8 April 2022 in IEEE Signal Processing Letters
Empirical mode decomposition (EMD) and its improved EMD-based algorithms are adaptive and nonlinear methods that decompose a nonstationary signal into several intrinsic mode functions (IMFs) through the sifting process. When the EMD-based algorithms are applied to decompose a signal polluted with noise in some regions, the noise will spread into the clean region and introduce errors in the IMFs. During the real-time computation of an EMD-based algorithm, the signal is partitioned into a series of overlapping time windows. Points outside the window are discarded, they act as an error source and the error will spread into the window. We will prove that the above two problems are mathematically identical. Due to the complex nature of the sifting iteration, a mathematical theory for the error analysis of the boundary effect is still lacking. Previous studies mainly rely on simulations, and the results show that the error seems to be confined locally near the boundary. Beginning with one sifting iteration, which is the kernel of an EMD-based method, we will theoretically analyze the boundary effects completely based on the sifting process without any approximations, and prove that the error will propagate into the interior domain and may be amplified and thus obscure the meaning of an IMF. Then, we will prove a sufficient condition for the exponential decay of the error. When the error is amplified, we propose a method to resolve this problem. Finally, numerical experiments are conducted to analyze multiple siftings and IMFs.
4 January 2021 in IEEE Transactions on Instrumentation and Measurement
Sample entropy (SpEn) is a measure of the underlying regularity or complexity of a system that is achieved by assessing the entropy of a time series recorded from the system. It is a powerful signal processing tool and has received increasing attention in recent years. SpEn has been successfully applied in biomedical measurements and other applications. In particular, many emerging applications require measuring the SpEn of signals in real-time embedded systems. However, the standard implementation of SpEn requires a computational complexity of $O(n^{2})$ , where $n$ is the data length, making it difficult to meet real-time constraints, especially for large $n$ . Moreover, power consumption and computation latency must be considered as well. The data length used in previous studies was approximately several hundred, and it remains a challenging task to operate on longer data lengths. In this article, we propose the assisted sliding box (SBOX) algorithm to accelerate the computation of SpEn without any approximation while maintaining a low memory overhead so that the algorithm can be executed in embedded systems for edge computing. We also develop an electroencephalogram (EEG)-based wearable device for comfortable overnight recording. The SBOX algorithm is then implemented in the system to measure the online SpEn of an overnight sleep EEG signal. The results show that, compared with the standard algorithm, the SBOX algorithm speeds up the computation time by a factor of 60, thereby reducing power consumption by 98% when measuring a 30-s epoch of sleep EEG with $n=7500$ and a 250-Hz sampling rate.
4 September 2021 in Journal of Medical and Biological Engineering
31 October 2022 in IEEE Access
Empirical mode decomposition (EMD) and its variants are adaptive algorithms that decompose a time series into a few oscillation components called intrinsic mode functions (IMFs). They are powerful signal processing tools and have been successfully applied in many applications. Previous research shows that EMD is an efficient algorithm with computational complexity O ( n ) for a given number of IMFs, where n is the signal length, but its memory is as large as (13+ mimf ) n , where mimf is the number of IMFs. This huge memory requirement hinders many applications of EMD. A physical or physiological oscillation (PO) mode often consists of a single IMF or the sum of several adjacent IMFs. Let mout denote the number of PO modes and, by definition, mout ≤ mimf . In this paper, we will propose a low memory cost implementation of EMD and prove that the memory can be optimized to (2+ mout ) n without aggravating the computational complexity, while gives the same results. Finally, we discuss the optimized memory requirements for different noise-assisted EMD algorithms.
26 August 2024 in IEEE Transactions on Instrumentation and Measurement
Empirical mode decomposition (EMD) is a nonlinear, data-dependent method that decomposes a nonstationary signal into a few oscillation modes, which are referred to as intrinsic mode functions (IMFs). However, EMD often suffers from modemixing phenomena. Therefore, many disturbance-assisted EMD (DA-EMD) algorithms, such as ensemble EMD (EEMD) and uniform phase EMD (UPEMD), have been developed to resolve this problem. DA-EMD algorithms are powerful signal processing tools with widespread use in numerous applications. In particular, many emerging industrial or biomedical applications require measuring features by the decomposition of a signal using DA-EMD algorithms in an online, real-time embedded microprocessor (MCU) of a wearable or standalone device. However, due to the high computational and memory costs of DA-EMD algorithms, it is difficult to implement a DA-EMD algorithm using a pure software approach in a resource-limited MCU. Moreover, the quality of the IMFs must be maintained. Computation latency and power consumption must be minimized as well. In this paper, we combine fast EMD and the recently developed low memory EMD (LMEMD) implementation to significantly reduce the run time and memory to meet the real-time constraints. We also develop the adaptive phase number UPEMD to further reduce the run time of UPEMD. In addition, we develop a theory to estimate the relation between data latency and boundary error so that we can determine the required data latency within the accepted boundary error. Different DA-EMD algorithms are realized in a wearable system to extract clean electrocardiogram (ECG) signals and ECG-derived respiration (EDR) for demonstration. The performances of different DA-EMD algorithms are also compared. The proposed software approach has the advantage of being able to be implemented on various existing embedded platforms as an available and practical signal decomposition solution for edge computing.