All of the filters mentioned above are “estimation filters” (EF). When talking about estimation filters, one can quickly get mired in alphabet soup.
A Kalman Filter (KF) is a linear quadratic estimation algorithm that operates recursively on noisy data and produces an estimate of a system’s current state that is statistically more precise than what a single measurement could produce.
An Extended Kalman Filter (EKF) is used generically to describe any estimation filter based on the Kalman Filter model that can handle non-linear elements. Almost all inertial estimation filters are fundamentally EKFs.
An Adaptive Kalman Filter (AKF), technically speaking, is also an EKF but it contains a high dependency on “adaptive” elements. “Adaptive” technology refers to the ability of a filter to selectively trust a given measurement more or less based on a “trust” threshold when compared to another measurement that is used as a reference. The 3DM GX4-25 and -15 rely on adaptive control elements to improve their estimations and hence we refer to the estimation filter used in those devices as an “AKF”. Technically speaking it is an “EKF with heavy reliance on adaptive elements” or possibly an “Adaptive Extended Kalman Filter”. We just call it an AKF.
An Auto-Adaptive Extended Kalman Filter (AA EKF) is an adaptive EKF that, like the AKF described above, has “adaptive” elements that selectively trust given measurements more or less based on comparison to reference inputs. The difference with the auto-adaptive filter is that the “trust” thresholds are automatically determined by the filter itself. The filter collects error metrics on all the measurements and uses this to determine appropriate trust thresholds. This feature makes tuning a Kalman Filter for optimum performance much easier than manually determining these thresholds. The GX5/CX5/CV5 series introduces the Auto-Adaptive feature whereas the GX4 series has fixed adaptive thresholds.
A Complementary Filter (CF) is commonly used as a term for an algorithm that combines the readings from multiple sensors to produce a solution. These filters usually contain simple filtering elements to smooth out the effects of sensor over-ranging or anomalies in the magnetic field.