Applications / UAV & Defense
EO/IR stabilization in high-vibration environments: control bandwidth requirements
Matching actuator bandwidth to UAV vibration spectra for sub-pixel image stability
A stabilized gimbal on a UAV has one fundamental job: keep the sensor line of sight stationary in inertial space while the aircraft moves beneath it. The difficulty of that job depends almost entirely on the vibration environment the gimbal must reject. Small rotary-wing UAVs generate broadband vibration from 10 Hz to well above 500 Hz, with dominant peaks at the blade-pass frequency and its harmonics. Fixed-wing platforms are kinder, but propeller-driven variants still produce significant energy in the 20 to 200 Hz band. The actuator's control bandwidth determines how much of that vibration spectrum the servo loop can attenuate. If the bandwidth is too low, vibration passes through to the sensor and degrades image quality.
This article examines UAV vibration environments quantitatively, establishes the control bandwidth needed for specified levels of image stabilization, compares the achievable bandwidth of brushless DC (BLDC) geared drives versus piezoelectric direct drives, works through detailed jitter budgets for multiple scenarios, and explores the control architecture differences that make bandwidth matter.
Image: Nanomotion Ltd.
UAV vibration environments
Rotary-wing platforms
Multi-rotor UAVs generate vibration primarily from motor and propeller imbalance. The dominant vibration frequency is the propeller blade-pass frequency:
f_bp = (RPM x N_blades) / 60
For a typical small quad-rotor with motors spinning at 6,000 RPM and two-blade propellers, f_bp = 200 Hz. Higher harmonics appear at 400, 600, and 800 Hz, though with decreasing amplitude.
Vibration amplitude depends on propeller balance quality, motor bearing condition, and airframe structural dynamics. Typical values measured at the gimbal mounting point on well-maintained small multi-rotors:
| Frequency band | Angular vibration amplitude | Primary source |
|---|---|---|
| 2 to 5 Hz (pendulum mode) | 10 to 50 mrad peak | Wind gusts, attitude control |
| 5 to 20 Hz (airframe sway) | 5 to 20 mrad peak | Control loop interactions, turbulence |
| 20 to 100 Hz (structural modes) | 0.5 to 5 mrad peak | Arm bending, frame resonances |
| 100 to 300 Hz (blade pass + harmonics) | 0.1 to 2 mrad peak | Propeller imbalance, motor vibration |
| 300 to 1,000 Hz (high-frequency hash) | 0.01 to 0.2 mrad peak | Bearing noise, electronic noise |
The total integrated angular vibration at the gimbal mount on a quad-rotor in forward flight at moderate speed typically measures 10 to 30 mrad RMS across the 5 to 500 Hz band. In gusty conditions or during aggressive maneuvering, peaks can reach 50 mrad or more.
The vibration spectrum is not static. It shifts with flight condition:
| Flight condition | Dominant frequency shift | Amplitude change |
|---|---|---|
| Hover (no wind) | Motor RPM nominal | Baseline |
| Hover (15 kt wind) | Motor RPM varies +/- 10% | +6 to +10 dB broadband |
| Forward flight (10 m/s) | Motor RPM increases 5 to 15% | +3 to +6 dB below 50 Hz |
| Aggressive maneuver | Motor RPM varies +/- 20% | +10 to +20 dB broadband, transient |
| Descent (vortex ring) | Motor RPM oscillates erratically | +15 to +25 dB at 2 to 10 Hz |
These dynamic shifts mean the gimbal servo must be robust across a wide frequency range, not just tuned for a single operating point.
Fixed-wing platforms
Propeller-driven fixed-wing UAVs exhibit lower overall vibration, but the spectrum is concentrated at the propeller frequency and engine harmonics. A gasoline-powered pusher-prop configuration at 5,500 RPM with a two-blade propeller produces a blade-pass frequency of 183 Hz. Vibration at the gimbal mount (typically in the nose or belly) measures:
| Frequency band | Angular vibration amplitude | Primary source |
|---|---|---|
| 1 to 5 Hz (phugoid / short period) | 2 to 15 mrad peak | Atmospheric turbulence, flight dynamics |
| 5 to 20 Hz (aeroelastic modes) | 1 to 10 mrad peak | Wing bending, fuselage torsion |
| 20 to 80 Hz (fuselage bending) | 0.2 to 2 mrad peak | Higher structural modes |
| 80 to 250 Hz (engine + prop) | 0.5 to 3 mrad peak | Combustion, propeller imbalance |
| 250 to 500 Hz (harmonics) | 0.05 to 0.5 mrad peak | Engine firing harmonics |
Electric fixed-wing platforms are significantly quieter, with total integrated vibration often below 5 mrad RMS. The absence of combustion eliminates a major vibration source, leaving only propeller imbalance and aerodynamic buffet.
Image: Nanomotion Ltd.
Helicopter platforms
Single-rotor helicopters produce the most challenging vibration environment. The main rotor blade-pass frequency is low (typically 10 to 30 Hz), placing large-amplitude vibration squarely in the band where human-perceptible image motion occurs. A 4-blade rotor at 400 RPM produces f_bp = 26.7 Hz with amplitudes of 5 to 30 mrad at the fundamental and 1 to 10 mrad at the 2x and 4x harmonics. The tail rotor adds high-frequency content at 80 to 150 Hz.
The helicopter vibration environment is particularly challenging because:
- The dominant vibration at ~27 Hz is too high for soft isolation mounts (which would require impractically large deflections) yet too low for most BLDC geared servo systems to reject effectively.
- The Nx harmonics (2N, 3N, 4N of blade pass) create a "picket fence" of discrete tonal peaks, each requiring individual attenuation.
- Maneuver-dependent vibration (blade flapping during turns) is transient and unpredictable.
Vibration PSD summary table
The following table consolidates the power spectral density (PSD) levels at the gimbal mount for the three platform types, expressed as angular acceleration spectral density in (rad/s^2)^2/Hz:
| Frequency (Hz) | Quad-rotor PSD | Fixed-wing PSD | Helicopter PSD |
|---|---|---|---|
| 5 | 1.0e-1 | 5.0e-3 | 2.0e-1 |
| 10 | 5.0e-2 | 2.0e-3 | 1.0e+0 |
| 20 | 2.0e-2 | 1.0e-3 | 5.0e+0 |
| 50 | 5.0e-3 | 5.0e-4 | 1.0e-1 |
| 100 | 2.0e-2 | 3.0e-3 | 5.0e-2 |
| 200 | 5.0e-2 | 1.0e-2 | 1.0e-2 |
| 500 | 5.0e-4 | 1.0e-4 | 5.0e-3 |
Image stabilization requirements
The pixel-rate criterion
The fundamental measure of stabilization quality is residual angular jitter at the sensor, expressed in microradians (urad) RMS. This must be evaluated against the sensor's instantaneous field of view (IFOV) per pixel:
IFOV = pixel_pitch / focal_length
For common sensor and lens combinations used in UAV gimbals:
| Sensor | Pixel pitch (um) | Focal length (mm) | IFOV (urad) | 0.5 IFOV target (urad) |
|---|---|---|---|---|
| 1080p CMOS (1/2.8") | 2.9 | 25 | 116 | 58 |
| 4K CMOS (1") | 3.76 | 50 | 75.2 | 37.6 |
| HD CMOS (2/3") | 5.5 | 50 | 110 | 55 |
| LWIR 640x512 (17 um) | 17.0 | 50 | 340 | 170 |
| LWIR 640x512 (12 um) | 12.0 | 75 | 160 | 80 |
| MWIR 640x512 (15 um) | 15.0 | 100 | 150 | 75 |
| SWIR 640x512 (20 um) | 20.0 | 50 | 400 | 200 |
Acceptable image blur depends on the application:
- General surveillance: residual jitter less than 1 IFOV. Images appear stable to the operator, though fine detail is slightly softened.
- Target identification: residual jitter less than 0.5 IFOV. Detail is preserved for classification tasks.
- Precision geolocation: residual jitter less than 0.25 IFOV. Required when combining imagery with IMU and GPS data for target coordinate extraction.
- Long-range ISR with telephoto optics (200 mm focal length, IFOV = 27.5 urad for 5.5 um pixel): residual jitter less than 0.25 IFOV = 6.9 urad. This is extremely demanding and represents the state of the art.
Translating jitter requirements to bandwidth
The relationship between control bandwidth and residual jitter depends on the vibration power spectral density (PSD) and the gimbal servo's disturbance rejection transfer function. For a well-designed servo loop with approximately -20 dB/decade roll-off inside the bandwidth and -40 dB/decade outside, the disturbance rejection at frequency f for a servo with bandwidth f_bw is approximately:
Rejection(f) = (f / f_bw)^2, for f << f_bw
At frequencies well below the servo bandwidth, vibration is attenuated by this quadratic factor. At frequencies above the bandwidth, no attenuation occurs (and amplification can happen near the bandwidth frequency if damping is insufficient).
For practical estimation, the "isolation frequency," below which the servo provides at least 20 dB (10x) attenuation, is approximately f_bw / 3.2. To achieve 40 dB (100x) attenuation at a given frequency, the servo bandwidth must be approximately 10x that frequency.
Worked example 1: A quad-rotor gimbal must stabilize a 50 mm lens camera (5.5 um pixel, IFOV = 110 urad) to 55 urad RMS (0.5 IFOV). The dominant vibration at the mount is 15 mrad peak at 15 Hz (airframe sway) and 1 mrad peak at 200 Hz (blade pass). Converting peaks to RMS (divide by sqrt(2) for sinusoidal sources):
- 15 Hz component: 10.6 mrad RMS input. Required attenuation to bring contribution below 30 urad RMS: 10,600/30 = 353x, or about 51 dB. This requires f_bw of at least 15 Hz x 18.8 = 282 Hz.
- 200 Hz component: 0.71 mrad RMS input. Required attenuation to bring contribution below 30 urad RMS: 710/30 = 23.7x, or about 27 dB. This requires f_bw of at least 200 Hz x 4.9 = 980 Hz.
This example illustrates a critical point: achieving sub-pixel stability on a multi-rotor requires control bandwidth in the hundreds of hertz. The high-frequency blade-pass vibration, though lower in amplitude, demands high bandwidth because it is already near the noise floor and requires less attenuation.
In practice, passive vibration isolation (elastomeric mounts) is used to attenuate frequencies above 50 to 100 Hz by 10 to 20 dB, reducing the bandwidth requirement. A well-designed isolator + servo system might need 80 to 150 Hz servo bandwidth to meet the 0.5 IFOV target on a multi-rotor.
Worked example 2: A long-range ISR gimbal on a fixed-wing UAV must stabilize a 200 mm telephoto lens (5.5 um pixel, IFOV = 27.5 urad) to 6.9 urad RMS (0.25 IFOV). The vibration environment is gentler (5 mrad RMS total), but the jitter tolerance is 8x tighter than the quad-rotor case.
-
Primary disturbance: 5 mrad peak at 8 Hz (wing aeroelastic mode). RMS: 3.5 mrad.
-
Required attenuation: 3,500 / 5.0 = 700x (57 dB).
-
Required bandwidth for 57 dB at 8 Hz: f_bw = 8 x sqrt(700) = 8 x 26.5 = 212 Hz.
-
Secondary disturbance: 2 mrad peak at 183 Hz (blade pass). RMS: 1.4 mrad.
-
Required attenuation: 1,400 / 3.0 = 467x (53 dB).
-
Required bandwidth: 183 x sqrt(467) = 183 x 21.6 = 3,953 Hz.
The blade-pass component at 183 Hz would require nearly 4 kHz bandwidth to reject passively through the servo. This is impractical; passive isolation must attenuate the blade-pass vibration by at least 30 dB (a factor of 31.6) before the servo sees it. With isolation reducing the 183 Hz input from 1.4 mrad to 0.044 mrad, the servo requirement drops to:
- Required attenuation: 44 / 3.0 = 14.7x (23 dB).
- Required bandwidth: 183 x sqrt(14.7) = 183 x 3.83 = 701 Hz.
Even with excellent isolation, the telephoto lens application requires a servo bandwidth well above 200 Hz. This is firmly in piezo direct-drive territory and beyond what any geared BLDC system can achieve.
Actuator bandwidth: BLDC + gearbox versus piezo direct drive
BLDC + gearbox bandwidth limitations
The control bandwidth of a BLDC motor driving through a gearbox is limited by several factors:
-
Gearbox compliance: Planetary gearboxes have finite torsional stiffness, creating a resonant mode between the motor inertia and the load inertia. For miniature planetary gearboxes (22 to 32 mm diameter), torsional stiffness ranges from 0.5 to 5 Nm/rad. With typical motor and load inertias, this places the structural resonance at 50 to 150 Hz. The servo bandwidth must remain below this resonance (typically at 1/3 to 1/2 the resonant frequency) to avoid instability.
-
Backlash: Even precision planetary gearboxes exhibit 1 to 3 arcminutes of backlash. Harmonic drives reduce this to under 1 arcminute but introduce their own compliance. Backlash creates a dead zone in the control loop that limits effective bandwidth for small-signal disturbance rejection.
-
Reflected inertia: The gearbox multiplies the motor's rotor inertia by the square of the gear ratio when reflected to the output. For a 100:1 ratio, a motor with 5 gcm^2 rotor inertia presents 50,000 gcm^2 (0.005 kgm^2) at the output. This reflected inertia must be accelerated by the output torque, limiting dynamic response.
-
Current loop bandwidth: The electrical time constant of the motor winding limits how quickly current (and therefore torque) can change. For small BLDC motors, this is typically 0.5 to 2 ms, supporting current loop bandwidths of 500 Hz to 2 kHz, usually not the limiting factor.
-
Gear train friction and cogging: The periodic variation in gearbox friction (meshing frequency) and motor cogging torque introduce disturbances within the control loop. These disturbances are at frequencies related to the motor speed and gear tooth counts, and they effectively reduce the usable bandwidth by consuming servo authority.
In practice, BLDC + planetary gearbox gimbal systems achieve closed-loop bandwidths of 15 to 40 Hz. With harmonic drives, 30 to 60 Hz is typical. High-performance systems with carefully designed structural loops can reach 80 Hz, but this requires extensive tuning and is difficult to maintain across temperature and aging.
Piezo direct-drive bandwidth
Piezoelectric ultrasonic motors eliminate the gearbox entirely, removing the dominant bandwidth-limiting element. The motor output shaft connects directly to the gimbal axis, and the system bandwidth is limited by different factors:
-
Motor mechanical response: The piezoelectric stator's vibration amplitude (and therefore output torque) can be modulated at rates determined by the mechanical Q-factor of the resonator and the drive electronics bandwidth. For typical traveling-wave motors with Q of 500 to 2,000, the mechanical response bandwidth is f_resonance / (2Q), typically 15 to 80 Hz for the stator's amplitude envelope. However, the output shaft responds to changes in frictional drive force, which has a bandwidth determined by the contact dynamics, typically faster than the stator amplitude modulation.
-
Drive electronics: Modern piezo motor controllers use direct digital synthesis (DDS) to generate the drive signals and can modulate amplitude and phase at rates exceeding 10 kHz. The electronics are rarely the bandwidth bottleneck.
-
Structural compliance: Without a gearbox, the first structural resonance is determined by the motor-to-gimbal mounting stiffness and the load inertia. With rigid mounting (bolted flange, no intermediate coupling), structural resonances typically appear at 300 to 800 Hz, well above the desired servo bandwidth.
-
Friction nonlinearity: The friction-based drive mechanism introduces nonlinear behavior at very small amplitudes (below the stiction threshold). This can limit small-signal bandwidth, though modern controllers use dither techniques to mitigate this effect.
Practical piezo direct-drive gimbal systems achieve closed-loop bandwidths of 80 to 200 Hz. State-of-the-art implementations with high-speed feedback sensors (fiber-optic gyroscopes or MEMS gyros with >1 kHz bandwidth) and advanced control algorithms have demonstrated bandwidths exceeding 300 Hz in laboratory settings.
Bandwidth comparison summary
| Parameter | BLDC + planetary | BLDC + harmonic | Piezo direct drive |
|---|---|---|---|
| Typical bandwidth (Hz) | 15 to 40 | 30 to 60 | 80 to 200 |
| Best demonstrated (Hz) | 50 | 80 | 300+ |
| Bandwidth-limiting factor | Gearbox compliance | Harmonic drive compliance | Friction dynamics, structural modes |
| Small-signal linearity | Poor (backlash) | Good | Moderate (stiction) |
| Bandwidth degradation with temperature | 5 to 15% over -20 to +50 C | 3 to 10% | 10 to 25% (friction coefficient changes) |
| Bandwidth degradation with age | 5 to 20% (gear wear) | 3 to 10% (harmonic cup wear) | 10 to 30% (contact surface wear) |
Phase margin and stability considerations
The achievable bandwidth is ultimately limited by phase margin in the servo loop. A stable control system requires at least 30 degrees of phase margin (45 degrees preferred) at the gain crossover frequency (the effective bandwidth).
Each element in the signal chain contributes phase lag:
| Element | Phase lag at 100 Hz | Phase lag at 200 Hz |
|---|---|---|
| MEMS gyro (1 kHz BW) | 18 deg | 36 deg |
| MEMS gyro (4 kHz BW) | 4.5 deg | 9 deg |
| FOG (10 kHz BW) | 1.8 deg | 3.6 deg |
| Controller computation (0.2 ms) | 7.2 deg | 14.4 deg |
| Controller computation (0.05 ms) | 1.8 deg | 3.6 deg |
| Piezo motor response (2 ms) | 7.2 deg | 14.4 deg |
| BLDC current loop (1 ms) | 3.6 deg | 7.2 deg |
| Gearbox compliance (first resonance 80 Hz) | 90+ deg | unstable |
| Gearbox compliance (first resonance 200 Hz) | 25 deg | 90+ deg |
| Power amplifier (0.1 ms) | 3.6 deg | 7.2 deg |
For a BLDC + planetary gearbox system with an 80 Hz gearbox resonance and a 1 kHz MEMS gyro, the total phase lag at 50 Hz is approximately 18 + 7.2 + 3.6 + 40 + 3.6 = 72.4 degrees, leaving only about 108 degrees of phase margin. At 80 Hz, the gearbox resonance contributes 90+ degrees, making the loop unstable. Practical bandwidth is therefore limited to approximately 30 to 40 Hz.
For a piezo direct-drive system with a 4 kHz MEMS gyro and 0.05 ms controller, the total phase lag at 150 Hz is approximately 6.8 + 2.7 + 10.8 + 5.4 = 25.7 degrees, leaving 154 degrees of phase margin. This is comfortable; the bandwidth can be pushed to 200 Hz or beyond before phase margin becomes the constraint.
Jitter budget analysis
A jitter budget decomposes the total residual line-of-sight error into contributions from each source, verifying that the sum meets the stabilization requirement.
Budget 1: LWIR gimbal on a quad-rotor
640x512 LWIR sensor, 17 um pixel pitch, 50 mm focal length, IFOV = 340 urad. Target: 0.5 IFOV (170 urad) residual jitter.
| Error source | Allocation (urad RMS) | Notes |
|---|---|---|
| Residual vibration (after servo rejection) | 100 | Dominant contributor; depends on servo bandwidth |
| Gyro noise (angle random walk) | 30 | For tactical MEMS gyro, 0.3 deg/rt-hr ARW |
| Encoder quantization | 15 | 17-bit encoder, 2.7 urad/LSB |
| Structural resonance excitation | 40 | Due to amplification near servo bandwidth |
| Cable torque disturbance | 25 | Flex cable restoring torque, partially rejected by servo |
| Bearing friction variation | 20 | For BLDC; near zero for piezo (no bearings on output) |
| Thermal drift (over 10-second window) | 30 | Structural and sensor thermal effects |
| Cross-axis coupling | 15 | Elevation disturbance coupling into azimuth |
| Wind load torque residual | 20 | Aerodynamic force on gimbal housing |
| RSS total | 128 | Below 170 urad allocation |
This budget has 42 urad of margin. With a BLDC + gearbox system at 25 Hz bandwidth, the residual vibration line item would increase to approximately 300 urad (because the lower bandwidth provides less attenuation of the 100 to 200 Hz vibration content), blowing the budget by nearly 2x.
Budget 2: HD daylight camera with 100 mm lens on a quad-rotor
HD CMOS sensor, 5.5 um pixel pitch, 100 mm focal length, IFOV = 55 urad. Target: 0.5 IFOV (27.5 urad) residual jitter. This is a demanding application.
| Error source | Piezo allocation (urad RMS) | BLDC allocation (urad RMS) | Notes |
|---|---|---|---|
| Residual vibration (servo) | 12 | 120 | Piezo at 150 Hz BW; BLDC at 30 Hz BW |
| Gyro noise (ARW) | 8 | 8 | High-grade MEMS, 0.08 deg/rt-hr |
| Encoder quantization | 3 | 3 | 19-bit encoder |
| Structural resonance excitation | 8 | 25 | Piezo: higher structural freq; BLDC: gearbox resonance |
| Cable torque disturbance | 6 | 15 | Piezo: better rejection at cable drag frequency |
| Bearing friction variation | 2 | 10 | Piezo: no output bearing friction |
| Thermal drift (10 s) | 5 | 5 | Both experience similar thermal effects |
| Cross-axis coupling | 4 | 8 | Piezo: tighter control reduces coupling |
| Wind load torque residual | 5 | 12 | Piezo: better rejection bandwidth |
| IMU alignment error | 3 | 3 | Mechanical alignment tolerance |
| Optical element vibration | 4 | 4 | Lens internal resonances |
| RSS total | 19 | 126 |
The piezo system meets the 27.5 urad requirement with margin. The BLDC system fails by 4.6x, even with a good-quality gyro. The residual vibration term dominates the BLDC budget because the 30 Hz servo bandwidth cannot adequately reject the vibration content between 30 and 200 Hz.
Budget 3: MWIR cooled sensor with 150 mm lens on a helicopter
This represents one of the most challenging stabilization scenarios. 640x512 MWIR sensor, 15 um pixel pitch, 150 mm focal length, IFOV = 100 urad. Target: 0.25 IFOV (25 urad) residual jitter.
| Error source | Allocation (urad RMS) | Actuator requirement |
|---|---|---|
| Residual main rotor vibration (27 Hz) | 8 | Servo must reject 15 mrad at 27 Hz by 63 dB |
| Residual 2N vibration (53 Hz) | 5 | Servo must reject 5 mrad at 53 Hz by 60 dB |
| Residual 4N vibration (107 Hz) | 4 | Servo must reject 2 mrad at 107 Hz by 54 dB |
| Residual tail rotor vibration (120 Hz) | 3 | Servo must reject 1 mrad at 120 Hz by 50 dB |
| Gyro noise (ARW) | 5 | FOG required: 0.01 deg/rt-hr |
| Encoder quantization | 2 | 21-bit encoder |
| Structural resonance excitation | 6 | Resonance must be above 400 Hz |
| Feed-forward compensation residual | 4 | Feed-forward on 1N and 2N |
| Cable torque disturbance | 4 | Low-stiffness cable with active compensation |
| Thermal drift (10 s) | 3 | Active temperature control |
| Cross-axis coupling | 3 | Decoupled gimbal axes |
| Cooler microphonics | 5 | Stirling cooler vibration at 60 Hz |
| RSS total | 16 | Below 25 urad allocation |
The 63 dB rejection at 27 Hz requires a servo bandwidth of: 27 x 10^(63/40) = 27 x 10^1.575 = 27 x 37.6 = 1,015 Hz. This is only achievable with a piezo direct drive and a fiber-optic gyroscope; it is physically impossible with any geared BLDC system. Even with feed-forward compensation providing 20 dB of additional rejection at the 1N frequency, the servo must still provide 43 dB, requiring f_bw = 27 x 10^(43/40) = 27 x 14.1 = 381 Hz. This remains beyond geared BLDC capability.
Worked example: specific camera/lens combination
Consider a complete worked analysis for a specific, commonly fielded sensor package: a 640x480 uncooled LWIR microbolometer with 25 mm f/1.0 lens, paired with a 1080p CMOS daylight camera with 75 mm f/2.8 lens, mounted on a two-axis gimbal on a 5 kg hex-rotor UAV.
Sensor parameters:
| Parameter | LWIR channel | Daylight channel |
|---|---|---|
| Pixel pitch | 17 um | 2.9 um |
| Focal length | 25 mm | 75 mm |
| IFOV | 680 urad | 38.7 urad |
| 0.5 IFOV target | 340 urad | 19.3 urad |
| Array size | 640 x 480 | 1920 x 1080 |
| FOV | 24.9 x 18.7 deg | 4.2 x 2.4 deg |
The daylight channel with its longer focal length is the limiting case. 19.3 urad is a very tight jitter requirement.
Vibration environment (hex-rotor at 5,400 RPM, 3-blade props):
f_bp = 5400 x 3 / 60 = 270 Hz
| Frequency | Amplitude at gimbal mount |
|---|---|
| 8 Hz (airframe pendulum) | 12 mrad peak |
| 22 Hz (arm bending mode) | 3 mrad peak |
| 65 Hz (frame torsion) | 0.8 mrad peak |
| 270 Hz (blade pass) | 0.4 mrad peak |
| 540 Hz (2x blade pass) | 0.08 mrad peak |
Vibration isolation: Two-axis elastomeric isolator with 15 Hz corner frequency, Q = 3.
Post-isolation vibration:
| Frequency | Isolation factor | Post-isolation amplitude |
|---|---|---|
| 8 Hz | 1.8x amplification (below corner) | 21.6 mrad peak |
| 22 Hz | 0.46x (2.2x corner ratio) | 1.38 mrad peak |
| 65 Hz | 0.053x (4.3x corner ratio squared) | 0.042 mrad peak |
| 270 Hz | 0.0031x | 0.0012 mrad peak |
| 540 Hz | 0.00077x | 0.000062 mrad peak |
Servo rejection requirement (for 19.3 urad total, allocating 12 urad to the vibration term):
| Frequency | Post-isolation (mrad RMS) | Required attenuation | Required BW (Hz) |
|---|---|---|---|
| 8 Hz | 15.3 | 15,300 / 8 = 1,913x (66 dB) | 8 x 43.7 = 350 |
| 22 Hz | 0.98 | 980 / 4 = 245x (48 dB) | 22 x 15.7 = 345 |
| 65 Hz | 0.030 | 30 / 3 = 10x (20 dB) | 65 x 3.2 = 208 |
| 270 Hz | 0.00085 | < 1x (already below noise floor) | N/A |
The binding requirement is approximately 350 Hz servo bandwidth. The isolator has eliminated the blade-pass problem but amplified the low-frequency content, making the 8 Hz pendulum mode the dominant challenge.
Solution assessment:
- BLDC + harmonic drive at 50 Hz bandwidth: residual vibration at 8 Hz = 15.3 mrad x (8/50)^2 = 15.3 x 0.0256 = 392 urad. Fails by 33x.
- Piezo direct drive at 150 Hz bandwidth: residual at 8 Hz = 15.3 x (8/150)^2 = 15.3 x 0.00284 = 43.5 urad. Fails by 3.6x.
- Piezo direct drive at 350 Hz bandwidth: residual at 8 Hz = 15.3 x (8/350)^2 = 15.3 x 0.000522 = 8.0 urad. Meets requirement.
This analysis demonstrates that the 75 mm daylight channel drives the system to a 350 Hz bandwidth requirement, achievable only with a state-of-the-art piezo direct-drive system using a high-bandwidth MEMS gyro and advanced control algorithms. The LWIR channel, with its 340 urad tolerance, is easily served by even a modest 50 Hz BLDC system.
Control architecture considerations
Rate loop versus position loop
Gimbal stabilization systems typically use a cascaded control architecture:
- Inner rate loop: uses gyroscope feedback to stabilize angular rate. This loop rejects vibration and disturbances. Its bandwidth determines stabilization performance.
- Outer position loop: uses encoder feedback to maintain commanded pointing angle. This loop is slower (typically 5 to 20 Hz bandwidth) and handles tracking commands.
For the inner rate loop, the gyroscope bandwidth must exceed the desired servo bandwidth by at least 3x to avoid introducing excessive phase lag. For a 150 Hz servo bandwidth, the gyro must have useful bandwidth to at least 450 Hz. For a 350 Hz servo bandwidth, the gyro needs bandwidth to 1,050 Hz or better.
MEMS gyroscopes suitable for high-bandwidth gimbal stabilization:
| Gyroscope | Bandwidth (Hz) | ARW (deg/rt-hr) | Bias stability (deg/hr) | Mass (g) | Suitable for BW up to (Hz) |
|---|---|---|---|---|---|
| Analog Devices ADXRS646 | 2,000 | 0.04 | 2.0 | 1.5 | 660 |
| TDK InvenSense ICM-42688 | 4,000 | 0.08 | 5.0 | 0.5 | 1,300 |
| Epson M-G370 | 2,000 | 0.06 | 3.0 | 2.0 | 660 |
| Sensonor STIM318 (3-axis) | 1,500 | 0.15 | 0.5 | 55 | 500 |
| KVH DSP-1760 (FOG) | 10,000 | 0.01 | 0.1 | 320 | 3,300 |
The choice of gyroscope often determines the practical bandwidth ceiling. A low-cost MEMS gyro with 500 Hz bandwidth limits the achievable servo bandwidth to approximately 160 Hz, regardless of the actuator's capability.
Feed-forward disturbance compensation
An increasingly common technique augments the feedback loop with feed-forward compensation using accelerometer data from the airframe. The accelerometers sense vibration at the gimbal base, and the controller generates a compensating torque command before the vibration propagates to the line of sight. This technique can provide an additional 10 to 20 dB of rejection at known frequencies (such as the blade-pass frequency) without requiring higher servo bandwidth.
Feed-forward works best with fast actuators that can respond to the compensation signal with minimal delay. Piezo motors, with their fast torque response and no gearbox backlash, are better suited to feed-forward control than BLDC + gearbox systems. The effectiveness of feed-forward depends on:
| Parameter | Effect on feed-forward performance |
|---|---|
| Accelerometer noise floor | Sets the minimum disturbance detectable; below 1 mg/rt-Hz needed |
| Actuator delay (transport lag) | Must be < 1/(4 x f_disturbance) for constructive cancellation |
| Transfer function accuracy | Model errors above 3 dB reduce cancellation to < 6 dB |
| Frequency tracking accuracy | For variable-speed props, must track within 2% of actual frequency |
For a piezo motor with 0.5 ms transport lag, the maximum frequency for effective feed-forward is approximately 1/(4 x 0.0005) = 500 Hz. For a BLDC + gearbox system with 5 ms effective transport lag (due to backlash and gearbox compliance), the limit drops to 50 Hz, making feed-forward ineffective against blade-pass frequencies above 50 Hz.
Advanced control algorithms
Beyond classical PID control, several advanced algorithms are used in high-performance gimbal systems:
H-infinity robust control: Optimizes worst-case disturbance rejection across a specified frequency range. Particularly effective for multi-rotor platforms where the vibration spectrum shifts with flight condition. Typically improves disturbance rejection by 3 to 6 dB compared to PID at the same bandwidth, at the cost of higher computational load.
Repetitive control: Adds a periodic signal model to the controller to achieve very high rejection at specific harmonics. Effective against blade-pass frequencies and their harmonics. Can achieve 40+ dB rejection at the target frequency with only modest servo bandwidth. Requires accurate knowledge of the disturbance frequency.
Adaptive notch filtering: Automatically tracks the dominant vibration frequency (which shifts with motor RPM) and places a deep notch in the sensitivity function. Reduces residual vibration at the blade-pass frequency by 20 to 30 dB. Combined with feed-forward, can achieve 40+ dB total rejection.
These advanced algorithms are most effective with piezo direct-drive actuators because the absence of gearbox nonlinearities (backlash, friction variation, compliance) allows the controller to operate closer to the theoretical performance limits.
Field performance data
Published and conference-reported stabilization performance numbers from operational gimbal systems illustrate the bandwidth advantage in practice:
| System type | Actuator | Bandwidth (Hz) | Platform | Residual jitter (urad RMS) | Sensor IFOV (urad) | Jitter/IFOV ratio |
|---|---|---|---|---|---|---|
| Commercial mini-gimbal | BLDC + planetary | 20 | Quad-rotor | 200 to 400 | 340 (LWIR) | 0.6 to 1.2 |
| Military small gimbal | BLDC + harmonic | 45 | Fixed-wing | 50 to 100 | 110 (visible) | 0.5 to 0.9 |
| Military micro-gimbal | Piezo direct drive | 120 | Quad-rotor | 30 to 80 | 110 (visible) | 0.3 to 0.7 |
| High-performance ISR | Piezo + voice coil hybrid | 200 | Helicopter | 15 to 40 | 55 (telephoto) | 0.3 to 0.7 |
| Lab demonstration | Piezo direct drive | 350 | Shaker table | 5 to 15 | 28 (telephoto) | 0.2 to 0.5 |
| DIRCM laser pointer | Piezo direct drive | 250 | Helicopter | 20 to 50 | N/A (laser) | N/A |
The progression is clear: higher bandwidth enables tighter stabilization, and piezo direct drive consistently achieves higher bandwidth than geared BLDC systems in the same size class.
When BLDC still wins
For completeness, there are gimbal applications where the BLDC + gearbox solution provides adequate bandwidth and the piezo motor's higher bandwidth is not needed:
- Large fixed-wing UAVs with low vibration and long-focal-length sensors where the IFOV is large (>500 urad). A 30 Hz bandwidth servo meets the jitter requirement with margin.
- Pan-only turrets where only azimuth pointing is stabilized and the elevation axis is locked. Single-axis simplifies the control problem.
- Very high slew rate requirements (>500 deg/s) where the BLDC motor's higher speed capability is needed for rapid repointing.
- Extreme cold environments (below -40 C) where piezo motor friction characteristics degrade significantly while BLDC motors with appropriate lubricants maintain performance.
- Applications with long duty cycles (>16 hours per day continuous scanning) where piezo motor wear life may be insufficient and BLDC motor life (50,000+ hours) provides more margin.
Even in these cases, the bandwidth headroom provided by piezo direct drive is valuable as a performance margin, allowing the system to maintain stabilization quality as components age or environmental conditions deteriorate.
Conclusion
Control bandwidth is the single most important actuator parameter for gimbal stabilization performance. On small UAV platforms, where vibration environments are harsh and payload mass is limited, the 3 to 5x bandwidth advantage of piezo direct drive over BLDC + gearbox translates directly into 3 to 5x better stabilization (measured as residual jitter reduction, due to the quadratic relationship between bandwidth and vibration rejection). For demanding applications requiring sub-pixel stability with compact sensors, this advantage is not incremental; it is enabling. The combination of high bandwidth, low mass, and low power consumption makes piezo direct drive the actuator of choice for modern small UAV stabilized payloads.
The jitter budget methodology presented here provides a quantitative framework for evaluating whether a given actuator technology can meet a specific stabilization requirement. The key steps are: characterize the vibration environment, define the sensor's jitter tolerance (in IFOV fractions), design the isolation system, compute the residual vibration the servo must reject, and verify that the actuator's achievable bandwidth provides sufficient rejection. When the numbers are run honestly, the bandwidth threshold that separates "adequate" from "inadequate" stabilization almost always falls between 60 and 400 Hz, placing it squarely in the domain where piezo direct drive outperforms geared BLDC by a wide margin.