Analysis of Multiple-class Brain Computer Interface using ESPRIT

  • Asma AlRashedi

Student thesis: Master's Thesis


A brain computer interface (BCI) is a system that provides a non-muscular communication channel between the human brain and a control device. The growing interest in the current BCI research is motivated by the detailed understanding of the brain dynamics, the availability of cheap imaging devices, and the awareness of the needs of physically disabled individuals. BCI has emerged as an alternative that has the potential to improve the quality of the lives of millions of people worldwide who suffer from severe muscles impairment that hinder normal communication with the outside world. Though its main application is to assist the severely disabled people, the development of BCI systems is expanded to other fields such as entertainment and video games. A non-invasive BCI system works by extracting control signals from externally observable correlates of neuronal function, and using these to direct the operation of computers, wheel chairs and other devices. The electroencephalogram (EEG), recorded from the surface of the scalp, is one of the most widely used scanning technologies with which brain dynamics and functions are studied. However, EEG signals are highly complex due to the continuous and rapid electroencephalographic recordings changes and as well as the large amounts of data that are being recorded from different locations on the scalp. Thus, analyzing EEG signals requires a set of algorithms, drawn from a variety of fields including machine learning, data mining, signal processing and neuroscience that could help in removing the noise such as eye blinks and muscles, improving the spatial resolution of the signal, extracting the patterns that encode desired messages and classifying different brain states. The traditional feature extraction techniques rely mainly on the analysis of the spectral power changes that are associated with mental tasks. These methods depend on the frequency based methods such as Welch algorithm or by modeling the signal using parametric algorithms like Burg. Although these are widely used and reported to yield good accuracy, they still have several limitations that degrade the overall performance of BCI; in particular, they assume that the signal is Gaussian and exhibits linear characteristics. In this study, a novel application of an existing feature extraction technique is proposed. It is hoped that this technique will allow the characteristics of the EEG signals to be better exploited, and will accommodate the practical limitations of the current BCI systems. The method is known as 'Estimating Signal Parameters using Rotational Invariance Techniques' (ESPRIT). Though, it is used in power system applications, the study showed that it is a promising approach for BCI application. The obtained results of four mental classes are comparable to the results obtained by Burg and Welch. In addition, the performance of ESPRIT was benchmarked against published results, obtained using a well-known results, which were published on a well-known BCI competition. The analysis has shown that the proposed algorithm is competitive with the published results that are highly optimized.
Date of Award2011
Original languageAmerican English
SupervisorWei Lee Woon (Supervisor)


  • Brain-Computer Interfaces
  • Computer Interfaces

Cite this