Bayesian versus Fourier spectral analysis of ion cyclotron resonance time-domain signals

The frequency-domain spectrum obtained by Fourier transformation (FT) of a time-domain signal is accurate only for a continuous noiseless time-domain signal of infinite duration. For discrete noisy truncated time-domain signals, non-FT (e.g., Bayesian analysis) methods may provide more accurate spectral estimates of time-domain signal frequencies, relaxation time(s), and relative abundances. In this paper, we show that Bayesian analysis of simulated and experimental ion cyclotron resonance (ICR) time-domain noisy signals can produce a spectrum with mass accuracy improved by a factor of 10 or more over that obtained from a magnitude-mode discrete fast Fourier transform (FFT) spectrum. Moreover, Bayesian analysis offers the useful advantage that it automatically estimates the precision of its iteratively determined spectral parameters. The main disadvantage of Bayesian analysis is its lengthy computation time compared to that of FFT (hours vs seconds on the same hardware for approximately 4K time-domain data points); the Bayesian computation time increases rapidly with the number of spectral peaks and (less rapidly) with the number of time-domain data points. Bayesian analysis should thus prove useful for those FT/ICR applications involving relatively few data points and/or requiring high mass accuracy.