Incomplete edge electric fields systematically distort touch coordinates

What it is

Projected capacitance estimates a centroid from the electric field on both sides of an electrode grid. The outermost ring of electrodes has no further neighbor; the field is truncated by the housing, the metal mid-frame, and air, so the left/right (or top/bottom) weights in the centroid formula are no longer symmetric. A finger in the last few millimetres reports a coordinate systematically pulled inward or drifting along the edge, not random jitter. That bias is sensor geometry, not the user’s aim.

Why it happens

Centroid estimation assumes comparable electrode responses around the blob. At the edge a slice of the response surface is missing, and the algorithm treats the remaining peak as complete, so the coordinate walks toward the side where the field still exists. A metal mid-frame acts as extra ground and bends field lines further. Firmware often pushes coordinates back out with an edge compensation table; the amount depends on contact area, pressure, and humidity. Over-compensation makes a coordinate “grow off-screen”; under-compensation leaves the cursor a gap inward while the finger is already on the bezel. During a slide along the long edge the distortion is a slowly varying bias field, not a constant pixel offset, so a constant patch calibrated at centre waxes and wanes at the edge.

Studying it

Walk a stepper or a graduated probe toward the edge in 0.5 mm steps along the surface normal. Record true position minus system coordinate and plot the normal error curve. Independent variables are which edge (top/bottom/left/right), contact area, case on or off, and charging state. Dependent measures are normal bias, tangential drift, and residual before versus after compensation. Human fingers repeat poorly; probes give the geometric curve, then a thumb sliding along the edge checks whether dynamic error matches the static curve. Do not declare whole-screen linearity after calibrating only the centre.

Where it stops holding

Optical touch and infrared frames get edge error from lamp layout and reflections, not capacitive truncation; the curve shape differs. An active stylus has its own electrodes and edge algorithm; finger millimetres do not transfer to the nib. A full-screen module that runs electrodes to the glass lip narrows the distortion band but does not zero it—the mid-frame is still there. Users also aim with an upward systematic bias that adds to edge-field distortion; if the experiment does not fix the true contact location, the two biases collapse into one number.

Applying it

• Treat the outer 5–8 mm as its own input zone for calibration and test; do not quote centre precision as whole-screen precision.
• Keep precision drags (selections, inking, slider thumbs) off the physical edge by default; when they must sit there, add edge snapping or button stepping.
• Accept with a probe or a graduated sticker: when the finger is on the bezel, check whether log coordinates still sit inward or overshoot the screen, and file compensation residual as hardware, not an application bug.

Related

• Same group: C2.21.2 Flush-to-edge controls hit-test less reliably than center ones · C2.21.3 Curved edges compound optical distortion with incomplete sensing · C2.21.4 Edge gestures are especially exposed to edge-field distortion
• Nearby: C2.06 Touch offset and aiming bias · C2.23 Screen materials, films, and cases
• Search terms: edge distortion · capacitive field · centroid bias