and the high band plays as tones on the trackpads.
• Steam mode: relays the controller's PCM haptic streams (OUTPUT 0x87-0x89), so audio-to-haptics apps work through the puck.

Changes

Eight commits, each builds on its own:

1. UAC1 4-channel audio haptics (mode_ps5_audio.*). 48 kHz 4-channel output plus a silent 2-channel mic, which PipeWire's stock DualSense profile requires. Channels 3 and 4 drive the actuators in one of four styles, set from the panel:

   • wave (default): the waveform itself, low-passed at 300 Hz and streamed to the grips as 4 kHz u-law PCM.
   • tone: 0x83 tones per actuator.
   • split: below 80 Hz as rumble, the rest as tones.
   • rumble: 0x80, merged with the game's normal rumble.

2. DualSense identity:

   • byte-for-byte report descriptor, pairing report tail, bcdDevice 1.00, no serial string in the single-HID mode;
   • feature reports answered at their declared sizes with a real pad's values, or stalled, never a short reply;
   • product string "DualSense Wireless Controller", as a real pad reports it.

3. PS trackpad click fix: the click pulse now buzzes the pad that was clicked. It always used side 1, which is the right pad.

4. Steam mode relay of OUTPUT 0x87-0x89. These are raw haptic sample streams, not settings reports, and the real puck forwards them. Changes:

   • relay payload raised to 63 bytes for haptic reports (feature commands stay at 60);
   • a second relay per poll cycle while the ring is backed up (an 8 kHz stream needs ~258 frames/s);
   • one same-PID resend of an unanswered PCM frame.

5. PCM stream + wave style + grip limiter:

   • shared hapticPcmStart / hapticPcmSend / hapticUlaw;
   • short first frames, so playback starts after 24 ms instead of 31 ms;
   • a soft-limit knee (default 85%) rounds off grip peaks instead of clipping them into pops.

6. Switch Pro HD rumble. Decoder:

   • tracks band frequencies;
   • maps amplitude code 100 to full scale, matching the public amplitude table and SDL's encoder (it used to play every level at 0.56x);
   • SPI 0x6012 reports device type 3, which Eden's direct Pro Controller driver needs.

   Renderer (in the main loop, not the USB callback):

   • grips: low + high band per side, summed and phase-continuous, as PCM;
   • trackpads: the high band as 0x83 tones, capped at -15 dB between 250 and 300 Hz;
   • silence stops it after a 500 ms hang.

   The decode and renderer are adapted from @Froggerdog's HD Rumble Emulation in Add optional Switch Pro profiles, trackpad D-pad and HD Rumble Emulation #303, without its profiles, trackpad D-pad and shortcut changes.

7. Panel: wave in the DualSense style button. A grip limiter select and an HD trackpad strength select in the Rumble card, hidden on older firmware.

8. Docs:

   • docs/DUALSENSE_HAPTICS.md: setup, Proton notes, hardware measurements;
   • PROTOCOL.md: OUTPUT 0x86-0x89 table, actuator routing, measured PCM behaviour, HD rumble (§9.5), new fields;
   • README and TESTING_GUIDE.

Update

Four things were added on top of the eight commits, after testing and the reports in the comments:

1. No click after HD rumble, and fewer dropped PCM frames. Stop messages now go only to actuators that may still be playing, since a -128 dB cut or 0x86 op 1 sent to an idle one plays as a click of its own. An unanswered 0x86-0x89 frame goes back to the front of the ring for up to two more poll cycles. This path is shared, so DualSense wave and Steam-mode 0x87-0x89 relay get the retry too.
2. Per-type haptic style and grip limiter for the Switch and the DualSense (fields 120-127). The Switch gets the four styles for its grips (wave, tone, split, rumble); both types start on wave. Fields 88 and 115 still work for older panels.
3. Per-type rumble strength (fields 108-111; 22 still edits the active type). The Switch starts at 100%, where its wave stops getting stronger; the other types start at 200%.
4. HD Rumble toggle for the Switch (field 128). Off plays plain motor rumble with its own strength (200%).

In the panel, each controller's tab now has Rumble strength, and the Switch and DS5 tabs have Haptics style and Grip limiter; the Switch tab also has HD Rumble and HD trackpad strength. The Rumble card keeps Style and Test rumble.

The status blob grows to p[202..214] (limiter and style per type, HD Rumble, strength per type). A config from before these starts every type from the old shared limiter, style and strength.

New tests: tests/type_haptics/run.py, and more cases in tests/hd_rumble/run.py. Hardware: Mario Kart in Eden feels fine.

Update 2: upstream merge and haptics fixes

Merged current main (no conflicts left) and brought over the fixes from testing in the comments (Jefemy's reports on HD rumble position and strength):

1. Wave frames are delivered once. The controller answers a PCM frame in its ESB ACK up to ~1 ms later, so the old 400 us window cut most replies off and the resends played twice. The puck now waits up to 1200 us, counts an empty ACK as delivered, and turns the radio around to RX in hardware, since in DualSense mode the USB audio work preempted a software RXEN and 83% of replies were missed (now 1%).
2. 0x86 goes out once per stream. Re-sending it every second made the grips stutter once a second.
3. Switch HD bands restart in step on each effect, so the same effect no longer swells or fades depending on where the bands' phases were.
4. DualSense wave ticks and underruns: an empty ACK after a PCM frame no longer counts as a dropped poll, which skipped a third of relay turns; and the ISO OUT endpoint is armed with a second buffer before processing, because the nRF driver drops a packet when none is armed.
5. USB audio survives settings saves. A flash write mid-stream wedged the ISO OUT endpoint until a bus reset. Saves now run from the loop task (also off the usbd task's 800-byte stack) and wait while the audio interface is open; a watchdog re-arms a stuck endpoint.
6. Defaults: grip limiter 85% (IMU: no measurable strength lost, 70 cost 3-5%); the Switch starts on wave; the Switch wave is capped at 100% grip strength and the DualSense wave at 250% gain, past which the grips grind.
7. Soft boost (fields 144-147, blob p[215..218]): an upward compressor that raises quiet wave passages such as footsteps, Off / 1.5x (default) / 2x / 3x / 4x.
8. "Test haptic rumble" (op 0x36): a tone panned across the grips, in every mode. Op 0x35 reads PCM stream counters (debug).

Behaviour changes worth a look

• Switch Pro renders HD rumble by default. The Switch tab's HD Rumble toggle (field 128) turns it off, which plays the console's rumble as plain motor rumble. The rumble style (field 39) applies in every other mode, and in Switch mode with HD Rumble off.
• Unanswered PS5 feature reports now stall, as on a real pad. That includes 0x03/0x08, which SDL probes for capabilities.
• Steam mode now forwards 0x87-0x89. Steam's own traffic uses them only for haptic streams.

Protocol / config

• Status blob v23 (upstream is at 22, with the RF recovery marker at p[196]) appends p[197..201]: audio gain/2, audio haptics on/off, audio style, grip limiter knee, HD trackpad strength/2.

• WebUSB fields:

  • 30 audio gain (percent/2, 0 = auto)
  • 31 audio haptics on/off
  • 88 audio style (0 rumble, 1 tone, 2 split, 3 wave)
  • 115 grip limiter knee (50-100, 100 = off)
  • 141 HD trackpad strength (percent/2)

  None of these collide with Add optional Switch Pro profiles, trackpad D-pad and HD Rumble Emulation #303. 141 is Add optional Switch Pro profiles, trackpad D-pad and HD Rumble Emulation #303's own field for the same setting.

• Later fields: 108-111, 120-127, 128 and 144-147, blob p[202..218] (see the updates above).

• cfg.bin: new fields are appended to the tail, after upstream's ext[] block, so existing files keep loading. A file from an earlier build of this PR loses its saved haptics settings once and starts on the defaults.

Related PRs

• Add optional Switch Pro profiles, trackpad D-pad and HD Rumble Emulation #303: this PR carries Add optional Switch Pro profiles, trackpad D-pad and HD Rumble Emulation #303's HD rumble core. If Add optional Switch Pro profiles, trackpad D-pad and HD Rumble Emulation #303 lands first, I'll rebase this onto it and drop the duplicate.
• Improve DualSense recognition in some games #295 (merged): the firmware-info payload here is identical; the merge keeps this branch's ps5GetCommon, which also stalls 0x03.

Testing

• make check (clang-format 18) passes; every commit builds.
• python3 tests/hd_rumble/run.py: HD decode, rendering, stop, transition, pan test and soft boost tests, built with ASan/UBSan and -Werror; python3 tests/type_haptics/run.py: per-type defaults and config loading.
• Hardware, this branch:
  • DualSense audio haptics (wave) in Control Resonant (GE-Proton, Steam Input off);
  • Switch Pro HD rumble in Super Mario Odyssey on Eden with the Direct Pro Controller driver.
• Hardware, earlier versions of this PR: Stellar Blade, Hi-Fi Rush and FFXIV. Per-game status and known issues are in docs/DUALSENSE_HAPTICS.md.

Linux setup

No WirePlumber config is required. Set the DualSense Direct sink's haptic channels to 100% once:

pactl set-sink-volume alsa_output.usb-Sony_Interactive_Entertainment_DualSense_Wireless_Controller-00.Direct__Direct__sink 40% 40% 100% 100%

make install-wireplumber installs an optional hook that does this automatically. libScePad games (Sony PC ports) need GE-Proton 11-6+ and Steam Input off.

AI Disclosure

Note

This feature was heavily developed with the assistance of LLMs/AI. Everything has been reviewed, built, and hardware-verified.

🤖 Generated with Claude Code

[ga2mer]

Why it uses regular rumble instead of audio-driven way? I think you could feed my code to your LLM Agent so it can come up with a better solution

[ripps818]

I've been working on this in my personal branch for a while. But my lack of knowledge on this is keeping me from getting it working correctly. It seems everytime a get one feature working, another stops working. If I get pcm haptics working, gyro and legacy rumble disappear. Something tells me this has do with hid-playstation feature flags, but I need to do some actual research now. I changed this to a draft until I figure it out.

[ripps818]

Okay, I refactored it a bit. If anybody wants to tear it apart, feel free.

[ripps818]

Yet another refactor. Updated the protocol to improve a bunch of haptic related protocols, improved dualsense audio haptics, and added switch pro hd rumble.

[kohupallintrax]

Working well for me in Arknights Endfield and Infinity Nikki in Dualsense mode. Touchpad haptics for touchpad "click" feels a bit stiffer than the typical steam controller experience.

[ripps818]

Try disabling audio haptics, and tell me if it still feels "stiff" I suspect it might latency from the audio stream queue.

[Jefemy]

Switch rumble feels much better, however it seems to rumble twice. It will fall off and then come back a bit at the end. Doesn't happen on #303. It also feels like it is 'dropping' the rumble a bit. Like a packet is being lost? On the Mario Kart World character select panning across each character doesn't feel consistent like on a real Pro controller. Might be a symptom of some interference with the rumble not ending properly.

[ripps818]

@Jefemy
Have you tried changing the grip limiter? I experienced harsh peaking with dualsense pcm haptics, and I adapted that system into the switch hd rumble. It's set to curve off peaks at 70%, but I left a slider to customize it. Move it to 90 or 100 (effectively disabling the limiter) and tell me if it still feels dropped off.

There very well could be dropped packets, so I'll look into it.

[Jefemy]

@Jefemy Have you tried changing the grip limiter? I experienced harsh peaking with dualsense pcm haptics, and I adapted that system into the switch hd rumble. It's set to curve off peaks at 70%, but I left a slider to customize it. Move it to 90 or 100 (effectively disabling the limiter) and tell me if it still feels dropped off.

There very well could be dropped packets, so I'll look into it.

Just tried messing around with all the settings on the panel using the built in web panel test rumble. On everything I tested there is a slight "click" after the rumble should end. Its power doesn't seem to be affected by any of the settings.

[ripps818]

are you using the docs/index.html in the pr, or the github.io config panel? Because the github.io one won't show my added options unless this PR gets merged.

[Jefemy]

I am using the panel and uf2 downloaded from the main branch of your repo. I also first tested it using a uf2 compiled from the branch pulled here.

[ripps818]

Oh, if your using the main in my ripps818/openpuck repo, it has tons of other stuff. You should make a new issue on my repo, not here. Because it might not be affecting this PR.

[Jefemy]

I only used the main branch to make sure it wasn't fixed by another commit. Both issues are still in this PR. I will make an issue on your repo to continue this though.

EDIT: Switch HD Rumble works great now. Have been using it for a few weeks now