Controls and software

Stable Pointing Without Added Weight for Gimbal Control Engineers

Independent analysis by Starro. Not client work.

A drone camera gimbal holding its camera steady

Objective

Notch Filter or Structural Change for a 166 Hz Yoke Mode

A camera gimbal keeps the line of sight steady while the aircraft moves. When the yoke between the motor and the camera resonates near the control band, the controller can push the yoke instead of the camera, and the image rings.

Should we fix the resonance with a notch filter or with the structure, and what does each cost?

Findings

A Box Crossbar Restores Near Rigid-Body Margins

On the baseline yoke the mode lifts the loop gain above one at 166 Hz. The loop is stable at nominal values only because of where the phase falls: peak sensitivity is 11.3 dB, it rings for about 100 ms, and it goes unstable in 18 of 135 variations. A 166 Hz notch cuts that to 1 of 135 but costs 10° of phase margin and raises overshoot to 56 %. Corner gussets also cut it to 1 of 135, yet leave peak sensitivity at 6.9 dB.

The box crossbar moves the mode to 237 Hz, where the loop no longer amplifies it: peak sensitivity 4.1 dB against 4.0 dB for a rigid frame, 11.7 dB gain margin, and stable in all 135 variations, with no added mass. All three modes still sit inside the two-per-revolution band of a two-blade propeller from 1,000 to 8,000 rpm, where each is amplified up to 25 times.

Peak sensitivity against the actual mode frequency with the controller fixed, with and without a 166 Hz notch. The loop is unstable below about 134 Hz and within 6 dB above about 202 Hz, where only the box crossbar lands.

Approach

One Gimbal Axis With a Single Structural Mode

One gimbal axis as a rigid body plus the yoke's first structural mode, taken from our companion finite-element study: 166 Hz on the baseline yoke, 191 Hz with 15 g of corner gussets, and 237 Hz with a closed box crossbar of the same mass as the original flat bar. The controller is fixed: integral action and phase lead for a 50 Hz crossover, with 0.3 ms of loop delay.

We compare the baseline with and without a 20 dB notch filter, the gusseted yoke and the box crossbar. Then we repeat the comparison across 135 combinations of mode frequency, modal participation, damping and loop gain.

Block diagram of one gimbal axis: controller, optional notch filter, motor and payload, the yoke mode at 166, 191 or 237 Hz depending on the design, and a gyro feeding the line of sight back with 0.3 ms delay.

Validation

Discrete-Time and Delay-Approximation Checks

How we checked the model:

  • The lead network peaks at 57.8° at 49.9 Hz, centered on the 50 Hz crossover, as the textbook formula predicts.

  • Gain margin is the same (4.65 dB) on frequency grids of 20,000, 60,000 and 200,000 points.

  • A digital version of the controller at 8 kHz matches the continuous model within 0.1° of phase margin and 0.04 dB of peak sensitivity.

  • Two levels of delay approximation give the same stability verdict in all 675 grid cases.

  • References: Skogestad and Postlethwaite, Multivariable Feedback Control (2005); Hilkert, IEEE Control Systems Magazine 28(1) (2008); Preumont, Vibration Control of Active Structures (2018).

Recommendations

Reshape the Structure Before Adding a Notch

Where the structure can still change, reshape it rather than add material: a closed-section crossbar of unchanged mass removes the problem at its source. Keep a notch filter for structures that are frozen, or as a backup that bounds the worst case. Neither fix stops propeller vibration in this speed range, so the gimbal still needs isolation from the airframe.

Tools used:

Linear loop model with one structural mode (python-control); discrete-time check at 8 kHz; 135-case parameter grid