RIFT: Entropy-Optimised Fractional Wavelet Constellations for Ideal Time-Frequency Estimation

Cambridge University researchers James Cozens and Simon Godsill have introduced RIFT (Reconstructive Ideal Fractional Transform), a breakthrough algorithm for analyzing complex, nonstationary signals. The method computes a constellation of Continuous Fractional Wavelet Transforms aligned to different local time-frequency curvatures, then combines them into a single optimized representation using a localized entropy-based sparsity measure. This approach is specifically designed to resolve auto-terms (true signal components) while attenuating cross-terms (artifacts that plague many time-frequency methods).

The algorithm applies a positivity-constrained Lucy-Richardson deconvolution with total-variation regularization to estimate the Ideal Time-Frequency Representation, achieving resolution comparable to the gold-standard Wigner-Ville Distribution but without its problematic cross-term interference. RIFT also produces an Instantaneous Phase Direction field that visualizes localized curvature in signals like speech or music, enabling component trajectory extraction through Kalman tracking schemes. Evaluation on both synthetic and real-world signals demonstrates superior time-frequency precision relative to competing methods, with the Spline-RIFT variant offering additional analytical capabilities.

This advancement represents a significant leap in signal processing methodology, providing researchers and engineers with a tool that delivers high-resolution analysis without the trade-offs traditionally associated with time-frequency representations. The fully derived Cohen's class convolutional kernels and optimization framework make RIFT both theoretically sound and practically implementable for demanding applications.