Inertial Device Noise Testing: Gyroscope and Accelerometer Noise Characterization

Inertial device noise testing quantifies the stochastic error characteristics of gyroscopes and accelerometers — the parameters that ultimately limit the accuracy of inertial navigation, stabilization, and guidance systems. Using Allan variance analysis, power spectral density estimation, and thermally controlled static test procedures in accordance with IEEE Std 952, IEEE Std 1293, and related standards, accredited third-party laboratories can measure angle random walk, bias instability, rate random walk, and other noise coefficients to validate sensor performance and support system-level design.

Technical Overview

Inertial devices, including gyroscopes and accelerometers, are the core sensing elements of inertial navigation systems (INS), attitude and heading reference systems (AHRS), and platform stabilization systems. Unlike deterministic errors such as scale factor error or axis misalignment, stochastic noise cannot be removed by simple calibration: it accumulates in the navigation solution over time and ultimately defines the drift behavior of the entire system. Inertial device noise testing therefore focuses on identifying, separating, and quantifying the random error components of the sensor output, including white noise, flicker (1/f) noise, random walk processes, quantization noise, and exponentially correlated noise.

The dominant methodology is time-domain Allan variance analysis, complemented by frequency-domain power spectral density (PSD) estimation. The principal reference documents are IEEE Std 952 (specification format guide and test procedure for single-degree-of-freedom gyroscopes), IEEE Std 1293 (linear single-axis nongyroscopic accelerometers), IEEE Std 647 (fiber optic gyroscopes), IEEE Std 1554 (inertial sensor test equipment, instrumentation, and data acquisition), and IEEE Std 528 (inertial sensor terminology). Testing is performed under controlled thermal conditions on vibration-isolated foundations, typically by accredited third-party laboratories with traceable calibration chains.

The principal noise characteristics extracted during inertial device noise testing include:

• Angle random walk (ARW) — white noise in the angular rate output, expressed in °/√h;
• Bias instability (BI) — the flicker noise floor of the output, expressed in °/h for gyroscopes and µg for accelerometers;
• Rate random walk (RRW) — the random walk process driving the rate output, expressed in °/√h³;
• Velocity random walk (VRW) — white noise in the accelerometer output, expressed in m/s/√h or equivalently µg/√Hz;
• Quantization noise, rate ramp, exponentially correlated (Markov) noise, and sinusoidal noise components.

Testing Samples

MEMS Gyroscopes and Accelerometers

Consumer-, automotive-, industrial-, and tactical-grade MEMS inertial sensors, either as discrete single-axis components or as integrated 6-axis and 9-axis modules. ARW values range from several °/√h for consumer-grade parts down to below 0.01 °/√h for high-end tactical devices, so the test duration, data resolution, and environmental control must be scaled to the noise floor of the specific grade.

Fiber Optic Gyroscopes (FOG)

Open-loop and closed-loop interferometric fiber optic gyroscopes used in navigation- and tactical-grade applications, with bias instability targets ranging from a few °/h down to below 0.001 °/h. Resolving such low flicker floors requires long-duration static data acquisition in a highly stabilized thermal environment.

Ring Laser Gyroscopes (RLG)

Mechanically dithered ring laser gyroscopes used in aviation and marine navigation. Noise characterization covers random walk coefficients, lock-in-related effects, dither injection noise, and residual bias instability under static and thermally varying conditions.

Quartz Flexure and Pendulous Accelerometers

Precision force-rebalance accelerometers with bias instability in the µg range and correspondingly low velocity random walk. Testing emphasizes long static records, multiple gravity-referenced orientations, and temperature-dependent noise behavior.

Inertial Measurement Units and Inertial Navigation Systems

Integrated IMUs, AHRS units, and complete INS products at board, module, or sealed-enclosure level. System-level noise testing verifies the combined contribution of the sensors, front-end electronics, ADC quantization, and internal digital filtering to the overall stochastic error model.

Testing Items

Depending on the device type and grade, a typical noise test program reports the following items, each derived from the Allan deviation curve and the noise power spectral density:

• Angle random walk (ARW), °/√h — gyroscopes;
• Velocity random walk (VRW), m/s/√h or µg/√Hz — accelerometers;
• Bias instability (BI), °/h or µg — the flat minimum region of the Allan deviation curve;
• Rate random walk (RRW) and acceleration random walk, °/√h³ and m/s/√h/√h respectively;
• Quantization noise coefficient, derived from the −1 slope region;
• Rate ramp / acceleration ramp (deterministic drift components);
• Exponentially correlated (first-order Markov) noise time constant and amplitude;
• Sinusoidal noise components (amplitude and frequency of periodic interference);
• Noise power spectral density over the operating bandwidth, (°/s)/√Hz or µg/√Hz;
• Temperature dependence of bias instability and random walk coefficients across the operating range.

Testing Methods

Static Allan Variance Test Procedure

The device under test is mounted on a precision, vibration-isolated rate table installed on a stable foundation, typically inside a thermal chamber. The test sequence follows IEEE Std 952 and IEEE Std 1293 practice:

• Mounting and alignment of the sensitive axes with respect to gravity and the local vertical, using defined orientations (single-axis up, or multi-position schemes);
• Thermal soak and electrical warm-up until the bias has stabilized, exceeding the manufacturer-specified warm-up time;
• Continuous acquisition of the raw, unfiltered output at a sufficient rate (commonly 100 Hz to 10 kHz) with high-resolution data acquisition hardware and synchronized time stamping;
• Recording duration of typically 4 to 24 hours, ensuring the record length is at least ten to one hundred times the longest correlation time of interest so that the bias instability minimum is clearly resolved;
• Data screening to remove start-up transients, saturation events, and external disturbances, taking care that detrending does not mask rate random walk;
• Computation of Allan deviation over cluster times spanning from roughly ten sample periods up to one tenth of the total record length.

Allan Deviation Curve Interpretation

In a log-log plot of Allan deviation versus cluster time, each noise process occupies a characteristic slope region, allowing the individual coefficients to be separated:

• Slope −1 at the shortest cluster times: quantization noise;
• Slope −1/2: white noise, giving ARW as σ(τ)/√τ for gyroscopes and VRW for accelerometers;
• Slope 0 (the flat minimum): bias instability, approximately the minimum value divided by 0.664;
• Slope +1/2: rate random walk, obtained from σ(τ)·√(3/τ);
• Slope +1: rate ramp, a deterministic drift;
• Local bumps or ripple: exponentially correlated (Markov) noise or periodic (sinusoidal) interference.

Temperature-Dependent Noise Testing

The static noise test is repeated at multiple temperature soak points, commonly from −40 °C to +85 °C, following thermal stabilization at each point. Controlled temperature ramps at defined rates (for example 0.5 to 2 °C/min) are additionally used to quantify noise and bias behavior under thermal transients, and repeated up-and-down profiles reveal thermal hysteresis.

Complementary Analysis Methods

Power spectral density estimation (for example Welch or periodogram methods) verifies the noise distribution in the frequency domain and identifies interference lines at power frequencies or dither frequencies. Autocorrelation analysis and wavelet-based variance estimation may be used to cross-check correlated noise components. For applications with strong dynamic environments, noise testing can be extended with shaker excitation to assess vibration-induced noise and vibration rectification behavior.

Equipment and Traceability

Testing is conducted with precision rate and positioning tables featuring sub-arcsecond indexing accuracy, environmentally controlled thermal chambers, low-noise power supplies, and high-resolution (24-bit or better) data acquisition systems. All measurement equipment is calibrated with traceability to national or international standards, and laboratories operating under ISO/IEC 17025 accreditation provide documented quality assurance of the results.

Applications

Noise test results directly feed the error models used in Kalman filter design, aiding sensor selection, and navigation performance prediction. Key application areas include:

• Aerospace and defense: inertial navigation for aircraft, missiles, guided munitions, and munitions fuze safety analysis;
• Satellite and launch vehicle attitude control and precision pointing;
• Marine navigation, shipborne stabilization, and underwater vehicle guidance;
• Autonomous driving and advanced driver assistance systems (ADAS);
• Unmanned aerial vehicles (UAV) and aerial surveying platforms;
• Oil and gas measurement while drilling (MWD), downhole surveying, and mining or tunneling guidance;
• High-precision surveying and geodesy with GNSS/INS integration;
• Robotics, structural health monitoring, and platform stabilization systems.

FAQ

Why is Allan variance preferred over a simple standard deviation for noise evaluation?

A standard deviation depends strongly on the averaging time and conflates all noise processes into one number. Allan variance characterizes noise as a function of cluster time, which allows white noise, flicker noise, random walk, and quantization effects to be separated and individually quantified as physically meaningful coefficients.

How long must the static data record be?

The record must be long enough for the longest correlation time of interest to be sampled repeatedly. As a practical rule, the record length should be at least ten times — preferably one hundred times — the cluster time at which the bias instability minimum appears. For navigation-grade devices this typically means 12 to 24 hours or more of continuous data; for lower-grade MEMS devices several hours may suffice.

Should raw or filtered output data be used?

Raw, unfiltered output must be acquired. Internal or external low-pass filtering modifies the noise power spectral density and distorts the Allan deviation curve, leading to optimistic or misleading random walk and bias instability values.

At what temperatures is noise testing performed?

A baseline test at controlled room temperature is always performed, followed by soak points across the operating range (for example −40 °C to +85 °C) and optionally controlled temperature ramps, so that the temperature dependence of each noise coefficient can be reported.

What deliverables are provided at the end of a test?

A typical report includes the test configuration and environmental conditions, the Allan deviation log-log curve with the fitted noise coefficients (ARW/VRW, bias instability, RRW, quantization, and correlated noise terms), power spectral density plots, temperature-dependence results, and observations on data quality and repeatability.

How should a testing laboratory be selected?

Choose an accredited third-party laboratory with ISO/IEC 17025 accreditation, vibration-isolated facilities, traceable rate table and thermal chamber equipment, and demonstrated experience executing noise characterization programs in accordance with IEEE Std 952, IEEE Std 1293, and the related inertial sensor test standards.