Jitter on a stationary tag is reduced by a "static filter". When a location event is less than the static
distance from the last one it is damped by blending it with the last event and damping it by a factor specified
by static alpha.
But if the tag moves more than the static distance away from the last event its location is accepted immediately.
If static distance is made too big (e.g. more than about 30cm) it will interfere with tracking in confusing ways.
Static distance and static alpha can be used as a cosmetic measure to reduce
jitter on stationary tags, but a large static distance value should never be used.
Filters explained
Terminology
In order to understand how to optimise the filter settings
for your application, you need to be aware of the following:
Filter modes: There are 2 filter modes: "stateless" and "stateful".
Stateless mode: Finds the tag position using data from sensors.
Stateful mode: Uses the filter state to reject outlier sensor measurements and help reduce errors.
Variance: In stateful mode, this is how certain we are of the location of the tag.
Large variance = not very certain.
Tracking the tag over time
In the stateful filter, the following happens:
As time passes, the variance will get bigger, because the tag might be moving.
This process is called filter state decay.
As good data from sensors is added to the filter state, the variance will get smaller.
When the variance is small enough, a location event is asserted.
Losing track of the tag
There are 2 ways in which the stateful filter can lose track of the tag:
Variance is so big that no outliers are rejected.
Variance is too small and grows too slowly over time so that all good data is rejected as outlier data.
There are a couple of terms used to describe the filter state:
When most sensors agree with the filter state, it has good support.
When most sensors do not agree with the filter state, it has low support.
After several events with low support we conclude that we have lost track of the tag.
Transitioning between modes
Stateless mode
There are 2 cases in which the stateless filter is required:
The tag is seen for the first time so there is no filter state.
Stateful mode has lost track of the tag.
The stateless filter uses a least-squares method to find the tag position
that fits the given data with a residual standard error.
You can specify the acceptable maximum standard error for each cell:
Pre-2.1.5 systems had typical standard error values around 0.05 - 0.1.
2.1.5 and all subsequent systems have an option to treat the standard
error as a distance with typical values around 0.2 - 1.0.
Configuring the location algorithms
The stateful filter parameters and some of the stateless parameters
are configured in the Location Engine Config "Filters" tab:
There are further stateless mode parameters in an individual cell's properties...
...and in the properties of the cell master sensor. This tab is only available
by typing Ctrl-Shift-A in the main LEC window. From 2.1.6, the max standard
error text box is repeated next to the "2.1.5 error mode" check box to remind
you to that if you change the error mode, you need to change the
max standard error as well:
Table 1: The stateful algorithms
Choose the model with the fewest dimensions that still accurately describes
your tracking scenario. This pushes knowledge of the world into the filter,
making outlier rejection better.
Table 2: Parameters describing object motion
Table 3: Filter state management parameters
Table 4: Stateless mode parameters
Step-by-step configuration process
Optimize tracking using stateless mode:
Make sure orientations and cable offsets are correct,
use 2.1.5 error mode and choose an acceptable max standard error.
Choose a filtering algorithm: If possible, use "static fixed height information filtering".
Refer to table 1 to check which one is suitable for your application.
Set static distance to zero: It can make results confusing when tuning.
Set parameters describing object motion: Refer to table 2.
Set the required update rate of the tag:
Remember that a low update rate means more filter decay between updates.
Specify state reinitialization behaviour: Refer to table 3. - Set both variance parameters low. - Set low support reset count to 1. - Increase min reset measurements to just below the value where you stop getting location events. - Set low support reset count back to its default value.
Specify thresholds for acceptable uncertainty: Refer to table 3. - Set the sighting trace flag on the cell master. - Set both variance parameters high. - Track the tag in normal motion and reduce max position variance to just above the value where
the trace (LEC log) shows frequent low support and filter resets (i.e. calls to the stateless filter). - Set max valid position variance less than or equal to max position variance and such
that the quality of asserted location events looks good.
Specify stateless mode sanity check parameters: Refer to table 4.
Set static distance to some small value.
Alternative configuration process
The default settings have been chosen to support a wide range of sensible
real-world situations, so it is always worth trying them first:
Optimize tracking using stateless mode:
Make sure orientations and cable offsets are correct,
use 2.1.5 error mode and choose an acceptable max standard error.
Choose a filtering algorithm: If possible, use "static fixed height information filtering".
Refer to table 1 to check which one is suitable for your application.
