Augmenting Extended Kalman Filters with Dynamic Temperature Derivative State Models
Augmenting Extended Kalman Filters with dynamic temperature derivative states eliminates dynamic thermal bias drift during rapid ramp conditions.
Multi-Rate Filtering
Sampling Conversion
Digital signal processing stages alter sampling frequencies across successive processing layers to balance computational workload and signal bandwidth. Implementation of multi-rate filtering allows high-speed sensor acquisition fronts to downsample digital outputs before transmitting processed bandwidth to navigation algorithms. Decimation stages cascade finite impulse response filters to attenuate out-of-band noise before reducing discrete sample rates.
Polyphase filter structures implement these rate conversions efficiently by performing arithmetic operations at the lower output sampling frequency.
Phase Distortion
Interpolation and decimation operations introduce digital phase distortion when anti-aliasing filter stages exhibit non-linear phase profiles. Spectral analysis tools evaluate passband ripple and stopband rejection to verify multi-rate filtering performance across dynamic signal bands. Quantization noise accumulates inside finite word-length registers when intermediate accumulator products undergo truncated rounding operations.
Calibration protocols pass synthesized multi-frequency test vectors through the digital signal pipeline to verify that phase delays remain linear across all operational sample rates.
Computational Workload
Group delay variations across filter transition bands produce frequency-dependent time lags in real-time control loops. High decimation ratios require steep transition bands that increase digital memory storage demands.
Bandwidth Limit
Verification of filter performance occurs by processing impulse and step signal inputs through automated hardware-in-the-loop simulation test benches. Digital clock jitter across asynchronous processing domains causes spectral leakage that degrades anti-aliasing attenuation thresholds. The sampling rate conversion process remains bounded by the Nyquist criterion at the lowest output sample rate.