Abstract- Recent research has shown that selecting specific frequency bands for the extraction of invariant characteristics specific to each brain state can significantly improve the performance and accuracy of a brain-computer interface (BCI). Subject-specific frequency band selection helps in filtering out the nonevent related data, hence, assists in concentrating on the band where the separability between different classes is maximum. On the other hand, the subject-specific frequency band selection adds a new parameter to be tuned explicit to each subject which is a limitation for an autonomous BCI system. To realize a BCI system it is desirable to have an algorithm to govern the frequency band selection. This paper presents Particle Swarm Optimization (PSO) as a search algorithm to identify subject-specific frequency bands. A modified PSO method is proposed utilizing two independent swarms of particles in order to identify frequency bands which are maximally separable between two classes of mental tasks and thus enhance the classification of the BCI. Experimental results obtained utilizing BCI competition III data are presented. The results are promising and indicate that particle swarm optimization-based approach presents superior performance and quick convergence, therefore, not only improves the performance but also speeds up the processing time by a factor of between 4-8, compared to a thorough heuristic search for the separable frequency bands. I. INTRODUCTION rain-Computer interface (BCI) is a direct communication pathway between brain and an external device. It involves transforming signals from the brain into control signals for transmission of messages or commands, thus offering a new communication pathway between the human brain and the computer system [1]. Patients suffering from motor impairments, severe cerebral palsy and spinal chord injuries (SCI) may use a BCI system as a substitute communication pathway which relies only on the mental imagination and not on neuromuscular control. The main difference between BCI techniques and human-computer interface (HCI) tasks lies in not relying on muscular response, but only on detectable signals representing responsive or intentional brain activity. Recently there has been a significant growth in BCI technology but there are a significant number of issues and areas that need to be improved [2] due to the complexity and ambiguity of the EEG signals recorded from the brain [3]. Many BCIs are based on EEG signals which are modified by motor related brain activity, as these signals exhibit significant and lateralized event related activity. Event related desynchronization (ERD) is the phenomenon which results in amplitude attenuation of certain EEG rhythms when an event is initiated or is taking place in the brain [3][4]. On the other hand, event-related synchronization (ERS) is an amplitude enhancement of a certain EEG rhythm when cortical areas are not specifically engaged in a given mode of activity at a certain instant of time [3].To capture only these rhythms, generally a band pass filter between 8-26 Hz can be applied to filter out the non-event related data. This paper presents a precise autonomous frequency band selection algorithm within the range of 8-26 Hz based on a modified PSO algorithm. The PSO algorithm has been analyzed and modified whilst adhering to BCI requirements. PSO is a stochastic, population-based evolutionary computer algorithm for parameter tuning and problem solving. It was invented by Kennedy and Eberhart1 in the mid 1990s while attempting to simulate the choreographed, graceful motion of swarms of birds as part of a sociocognitive study investigating the notion of “collective intelligence” in biological populations [5]. It is analogous to a genetic algorithm (GA) in that the system is initialized with a population of random solutions and the potential solutions, called particles, are then “flown” through the problem space in search of the optimal solution [6]. Since the mid 1990’s, PSO has been used in numerous applications such as evolving artificial neural networks, analysis of human tremor, milling metal removal applications, optimization of reactive power and voltage control and ingredient mix optimization [6]. In this work, for the first time PSO has been applied in a BCI for parameter tuning - in this case frequency band selection. Results indicate that PSO improves the performance and significantly decreases the computational time overhead for an optima search by a factor of 4-8. The remainder of the paper is organized into four sections. Section II contains details on the BCI Competition III multichannel/multiclass datasets and acquisition procedure [7][8]. Section III discusses the PSO method and the modifications applied to it. Results are plotted and discussed in section IV. Section V provides the conclusion. Optimal Frequency Band Selection with Particle Swarm Optimization for a Brain Computer Interface Abdul Satti, Damien Coyle, Member, IEEE and Girijesh Prasad, SrMember, IEEE B The authors are with the Intelligent Systems Research Center, School of Computing and Intelligent Systems, Faculty of Computing and Engineering, Magee Campus, University of Ulster, Northland Road, Derry, Northern Ireland, BT48 7JL, UK. (phone: +44 (0)28 7137 5170; fax: +44 (0)28 7137 5470; email: dh.coyle@ulster.ac.uk ). Workshop/Summer School on Evolutionary Computing Lecture Series by Pioneers, August 18-22, 2008, Londonderry Copyright IEEE 2008 72