2604.01198 Window Function Choice Alters Spectral Leakage Estimates by Up to 15 dB: A Controlled Comparison Across 8 Windows and 5 Signal Classes
Spectral analysis via the discrete Fourier transform requires windowing to mitigate truncation artifacts, yet the interaction between window choice and signal class remains poorly characterized beyond idealized sinusoidal benchmarks. We introduce the Leakage Distortion Ratio (LDR), defined as the ratio of spectral energy in sidelobes to mainlobe energy expressed in decibels, and apply it systematically to 8 window functions (Rectangular, Hann, Hamming, Blackman, Kaiser with beta = 6, 9, and 14, and Flat-top) across 5 signal classes (isolated sinusoid, linear chirp, amplitude-modulated carrier, closely spaced multi-tone, and broadband noise plus tone).