$Id: README,v 1.4 2004/07/21 19:12:58 dlowder Exp $
The idea of tslib is to have a core library that provides standardised
services, and a set of plugins to manage the conversion and filtering as
The plugins for a particular touchscreen are loaded automatically by the
library under the control of a static configuration file, ts.conf.
ts.conf gives the library basic configuration information. Each line
specifies one module, and the parameters for that module. The modules
are loaded in order, with the first one processing the touchscreen data
first. For example:
module variance delta=30
module dejitter delta=100
These parameters are described below.
With this configuration file, we end up with the following data flow
through the library:
raw read --> variance --> dejitter --> linear --> application
module module module module
You can re-order these modules as you wish, add more modules, or remove them
all together. When you call ts_read(), the values you read are values that
have passed through the chain of filters and scaling conversions. Another
call is provided, ts_read_raw() which bypasses all the modules and reads the
raw data directly from the device.
There are a couple of programs in the tslib/test directory which give example
usages. They are by no means exhaustive, nor probably even good examples.
They are basically the programs used to test this library.
TSLIB_TSDEVICE TS device file name.
Default (non inputapi): /dev/touchscreen/ucb1x00
Default (inputapi): /dev/input/event0
TSLIB_CALIBFILE Calibration file.
TSLIB_CONFFILE Config file.
TSLIB_PLUGINDIR Plugin directory.
TSLIB_CONSOLEDEVICE Console device.
TSLIB_FBDEVICE Framebuffer device.
Module Creation Notes
For those creating tslib modules, it is important to note a couple things with
regard to handling of the ability for a user to request more than one ts event
at a time. The first thing to note is that the lower layers may send up less
events than the user requested, because some events may be filtered out by
intermediate layers. Next, your module should send up just as many events
as the user requested in nr. If your module is one that consumes events,
such as variance, then you loop on the read from the lower layers, and only
send the events up when
1) you have the number of events requested by the user, or
2) one of the events from the lower layers was a pen release.
Variance filter. Tries to do it's best in order to filter out random noise
coming from touchscreen ADC's. This is achieved by limiting the sample
movement speed to some value (e.g. the pen is not supposed to move quicker
than some threshold).
This is a 'greedy' filter, e.g. it gives less samples on output than
receives on input.
Set the squared distance in touchscreen units between previous and
current pen position (e.g. (X2-X1)^2 + (Y2-Y1)^2). This defines the
criteria for determining whenever two samples are 'near' or 'far'
to each other.
Now if the distance between previous and current sample is 'far',
the sample is marked as 'potential noise'. This doesn't mean yet
that it will be discarded; if the next reading will be close to it,
this will be considered just a regular 'quick motion' event, and it
will sneak to the next layer. Also, if the sample after the
'potential noise' is 'far' from both previously discussed samples,
this is also considered a 'quick motion' event and the sample sneaks
into the output stream.
Removes jitter on the X and Y co-ordinates. This is achieved by applying a
weighted smoothing filter. The latest samples have most weight; earlier
samples have less weight. This allows to achieve 1:1 input->output rate.
Squared distance between two samples ((X2-X1)^2 + (Y2-Y1)^2) that
defines the 'quick motion' threshold. If the pen moves quick, it
is not feasible to smooth pen motion, besides quick motion is not
precise anyway; so if quick motion is detected the module just
discards the backlog and simply copies input to output.
Linear scaling module, primerily used for conversion of touch screen
co-ordinates to screen co-ordinates.
interchange the X and Y co-ordinates -- no longer used or needed
if the new linear calibration utility ts_calibrate is used.