Complex Orthogonal Decomposition (C.O.D.)

using Python · 2026

Read Complex Orthogonal Decomposition (C.O.D.) by using Python (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-13084.txt.

Introduction

Vacher, Perrard, and Ramananarivo apply Complex Orthogonal Decomposition, a signal analysis method from Feeny's work on wave motion, to a simple spatio-temporal signal. The method extracts spatial and temporal modes, suiting oscillatory signals where phase matters and spatial forms are unknown. Two theoretical chapters lay out the mathematics behind the method, with examples and Python scripts demonstrating its efficiency and characteristic features.

Catalog record

CREATORusing Python
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE47833
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.13084

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Complex Orthogonal Decomposition (C.O.D.) using Python Abstract Abstract This work presents the application of the Complex Orthogonal Decomposition (C.O.D.) to a simple spatio-temporal signal. C.O.D. has been introduced first in the article of B. Feeny, entitled "A Complex Orthogonal Decomposition for Wave Motion Analysis" [ 4 ] , published in the Journal of Sound and Vibration . The purpose of this signal analysis method is to extract spatial and temporal modes out of a signal. This approach is especially suited to deal with oscillatory signals where phase information is important and where spatial forms are unknown. We provide two theoretical chapters presenting the main mathematical concepts behind C.O.D. and a series of example (with associated Python scripts) to demonstrate the efficiency of the method and some characteristical features. I Motivation and presentation The main idea of Complex Orthogonal Decomposition (C.O.D.) is to analyse a signal of time and space s(t,x) representing a physical quantity that oscillates. To introduce it, we look at a real example that demonstrates its purpose and how it works : fish swimming. Indeed, many fishes have oscillatory motions when they swim, with definite frequencies linked to their species, their size or if they swim in schools or not for example [ 3 ] . However, this wave motion is not trivial : if we analyse their motion using the centreline of their body, researchers have shown that this oscillatory motion is not spatially periodical at all. It means that the spatial form of the wave is not trivial, as we can see on Fig. 1 . (b)-(e). Figure 1: Figure from Chen et al. [ 3 ] - Fish body wave curve: (a) Diagram of the fish body wave curve. The blue area is the outline of a BCF fish, and the red line segment is the fish body wave curve. (b)-(e) Fish body wave curves of four different species of fish. Curves of different colors represent the body midline displacement of fish at 20 equally spaced time intervals during a single tail beat cycle. Mathematically, it means that the signal s(t,x) (being the amplitude of oscillation of the fish, t being time and x a position along its centerline) does not decompose well on a 2D space-time Fourier basis. Consequently, the following decomposition s(t,x)=\sum_{j=1}^{+\infty}A_{j}\cos(k_{j}x-\omega_{j}t+\phi_{j}) (1) cannot be truncated at relatively low index. The space Fourier basis do not represent well the motion in this case. In this context, C.O.D. is the appropriate tool to analyse the signal. In a sense, it allows you to have your signal decomposed as : s(t,x)=\sum_{k=1}^{+\infty}T_{k}(t)\,X_{k}(x), (2) with X_{k} being spatially orthogonal to each other. In the case of the fish, it is interesting to see if some frequencies are associated with specific unknown body motions that would enhance the efficiency of the swim. Additionally, the C.O.D. provides a tool called the travelling index, a number that quantifies how much a given mode resembles a travelling wave versus a standing wave. It is typically an indicator that can give clues on the swim efficiency [ 2 ] . From a mathematical point of view, the C.O.D. is not that different from the classical Proper Orthogonal Decomposition (P.O.D.) in terms of conceptual ideas [ 4 ] but is not used as much in studies. This paper aims at recalling the theory concerning Complex Orthogonal Decomposition, with applicative examples, and demonstrate its efficiency. Python scripts are attached to support these examples and give easy access to the method for research or pedagogical purposes. Are attached to this document: • a Python package pack_COD , including all the related functions performing the Complex Orthogonal Decomposition, plus other verification and plotting tools. • test_COD_ex_1.py : a first example to illustrate the main principles of the C.O.D. and how the package works. The theoretical analysis that goes along this case is developed in section IV . • test_COD_ex_2.py : this example shows that in the case of non perfectly periodical signals, some approximation in the method can lead to errors. We explain quantitively their origin in section V • test_COD_ex_3.py : this example shows the efficiency of C.O.D. if one spatial form is associated with a multi-frequency oscillations. We use a frequency modulated signal to demonstrate this fact. It corresponds to section VI In addition to each of these four examples, videos of the 2D spatio-temporal signals are associated. Three other examples are added : • test_COD_ex_1_nu.py : this example shows how the C.O.D. works in the context of non-uniform space grids. It corresponds to the Appendice. • test_COD_ex_your_data.py : this script allows you to try the C.O.D. on your own data, put into the form of two .csv files. • test_COD_ex_noise.py : it is an example of how C.O.D. deals with noise. This example is not discussed since it is not the main purpose of the document here, yet interesting things could be analysed. II Spatio-temporal signal processing and C.O.D. Let s(t,x)\in\mathbb{R} be a real-valued spatio-temporal signal, smooth in both time t\in\Omega_{t} and space x\in\Omega_{x} . \Omega_{t}=[0,T] , with T\in\mathbb{\overline{R^{+}}} and \Omega_{x}=[x_{lim_{1}},x_{lim_{2}}] , with (x_{lim_{1}},x_{lim_{2}}) in \overline{\mathbb{R}}^{2} . The core idea of C.O.D. is to express the real analytic signal as the real part of sum of separable complex spatio-temporal modes: s(t,x)=\Re\left(\sum_{j=1}^{+\infty}a_{j}(t)\,\phi_{j}(x)\right) (3) where: • \phi_{j}(x)\in\mathbb{C} are orthonormal spatial modes, • a_{j}(t)\in\mathbb{C} are complex temporal coefficients (also called complex ortogonal coordinates, C.O.C. [ 4 ] ). To do that, we need to have an equivalent complex signal. We use the Hilbert transform to achieve this goal. 1. Complex analytic signal and Hilbert transform Given a real-valued signal s(t,x) , we define its Hilbert transform in time at each spatial position x . This transform captures the oscillatory structure of the signal in a way that separates amplitude and phase. Formally, the Hilbert transform \mathcal{H}\{s(t,x)\} is defined by convolution in time [ 6 ] : \widetilde{s}(t,x)=\mathcal{H}\{s\}(t,x)=\frac{1}{\pi}\,\mathrm{p.v.}\int_{\Omega_{t}}\frac{s(\tau,x)}{t-\tau}\,\mathrm{d}\tau, (4) where “p.v.” denotes the Cauchy principal value. The complex analytic signal is then: s_{c}(t,x)=s(t,x)+i\,\widetilde{s}(t,x). (5) Relation to the Fourier transform : In the frequency domain, constructing the complex signal corresponds to suppressing the negative frequency components and doubling the positive ones. This is efficiently done using the (complex) Fourier transform. Let \mathcal{F}_{t}\{s\}(f,x)=\hat{s}(f,x) be the temporal Fourier transform of s(t,x) . Then the complex part of the signal can be obtained via: \hat{s_{c}}(f,x)=\begin{cases}2\hat{s}(f,x),&f>0,\\ \hat{s}(0,x),&f=0,\\ 0,&f<0.\end{cases} (6) The complex analytic signal s_{c}(t,x) is then recovered by taking the inverse Fourier transform of \hat{s_{c}}(f,x) : s_{c}(t,x)=\mathcal{F}_{t}^{-1}\{\hat{s_{c}}(f,x)\}. (7) This frequency-domain filtering approach is equivalent to performing the Hilbert transform in the time domain, and is especially useful in numerical computations. It ensures that the resulting s_{c}(t,x) contains only positive frequencies, making it ideal for phase and envelope analysis. 2. Modal decomposition via C.O.D. The spatial modes \phi_{j}(x) are obtained by solving the eigenvalue problem of the spatial covariance operator: \mathcal{R}[f](x)=\frac{1}{T}\int_{0}^{T}\left(\int_{x_{lim_{1}}}^{x_{lim_{2}}}s_{c}(t,x^{\prime})f(x^{\prime})^{*}\mathrm{d}x^{\prime}\right)s_{c}(t,x)\,\mathrm{d}t, (8) where ∗ denotes complex conjugation. This operator is Hermitian and positive semi-definite, ensuring a complete orthonormal set of eigenfunctions. We call : • \lambda_{j}^{e} the eigenvalues (that are real and positive, because of the positive semi-definite characteristics of the operator) 1 1 1 We use the exponent e to avoid any confusion with \lambda - the spatial wavelength. . • \phi_{j} the associated eigenvectors. The temporal coefficients a_{j}(t) can be retrieved

Read the complete text