Analysis of EEG using Ensemble Empirical Mode Decomposition
This work provides the back ground for analysis of non-linear and non-stationary brain signals by using Ensemble Empirical Mode Decomposition. The electroencephalogram (EEG) is a representative signal which contains information about the condition of the human brain. The disorder in brain is characterized by recurrent electrical discharge of the cerebral cortex. Detection of such disorders by visual scanning of EEG signal is a time consuming task and it may be inaccurate, particularly for long recording data set. In this paper an algorithm is presented which is based on the concept of Ensemble Empirical Mode Decomposition (EEMD). The main idea is White Gaussian Noise is added to the original signal and then EMD is performed. This realization is done several times and by averaging the modes we obtain the true values of modes. Empirical Mode Decomposition is being done over an ensemble of the Gaussian white noise plus signal. Hence Mode mixing problem is being resolved by populating the time frequency space plane.
The electroencephalogram (EEG) is a representative signal which contains information about the conditions of the human brain. EEG is non-linear and non-stationary signal. Most of the methods developed in the literature for EEG signal analysis and classification are based on time domain, frequency domain, and time-frequency domain. Several individual processing techniques and also combinations of those were employed and refined for analysis, quantification and recognition. Neural Networks (NN) have been used to detect abnormal patterns in the EEG . Wavelet Transform is also widely used for disorder detection .Others concepts combine Approximate Entropy and Lempel-Ziv Complexity , and Time Frequency Distributions. Recently a new technique for analysis of nonlinear and non-stationary EEG has been introduced which is based on the Ensemble Empirical Mode Decomposition (EEMD) of EEG.
Ensemble Empirical Mode Decomposition (EEMD)  is data adaptive method which is used for the analysis of non-stationary and non-linear signals. It comprises of a local and completely data-driven separation of a signal into slow and fast oscillations. It performs the EMD over an ensemble of the signal plus Gaussian white noise. This methodology resolves the problem of mode mixing by populating the whole time-frequency space. The main focus of this approach is sifting an ensemble of white noise and added signal, hence mean value obtained is treated as final result. Mode mixing is defined as a single Intrinsic Mode Functions (IMFs)  consisting of signals of widely disparate scale or a signal of un-comparable scale residing in different IMFs. Mode mixing  is a consequence of intermittency. Intermittency means signal is stopped or cease for some time. Intermittency can cause aliasing problems and also responsible for losing the meaning of IMFs. As we go from lower order IMFs to higher order IMFs, the scaling increases. This means scaling factor increases hence the signal decomposes. With this ensemble mean concept, we can separate scales without any a priori subjective criterion selection.This new approach utilizes the advantage of the statistical characteristics of white noise which slightly modifies the signal in its true solution neighborhood, and to cancel itself out after serving its purpose.