MEMS Inertial Measurement Unit

Legacy context

The documented heritage of this site traces back to the Applied Analytics Position and Attitude Data System (PADS) and the GPS Inertial Data Simulator (GIDS). PADS was engineered as a high-accuracy, near real-time airborne direct georeferencing system, integrating a Kearfott KI-4901 Inertial Measurement Unit with GPS observations through a Multi-State Kalman filter. GIDS complemented this by providing a bench-top simulation environment, generating real-time GPS and IMU data streams to test systems without physical movement.

This lineage established a core competency in the precise integration of inertial sensors with satellite navigation. The foundational work on the Real Time Navigator (RTN) software, which provided explicit interfaces for any IMU and GPS hardware, reflects a long-standing focus on sensor fusion and system flexibility.

That engineering focus naturally extends to the modern landscape of the mems inertial measurement unit. The transition from the larger, high-end units used in early PADS configurations to the compact, low-power MEMS designs represents a significant evolution in the field. The principles of calibration, alignment, and data integration developed for legacy systems remain directly applicable to the challenges and opportunities presented by contemporary MEMS-based inertial technology.

Scale of the Problem: What MEMS IMUs Actually Provide

Micro-Electro-Mechanical Systems (MEMS) technology has made inertial position and attitude estimation "affordable and increasingly accurate as sensors have become smaller and cheaper" [6]. That affordability comes with a specific performance envelope that navigation and avionics engineers must quantify before selecting or integrating a unit. The key numbers to internalize are not marketing figures but operational parameters drawn from flight-tested systems.

A MEMS inertial measurement unit (IMU) typically contains three orthogonal accelerometers and three orthogonal gyroscopes, though redundant configurations exist. A redundant IMU (RIMU) has more than three active sense axes; three-dimensional body rates are mapped to those axes through a 3×n body-to-gyro matrix, and a three-dimensional measurement is recovered using the pseudo-inverse of that matrix [2]. That redundancy creates a null space—information present in the n measurements that is not observable in the computed three-dimensional result—which means physical calibration parameters are not fully observable without special handling [2]. For avionics work, this is not an abstract concern: an Extended Kalman Filter (EKF) designed for calibration of a three-axis IMU must account for this null space, or the calibration will silently diverge [2].

The Numbers That Matter: Bias, Scale Factor, and Misalignment

The dominant error sources in a MEMS IMU are accelerometer bias, scale factor error, and non-orthogonality (misalignment). These are not abstract concepts; they are explicit terms in the measurement compensation equation used in flight software. The compensated accelerometer measurement in the inertial frame is computed as:

Δv̂ᵢ = Tᵢ_c [(I₃ − Ξ̂ₐ − Ŝₐ) Δv_c − b̂ₐ]

where b̂ₐ is the estimated accelerometer bias vector, Ŝₐ is the diagonal scale factor matrix, and Ξ̂ₐ is the skew-symmetric misalignment matrix populated by terms such as ξ_xy, ξ_xz, ξ_yx, ξ_yz, ξ_zx, ξ_zy [4]. This formulation, used in the Orion Exploration Flight Test 1 navigation analysis, shows that the raw velocity increment Δv_c must be corrected for all three error classes before it is accumulated in the inertial frame [4].

The practical consequence: a MEMS accelerometer with a bias of, say, 1 mg (0.001 g) will produce a velocity error of approximately 0.0098 m/s after one second, but that error grows linearly with time in velocity and quadratically in position. Scale factor errors, typically expressed in parts per million (ppm), multiply the measured acceleration and therefore produce errors proportional to the actual dynamic acceleration. Misalignment terms, expressed in microradians or arcseconds, couple acceleration along one axis into another. All three must be estimated and compensated at a rate commensurate with the platform dynamics—the Orion filter used 1 Hz estimates of bias, scale factor, and misalignment [4].

Calibration Procedures and Their Time Scales

Calibration is not a one-time factory event; it is an ongoing operational procedure with specific durations and orientations. In the Shuttle Avionics Integration Lab testing of a high-accuracy inertial navigation system, the IMUs were placed in two orientations differing by a 90-degree rotation about the up axis [7]. Data was collected for 90 seconds by the accelerometers to remove misalignment due to reorientation—the accelerometers were used because their accuracy is much better than the resolvers, and the acceleration due to Earth's rotation is precisely known, so any unexpected acceleration is attributable to IMU misalignment [7].

After misalignment was nulled, the platform was torqued about the north axis to compensate for Earth's rotation, and data was collected for ten minutes to measure platform drifts [7]. This sequence was repeated at the second orientation, and relative attitude errors for each IMU pair were computed using resolver data [7]. These time scales—90 seconds for alignment verification, ten minutes for drift measurement—are not arbitrary; they reflect the physical reality that gyro drift rates on the order of Earth's rotation rate (approximately 15 degrees per hour) require integration times long enough to separate systematic drift from noise.

For a MEMS-grade unit, the same logic applies but with different magnitudes. A MEMS gyro with a bias stability of, say, 10 degrees per hour will require a calibration integration time long enough to average out angle random walk, which is typically specified in degrees per root-hour. The ten-minute drift measurement used in the Shuttle program would need to be extended for a MEMS unit with higher noise density, or the calibration uncertainty will be dominated by random walk rather than by the systematic bias you are trying to estimate.

Integration Architectures: Where MEMS Fits

MEMS IMUs are rarely used alone in avionics; they are typically fused with GPS or other aiding sensors through a Kalman filter. The inertial sensor is unique among navigation sensors in that it produces data "solely dependent on the motion of the platform and local gravity" and is "more or less unaffected by its surroundings" [5]. This makes the inertial sensor the natural primary sensor in an integrated architecture, with GPS, laser scanners, and cameras providing corrections through a complementary Kalman filter that estimates inertial errors and feeds them back to the strapdown calculations [5].

The tutorial literature describes a simple two-dimensional case where attitude estimates derived from sensor data are used in dead-reckoning position and velocity estimation, with GPS updates through a Kalman filter [6]. The key architectural point for avionics engineers: the MEMS IMU provides the high-rate propagation (typically 100 Hz to 1 kHz), while GPS provides the low-rate absolute correction (typically 1 Hz to 10 Hz). The Kalman filter's role is to estimate the inertial error states—bias, scale factor, misalignment—and feed them back so that the strapdown integration does not accumulate unbounded error between GPS updates.

Practical Limits and Design Trade-offs

The performance limits of a MEMS IMU are set by physics, not by software. The accelerometer bias and scale factor errors must be calibrated to the level required by the mission's navigation accuracy budget. For a low-cost miniature accelerometer array fused with GPS, the estimator must be designed to handle the higher noise and larger biases typical of MEMS devices [8]. The trade-off is qualitative but real: lower-cost MEMS units have higher noise and bias, requiring more frequent aiding and more sophisticated filtering, while higher-grade units (fiber optic or ring laser gyros) have better raw performance but at higher cost, size, and power [8].

For avionics engineers, the practical rule is: specify the MEMS IMU's bias, scale factor, and misalignment requirements from the navigation error budget, then verify those parameters through calibration procedures with known durations and orientations. The 90-second and ten-minute calibration intervals from the Shuttle program [7] provide a starting point, but the actual numbers must be derived from the specific MEMS unit's noise characteristics and the mission's accuracy requirements. The compensation equation from Orion [4] provides the mathematical framework; the calibration observability analysis from the RIMU work [2] reminds you that redundant axes require null-space updates for full observability. No single number defines a "good" MEMS IMU—the numbers only have meaning in the context of your integration architecture and error budget.