Direct Band-Energy Estimation from Finite Samples via Continuous Slepian Multitapers

Abstract

In this work, a direct band-energy estimation approach using finite sampled data under explicit time- and frequency-domain regularity and concentration assumptions is proposed. Standard two-stage methods always estimate the power spectral density and then integrate it over the target band, thereby accumulating error through spectral smearing, leakage, and numerical band integration. Instead, we formulate the band-energy estimation as a finite-sample approximation problem. We then construct a computable direct estimator based on sampled continuous prolate spheroidal wave functions (PSWFs) in a multitaper-type framework, which links continuous-time/frequency assumptions to discrete observations. The estimator is obtained by first constructing a band $L^2$-norm estimator and then squaring it to estimate the band energy. We derive deterministic and non-asymptotic error bounds that explicitly separate four sources of approximation error: time limitation, sampling, band limitation, and finite-term truncation. In addition, several theories are derived to show how the accuracy of the estimation depends on the length, bandwidth, sampling step, time-frequency smoothness, and concentration. We also establish deterministic stability under additive noise, analyze computational complexity, and show that the method admits an efficient multiple-band implementation through offline reuse of PSWF-related computations. Numerical experiments across multiple observation lengths, bands of interest, and noise levels are conducted to show that the proposed approach achieves excellent performance compared to integrated-periodogram-type methods.