When a new calibration request is received, all the calibration requests
of the corresponding reference point are now cleared prior to add the
new request.
We should probably use a timeout to delete only the old calibration
requests and keep the recent ones.
The option positioning.accept-new-calibration-requests allows the
calibration requests sent during the positioning phase to be added to
the Stock. They are added to the calibration requests read by
InputDataReader during the start-up phase.
In InputUDPSocket, update the AP's coordinates when receiving a
calibration or autocalibration request that contains the coordinates.
This is done if the option positioning.update-ap-coordinates-online is
activated.
Add double Stock::ap_matrix_get_ss(transmitter, receiver).
This is the first step to handle self-calibration in owlps-positioning.
The idea is to simulates a square matrix of the listeners, with the
transmitters in one dimension and the receivers in the other dimension.
matrix[transmitter][receiver] is the signal strength received by
'receiver' from 'transmitter'.
The purpose of Stock::ap_matrix_get_ss() is precisely to access the
entry matrix[transmitter][receiver].
Note that the access is read-only, as the pseudo-matrix is (or will be)
filled by the standard mechanism of (auto)calibration requests and
measurements stored in Stock.
Add double ReferencePoint::average_measurements(string mac_transmitter),
that allows to get the average of all the Measurements sent by a given
transmitter.
Add, as protected functions, get_all_measurements(string
mac_transmitter) and get_requests(string mac_transmitter), that do the
same as get_all_measurements(void) and get_requests(void), but filtering
on the transmitter of the packets.
-lrt is now added as needed in each module Makefile, but not directly in
the library. Furthermore, the Makefiles add -lrt only if the platform is
Linux.
Accept 0 as a "valid" value for the constructor and operator=()
(constructing a Direction with 0 value was possible any way, using the
default constructor).
This fixes the reception of requests on UDP.
For RADAR, the default is now to compare the current positioning request
to each stored CalibrationRequest instead of each ReferencePoint
(averaging the calibration requests associated with the reference
point).
The new option --positioning.radar-average-reference-points allows to
activate the old behaviour (compare to reference points).
Note: the bug fixed in the previous commit caused the creation of
a ReferencePoint for each CalibrationRequest, implying the 'new'
default behaviour (one ReferencePoint was associated to only one
CalibrationRequest, so comparing each ReferencePoint or each
CalibrationRequest was giving the same result). Anyway, there was
(hopefully) no such bug in OwlPS v0.8, so it is actually the old
behaviour :-)
For each calibration request read, a new ReferencePoint was created in
Stock. This is because of the behaviour of unordered_set: as pertaining
to unordered_set, two elements are not considered equal if the hashes
alone are identical, they must also be actually equal (operator==).
The problem was fixed by using a custom equality check function that
checks the equality only on the coordinates of the ReferencePoint (as if
it was a Point3D).
ReferencePoint::complete_with_dummy_measurements(
map<string, Measurement>,
map<string, Measurement>)
becomes PosUtil::complete_with_dummy_measurements(
unordered_map<string, Measurement>,
unordered_map<string, Measurement>)
ReferencePoint::compute_ss_square_distance(
map<string, Measurement>,
map<string, Measurement>)
becomes PosUtil::ss_square_distance(
unordered_map<string, Measurement>,
unordered_map<string, Measurement>)
map<string, Measurement> ReferencePoint::get_all_measurements_sorted()
is now private, returns an unordered_map instead of a map, and is
renamed ReferencePoint::get_all_measurements().
Note that we now use unordered_maps for these functions. The use of
sorted lists (maps) was useless, as stated in the commit message of
e9eab687 (Wed Jun 2 12:37:43 2010 +0200).
If the real coordinates are passed to the constructor, the distance
error is computed. One can also use Result::compute_error() on a
constructed object.
Matteo Cypriani