Implementation of empirical mode decomposition with sifting optimizations for real-time de-noising

Empirical mode decomposition (EMD) splits a signal into oscillatory components without a fixed basis, which makes it attractive for de-noising biological signals. In its usual form, however, it needs the whole record in memory and repeats the sifting step until a stopping rule is met, so the processing time cannot be predicted. This paper describes a frame-based Simulink implementation of EMD and tests three changes to the sifting process aimed at real-time use: a fixed number of sifting passes, linear envelopes in place of cubic splines, and overlapping frames of which only the centre is kept. The model was tested on a two-tone synthetic signal sampled at 1 kHz with white Gaussian noise added (input signal-to-noise ratio, SNR, 3.98 dB). Removing the first three intrinsic mode functions (IMFs) gave the best output SNR. With that setting, three sifting passes reached 13.66 dB and four passes 13.80 dB, while eight and twelve passes gave 12.98 and 12.42 dB, so extra passes lowered the SNR. Linear envelopes beat cubic splines by about 2 dB (13.66 against 11.63 dB), and splines also produced a large start-up transient in the first output frame. An overlap of 64 samples or more improved the SNR by about 0.2 dB compared with no overlap, with no further gain up to 256 samples, while latency fell as overlap grew. On a 20 s electrocardiogram (ECG) segment from MIT-BIH record 100 with white noise added at an input SNR of about 4 dB, removing only the first IMF raised the SNR to 8.10 ± 0.20 dB (ten noise realizations), and R peaks kept 0.95 of their height on average, but the P and T waves were not recovered. The synthetic results come from one signal and one noise realization, the ECG results from one record segment, and execution on hardware was not measured.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-18
DOI
https://doi.org/10.21203/rs.3.rs-11771797/v1
Primary Topic
Image and Signal Denoising Methods
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Implementation of empirical mode decomposition with sifting optimizations for real-time de-noising

Md. Rezwan Islam, Akib Ikbal
Research Square
Image and Signal Denoising Methods
preprint

Implementation of empirical mode decomposition with sifting optimizations for real-time de-noising

Md. Rezwan Islam, Akib Ikbal
preprint en

Abstract

Empirical mode decomposition (EMD) splits a signal into oscillatory components without a fixed basis, which makes it attractive for de-noising biological signals. In its usual form, however, it needs the whole record in memory and repeats the sifting step until a stopping rule is met, so the processing time cannot be predicted. This paper describes a frame-based Simulink implementation of EMD and tests three changes to the sifting process aimed at real-time use: a fixed number of sifting passes, linear envelopes in place of cubic splines, and overlapping frames of which only the centre is kept. The model was tested on a two-tone synthetic signal sampled at 1 kHz with white Gaussian noise added (input signal-to-noise ratio, SNR, 3.98 dB). Removing the first three intrinsic mode functions (IMFs) gave the best output SNR. With that setting, three sifting passes reached 13.66 dB and four passes 13.80 dB, while eight and twelve passes gave 12.98 and 12.42 dB, so extra passes lowered the SNR. Linear envelopes beat cubic splines by about 2 dB (13.66 against 11.63 dB), and splines also produced a large start-up transient in the first output frame. An overlap of 64 samples or more improved the SNR by about 0.2 dB compared with no overlap, with no further gain up to 256 samples, while latency fell as overlap grew. On a 20 s electrocardiogram (ECG) segment from MIT-BIH record 100 with white noise added at an input SNR of about 4 dB, removing only the first IMF raised the SNR to 8.10 ± 0.20 dB (ten noise realizations), and R peaks kept 0.95 of their height on average, but the P and T waves were not recovered. The synthetic results come from one signal and one noise realization, the ECG results from one record segment, and execution on hardware was not measured.

Research Square
Varendra University (BD)
Image and Signal Denoising Methods
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon