Time-Domain Spectroscopy Tools 🧰

A small library for high-precision material parameter extraction from time-domain THz signals, written in Wolfram Language 🐺 and OpenCL 🏎️

Developed at University of Augsburg, Germany 🇩🇪

Beta testing ♲

Processed 3411 THz spectra from 2021 to 2024

acquired using TDS-THz 200ps Spectrometer Toptica TeraFlash 📸

Features

• Fabry-Pérot deconvolution ⭐️
• Informed phase unwrapping
• High precision for $n$, $\kappa$, and $\alpha$ solving
• Works for both thin and thick samples
• GPU Acceleration (OpenCL) ⭐️
• Functional approach, no hidden state
• Syntax sugar with data preview ⭐️
• Jitter-proof

See it in action! ⏱️

An online example on how to work with it.

Examples of processed TDS

GaGe 0.4mm semiconductor

Fe2Mo3O8 0.4mm dielectric

Supported configurations 📸

• Transmission
• Reflectivity

Installation

Option 1

Install locally using LPM

PacletRepositories[{
    Github -> "https://github.com/JerryI/wl-tds-tools" -> "master"
}]

Option 2

Clone this repository into your project folder and run:

    PacletDirectoryLoad["wl-tds-tools"]

Library Interface

We separate the toolbox into 3 contexts:

• JerryI`TDSTools`Trace` : operates with raw time-traces.
• JerryI`TDSTools`Transmission` : constructs transmission objects from sample and reference time-traces.
• JerryI`TDSTools`Material` : extracts material parameters from transmission objects.

Notes on GPU Acceleration

Test were performed on Mac Air M1 (2021), where 4 cores are used for CPU calculations

• single spectrum wall-time

GPU: 0.01 sec
CPU: 0.46 sec

• 286 spectra wall-time

GPU: 6.05 sec
CPU: 148.3 sec

• 1071 spectra wall-time

GPU: 22.1 sec
CPU: 550.3 sec

The peformance can still be improved by moving more calculations on GPU side. It is in TODO list.

Time-traces ⤵️

To construct a time-trace object from a table, use:

TDTrace

TDTrace[q_QuantityArray] _TDTrace

This accepts a QuantityArray. The units for the time-step must be provided. For example

tds = {
    {3010.1, 0.1},
    {3011.2, 0.2},
    ...
};

tds = TDTrace[QuantityArray[tds, {"Picoseconds", 1}]]

Properties

Any property can be accessed as a down-value

tds["Properties"]

Here is a list

• "Properties" : lists all available keys
• "Spectrum" : return QuantityArray with a complex amplitude spectrum (automatically resampled in a case of jitter issues). Read-only
• "PowerSpectrum" : returns QuantityArray of power spectrum. Read-only
• "Trace" : return QuantityArray of the original time-trace. Read-only
• "FDCI" : a range of frequency-domain confidence interval. It is generated according to the powerspectrum and provides a coarse estimation of the working range in wavenumbers. This interval will be used later for optimizing material parameters. Read-only

Reconstructing transmission function ⤵️

On the next step we can derive an initial information about the transmission based on the sample and reference time-traces

TransmissionObject

A constructor

TransmissionObject[sam_TDTrace, ref_TDTrace, "Thickness"->_Quantity, opts___] _TransmissionObject

it creates a special object, calculates fourier images, fixes possible jitter issues and estimates the transmission function, i.e.

$$ \hat{t} = \frac{E_{sam}(\omega)}{E_{ref}(\omega)} $$

Options

One can provide the following opts to the constructor

• "Thickness" : mandatory field. Must be _Quantity
• "Gain" : a scaling factor for the amplitude of sam. By the default is 1.0
• "PhaseShift" : a constant offset in a multiples of 2Pi to the whole phase of the transmission function. Must be _Integer. By the default is 0.

Properties

• "Properties" : lists all keys
• "Thickness" : returns Quantity object with a thickness of the material
• "\[Delta]t" : returns estimated time-delay between the sample and the reference signals. Read-only
• "Gain" : returns a multiplier for sam signal
• "PhaseShift" : returns a constanst offset of the phase in multples of 2Pi

---

• "Frequencies" : returns QuantityArray of the whole range of frequencies. Read-only
• "Transmission" : returns the power transmission QuantityArray, i.e. array of $|\hat{t}(\omega)|^2$. Read-only
• "Phase" : returns the argument of $\hat{t}$ sampled as a function of frequency in a form of QuantityArray. After the creation is wrapped and cannot be used before unwrapping process (see later). Read-only

---

• "Domain" : returns the ranges of frequencies. Read-only
• "FDCI" : a range of frequency-domain confidence interval. It is generated according to the powerspectrum and provides a coarse estimation of the working range in wavenumbers. This interval will be used later for optimizing material parameters. Read-only
• "FDCI2" : it serves the same purpose as FDCI (as a criteria parameter), but relies on the phase information. It cannot be with a wrapped phase. Read-only

How to update transmisstion object

One can update properties of an object without recreating it fully using Append

new = Append[tr, "Thickness"->Quantity[1.2, "Millimeters"]]

or multiple properties

new = Append[tr, {"Thickness"->Quantity[1.2, "Millimeters"], "Gain"->0.9}]

It returns a new object, while the raw data is shared between them internally.

Phase unwrapping

A key challenge in digital signal processing is managing the phase of the signal. To address this, phase unwrapping is employed, which transforms a discontinuous phase signal into a continuous one

TransmissionUnwrap[t_TransmissionObject, type_String, opts___] _TransmissionObject

it performs phase-unwrapping procedure on t object and returns a new TransmissionObject with a modified phase. There is only one sheme specified by type for unwrapping is available for now

• "Basic" : uses informed phase unwrapping based on a time-delay between the sample and the reference signals

Options

• "PhaseThreshold" : sets the treshold for a detector to remove 2Pi jump. By the default is 5.6
• "PhaseShift" : overrides a constant phase offset parameter to be used later in material parameters extraction

Material Parameters ⤵️

This the main goal of all optical spectroscopy: use EM waves to probe the responce function of the material to extract all useful properties. Here we focus on $n$ and $\kappa$ (or their derivatives).

MaterialParameters

A constructor and executor

MaterialParameters[t_TransmissionObject, opts___] _MaterialParameters

return MaterialParameters object, which contains all information about n, k, \[Alpha] (absorption coefficient). Depending on provided options it can use different methods to extract them

It can also accept a list with multiple transmission objects and process them as a batch

MaterialParameters[{t__TransmissionObject}, opts___] _List

Options

• "Target" : specifies the target device to be used "CPU" or "GPU" (OpenCL). By the default is "CPU"
• "SolveNKEquations" : enables high-persicision $n$ and $\kappa$ extraction (by solving iteratively equations compling them with transmission and phase). The default value is True.
• "NKCycles" : number of interations for solving nk-equations. The default is 30.
• "MovingAverageFilter" : enables moving average filter (kernel is 2). By the default is True.
• "FabryPerotCancellation" : enables deconvolution procedure asumming a normal incidence and a perfect slab-like sample. By the default is True. DOI: 10.1117/12.612946
• "FabryPerotCycles" : number of iterations for deconvolution procedure. The default is 8.

Notes on GPU target

It uses OpenCL API to work with GPU devices and picks up the fastest available. If the device does not support OpenCL (or problems with drivers), library will fall back to CPU target.

Properties

All properties are read-only.

• "Domain" : frequency range.

---

• "Frequencies" : returns QuantityArray of frequencies
• "Transmission" : returns QuantityArray of a power transmission. If FabryPerotCancellation is applied, it will return deconvoluted spectrum.
• "Best Transmission" : returns "Transmission" selected in the region of "FDCI"
• "Phase" : returns QuantityArray of the phase of the transmissino function.
• "Best Phase" : returns "Phase" selected in the region of "FDCI"

---

• "\[Alpha]" : returns QuantityArray of extracted absorption coefficient.
• "Raw \[Alpha]" : returns "\[Alpha]" without FP cancellation applied.
• "n" : returns extracted real part of refractive index as QuantityArray.
• "k" : returns extracted imaginary part of refractive index as QuantityArray.
• "Raw k" : returns "k" without FP cancellation applied.
• "Best \[Alpha]" : returns "\[Alpha]" in the region of "FDCI".
• "Best Raw \[Alpha]" : returns "\[Alpha]" without FP cancellation applied in the region of "FDCI".
• "Best n" : returns "n" in the region of "FDCI".
• "Best k" : returns "k" in the region of "FDCI".
• "Best Raw k" : returns "k" without FP cancellation applied in the region of "FDCI".

---

• "Thickness" : returns the thickness of the material
• "Gain" : returns gain of the sample signal (or an amplitude transmission function).
• "PhaseShift" : returns a constanst offset of the phase in multples of 2Pi

---

• "FDCI" : a range of frequency-domain confidence interval. It is generated according to the powerspectrum and provides a coarse estimation of the working range in wavenumbers. This interval will be used later for optimizing material parameters.
• "FDCI2" : it serves the same purpose as FDCI (as a criteria parameter), but relies on the phase information. It cannot be with a wrapped phase.
• "FPReduction" : a quality factor of Fabry-Perot fringes reduction. 1.0 or close to it means the deconvolution procedure was not succesfull. The higher - the better.

Optimizing parameters

The main issue is inaccuracy in "Thickness" and "Gain", which comes naturally in a measurements. In this case a slight misaligment can make your results look much worse, when FabryPerotCancellation is enabled.

In order to fix this, one can vary those parameters in a fixed range and plot a contour map of FPReduction factor. For example

...
tr = TransmissionObject[...];
utr = TransmissionUnwrap[tr, "Basic"];

(* vary thickness and gain *)
map = Table[With[{
  t = Append[utr, {
    "PhaseShift"->1, 
    "Gain"->gain, 
    "Thickness"->Quantity[thickness, "Millimeters"]
    }]
},
  {thickness, gain, t}
], {thickness, 0.4-(15 0.01), 0.4+(15 0.01), 0.01}, {gain, 0.6, 1.2, 0.1}];

map = Flatten[map,1];

(* extract nk and estimate FPReduction *)
map[[All,3]] = #["FPReduction"] &/@ MaterialParameters[map[[All,3]]];

ListContourPlot[map, PlotLegends->Automatic]

(* find the best parmaeters *)
MaximalBy[map, Last]

as a result you will see a streched spot on the contour plot, which corresponds (probably) to the best pair

The processed data will look like

Bath processing (GPU only) 🪈

If you vary the parameters of the same TransmissionObject to avoid extra transactions to GPU, one can use provide the set as a list to MaterialParameters contructor

MaterialParameters[map[[All,3]], "Target"->"GPU"]

this function will group them by Phase arrays (comparison by the internal reference will be used, no actual evaluation) and then feed to GPU with shared initial data.

This also takes advantage of non-blocking execution model of OpenCL and loads all data at once.

Acknowledgments 💛

The work of K.V. (@JerryI) was supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)-TRR 360-492547816

I am deeply grateful to Dr. Joachim Deisenhofer for countless discussions, endless support, and deep insights into THz spectroscopy.

References 📔

We rely on well-known methods for material parameter extraction:

• Material parameter extraction for terahertz time-domain spectroscopy using fixed-point iteration. W. Withayachumnankul, B. Ferguson, T. Rainsford, S. P. Mickan, and D. Abbott. Photonic Materials, Devices, and Applications (2005) DOI: 10.1117/12.612946.
• Phase Retrieval in Terahertz Time-Domain Measurements: A “How To” Tutorial. Peter Uhd Jepsen. Journal of Infrared, Millimeter, and Terahertz Waves (2019) DOI: 10.1007/s10762-019-00578-0.

Published Papers

We use these tools for processing THz data in our department Experimentalphysik V. Below is a list of our publications where we used this tool to process data:

• Optical magnetoelectric effect in the polar honeycomb antiferromagnet Fe₂Mo₃O₈. K. V. Vasin, A. Strinić, F. Schilberth, S. Reschke, L. Prodan, V. Tsurkan, A. R. Nurmukhametov, M. V. Eremin, I. Kézsmárki, and J. Deisenhofer. Phys. Rev. B 110, 054401 – Published 1 August 2024.
• Magnetization reversal through an antiferromagnetic state. Somnath Ghara, Evgenii Barts, Kirill Vasin, Dmytro Kamenskyi, Lilian Prodan, Vladimir Tsurkan, István Kézsmárki, Maxim Mostovoy & Joachim Deisenhofer. Nature Communications volume 14, Article number: 5174 (2023).

Supported Platforms

• WLJS Notebook
• Wolfram Mathematica
• WolframScript

License

MIT