Adaptive filtering of EEG/ERP through Bounded Range Artificial Bee Colony (BR-ABC) algorithm

Abstract

In this paper, the Artificial Bee Colony (ABC) algorithm is applied to construct Adaptive Noise Canceller (ANC) for electroencephalogram (EEG)/Event Related Potential (ERP) filtering with modified range selection, described as Bounded Range ABC (BR-ABC). ERP generated due to hand movement is filtered through Adaptive Noise Canceller (ANC) from the EEG signals. ANCs are also implemented with Least Mean Square (LMS) and Recursive Least Square (RLS) algorithm. Performance of the algorithms is evaluated in terms of Signal-to-Noise Ratio (SNR) in dB, correlation between resultant and template ERP, and mean value difference. Testing of their noise attenuation capability is done on contaminated ERP with white noise at different SNR levels. A comparative study of the performance of conventional gradient based methods like LMS, RLS, and ABC algorithm is also made which reveals that ABC algorithm gives better performance in highly noisy environment.

Introduction

As per need of optimization in every field, the emerging technologies play an important role to benefit the science and engineering application. The rising complexity has forced researchers to discover possible ways of easing solution of the problems. This motivates the researchers to grasp ideas from the natural organism, and implant it in engineering and sciences. Evolutionary Computation (EC) is a form of stochastic optimization search. It includes Swarm Intelligence (SI) based algorithms and evolutionary algorithms. EC is also applicable in area of Computational Intelligence (CI). Algorithms such as Genetic Algorithm (GA), Ant Colony Optimization (ACO), Particles Swarm Optimization (PSO), and Artificial Bee Colony (ABC), etc., are all inspired from natural organisms and phenomenaʼs [1], [2].

There are many application areas in which bio-inspired/swarm intelligence techniques are proven effective than the conventional gradient based techniques. Adaptive signal processing and filter design also gets benefited through use of swarm intelligence and evolutionary techniques [2], [3]. Filtering of ERP from EEG is one of the well-known application area of adaptive filtering, under adaptive noise cancellation scheme [4], [5], [6]. EEG and ERP signals are more specifically discussed below along with the review of previous work and proposed combination of ABC algorithm with adaptive filtering.

EEG signals, which are multichannel signals, recorded as brain wave in form of electrical signals, reflect the response of stimulation or a task, known as Evoked Potential (EPs), or Event-Related Potentials (ERPs). Stimulation for ERP generation has various types like visual, auditory, and motor movement, etc. [6], [7]. ERPs are weak signals buried in signals of spontaneous EEG with very low Signal-to-Noise Ratio (SNR) [7]. ERP in EEG are enhanced and extracted by simple linear methods based on synchronized averaging, power spectral analysis and rectified averaging [8]. Presently, ERP analysis has become a major part of the brain research. These ERPs play an important role in design and development of Brain–Computer Interface (BCI) [8], [9]. EEG classification also motivates researchers to explore various states reflected in EEG [10]. Extreme Leaning Machine (ELM) is the famous techniques proposed to overcome the slow learning of neural networks; basically neural network is trained with gradient based algorithm [11], [12], [13]. Neural network trained with ELM was successfully tested on EEG classification for five mental tasks [10]. Effectiveness of ERP analysis depends only on EEG signal of high SNR value. EEG signals are noisy and non-stationary due to its process of generation from group of neurons. EEG signals are contaminated by artifacts due to line noise, muscle movements, sometime with cardiac signals (ECG), eye blinking and eyeball movements also [7], [14]. Therefore, during the past decades, several techniques have been developed for artifact removal from EEG signals [14].

The simplest and most widely applied method for analysis of ERPs is averaging of the measurements over an ensemble of trials, also known as Ensemble Averaging (EA). It is an optimal way to improve signal-to-noise ratio (SNR) when underlying model of the observations is assumed that ERP is a deterministic signal independent of additive background noise. Major drawback of averaging technique is its dependency on number of trials or more trials are needed for better results [15], [16]. Filtering is another common method used for the single trial analysis of ERP, through which the contamination due to on-going background activity can be attenuated from ERP. A major disadvantage of filtering method is very low SNR, and the performance of filter in detection of signals depends on statistical properties of the signal, which is to be processed. To overcome these problems, concept of adaptive filters and its applications as noise canceller was introduced by Widrow et al. [17]. Since then, adaptive noise cancellation techniques have been used in many engineering applications [17], [18], [19], [20].

Literature review explores that various types of algorithms or error estimation methods have been exploited in adaptive filters to adjust the weights of filter, and error estimation according to EEG signals and noise property [4], [5], [6], [7]. Most efficient gradient based algorithms are LMS, RLS and their different variants used for adaptive filtering of EEG/ERP. Recently, swarm intelligence and evolutionary techniques have emerged as robust tool for solving linear and nonlinear equations. These techniques use the concept of random population generation which acts as possible solutions. There are very few references available in which ABC has been employed for adaptive filter design [3], [20], [21]. Adaptive filtering implementation is previously done with many of evolutionary techniques, swarm and computational intelligence techniques (GA, PSO, ANN, etc.), and applied in many areas [22], [23], [24], [25], [26], [27], [28]. As such there are no references available on ABC based ANC for EEG/ERP filtering. Main contribution of the proposed work is the methodology that explains the design and application of ANC through ABC algorithm and EEG/ERP noise removal respectively.

The “Bounded range” term indicates the range declaration method. Normally, range of search space (solution set) is declared as “±R” (R integer or float number) that covers the random number generation and updation during execution of algorithm. It is a conventional method. The proposed idea or modification in ABC is the modification of range declaration method as “$\mathrm{R}\pm \mathrm{C}$” (C is also an integer or float value) used, in which “C” acts as the control or bound (bounded range) over “R”. Thus, for the first time adaptive noise canceller with ABC algorithm for ERP filtering is proposed with modification in range selection method of bees using a constant (C), and behavioral analysis of ABC is done with increasing colony size and Bees range. Results of LMS and RLS methods are also illustrated only to observe the difference in effectiveness between gradient based and ABC based adaptive filtering. This proposed method provides a novel idea based on the applied research of computational intelligence method in the field of EEG signal processing.