Statistical models for averaging of the pump-probe time traces: example of denoising in terahertz time domain spectroscopy

TL;DR

This paper introduces four statistical models for denoising pump-probe time traces in terahertz spectroscopy, demonstrating improved accuracy over simple averaging through rigorous algorithms and experimental validation.

Contribution

The paper presents novel statistical models and algorithms specifically designed for noise reduction and trace averaging in terahertz time domain spectroscopy.

Findings

Advanced models outperform simple averaging in reducing fitting errors

Algorithms effectively characterize different types of experimental noise

Statistical models enhance the quality of THz-TDS data analysis

Abstract

In this paper, we first discuss the main types of noise in a typical pump-probe system, and then focus specifically on terahertz time domain spectroscopy (THz-TDS) setups. We then introduce four statistical models for the noisy pulses obtained in such systems, and detail rigorous mathematical algorithms to de-noise such traces, find the proper averages and characterise various types of experimental noise. Finally, we perform a comparative analysis of the performance, advantages and limitations of the algorithms by testing them on the experimental data collected using a particular THz-TDS system available in our laboratories. We conclude that using advanced statistical models for trace averaging results in the fitting errors that are significantly smaller than those obtained when only a simple statistical average is used.