Config Table

Config Table — Reference Values and Linux Mapping

Configuration values the Linux driver ports from the reference touch stack, with adaptation where noted. Cross-reference: Touch Pipeline.

Association radii

Maximum distance a blob may move between frames, scaled by the number of tracked fingers. Linux: effective_radius = blob_max_distance × multiplier.

Reference value	Meaning
0.545009	2-finger radius multiplier
1.218098	Single-track continuity radius
1.549246	3-finger radius multiplier
1.845492	4-finger radius multiplier
2.161228	5+-finger radius multiplier

Edge contact weights

Finger weight is reduced at panel edges (partial TX/RX coverage). Linux multiplies first/last row/column cells before centroid computation: HEATMAP_EDGE_PENALTY_TOP = 97 (0.966512 × 100), BOTTOM = 23 (0.228140 × 100).

Reference value	Linux
0.966512	top/side penalty ×0.97
0.228140	bottom penalty ×0.23

Coalescing threshold

Reference value	Linux
36.0	ghost_dist = sqrt(36.0) = 6 cells

Pre-association filter

Rejects blobs below a fraction of the strongest blob's weight.

Reference value	Meaning
0.611209	Primary ratio threshold
0.754732	Secondary ratio threshold

Disabled in Linux (pre_assoc_ratio=0): testing showed it is too aggressive for multi-finger scenarios without the Mahalanobis classifier.

Noise floor

Reference value	Linux
0.04	cells with c590 < 400 (0.04 × 10000) suppressed

Applied during baseline processing, not after.

Hold policy

Reference value	Linux
0xB3 (enabled)	hold_frames=0 — disabled

The reference enables hold only for tracks passing strict quality checks (duration ≤ 2 frames, pixel count in bounds, eigenratio < 4.0, signal > 0.095). The initial Linux placeholder lacked these checks and caused scroll-brake behavior, so hold is off by default.

Touch detection threshold

Reference value	Meaning
0.17	Touch detection as a fraction of peak signal

Not standalone — requires the per-cycle gain adaptation pipeline, which is runtime-populated from device firmware. Cannot be used directly in Linux.

Peak isolation gate

The detector qualifies each candidate against the raw byte plane (see Touch Pipeline): it is rejected when any of the four cells five cells away on the cardinal axes reads below 135. The threshold and the distance are hard-coded in the detector, not read from the table.

Encoding	Value
raw byte (u8)	135
fixed point (u16)	1000 (= 0.1 × 10⁴)
signal domain (float)	0.1

A neighbour at or above 135 is quiet (c590 ≤ 0.10025, i.e. ≤ 1002 fixed point); the test rejects the candidate when the deflection has not decayed below that level within five cells. Samples that fall outside the raster (x < 5, x + 5 ≥ columns, y < 5, y + 5 ≥ rows) pass — edge candidates are never rejected by this gate.

Peak acceptance and scoring kernel

Surviving candidates are scored with a 11 × 11 float kernel (121 floats, Σ = 1.5513; positive ring, negative corners, near-zero core): score = Σ kernel·c590 over the window, where columns outside the raster contribute 0 and the window rows must fit inside the raster.

Value	Meaning
0.04	Peak acceptance in the alternate state
0.05	Peak acceptance, default

Only the frame's highest-scoring candidate is published, and only when its score reaches the threshold. The pipeline publishes one peak per frame; multi-contact logic happens in the tracker that consumes it (see Multi-touch (Experimental)).

Published heat map

The reference stack publishes a decimated copy of the raster for the OS:

Value	Effect
4	Decimation factor D; rows and columns are divided by it (SL4: 12 × 18), shift = log2(D) = 2
1	Enable flag (the decimated copy is skipped when zero)
derived	Cell values are divided by 10 before publication

The destination is a 5616-byte buffer with a 72-byte row stride, zeroed before each write.

Tracking constants

| Source | Value | Meaning | |---|---|---:|---| | reference constant | 36.0 | Squared handover radius: a new frame peak within 6 cells of a track's prediction takes the track over; beyond that the track keeps its position | | reference constant | 0.04 | Per-track liveness: a track that was not updated survives while its own cell reads c590 ≥ 0.04 (400 fixed point), an absolute test — not a comparison with the strongest contact | | coalescing constant | 36.0 | Coalescing / merge threshold (same value, see above) |

Extended re-search (present, inactive)

The reference record carries a second-search family that widens the search around the last published peak when a frame yields nothing:

Value	Meaning
2	Maximum candidate count that allows the re-search
3	Minimum consecutive empty frames before it arms
5	Maximum consecutive empty frames it tolerates
0	Enable flag — 0 in every device record, so the path arms nothing

Runtime-populated values (unavailable)

Three regions are zero in the static reference image and populated at runtime after gain calibration, so they are not portable without equivalent device communication: per-cycle gain calibration data, the Mahalanobis classification matrix (10×11 floats), and background noise tracking.

Module parameters

Every Linux module parameter, its class and its load-time default. Defaults are read from the driver source; the raw controls are load-time only, so every profile is a fresh module load. The classes match the parameter contract in docs/PARAMETERS.md.

Parameter	Class	Default	Notes
raw_mode	Standard safety	0	The only normal profile control; 0 selects standard HID and skips the raw activation/pipeline
sl4a_debug_level	Diagnostic	0	Log verbosity: 0=errors, 1=transitions, 2=per-frame, 3=full hex
debug_trace	Diagnostic	0	Controller-hop trace: 0=off, 1=lifecycle, 2=SPI segments, 3=read data
raw_input_beta	Experimental activation	0	Publish beta multi-touch from raw CapImg frames
skip_getfeat	Experimental activation	1	Skip the standard-mode feature-read handshake (the raw Report ID 6 read still runs)
getfeat_delay_ms	Experimental activation	0	Delay (ms) between RPT_DESC and GET_FEATURE
setfeat_speed_hz	Experimental activation	0	SPI clock (Hz) for the SET_FEATURE write only; 0 = bus default
wire_double_opcode	Experimental activation	0	1 sends the legacy doubled leading opcode on every sequencer frame
setfeat_no_double	Experimental activation	0	Deprecated alias: 1 sends SET_FEATURE without the doubled opcode
read_frame_variant	Experimental activation	1	Read-approval shape: 0=reference, 1=legacy five-byte, 2=both
acpi_probe_power_cycle	Experimental activation	0	ACPI _PS3→_PS0 power cycle at probe
sync_timeout_ms	Experimental activation	6000	Timeout (ms) for synchronous requests; clamped to [100, 60000]
stream_watchdog_ms	Experimental activation	2000	Runtime streaming watchdog interval (ms); 0=disable
stream_watchdog_max_retries	Experimental activation	3	Max re-init retries before giving up
skip_vendor_stop	Experimental activation	0	Omit the pre-DESCREQ vendor_stop + D2/D0 preamble
raw_fallback_on_reset	Experimental activation	0	First poller RESET_RSP gives up to the hardcoded fallback descriptors
raw_pre_desc_reg0	Experimental activation	0	Pre-DONE reads go to register 0 (input_register) instead of the {3, 0x0A} selector
raw_b1f8109_preset	Experimental activation	0	One switch restoring the b1f8109 raw dialect
get_noread	Experimental activation	0	GET Report ID 6 write-only: skip the reply read-back
raw_handshake_first_ms	Experimental activation	2000	First raw post-DONE heatmap wait (ms); later retries keep the default
raw_no_enable	Experimental activation	0	Skip the 0x56 stream enable at DONE (SET_FEATURE ID5 owns the stream)
raw_watchdog_teardown	Experimental activation	1	Raw handshake-watchdog retries send STOP+D2/D0 before DESCREQ (0 = DESCREQ only)
std_raw_transition	Experimental activation	0	Run the GET ID6 + SET ID5 heatmap transition in standard mode
std_liveness_ms	Experimental standard-mode recovery (issue #4)	0	Standard-mode startup liveness window (ms); logs only, never recovers
std_liveness_recover	Experimental standard-mode recovery (issue #4)	0	Run ACPI recovery when the liveness check sees no activity after DONE (needs std_liveness_ms)
skip_std_getfeat	Experimental standard-mode recovery (issue #4)	0	Answer feature GET_REPORT with -EOPNOTSUPP in standard HID
wait_reset_kick_ms	Experimental standard-mode recovery (issue #4)	0	Kick discovery (one DESCREQ) when no IRQ edge since power-up/resume; 0=disable
blob_min_weight	Experimental raw pipeline	1000	Minimum blob signal rise
ema_alpha	Experimental raw pipeline	2	Position EMA smoothing
dfa_data_offset	Experimental raw pipeline	0	DFT antenna frame data offset in bytes (0 = decoded raster)
ghost_dist	Experimental raw pipeline	6	Coalescence radius (cells)
grid_cols	Experimental raw pipeline	0	Grid columns; 0 = per-device default
grid_rows	Experimental raw pipeline	0	Grid rows; 0 = per-device default
blob_debounce	Experimental raw pipeline	3	New-contact debounce (frames)
blob_lift_frames	Experimental raw pipeline	3	Missed frames before a contact lifts
hold_frames	Experimental raw pipeline	0	Contact hold grace frames (disabled)
pre_assoc_ratio	Experimental raw pipeline	0	Pre-association ratio ×1000 (disabled)
blob_max_distance	Experimental raw pipeline	3	Slot reassignment distance (cells)
invert_x	Experimental raw calibration	0	Invert the X axis
invert_y	Experimental raw calibration	0	Invert the Y axis
swap_xy	Experimental raw calibration	0	Swap the X/Y axes
calib_scale_x	Experimental raw calibration	0	X scale ×1000 (0 = derived from the grid)
calib_scale_y	Experimental raw calibration	0	Y scale ×1000
calib_offset_x	Experimental raw calibration	0	X pixel offset
calib_offset_y	Experimental raw calibration	0	Y pixel offset

Per-device geometry

The reference values were taken from the SL4 (MSHW0231) pipeline. The SL3 AMD controller (MSHW0162) runs the same pipeline with a different sensor grid selected by ACPI ID at probe: 78×52 (4056 CapImg samples) instead of 72×48 (3456), and a 33-frame resting baseline instead of 30. See Hardware and Multi-touch (Experimental).