BWView -- Recorded brain-wave viewing app
=========================================
Copyright (c) 2002 Jim Peters, . All rights
reserved. Released under the GNU General Public Licence version 2.
See the file COPYING for details, or visit .
FFTW code is Copyright (c) 1997-1999 Massachusetts Institute of
Technology, released under the GNU GPL. See
for the original sources.
The SDL library code is released under the GNU Lesser General Public
Licence version 2 or later. See file COPYING.LIB for details. See
for more information on this project.
"This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License
version 2 as published by the Free Software Foundation."
"This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details."
"You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
USA"
Note that this software is released as freeware. However, if you want
to give me some money I'm not going to argue with you. See my pages
here for details:
http://uazu.net/money/
Brief spec
----------
This program allows the analysis of multiple-channel brain-wave data
into log-frequency against time amplitude plots. It is designed for
navigating around the data and getting a visual understanding of it as
quickly as possible. It is also designed to be accurate and
predictable, like a good engineering tool. The algorithm is based on
a FFT-accelerated convolution of the signal with a series of complex
wavelets. Each wavelet is formed by combining a Blackman window with
a complex 'corkscrew' at the desired frequency (there must be a
mathematical term for this). The window-width used is proportional to
the wavelength, and this ratio can be varied within the program to
change the display to show different aspects of the data.
(TODO: The program also uses phase information to give accurate
measurements of the centre-frequency of any trace on the display.)
Controlling the program
-----------------------
The following keys allow navigation around the file:
PageUp Back a screen
PageDn Forward a screen
BackSp Back a screen
Space Forward a screen
Left Back a half-screen
Right Forward a half-screen
Up Move up an octave in frequency
Down Move down an octave in frequency
Home Go to the start of the file
End Go to the end of the file
You should run the tool with Caps Lock OFF. The following shifted
letters and other keys give special functions:
shift-O Optimise the FFTW engine. This makes sure the FFTs are
running as fast as possible on your machine. The
optimisation information is saved into a 'wisdom' file, so
you only need to do it once, if you do it at all (FFTW seems
to work well in any case, with or without optimisation).
ESC Exit the program.
Q Exit the program.
. The dot (or 'period') key causes BWView to recheck the file
to see if any further data has been written since it last
looked and update the display. This can be used to analyse
a file whilst it is being recorded.
shift-F Turn 'follow' mode on or off. When turned on, the display
updates to the current end-of-file every second. This makes
it easier to automatically follow a file whilst it is being
written to. Other commands continue to work normally
whilst the file is being followed.
There are many settings that affect what is displayed, and these are
summarised in the top-left corner of the screen at all times. If you
press any of the letters listed there, a short description of the
parameter will appear on the status line, and the current value will
be shown at the very top-left of the screen. You can now adjust this
value, either using the number keys or the '+' and '-' keys. The
number keys provide preset values that are quick to access, and the
'+' and '-' keys can be used for fine-tuning if required. The presets
are all read from a configuration file, which you can customise if you
want to. See later for that.
Here are the keys related to this:
a-z Select a setting to adjust
0-9 Change the setting to one of its preset values
+ Increase a setting by a small amount
- Decrease a setting by a small amount
= Same as '+' (for UK/US users, so they don't have to
hold shift)
Moving the mouse cursor over the main area of the display allows you
to read off time, frequency and magnitude information. It also shows
you the nearest peak frequency on the display, and its magnitude.
Clicking and holding the mouse button replaces the normal signal
display with a version that shows the windowed signal that has been
used to calculate the point the mouse cursor is over. This is useful
to get an understanding of the analysis method, and to see which
features in the signal data might be affecting an area in the main
display that you are interested in.
Note that, for very large window widths, you won't be able to see the
whole of the window used to calculate that point within the width of
the display. Adjusting the 't' setting will solve this problem.
Getting started with some data
------------------------------
Try loading a file using one of these command-lines:
bwview jm2/140 log001.dat
bwview jm4/130 example.dat
When the data first appears, you probably won't see too much. You
need to start by adjusting the brightness. Press 'b', and then '9' or
'0' and work down the digits until it looks right. You'll also
probably need to adjust the gain for the signal level display so that
you can see it properly. Use 's' and then digit keys until that looks
okay.
It's worth bearing in mind that if you see a bright white area in the
main area of the display, the signal is probably saturated (meaning
the values are too large to be displayed properly), and you're
probably missing a lot of detail. Reduce the brightness a bit and
you'll see.
If you see red lines in the signal display area, these mean that there
were errors in the data file, probably sync errors. Bear this in mind
when interpreting data around these points. If it is all red, then
you have probably chosen the wrong file format!
If updates are very slow on your machine, you can reduce the
resolution using 'v'. This makes it quicker to navigate -- for
example using 'v4' to move around, and then switching back to 'v1'
once you've spotted something you want to look at. You can also make
the window smaller by resizing it, or even run in full-screen mode at
640x480 or 800x600 if your machine supports that (see later section).
You can also adjust the font -- 'f1' gives small text, and 'f2' gives
large text.
Once you've got used to adjusting brightness and moving around, it's
time to understand a bit more about how the analysis works. At each
moment of time for each frequency, the signal is analysed by testing a
'window' of the original sampled signal. The window used is a
'Blackman' window which gives good rejection of other frequencies.
Click on a part of the main display to see the window used to
calculate that point. The window used for a high frequency is
narrower than that used for a low frequency, because the wavelength of
a high frequency wave is much shorter; using a narrow window at a low
frequency would give very little information.
The size of the window used is therefore measured in wavelengths, and
when expressed in wavelengths, it is constant over the whole display.
The default presets allow you to adjust the window from 0.5
wavelengths up to 16 wavelengths. Using a narrow window (small
numbers) gives a lot of time-information, but frequency information is
very blurred. Using a wide window (larger numbers) gives precise
frequency information, but the traces are blurred in time (left to
right).
For instance, if you had a pure sine-wave as input, using a narrow
window would show a wide blurry line across the trace. As the window
is made wider, the line becomes thinner and thinner, more and more
precise. However, if the sine-wave changes in frequency then with the
very wide window you would see a lot of blurring in the areas where
the frequency changed, because the window is effectively averaging a
changing frequency. A shorter window would show less blurring in this
case.
Now you have a basic idea of the effect of the window width on what
you are viewing, have a play with adjusting it. Press 'w', and use
the digits 0-9.
Have a look at some of the test files, for example 'log001.dat' from
Jim Meissner. This file in particular is good to look at as it
contains stable tones, sweeps, as well as various types of modulation
(EXAMPLES.TXT contains some additional notes on these files). At one
point in that file there is some amplitude modulation of a carrier.
By changing the window width 'w', you can change the display to show
this signal either as the individual pulses or as a carrier and
side-bands! These are both valid interpretations of the data.
Remember, if you find something interesting or unusual in a file, it
is worth looking at it with a few different window widths. That way
you're likely to have a better understanding of what you're seeing in
front of you.
The other settings you will need to use are as follows: 'n' allows you
to adjust the number of octaves displayed on the screen, i.e. the
frequency range in the left margin. 'o' allows you to change the top
octave, i.e. the highest frequency shown, which is also adjusted by
the up/down keys. 'c' allows you to change between different channels
if the input file has more than one. 't' lets you change the
time-base, which lets you fit more into the screen horizontally. 'f'
allows you to change the font size. 'm' allows you to change the
display mode.
Here are a summary of the display modes currently available with the
'm' setting:
m1: Greyscale display of magnitude.
m4: Colour display of magnitude -- the magnitude is mapped to a range
of colours which makes it easier to see areas of similar
magnitude in the display. The brightness setting ('b') controls
the display just as for 'm1'.
m3,m5: A 'sideways waterfall' display, with magnitude controlling both
colour and 'elevation' of the peaks. m3 is left-leaning, m5 is
right-leaning.
m2,m6: Another colour 'sideways waterfall' display, only this is
sliced up every 8 pixels to make the shape of the contours more
obvious. Again there are left and right-leaning versions so you
can look behind tall peaks if necessary.
m7,m8: Two experimental displays using phase information to enhance
the display. See NEWS.TXT for details.
Command-line arguments
----------------------
Run 'bwview' with no arguments to see an up-to-date summary of all the
options. Also run 'bwview -f' to see the current list of supported
file formats.
The simplest command-line includes just the format of the file and the
filename itself. For example:
bwview jm4/130 example.dat
This opens a window and enters the program. Use 'shift-Q' to quit.
If you want to run the program full-screen, then you need to use the
-F option. This requires a screen-size as specification. Here are
some examples:
bwview -F 640x480x16 jm4/130 example.dat
bwview -F 800x600x32 jm4/130 example.dat
bwview -F 1024x768x16 jm4/130 example.dat
Note that the last part of the screen-size is either x16 or x32.
'x16' gives a 16-bit display (32768 or 65536 colours). 'x32' gives a
24-bit display (16 million colours). If the screen size is not
available, then you will either get an error message, or the SDL
library will emulate the mode for you somehow.
If you do not wish to run full-screen, you can set the initial size of
the window using -W, for example:
bwview -W 900x600 jm4/130 example.dat
You can also supply settings on the command-line using the -c option.
For example, if you wanted to use 1024x768 full-screen, switching to
display mode six and a 4-octave frequency range automatically, use
this:
bwview -F 1024x768x16 -c f2n4 jm4/130 example.dat
You can use any sequence of letters and digits, as well as '+' and '-'
in this string, just as you would when using the program directly.
Note that it is also possible to preload settings using the config
file (see below).
Notes on some of the supported file formats
-------------------------------------------
'jm2' and 'jm4' are 2 and 4-channel Jim Meissner files, with one 0x03
sync byte followed by 2 or 4 unsigned data bytes. Sync loss can be
detected thanks to the sync bytes.
'bm1' and 'bm2' are 1 and 2-channel BrainMaster files at 120Hz. The
1-channel file is just the same as raw/120:b, except 'bm1' is more
convenient. The two-channel version has a sync byte, permitting sync
errors to be detected.
'mod' support 6-channel modularEEG files as output by the modularEEG
design from the OpenEEG project (openeeg.sf.net). Follow it by the
sampling rate, which is normally 256Hz: 'mod/256'. The channels
appear as 'c1' to 'c6'. The four digital outputs also appear as
channels 7 to 10.
'mod0' supports the old (pre Apr-2003) modularEEG data format, but is
the same as 'mod' otherwise.
'raw' allows raw files of various formats to be read. The spec
consists of a sample-rate and a format-spec, for example
'raw/256:ssss'. There is no detection of sync-loss.
The format-spec is made up of characters, each of which consumes bytes
in the input file. All except '_' (which represents dummy bytes) also
represent different channels in the input file. The format characters
are:
_ Dummy byte (byte is skipped and ignored)
b Unsigned byte
w,W Little and big-endian unsigned 2-byte words
c Signed byte (aka 'char' in C)
s,S Little and big-endian signed 2-byte words (aka 'short' in C)
f Machine-format 32-bit floating point number
For example 'raw/128:_bbbb' could be used to read a 'jm4' file above,
assuming the sync was perfect all the way through the file.
'raw/128:ss' can be used to read 16-bit headerless stereo sound files.
It is fairly easy to add support for other file formats -- programmers
can see the file "src/file_formats.inc" for instructions. Please let
me have the source code for any new formats you add so that I can
incorporate them into the main release.
Adjusting presets in the configuration file
-------------------------------------------
All of the presets accessed by using the letter keys followed by
digits in the main program can be adjusted in the configuration file
'bwview.cfg'. This is useful if you find that the presets don't suit
the values that you use most often. It is also possible to set up an
initial set of presets that will be used when the program starts.
The file is a plain-text file. Lines starting with '#' are taken as
comments, and blank lines are ignored. A preset line consists of the
preset name (e.g. 's6') followed by white space and the value for that
preset. There should be nothing else on the line.
The initial setup is specified using a line consisting of 'init'
followed by white space and the startup string enclosed in double
quotes, for example:
init "f2t4n6o2b6s9--"
Within the string, all letters, digits and +/- keys may be used
exactly as in the main program.
Wisdom file
-----------
The FFT engine used in this program "FFTW" (see www.fftw.org) can be
optimised for your particular processor and system. To do this, press
'O' (shift-O) within the main program. After some time, this will
finish, and a 'wisdom' file "bwview.wis" will have been created on
your disk. This wisdom will be loaded up whenever you run the program
again, so you won't need to optimise the engine again. However, if
you do move to a new machine, it would be a good idea to re-optimise
the engine and/or delete the wisdom file.
Note that this optimisation step isn't 100% necessary -- FFTW performs
very well in any case.
Features for experimenters and filterbank designers
---------------------------------------------------
By invoking BWView with the -x option, two further Window functions
become available. The 'x' setting switches between them (x1,x2,x3).
These additional windows are based on IIR low-pass filters. These are
quick to execute, and are suitable for real-time use as a filterbank.
I put them into BWView to allow me to better understand their
behaviour, and to select suitable parameters for building filterbanks.
The two filters are as follows:
'x2': IIR with Q=0.5. This Q-factor gives the fullest bell-shaped
frequency response curve without the window tail ever crossing zero.
'x3': IIR with Q=0.72. This Q-factor gives the squarest frequency
response peak, without there being a dip in the middle of the
response. However, the window tail does cross the zero line (although
I don't how that might be significant).
By clicking on the main area, as usual, the window shape can be seen
and comparisons made between the different windows. Note that the IIR
filters are not symmetrical, and so features in the main display are
delayed in time compared to the original signal. The delay becomes
noticeable and significant for lower frequencies.
The 'window-width' can be varied as for the Blackman window. Since
these IIR windows never actually end, the window-width is taken to be
the width that contains 95% of the volume between the curve and the
axis. The IIR window is *not* truncated at the point (as you can
see). For the purposes of calculation, the 99.9% point is used
(i.e. expect values to be correct to 3 significant figures).
Script for modularEEG users
---------------------------
The 'rec-modeeg' script allows Linux users to easily record from the
modularEEG and start a BWView session running, automatically set to
follow the end of the data as it is written. Run it with no arguments
if your serial device is /dev/ttyS0, or else give it the serial device
as the first argument. See the script itself for more details.
// END //