A transform with at most 25 code lines, proved so far a convenient substitute for more sophisticated transforms (e.g. MFCCs) in various signal recognition problems. No multipliers, simple algorithm, one can apply it for various signals by optimzing the delays in the algorithm. iRDT stands from improved RDT (optimized for high speed using JIT compiler)
Particularly useful for HW-oriented devices (MCU, FPGAs) in the Tiny-ML context. iRDT is implemented here as function nrdt_2025z()
Copyright Radu and Ioana DOGARU radu.dogaru@upb.ro Last update: 23 oct. 2025.
It represents a revised, optimized and simplified code replacement for the older RDT Python code provided in https://github.com/radu-dogaru/NL-CNN-RDT-based-sound-classification- including JIT and fixed-point (INT32).
For facile access to datasets in examples, it is prefferable to run the notebook on Kaggle
Relevant papers
[0] R. Dogaru and I. Dogaru, "Why RDT ?, "Computational aspects in favor of the RDT transform", oct. 2025 here [https://github.com/radu-dogaru/rdt_transform_for_tiny_ml_signal_classifiers/blob/main/why_RDT.pdf]
[1] R. Dogaru and I. Dogaru, "A low complexity solution for epilepsy detection using an improved version of the reaction-diffusion transform," 2017 5th International Symposium on Electrical and Electronics Engineering (ISEEE), Galati, Romania, 2017, pp. 1-6, doi: 10.1109/ISEEE.2017.8170678. https://ieeexplore.ieee.org/document/8170678
[2] R. Dogaru and I. Dogaru, "State of the Art Recognition of Emotions from Speech, Using a Low Complexity Solution Based on Reaction-Diffusion Transform," 2022 International Symposium on Electronics and Telecommunications (ISETC), Timisoara, Romania, 2022, pp. 1-4, doi: 10.1109/ISETC56213.2022.10010234. https://ieeexplore.ieee.org/document/10010234
The RDT idea emerged in the 2000s inspired from cellular-automata works in the context of finding some convenient measure to quantify various emergent behaviors. See more in: https://ieeexplore.ieee.org/document/1630267
Later, in 2007 we first expanded it succesfully (introducing the idea of sub-sampling in some m-channels) to sound recognition problems. See more in: https://ieeexplore.ieee.org/document/4410603
Since then, it proved to be a convenient feature extractor for either audio or bio-medical signals, competing well against traditional spectrogram algorithms. See more here: https://ieeexplore.ieee.org/search/searchresult.jsp?action=search&newsearch=true&matchBoolean=true&queryText=(%22Full%20Text%20Only%22:%22reaction-diffusion%20transform%22)%20AND%20(%22Authors%22:Dogaru)