Dual Readhead Angular Error Cancellation: Measuring Yaw Out of the Position Signal

Why one readhead cannot see yaw

A linear encoder measures displacement along the scale between the readhead and the scale's reference. The measurement is exact for the point where the readhead senses. When the carriage rotates by a small angle about a vertical axis somewhere on its body, every point on the carriage moves by a different amount along the direction of travel—and the readhead faithfully reports its own point's displacement, which includes a component from the rotation. The controller receives a position that mixes translation and rotation and cannot separate them, because one input cannot resolve two degrees of freedom.

The consequence scales with geometry. The rotation that contaminated the encoder signal also displaces the tool or workpiece point, and the displacement there equals the rotation multiplied by the distance between the rotation center and the tool point—the Abbe arm. A readhead near the rotation center reports almost no rotation-induced error, while the tool point tens or hundreds of millimeters away sees the full effect. This is why an axis can 'hold position' at the encoder while failing its accuracy specification at the workpiece: the feedback point and the error point are different points.

Two readheads at different positions along the yaw axis see different rotation-induced displacement components. Their signals now contain enough information to separate translation from rotation—one linear equation in one unknown becomes two equations in two unknowns. The same principle, applied to two readheads diametrically opposed on a rotary scale, cancels eccentricity; applied to readheads stacked to sense pitch, it extends to a second angular axis. The architecture generalizes wherever two sensing points sample a rigid body's motion.

The averaging and differencing scheme

With two readheads at positions separated by a known span along the yaw axis, each reports position plus a yaw-induced component proportional to its distance from the rotation center. The arithmetic is the point of the architecture: the average of the two positions equals the translation, because the rotation-induced components are equal and opposite about the center and cancel; the difference of the two positions, divided by the known span, equals the yaw angle itself.

Implementation details decide whether the arithmetic survives contact with hardware. The two readheads must be sampled in the same servo cycle, because a difference computed from unsynchronized samples measures velocity, not yaw. Their interpolation settings must be identical, because a resolution mismatch between channels biases the average. And the controller must either run the position loop on the averaged signal or run two loops with yaw feedforward—the choice follows the machine's dynamics and the platform's capability.

The difference signal has value beyond correction. Logged over travel, it is a continuous yaw map of the stage—a diagnostic that reveals guide wear, scale-liveling drift and bearing anomalies as trends in an axis that previously reported nothing about its own rotation. Machines that installed dual readheads purely for accuracy often find the angular telemetry becomes the more valuable output in service.

• Average = translation; half-difference ÷ readhead span = yaw angle
• Synchronized same-cycle sampling is mandatory
• Identical interpolation and resolution settings on both channels
• Difference signal doubles as a continuous inline yaw diagnostic

Installation: spacing, datum and matching

The cancellation quality is set at installation. The two readheads should be spaced as widely as the carriage and scale permit, because yaw measurement resolution from the difference signal is inversely proportional to the span—doubling the spacing halves the angle uncertainty for the same position noise. Both readheads must track the same scale with the specified air gap and lateral alignment, and the scale itself must be straight within its specification, since the readheads now measure against a common geometric reference rather than independent ones.

Matching matters as much as placement. Two readheads of the same model can still differ in phase, amplitude and interpolation offset if configured differently or mounted with different local geometry; the commissioning procedure should verify that with the stage stationary, the difference signal is stable and small, and that during a slow full-travel move the difference signal follows a smooth, repeatable curve. An abrupt step or sign flip in the difference signal at some position usually indicates a readhead tracking problem, a scale joint, or a local contamination event—not real yaw.

Cabling and diagnostics should follow the same discipline as any dual-channel feedback system: separate routing to the controller, alarms configured for signal-loss or mismatch on either channel, and a startup check that both channels report consistent position before motion is enabled. A dual-readhead axis that runs when one channel is dead is averaging garbage; the startup interlock is what makes the architecture trustworthy.

What cancellation does and does not fix

Dual readheads cancel angular error of the carriage relative to the scale. They do not make the scale more accurate; scale deviation still appears identically in both channels and passes through the average untouched. They do not correct straightness of the guides, squareness of the axis stack, or thermal expansion of the scale—those need their own budget lines. The technique addresses exactly one error source: the yaw (or pitch, or eccentricity) component that a single readhead hides inside its position report.

The payoff should therefore be evaluated against the error budget, not against a hope. On a long-travel granite metrology stage where yaw × Abbe arm is the dominant term after scale accuracy, dual readheads can remove the majority of the tool-point error and shrink the residual enough that simple error mapping closes the gap. On a short-travel compact stage with a stiff carriage and a millimeter-scale Abbe arm, the same hardware adds cost and failure modes while removing an error that was already negligible. The architecture is powerful where geometry makes it necessary—and geometry is the test.

A commissioning workflow for a dual-readhead axis

Start by confirming the feedback architecture on paper: both readheads' configurations, the controller's dual-input mode, the sampling synchronization and the intended processing—average to the loop, difference to compensation or monitoring. Then commission in order: verify each channel independently against the scale datum, check the stationary difference signal, run a slow full-travel move and record the difference curve, and compare the derived yaw against an independent angular measurement—an autocollimator or electronic level—over several positions.

Only after the angular signal is validated should compensation be enabled, verified as every encoder compensation is: positioning measurements at the tool point, before and after, over full travel, in both directions. The before-and-after pair is the evidence that the cancellation works at the point that matters, and it becomes the baseline for the machine's future health checks.

• Confirm configurations, synchronization and processing mode on paper first
• Validate each channel, then the stationary difference, then the travel yaw curve
• Cross-check derived yaw against an autocollimator or level
• Verify tool-point accuracy before and after compensation, bidirectionally