Guide to the code.

The code comes in nine java files. I will go through them briefly.

The main file,, represents the main EEG worker window and implements all the panels, boxes, menus and button commands.

The Brain Map

Graph of the electrode nodes and neighbours. codes the shape of the display of the brain patterns in the left hand panel of the main window. It loads a set of 129 electrode positions from the file "data/129.CartAscii.txt". These 3D points are projected into a flat plane and connected into a nearest neighbour graph which is used to generate the interlocking polygons that colour each node, and to generate the NodeProbe patterns. contains the array of colours used for painting the nodes to represent the different intensities. If you are wanting to tinker with the program and make it use more attractive colours than simply red and blue, this is a good place to start.

Drawing showing polygons around the nodes painted in. In the released version, the wireframe shape is omitted for clarity. represents an actual drawing of the brain pattern as it is shown on the left hand panel of the main window. This is composed of a reference to the Ipos and the Imapcols, together with an array of 129 node intensities to represent. is the second window with the row of brain patterns. Consequently, it contains an array of Imaps.

The intensities data.

The data is loaded by the functions in This is of interest to anyone writing programs for reading output from EGI's machines. Their file comes in a simple format where all the electrode readings are recorded as short 16 bit integer numbers. There are functions for saving out a file in the same format. These should be ignored since they were merely used to trim down the example files I did have into one that was shorter. is where the calculations are controlled. The data from a single experiment, or the output from the fake data generator, is copied into here. This stores an array of, in the sample example, 850 129-dimensional vectors.

The most important function here is MakeSubspaceVex() which, given a series of vectors, electrodesub[0], ... electrodesub[nv - 1], loops through the 850 vectors electrodeins[0], ..., electrodeins[849] and for each i finds values of subspacescalars[i][0], ..., subspacescalars[i][nv - 1] such that the magnitude of the vector diff = electrodeins[i] - (subspacescalars[i][0] * electrodesub[0] + ... + subspacescalars[i][nv - 1] * electrodesub[nv - 1]) is minimized.

This latter equation can be expressed as: d = e - S * s where S is the matrix of subspace vectors (nv columns and 129 rows), and s are the scalars we want to find out. Left-multiplying both sides by Transpose(S) will set the left hand side to zero (if d is minimal), and leave the right hand side as an nv row column-vector minus a square matrix times another column vector. This can be solved using standard matrix arithmetic to derive a solution for s and thus a value for d.

When you click on the NextVex button, the program simply selects the vector with the highest magnitude of d as a good candidate for finding the widest support for the EEG space using the fewest basis vectors. contains the matrix functions, including matrix solution of the above matrix equation by Gaussian elimination. This code could be used in other applications.

The EEG time graph of one of the probes superimposed on a time graph of one of the subspacescalar values. represents the time graph of the electrode readings in the right hand panel of the main window.

This code is protected by the GNU General Public license which means you are free to use, modify and redistribute it as long as it does not get included as part of a program that is not also protected by an Open Source license.

Julian Todd 11/11/2001.