RU2774300C2 - Method and system for geolocation of terminal occupying certain geographical position - Google Patents

Abstract

FIELD: wireless communication.

SUBSTANCE: using messages transmitted by calibration terminals (72) at calibration stage (20), set (26) of support elements is determined, each of which contains a radio signature and a geographical position of a calibration terminal at the moment of message transmission; optimization stage (30) allows for the purification and/or enrichment of set (26) of support elements by means of identification of points of interest corresponding to preferable positions of message transmission by terminals; at search stage (40), position of terminal (70) of interest is determined based on its radio signature and modified set (36) of support elements.

EFFECT: increase in the accuracy of terminal geolocation.

15 cl, 14 dwg

Description

The field of technology to which the invention belongs

The present invention relates to the field of geolocation. In particular, the invention relates to a method and system for geolocation of a terminal of a wireless communication system. The invention is successfully applied, in particular, to the geolocation of smart objects such as the "Internet of Things" (Internet of Things, IoT).

State of the art

Over the past few years, the growing use of wireless communication systems has naturally led to the development of services based on the geographical location of the object and used, for example, to assist in navigation, in traffic control, control of the transport of goods and, in general, for remote reading of information, etc. .d.

Satellite positioning systems such as GPS ("Global Positioning System") are some of the best known geolocation technologies. These systems are based on the use by the terminal of a receiver of radio signals transmitted by specially designated satellites. Satellite geolocation is very accurate, but suffers from many disadvantages, in particular, the cost and power consumption when integrated into the object of the GPS receiver, as well as the lack of the ability to work in closed areas.

There are other technologies for geolocation of the terminal based on the exchange of signals with the base stations of the access network with which it is connected. In cellular networks, such as GSM (Global System for Mobile Communications), as is known, the position of the terminal is estimated relative to the position of the base station with which it is currently associated. This method of geolocation has, however, mediocre accuracy, since the coverage area of the base station can reach within a radius of several kilometers or even several tens of kilometers.

Other methods include estimating distances separating a terminal from a plurality of base stations by calculating arrival times, propagation angles, or frequency differences for signals exchanged between the terminal and base stations. All of these different methods, however, have the disadvantage of requiring special hardware and software. In addition, they often require costly synchronization of various base stations acting as lookouts. Finally, they are particularly sensitive to a phenomenon called multipath (propagation of the same radio signal over multiple paths due to the phenomena of reflection, refraction, and diffraction from obstacles encountered).

Other geolocation methods are based on the received power level (abbreviated as RSSI, from "Received Signal Strength Indicator") of the signal exchanged between the terminal and the base station. These techniques are particularly well adapted to wireless communication systems such as cellular (eg, such as GSM) for which the RSSI information is directly available as it is used by the communication system itself. These methods are based on the fact that the radio signal is attenuated in the atmosphere and thus the RSSI level of the signal received by the receiver varies according to the distance separating the receiver from the signal source. Thus, it is possible to determine the geographic location of a terminal using trilateration by estimating the distance from the terminal to the various base stations that surround it based on the RSSI levels measured by the base stations. The disadvantage of this trilateration geolocation method based on RSSI levels is the lack of accuracy due to the fact that numerous parameters affecting signal attenuation (obstacles, radio interference, terminal movement, etc.) make the distance function based on the RSSI level very complex.

Thus, new methods of geolocation based on RSSI levels have been developed. These new ways are based on machine learning technologies. Specifically, this includes building, at the calibration stage, a database that is associated with known geographic locations, a radio signature corresponding to all RSSI levels measured for a terminal at a given location for a set of system base stations. Then, in the search phase, the radio signal observed by the terminal located at the unknown position is compared with all database signatures to estimate the position of the terminal based on the position(s) corresponding to the most similar signature(s).

Such methods, however, have several disadvantages, in particular in the case where the geographical area to be covered is very large, for example if an entire country or even a continent is to be covered. The first drawback is related to the accuracy of geolocation. Indeed, it happens that two radio signatures corresponding to two adjacent geographic locations are significantly different, or two very similar radio signatures correspond to two geographic locations separated by terrain. Establishing a relationship between a radio signature and its associated geographic location thus becomes difficult due to the noise generated in such situations. Another disadvantage is related to the complexity of the machine learning algorithm used. Indeed, to obtain sufficient accuracy, a large number of elements must be inserted into the database created at the calibration stage. In addition, if the area to be covered is large, a large number of base stations must be considered. All this contributes to a significant increase in data when entering a machine learning algorithm and, thus, creating restrictions on the computation time and computing power. In order to perform the calibration step, it is known to place on board a group of vehicles moving through the area to be covered, devices adapted to accurately provide geolocation and RSSI levels for base stations of the communication system at various points (the English term for qualification this stage is "war-driving", "wardriving"). The larger the number of points, the better the performance of the geolocation method in terms of accuracy, but the longer and more expensive the calibration step is.

The essence of the invention

The object of the present invention is to overcome all or part of the disadvantages of the prior art, in particular those disclosed above.

To this end, and in accordance with the first approach, the present invention proposes, in a wireless communication system, a method for geolocation of a terminal called "terminal of interest". The method contains steps in which:

- determine on the basis of messages transmitted by other terminals, called "calibration terminals" (calibration terminal), and received by the base stations of the mentioned wireless communication system, a set of reference elements, each reference element containing a radio signature associated with the geographical position of the calibration terminal at the time when a message is transmitted, said radio signature corresponds to a set of values representing the quality of wireless links existing between said calibration terminal and base stations,

- determining, on the basis of a message transmitted by said terminal of interest located at the geographic location to be estimated, the radio signature of said terminal of interest,

- identify specific geographic locations, called "points of interest" based on geographic locations of reference elements from a set of reference elements,

- modifying the set of reference elements based on the identified points of interest,

- estimating the geographic location of said terminal of interest based on the radio signature of said terminal of interest and the modified set of reference elements.

Estimation of the geographic location of the terminal of interest is performed, for example, by a machine learning algorithm that aims to establish a relationship between a radio signature and geographic location.

The point of interest corresponds to the preferred point of transmission of messages, in which, for the terminals of the considered wireless communication system, there are, on average, more messages transmitted from or near the point of interest than from another geographic location.

For example, a point of interest may correspond to a particular geographic location where the likelihood of a terminal of the system in question being present is greater than elsewhere. In particular, in the case of transporting pallets of goods, equipped with terminals and moving from warehouse to warehouse with a relatively short travel time relative to the time spent in the warehouse, said warehouses can correspond to points of interest.

According to another example, the point of interest may correspond to the convergence point of the system terminals. In particular, airports may correspond to points of interest for baggage provided with system terminals.

Points of interest can be dynamically identified based on reference elements provided with calibration terminals. In particular embodiments, at least a portion of the points of interest may also be known in advance.

Modification of a set of datums corresponds, for example, to an optimization step during which undesired, disparate datums (e.g., those judged to be far from the points of interest) are removed and/or in which corresponding datums (e.g., those that are considered to be close to the points of interest) additional information is added to refine and/or enrich the set of pivots. Such measures provide gains in terms of geolocation accuracy since this optimization step amplifies the relationship that exists between the radio signature and geographic location, taking advantage of the existence of points of interest where the terminal is more likely to be present than elsewhere.

In addition, the removal of disparate elements helps to reduce the number of pivots used by the learning algorithm and thus reduce its complexity, which translates into an improvement in terms of computation time and computing power.

Finally, these measures can make it possible to reduce the size of the radio signature, for example, considering only base stations that are located near the points of interest, which further reduces the complexity of the learning algorithm.

In specific embodiments, the implementation of the invention may further contain one or more of the following features, considered separately or in accordance with all technically possible combinations.

In particular embodiments, the geographic location of the terminal of interest is estimated in the form of a value representing the probability of said terminal of interest being located at the identified point of interest.

Such measures make it possible, for example, to conclude that the terminal of interest is located at the point of interest for which the probability value is the largest, if this value is greater than a certain threshold, or to conclude, if this value is below the mentioned threshold, that the terminal of interest is not located near which -or point of interest (located, for example, at the transition between two points of interest).

In specific embodiments, the identification of points of interest comprises the steps of:

- dividing the geographical area to be covered into a plurality of cells having a given size,

- calculate for each cell values representing the set of reference elements located in said cell,

- if the calculated value satisfies the given criterion, then the point of interest corresponding to the said cell and/or the geographic locations of the reference elements located in the said cell is determined.

In specific embodiments, the identification of points of interest comprises the step of calculating for each reference element the distance separating said reference element and each other reference element from each other, as well as subsequent steps performed periodically on a temporary set of reference elements, initially corresponding to the set of reference elements, until stop condition is reached:

- for each reference element of the temporary set of reference elements, a value is calculated representing the set of reference elements, called "adjacent elements", located at a distance less than a given threshold,

- if the largest computed value for an element called "element of interest" is greater than a given threshold:

- determine the point of interest according to the geographic locations of the element of interest and/or its neighboring elements,

- exclude the element and its neighboring elements from the temporary set of reference elements.

In particular embodiments, the geolocation method further comprises the step of filtering points of interest identified according to their geographic location and/or value representing a plurality of reference elements of a set of reference elements located at a distance less than a predetermined threshold.

In particular embodiments, a value representing the plurality of anchors is computed according to a weighting factor assigned to each anchor and representing the time that has elapsed between the time the message was sent by the calibration terminal and the time at which the points of interest are identified.

Such measures make it possible, in particular, to promote recent points of interest by giving more importance (ie, by assigning a larger weight) to reference elements obtained at a time close to the time of estimating the position of the terminal of interest.

In specific embodiments, changing the set of reference elements for each reference element of the reference element set comprises the following steps, in which:

- calculate the smallest distance separating said reference element and one of the identified points of interest,

- if the smallest calculated distance is less than the given threshold:

- associating the corresponding point of interest with said reference element,

- otherwise:

- exclude the said support element from the set of support elements.

In specific embodiments, modifying a set of support elements comprises, for each support element from the set of support elements, the following steps, in which:

- calculating for each identified point of interest a value representing the distance separating said reference element from said point of interest,

- associate the values thus calculated with said reference element.

In particular embodiments, the wireless link is uplink towards a base station of the wireless communication system.

In particular embodiments, determining the radio signature of a terminal comprises the steps of:

- transmitting by means of the terminal a message in the direction of the base station,

- measuring for each base station a value representing the quality of the wireless link existing between said terminal and the base station in question, based on a message received from said terminal,

- form by means of a server connected to the base stations, the radio signature of the terminal based on the measured values.

In specific embodiments, the calibration terminal is provided with a positioning system, and determining the radio signature of said calibration terminal further comprises the steps of:

- inserting into the message transmitted by said calibration terminal the current geographic position of said calibration terminal measured by the positioning system,

- retrieve by the server the geographic location contained in the said message.

Such measures make it possible, in particular, to be able to easily enrich the set of reference elements with new reference elements coming from the calibration terminals, which makes it possible to detect new points of interest that may appear in time and/or to remove points of interest that have become obsolete.

The calibration step comprising determining or enriching the set of reference elements may be performed once at the beginning, or may be repeated continuously (for example, periodically or in a manner controlled by sending messages to the calibration terminals) or may be performed continuously (for example, if the calibration terminals regularly transmit messages containing their current geographic location).

Thanks to the existence of some specific terminals equipped with a positioning system, it is possible to fully automate the calibration step, which includes the definition of a set of reference elements. In addition, the implementation of the geolocation method in a wireless communication system is particularly simple and inexpensive, since it does not require additional special hardware at base stations or at most terminals (hardware and/or software modifications may only be required at terminals operating as calibration terminals).

In particular embodiments, a value representing the quality of a wireless link existing between a terminal and a base station is the received power level of a radio signal exchanged between said base station and said terminal.

In particular embodiments, the wireless link is an ultra-narrowband link.

The system of ultra-narrow band wireless communication ("Ultra Narrow Band" or UNB in Anglo-Saxon literature), in particular, is adapted to the use of IoT type devices. Of course, it allows communication with low bit rate, long range and low power consumption (LPWAN type system, in English "Low Power Wide Area Network").

In particular embodiments, the estimation of the geographic location of a terminal of interest based on the radio signature of said terminal of interest and a modified set of reference elements is performed using a decision tree based machine learning classification algorithm.

In accordance with the second approach, the present invention relates to a wireless communication system containing terminals and an access network containing base stations and a server connected to said base stations. Said system comprises a database storing a set of reference elements formed on the basis of messages transmitted by terminals, called “calibration terminals”, and received by at least one base station. Each reference element contains a radio signature associated with the geographic location of the calibration terminal at the time the message is transmitted. Said radio signature contains values representing the qualities of wireless links existing between said calibration terminal and base stations. In addition, the access network is configured to:

- measuring values representing the quality of the wireless links existing between the base stations and the terminal whose geographic location is to be estimated, called the "terminal of interest",

- generating a radio signature corresponding to all the values measured for said terminal of interest,

- identify specific geographic locations, called "points of interest", based on a set of reference elements,

- modifying the set of reference elements based on the identified points of interest,

- estimate the geographic location of the terminal of interest based on the radio signature of the terminal of interest and the modified set of reference elements.

Brief description of the drawings

The invention will become better understood upon reading the following description, given by way of example, which is not intended to be limiting in any way, and created with reference to FIG. 1-14, on which:

fig. 1 - a wireless communication system containing a calibration device;

fig. 2 shows the main steps of the conventional method for geolocation of a terminal of a wireless communication system;

fig. 3 shows the main steps of the geolocation method according to the invention;

fig. 4 shows a preferred embodiment in which the calibration device is a terminal equipped with a positioning system;

fig. 5 shows the main steps of the preferred embodiment of the geolocation method calibration step;

fig. 6 shows the main steps of the preferred embodiment of the geolocation method search step;

fig. 7 shows the main steps of the first specific embodiment of the point of interest identification step;

fig. 8 is a first specific embodiment of the step of identifying points of interest;

fig. 9 shows the main steps of a second specific embodiment of the point of interest identification step;

fig. 10 is a second specific embodiment of the step of identifying points of interest;

fig. 11 shows the main steps of the first specific embodiment of the base set modification step;

fig. 12 shows the main steps of a second specific embodiment of the base set modification step;

fig. 13 are graphs of the distance error distribution for the conventional geolocation method;

fig. 14 are graphs of the distance error distribution for the geolocation method according to the invention.

In the drawings, like references throughout the drawings refer to the same or like elements. For purposes of clarity, elements are not necessarily shown to scale unless otherwise noted.

Detailed description of embodiments

As stated above, the present invention is directed to estimating the geographic location of a terminal, referred to as a "terminal of interest", in a wireless communication system using a learning method based on information representing the quality of wireless links existing between said terminal and base stations of said wireless communication system.

In FIG. 1 is a schematic representation of a wireless communication system 60 comprising a plurality of terminals 70 and an access network 80 comprising a plurality of connected base stations 81 and a server 82 connected to said base stations 81. In such a system 60 communication may typically be bi-directional, i.e., data may be transmitted from the access network 80 to the terminal 70 on the wireless downlink, or from the terminal 70 to the access network 80 on the uplink wireless communication. Thus, the measurement of the quality of the wireless link existing between the terminal 70 and the base station 81 of the access network 80 may be performed by one or the other of these entities, for example.

Thus, one or more calibration devices 71 adapted to accurately determine the current geographic location, for example, using a positioning system such as a GPS receiver, can be introduced into the wireless communication system 60. This calibration device 71 further has means configured to allow measurement of a value representing the quality of the wireless link existing between it and the base stations 81 of the access network 80. As indicated above, this measurement can be performed, for example, by the calibration device 71 itself on the downlink, or by the base station 81 of the access network 80 on the uplink. If the measurement is performed by the downlink calibrator 71, this measurement can be made, for example, on radio signals destined for system terminals 70 other than the calibrator 71.

Values representing the quality of the wireless link existing between the terminal 70 or calibration device 71 and the base station 81 may be transmitted to the server 82 to implement certain steps of the geolocation method.

In FIG. 2 shows the two main steps of a method 10 for geolocation of a terminal 70 of interest of a wireless communication system 60 based on a learning method.

The first stage 20 calibration contains:

- a step 22 of determining radio signatures respectively associated with known geographic locations, each radio signature corresponding to a set of values representing the quality of the wireless links existing between the calibration device 71, the geographic location that is known, and the plurality of base stations 81 of the access network 80 of the wireless communication system 60 , and

- step 24, in which each radio signature and the associated geographic location of the calibration device 71 are stored in the set of 26 reference elements.

In the rest of the description, "geographic location" means a set of two coordinates corresponding to latitude and longitude, respectively. It should be noted that alternatives to determining geographic location are possible. For example, a third coordinate corresponding to the height relative to mean sea level may also be taken into account.

In short, this calibration step 20 comprises the creation of a kind of radio map of the geographic area in question.

The search stage 40 then comprises steps in which:

- at step 42, the radio signature of said terminal of interest 70 located at the geographic location to be estimated is determined,

- in step 44, the geographic location of said terminal of interest 70 is estimated based on the determined radio signature and the set of reference elements 26 collected during calibration step 20.

The geolocation method 10 of the invention introduces an additional step 30 called "optimization step 30", which aims to clean up the datum set 26 (i.e., remove unwanted datums) and/or enrich it (i.e., add information to the stored datums). elements) to improve the geographic location estimate 44 of the reference terminal 70.

This optimization stage 30 is based on the idea that terminals 70 may tend to pass through certain geographic convergence locations of terminals 70 of system 60 (meaning geographic locations through which a large number of terminals 70 tend to pass). In such specific geographic locations, the number of messages transmitted by the terminals 70 is thus, on average, greater than in any other location. In some cases, terminals 70 may also preferentially occupy a finite set of specific geographic locations (in other words, at a given time, terminal 70 may statistically have a higher probability of being located in one of these specific geographic locations than elsewhere). In the description, "point of interest" refers to such a specific geographic location.

This situation occurs, for example, in the case of transporting goods moving from warehouse to warehouse with a travel time relatively shorter than the time spent in the warehouse. For example, it is possible to provide pallets used for transporting goods with terminals 70 of the wireless communication system 60 in order to be able to geolocate them. The various warehouses in which the cargo pallets are successfully stored thus correspond to the points of interest of the geolocation method 10 . A truck transporting said cargo pallets can, for example, carry a calibration device 71 which makes it possible to form a set of support elements 26 .

To estimate the geographic location of a terminal of interest 70, instead of estimating the position of a terminal 70 over the entire geographic coverage area, the geolocation method 10 of the invention estimates the position of a terminal of interest 70 according to identified points of interest. In other words, instead of using the traditional regression method, which may require a very large number of pivots (and thus very large computational power and computation time) to obtain reasonable accuracy, the geolocation method 10 according to the invention uses a classification algorithm machine learning, which includes, for example, assigning the terminal of interest 70 a location probability at each of the identified points of interest. The estimated geographic location of the terminal of interest 70 may thus, for example, be defined as the geographic location of the point of interest where it has the highest probability of being located. According to the value of the returned probability (eg, if it is less than a predetermined threshold), it can also be determined whether the terminal 70 is located at the location of interest (eg, if it is a terminal 70 on a pallet that moves between two warehouses).

In FIG. 3 schematically shows the main steps of the geolocation method 10 according to the invention. In addition to the calibration step 20 and the search step 40 already described with reference to FIG. 2, the geolocation method 10 includes an optimization step 30 .

This optimization step 30 comprises a first point of interest identification step 32 which, based on the set of 26 reference elements collected during the calibration step 20, allows a set of 35 points of interest to be determined. This step will be described in detail later with reference to FIG. 7-10.

Then, using the set of identified points of interest 35, step 34 modifying the datum set 26 provides a modified datum set 36, which is thus used in the search step 40 to determine the geographic location of the terminal of interest 70. This step 34 of modifying 34 will later described in detail with reference to FIG. 10 and 11.

Of course, it is preferable to identify points of interest and assign greater importance to datums whose geographic location is close to the position of the point of interest. For example, you can exclude datums that are geographically more distant from the point of interest than a specified threshold. Moreover, it is preferable to associate with each anchor element the nearest point of interest or a value representing the distance that separates the geographic location of said anchor element from each point of interest. Such measures, of course, allow to reduce the "noise" generated by elements that are not grouped together near the points of interest, in the ratio that the machine learning algorithm seeks to establish between the radio signature and the point of interest.

Here, "distance" means a distance calculated in a traditional way between two geographic locations expressed using longitude and latitude. Thus, "distance between two anchors" means the distance between the geographic locations of said two anchors. The same applies throughout the description unless otherwise noted.

In the remainder of the description, by way of example and in a non-limiting manner, the case of an ultra-narrowband wireless communication system 60 is considered. "Ultra Narrow Band" or UNB in Anglo-Saxon literature means that the instantaneous frequency spectrum of the radio-electric signals transmitted by the terminals has a bandwidth of less than two kilohertz, or even less than one kilohertz. "Radioelectric" means an electromagnetic wave propagated by wireless means, the frequencies of which are contained in the conventional spectrum of radioelectric waves (a few Hertz to a few hundred Gigahertz). Such UNB wireless communication systems, in particular, are adapted to the use of devices of the IoT type or the M2M type (Anglo-Saxon short for "Machine to Machine", machine-to-machine).

In such a wireless communication system 60, communication is essentially unidirectional, in this case uplink, from the terminals 70 to the access network 80 of said wireless communication system 60. To minimize the risks of losing a message transmitted by terminal 70, scheduling of the access network 80 is often performed such that a given geographic area is simultaneously covered by multiple base stations 81 so that a message transmitted by terminal 70 can be received by multiple base stations 81.

Each base station 81 adapts to receive messages from terminals 70 that are located within range. Each message thus received is, for example, transmitted to the server 82 of the access network 80, optionally accompanied by other information such as the identity of the base station 81 that receives it, a value representing the quality of the radio signal carrying the message, the center frequency at which it was the message was received, the date the message was received, etc. The server 82 processes, for example, all messages received from the various base stations 81. The server 82 can, in particular, be used to implement the geolocation method 10 of the terminal 70 of the system 60.

In FIG. 4 schematically shows a preferred embodiment of such a system 60, in which some terminals 72 of the wireless communication system 60, referred to as "calibration terminals 72", have a positioning system (such as a GPS receiver) that can accurately obtain the geographical position of the terminal 72 (for example, when the last is located in a place where it can receive signals from the satellites of said positioning system). These calibration terminals 72 act like the calibration devices 71 mentioned above, in addition to their day-to-day functions in the wireless communication system 60.

It should be noted that other methods of obtaining the geographic location of the calibration terminal 72 are possible. to which said calibration terminal 72 is connected if the geographic location of said access point is known. The choice of a specific method for determining the geographic location of the calibration terminal 72 is simply an alternative implementation of the invention.

In FIG. 5 shows the main steps of a preferred embodiment of the calibration step 20 of the geolocation method 10.

For this calibration step 20, instead of deliberately moving through a geographic area that is to be covered by a special device adapted to determine the geographic location and measure a value representing the quality of the wireless link at that location, it is preferable to use the available computing power to fulfill this role. some terminals of the system and in this case it concerns the calibration terminals 72.

Thus, the first step of the calibration step 20 comprises the transmission 220 by the calibration terminal 72 of a message containing the current geographic location to the access network 80 of the wireless communication system 60 . It should be noted that this message may be transmitted independently of the geolocation method 10 . For example, it may be a traditional message for remote reading information containing the current geographic location, and not have the main purpose of participating in the calibration step 20 of the geolocation method 10.

The base stations 81 of the access network 80 which receive the signal containing said message thus perform the measurement 221 of the quality of the wireless link over which the message was transmitted. In the preferred embodiment, and in the rest of the description, by way of non-limiting example, the value representing the quality of the wireless link used is the base station 81's average received power level (RSSI), expressed in decibels, for example, for a signal carrying said message.

It should be noted that other values representing the quality of a wireless link can be used, such as signal attenuation, signal-to-noise ratio (SNR or "Signal on Noise Ratio" in Anglo-Saxon literature), or link quality indicator (CQI or " Channel Quality Indicator" in English).

In radio signature generation step 222, the server 82 thus collects the RSSI levels measured by the various base stations 81 and inserts them into the radio signature thus determined for the calibration terminal 72. For the base stations 81 that did not receive the message, for example because that terminal 72 is too far away and not in their radio coverage area, for example, the default value is -160 dB.

It should be noted that for the choice of base stations taken into account in the formation of the radio signature, many options are possible. According to the first example, conventionally, all base stations of the wireless communication system 60 are considered. According to another example, the number of base stations can be limited to take into account, for example, only a limited geographic area of interest.

The server 82 thus performs the retrieval 224 of the geographic location information contained in the message transmitted by the calibration terminal 72 (as a reminder, this message was previously transmitted to the server 82 by the base stations 81 that received it).

Finally, a pair of pieces of information, formed by the geographic location of the calibration terminal 72 and the accompanying radio signature, is added to the set 26 of reference elements collected during the calibration step 20. This step, for example, comprises storing 24 pairs of pieces of information generated by the geographic location and the accompanying radio signature in a database stored in the server 82.

In particular embodiments, the steps of the calibration step 20 described above are repeated over a predetermined amount of time, such as more than a few days, or even a few weeks or a few months, for the calibration terminals 72 of the wireless communication system 60 to obtain a set of reference elements 26, containing a significant number of pieces of information, that is, an accurate map of the area to be covered. In one alternative, the steps of step 20 of the calibration are repeated until the set 26 has a predetermined number of reference elements.

In a preferred embodiment, the set of reference elements 26 is continuously enriched with new reference elements coming from the calibration terminals 72. Such measures make it possible, in particular, to detect new points of interest that may appear in time and/or to forget points of interest that have become obsolete.

It should be noted that just like the terminals of interest 70, the calibration terminals 72 are mobile and can thus provide reference data corresponding to different geographic locations.

In particular embodiments, anchor element set 26 is also continuously enriched with elements of each geographically appropriate terminal of interest 70 associated with the radio signature of said terminal of interest 70.

It should also be noted that, in accordance with other embodiments, information about the current geographic location of the calibration terminal 72 may be transmitted by a communication system other than that in which RSSI measurements are made to determine the radio signature of said calibration terminal 72. For example, the current geographic location of the calibration terminal 72 may be transmitted to the server 82 via a mobile telephone network such as GSM, UMTS or LTE, while the message for which RSSI measurements are performed is transmitted to the access network 80 of the UNB wireless communication system 60. For ease of understanding, the server 82 can associate the radio signature defined for the calibration terminal 72 with its current geographic location, for example, with a timestamp of a message carrying information about the current geographic location and a message transmitted by the UNB wireless communication system.

In FIG. 6 shows the main steps of the preferred embodiment of the step 40 of searching for the geolocation method 10.

This search stage 40 is initiated by the transmission 420 of a message of interest to the terminal 70 of the access network 80 of the wireless communication system 60 . This message may be any given message transmitted regardless of the terminal's geolocation method 10 . For example, this message may be sent for the purpose of remotely reading information that has nothing to do with the geolocation of terminal 70. Alternatively, this message may be sent intentionally for the purpose of geolocation of terminal 70. In all cases, the content of the message is not necessarily important at stage 40 of the search.

Similar to what is performed for the calibration step 20, the base stations 81 of the access network 80 that have received said message perform a measurement 421 of the RSSI level of the signal carrying the message.

In radio signature generation step 422, the server 82 thus collects the RSSI levels measured by the various base stations 81 and includes them in the radio signature thus determined for the terminal 70 of interest.

The estimation 44 of the geographic location of the terminal 70 of interest that transmitted the message is thus performed on the one hand based on the reference element set 36 modified in the optimization step 30 and on the other hand based on the radio signature determined for the terminal of interest 70, the geographic whose position is to be assessed.

In FIG. 7 schematically shows the main steps of the first specific embodiment of step 32 of identifying points of interest. This step 32 of identifying points of interest is implemented by the service 82 in the optimization step 30 based on the reference element set 26 collected in the calibration step 20 .

The first step corresponds to dividing 320 the geographic area to be covered into a plurality of predetermined cells. This division may correspond, for example, to the division of the geographical area to be covered by means of a regular grid, the cells thus corresponding to the same squares. According to another example, the cells may be circular in shape with a predetermined fixed diameter and partially overlap each cell with at least one other cell to cover the entire area of a geographic area. According to another example, the cells may have various shapes and/or sizes, determined, for example, in accordance with certain pieces of information known prior to the division of the terminals 70 by the geographic area to be covered.

Server 82 then performs, for each cell, step 322 which calculates the number N _C of reference elements whose geographic locations are within said cell.

If for a given cell the number N _C thus calculated is greater than or equal to a predetermined threshold N ₁ , then the point of interest is identified for said cell. The point of interest can, for example, be defined as being the center of a cell. In one alternative, the point of interest may be defined as the average of the geographic locations of the reference elements located in the cell.

When all cells are taken into account, the step 32 of identifying points of interest ends. A set of 35 identified points of interest is thus available and stored in the server 82.

In FIG. 8 shows a first specific embodiment of the point of interest identification step 32 described with reference to FIG. 7. Geographical area 50 is divided into a grid of 16 identical square cells. Each cell is identified by a pair "L-C", where L is a letter from A to D and C is a number from 1 to 4. The reference elements 51 obtained during the calibration step 20 are indicated by crosses. Each cross represents a geographic location associated with a datum 51 obtained in the geographic area 50 in question. In this example:

- N _C = 0 for cells A-1, D-1, C-2 and B-3, N _C = 1 for cells C-1, D-2, A-3, C-4 and D-4,

- N _C = 2 for cells A-2, B-2, D-3 and B-4, N _C = 5 for cell A-4,

- N _C = 6 for cell C-3, and

- N _C = 7 for cell B-1.

In this example, the threshold N ₁ is set to 5 and the point of interest 52 is determined to be the center of the cell for which N _C ≥ N ₁ . Each point of interest 52 thus determined is shown in FIG. 7 black dot. In this example, three points of interest 52 are identified. They correspond to the center of cells B-1, C-3 and A-4, respectively.

This first particular embodiment of point of interest identification step 32 is more or less accurate according to cell sizes. For example, in FIG. 8, it is clear that the center of cell C-3 is relatively far away from the more likely position of the point of interest, which may be located in the lower right part of the cell, where most of the reference elements 51 located in said cell are located.

In FIG. 9 schematically shows the main steps of a second specific embodiment of step 32 of identifying points of interest.

For this second particular embodiment of point of interest identification step 32, operations are performed on a temporary datum set 27 that initially corresponds to the datum set 26 obtained in the calibration step 20 . Thus, the reference feature set 26 obtained during the calibration step 20 is not modified in this second embodiment of the point of interest identification step 32 .

In the first step, the server 82 calculates 321 the distances separating the two reference elements of the set 27.

In the second step, for each reference element of the temporary set of reference elements 27, the server 82 performs in step 323 the calculation of the number N _V of reference elements, called "neighbors", whose distance from the considered reference element is less than or equal to a given threshold D ₁ .

There is then a step 325 to determine the amount of N _Vmax corresponding to the largest value of the previously calculated amounts of N _V .

If N _Vmax is greater than or equal to the predetermined threshold value N ₂ , the server 82 then performs step 327 to determine the point of interest based on the reference element for which the number of neighbors is equal to N _Vmax , called the "element of interest", then step 329 to exclude said element of interest and its neighboring elements of the time set 27. A point of interest is defined, for example, as the average of the geographic locations of the element of interest and its neighboring elements. According to another example, a point of interest may be defined simply as the geographic location of the element of interest.

The stop condition is then evaluated. Until said stop condition is reached, there is periodic repetition starting at step 323 of calculating the number of neighboring elements remaining in temporary set 27.

If the number N _Vmax calculated in step 325 is less than N ₂ , then the stop condition is evaluated directly.

The stop condition is reached, for example, if a predetermined number of points of interest have been identified, or if the number N _Vmax becomes less than a predetermined threshold, or if there are no more elements in the time set 27 . When the stop condition is reached, the point of interest identification step 32 ends. A set of 35 identified points of interest is thus available and stored in the server 82.

In FIG. 10 shows a second embodiment of the point of interest identification step 32 described with reference to FIG. 9.

In FIG. 10 shows a geographic area 50 and support elements 51 identical to those shown in FIG. 8. Circles 530, 531, 532, shown in dotted lines, respectively group together elements of interest 510, 511, 512 and their adjacent elements. In this example, the first element of interest 512 is determined with 5 neighboring elements. Point of interest 522 is thus defined as the average of the geographic locations of said element of interest 512 and its neighboring elements (located within circle 532). Then, while iterating, a second element of interest 511 with 4 neighboring elements is determined. The point of interest 521 is thus determined. Finally, at the third and last iteration, based on the element of interest 510 also having 4 adjacent elements, the third point of interest 520 is determined. _Vmax neighbors, then the choice of one or another element of interest is arbitrary.

In particular embodiments, the point of interest identification step 32 described with reference to FIG. 7 and 9 may further comprise the step of grouping and/or filtering the identified points of interest. For example, if two points of interest are located at a distance less than a given threshold, then they are grouped into a single point of interest. According to another example, if for a given point of interest, the number of elements of the set of 26 reference elements located at a distance from said point of interest less than a given distance is less than a given threshold, then said point of interest is excluded.

In specific embodiments, instead of calculating an integer number of datums (e.g., at steps 322 and 323), a value representing the number of datums calculated according to the weighting factor assigned to each datum and representing the time elapsed between when the lines were used wireless communication (the measured RSSI levels of which are used to determine the radio signature) and the moment when points of interest are identified. Such measures make it possible, in particular, to promote recent points of interest by giving more importance (i.e., assigning a larger weight) to recent pivots, i.e., pivots corresponding to messages recently transmitted by calibration terminals 72. For example, the weighting factor can be determined for the reference element depending on time:

where:

A and σ are constant parameters,

t is the time elapsed since the received message transmitted by the calibration terminal 72 and used to determine the reference element,

exp() is an exponential math function.

With this definition of the weighting factor, the longer the time that has elapsed since the reception of the message that was used to determine the pivot, the smaller the value of the corresponding weighting factor and, thus, the less value given to the anchor compared to other anchors for which the corresponding the message was received later. Thus, it is possible to speed up the discovery of new POIs and ignore outdated POIs.

In FIG. 11 schematically shows the main steps of a first specific embodiment of step 34 of modifying a set of datums 26 based on a set of 35 identified points of interest.

In the first step, the server 82 performs for each element of the set of reference elements 26 the calculation 340 of the smallest distance D _min between said reference element and the point of interest of the set 35 of identified points of interest.

If this distance D _min is less than or equal to a predetermined threshold D ₂ , then the server 82 communicates 342 between the point of interest corresponding to the distance D _min and the reference element. In other words, the pivot is enriched with a unique identifier or label of the point of interest. The reference element thus enriched contains a radio signature, a geographic location, and a tag corresponding to the point of interest with which it is associated.

If, however, D _min > D ₂ , then server 82 performs pivot elimination at step 344 .

When all the anchor elements have been accounted for, the modification step 34 ends and the modified anchor element set 36 is thus available. This set 36 contains only datums that could be associated with the point of interest, that is, datums sufficiently close to the point of interest. Set 36 contains, for each point of interest, a certain number of associated radio signatures.

In FIG. 12 schematically shows the main steps of a second specific embodiment of step 34 of modifying a set of datums 26 based on a set of 35 identified points of interest.

The server 82 thus, in step 343, associates the probability values thus calculated with the reference element. In other words, the pivot is enriched with a set of values, with each value representing a probability of being associated with the point of interest. The reference element thus enriched contains the radio signature, geographic location, and probability value for each identified point of interest.

When all reference elements have been taken into account, the modification step 34 ends and the modified reference element set 36 thus becomes available. Set 36 thus contains, for each reference element radio signature, a probability of association with the identified point of interest.

In the search stage 40, the geographic location estimate 44 of the terminal 70 of interest is performed, for example, using a standard decision tree classification machine learning algorithm. The fact that the pivots present in set 36 have been enriched does indeed make it possible to transform a problem that usually seeks a solution in the form of a regression into a classification problem. Indeed, during the optimization step 30, each point of interest is assigned a label and the radio signature associated with each reference element is "classified" according to the probability of belonging to a particular label. During the search stage 40, based on the radio signature of the terminal of interest 70 and using the pivots present in the modified set 36, the classification algorithm is thus able to provide a probability of said radio signature belonging to a certain label. The geographic location of the terminal of interest 70 is thus typically estimated as the geographic location of the point of interest corresponding to the label with the highest probability value. If the largest probability value is less than a predetermined threshold, it can also be concluded that the terminal of interest 70 is not located at any of the identified points of interest (and in such a case, the terminal of interest 70 is likely to be on a transition between the two points of interest).

Numerous simulations show that the fact that the pivot set 26 was enriched and/or refined in the optimization step 30 gives the classification algorithm a noticeable gain in accuracy. The results of these simulations are shown in FIG. 13 and 14.

To perform these simulations, a reference element set 26 is assembled during the calibration step 20 and used for each simulation. In addition, elements called "test elements" are available, which are different from reference elements. Each test element contains a radio signature and a known geographic location. For each test item, a geographic location is estimated and a distance error corresponding to the distance between the estimated geographic location and the known geographic location is calculated. The graphs thus present curves 90-96 representing the distribution of the distance error for all test elements or only for a sample of test elements.

These simulations use 200,000 reference elements and 200,000 test elements. The size of the geographical area under consideration corresponds approximately to the size of a large country like France.

For the simulations whose results are shown in FIG. 13, a traditional geolocation method based on a regressive machine learning algorithm is used. The geographic location is estimated by the algorithm for each test element based on the radio signature of said test element and the reference element set 26 obtained during the calibration step 20 .

Plot 90 shows the distribution of distance error for all available test elements.

With respect to plot 91, it shows the distribution of the distance error for a sample corresponding to test items sufficiently close to the identified points of interest. In the present example, points of interest are identified based on the reference element set 26 using a second embodiment of the point of interest identification step 32 as described with reference to FIG. 9, and a test item is considered close enough to a point of interest if the shortest distance (D _min ) between the known geographic location of said test item and the identified point of interest is less than or equal to a threshold (D ₂ ). For these simulations, the D ₂ value is set to 250 meters.

Thus, from FIG. 13, it is clear that the regression algorithm is more accurate when it estimates the geographic location of a terminal of interest 70 located near one of the identified points of interest. Intuitively, this is due to the fact that the learning algorithm has better knowledge of the relationship between radio signature and geographic location for features close to the point of interest, since in set 26 the geographic density of the reference features is closer to the points of interest. The accuracy of the regression algorithm, however, is not satisfactory, eg the distance error is less than 0.5 km only about 50% of the time.

In FIG. 14 shows the results obtained by the geolocation method 10 according to the invention. A classification machine learning algorithm is used for the search stage 40 and uses the pivot set 36 modified in the optimization stage 30 . In the present example, the optimization stage 30 comprises a point of interest identification stage 32 in accordance with the specific embodiment described with reference to FIG. 9 and the step 34 of modifying the support element set 26 in accordance with the specific embodiment described with reference to FIG. 11. The classification algorithm thus provides, for the test item in question, a probability value for said test item associated with the identified point of interest. The estimated geographic location of the test item is thus defined as the geographic location of the point of interest for which said probability value is the largest. For these simulations, the value of D ₁ is also set to 250 meters and the threshold value of N ₂ is set to 300.

Plot 92 shows the distribution of distance error for all available test items. In this case, the distance error is less than 0.5 km approximately 70% of the time.

Plot 93 shows the distance error distribution for the sample of test items used for plot 91 corresponding to the test items under consideration, which should be close enough to the identified points of interest (point of interest distance less than or equal to threshold D ₂ ). For this sample, the distance error is less than 0.5 km approximately 90% of the time.

Plot 94 shows the distance error distribution for a sample of test items for which the largest probability value associated with the point of interest is greater than or equal to 0.9. For this sample, the distance error is less than 0.5 km approximately 87% of the time. This sample groups together 56% of test items. In this case, 32% of the test elements, considered as being close to the point of interest (distance less than or equal to the threshold D ₂ ), were "erroneously" removed from the sample.

Plot 95 shows the distance error distribution for a sample of test items for which the largest probability value associated with the point of interest is greater than or equal to 0.7. For this sample, the distance error is less than 0.5 km approximately 82% of the time. This sample groups together 76% of test items. In this case, 15% of the test items that are considered close to the point of interest (the distance to the point of interest is less than or equal to the threshold D ₂ ) were "erroneously" removed from the sample.

Plot 96 shows the distance error distribution for a sample of test items for which the largest probability value associated with the point of interest is greater than or equal to 0.4. For this sample, the distance error is less than 0.5 km approximately 73% of the time. This sample groups together 96% of the test items. In this case, 1.5% of test items believed to be close to the point of interest (distance to the point of interest less than or equal to the D ₂ threshold) were "erroneously" removed from the sample.

In the various cases shown in graphs 92-96, the distance error is less than 0.2 km in about 50% of the cases.

Thus, it is clear that the geolocation method 10 according to the invention, the results of which are shown in graphs 92-96 in FIG. 14 is clearly more accurate than the traditional geolocation method 10, the results of which are illustrated by graphs 90 and 91 in FIG. 13.

The results in FIG. 14 show that with the geolocation method 10 according to the invention, the geographic location of a terminal 70 of interest located near the point of interest can be estimated with good accuracy. It is also possible to determine if the terminal of interest 70 has a high probability of being en route between two points of interest (eg, if the largest value of the returned probability is less than a predetermined threshold).

The probability threshold allows you to determine whether the position of the terminal of interest 70 can be associated with the position of the point of interest, or, if on the contrary, it should be considered that the said terminal of interest 70 is on the way between the points of interest. This probability threshold may be determined, for example, according to a curve connecting the quality of the forecast and the confidence level of that forecast (for example, based on an inflection point or Youden index for an ROC curve (short for "Receiver Operating Characteristic")). According to another example, it can be chosen arbitrarily in accordance with the restrictions associated with the area of use.For example, if it is necessary to have a very high accuracy of the geolocation of the terminal of interest 70, at the risk of not being able to estimate the position of a significant number of terminals of interest 70, should be assigned a threshold value with high probability, such as a value of 0.9, as in the example shown in graph 94. If, on the contrary, it is necessary to be able to geolocate a significant proportion of terminals of interest 70, at the risk of reducing the accuracy of geolocation, then a lower threshold should be assigned zna probability value, for example, a value of 0.4, as in the example shown in graph 95. For the example under consideration, namely, the method of geolocation of cargo pallets of goods over a large area of geographical surface corresponding, for example, to a country or continent, the probability threshold is approximately 0 .7 seems to be a good compromise.

The above description clearly shows that due to its various features and advantages, the present invention achieves its intended objectives and also provides additional advantages.

In particular, the geolocation method 10 according to the invention significantly improves performance in terms of accuracy relative to standard geolocation methods based on learning methods. This accuracy gain is due, in part, to the optimization stage 30, which makes it possible to strengthen the relationship between the radio signature and geographic location by taking advantage of the existence of points of interest where the probability of the presence of terminals 70, 72 is greater than elsewhere. The optimization stage 30 also allows the elimination of disparate elements that do not match the point of interest and that contribute to the formation of "noise" that limits the performance of the learning algorithm.

In addition, the elimination of disparate elements helps to reduce the number of pivots used by the learning algorithm, and thus reduce its complexity, which translates into a gain in terms of computation time and computing power.

In addition, the exclusion of disparate elements can help to reduce the size of the radio signature (ie, the number of RSSI measurements of the radio signature) by considering only base stations that are located near the points of interest, further reducing the complexity of the training algorithm.

The fact that the list of points of interest for the terminals 70, 72 of the system 60 is known may also be of interest per se, for example to know the list of warehouses used in the transport of goods. Another advantage is based on the ability to assign a weight to each pivot according to the time elapsed since the message was transmitted by the calibration terminal 72 in order to enrich the pivot set 26 . Of course, this speeds up the discovery of new POIs and facilitates forgetting relatively old POIs.

Thus, adopting a geolocation method 10 such as the one described by way of example in the wireless communication system 60 is not particularly costly. Indeed, the method does not require the use of additional special hardware at multiple terminals 70 or base stations 81, and by having several special terminals 72 equipped with a positioning system in system 60, the method makes it possible to fully automate the calibration step 20.

In general, it should be noted that the embodiments discussed above have been described by way of non-limiting examples, and that other alternatives are therefore possible.

In particular, embodiments of step 32 of identifying points of interest and step 34 of modifying reference element set 26 have been given as an example, and other methods are possible as alternatives to the invention.

The invention has been described for an ultra-narrowband wireless communication system 60 adapted to use devices of the IoT type, but nothing prevents the geolocation method 10 of the invention from being implemented for other types of communication systems, such as, for example, GSM, UMTS, LTE, Wi-Fi cellular networks. etc.

Measurements representing the quality level of the wireless link used to generate the radio signature may be performed by base stations on the uplink of the wireless link as well as by terminals on the downlink of the wireless link. The results of the measurements are transmitted, for example, to a server, which then performs the other steps of the geolocation method. According to other examples, measurements representing the quality level of a wireless link may be performed directly by the server based on information received from terminals and/or base stations.

As stated above, the value representing the quality of the wireless link may be different from the RSSI level. This can be, for example, the level of signal attenuation, signal-to-noise ratio or another indicator of the quality of the radio channel.

In this example, the reference element corresponds to the geographic location obtained by the GPS receiver of the calibration terminal 72 associated with the radio signature of said calibration terminal 72 at said geographic location. In particular embodiments, anchor element set 26 may be further continuously enriched with anchor elements, each of which corresponds to a geographic location of a terminal of interest 70 estimated during the search phase and associated with the radio signature of said terminal of interest 70 located at that location.

However, other ways of defining the set of 26 support elements are also possible. For example, the geographic location of the calibration terminal 72 may be estimated at the calibration stage by a conventional geolocation method based on a regressive machine learning algorithm. According to another example, the geographic position of the calibration terminal 72 may be estimated at the calibration stage using another geolocation method, such as a TDOA (Time Difference Of Arrival) type method based on signal propagation time measurements. when exchanging between the terminal and different base stations.

During the search stage, various classification machine learning algorithms may be used to estimate the geographic location of the terminal 70 of interest based on its radio signature and the modified reference element set 36 . The fact that one or the other of these algorithms is used merely states an alternative embodiment of the invention.

The invention has been described considering the field of transporting goods. Nothing, however, precludes consideration of other areas of use of the geolocation method 10 of the invention, in particular areas to which geolocation-suitable terminals are able to preferentially occupy certain geographic locations or pass through certain geographic locations through which a large number of terminals tend to pass. 70 systems 60 (geolocation of baggage at airports, geolocation of packages, etc.).

In the example described, the points of interest are dynamically identified based on the references provided by the calibration terminal 72. Nothing prevents, however, that in particular embodiments, all or part of the points of interest are known a priori and the radio signatures associated with these points of interest are determined using the calibration devices 71 that can intentionally move to said points of interest known in advance.

Claims (45)

1. Method (10) for geolocation of a terminal, called "terminal of interest (70)" of a wireless communication system (60), comprising the steps of: based on messages transmitted by other terminals, called "calibration terminals (72)", and received by base stations (81) of said wireless communication system (60), a set (26) of reference elements is determined, each reference element containing a radio signature associated with a geographic position of the calibration terminal when the message is transmitted, said radio signature corresponds to a set of values representing the quality of the wireless links existing between said calibration terminal (72) and base stations (81), determining (42) based on the message transmitted by said terminal of interest (70) located at the geographic location to be estimated, a radio signature for said terminal of interest (70), wherein said geolocation method (10) is characterized in that it further comprises the steps of: identify (32) specific geographic locations, called "points of interest", based on the geographic locations of the reference elements of the set (26) of reference elements, modifying (34) the set (26) of reference elements based on the identified points of interest, estimate (44) the geographic location of said terminal of interest based on the radio signature of said terminal of interest (70) and the modified set (36) of reference elements. 2. The geolocation method (10) of claim 1, wherein the geographic location of the terminal of interest (70) is estimated in the form of a value representing the probability of finding said terminal of interest (70) at the identified point of interest. 3. The geolocation method (10) according to claim 1, wherein the identification (32) of points of interest comprises the steps of: divide (320) the geographical area (50) to be covered into a plurality of cells having a given size, calculate (322) for each cell a value representing the set of reference elements located in said cell, if the calculated value satisfies the predetermined criterion, a point of interest is determined (324) corresponding to said cell and/or geographic locations of reference elements located in said cell. 4. The geolocation method (10) of claim 1, wherein identifying (32) points of interest for each reference element comprises a step (321) calculating distances separating said reference element and each other reference element from each other, and subsequent steps performed repeatedly on a temporary set (27) of reference elements, initially corresponding to a set (26) of reference elements, until a stop condition is reached, on which: calculate (323) for each reference element of the temporary set (27) of reference elements, a value representing the set of reference elements located at a distance less than a given threshold, called "neighboring elements", if the largest computed value for an element called "element of interest" is greater than a given threshold: determining (327) a point of interest corresponding to the geographic locations of the element of interest and/or its neighboring elements, removing (329) the element of interest and its neighboring elements from the temporary set (27) of reference elements. 5. The geolocation method (10) according to any one of claims 1 to 4, further comprising the step of filtering points of interest identified according to their geographic locations and/or a value representing a plurality of reference elements of a set (26) of reference elements located at a distance less than the specified threshold. 6. Method (10) geolocation according to any one of paragraphs. 3, 4, in which a value representing a plurality of reference elements is calculated in accordance with a weight coefficient assigned to each reference element and representing the time that elapses between the moment the message is transmitted by the calibration terminal (72) and the moment at which the points of interest are identified. 7. Method (10) geolocation according to any one of paragraphs. 1-4, in which the modification (34) of the set (26) of support elements for each support element of the set (26) of support elements comprises the following steps, in which: calculate (340) the smallest distance separating said reference element and one of the identified points of interest, if the smallest calculated distance is less than the given threshold: linking (342) the corresponding point of interest to said reference element, otherwise: remove (344) said support element from the set (26) of support elements. 8. Method (10) geolocation according to any one of paragraphs. 1-4, in which the modification (34) of the set (26) of support elements for each support element of the set (26) of support elements comprises the following steps, in which: calculate (341) for each identified point of interest a value representing the distance separating said reference element from said point of interest, linking (343) the values thus calculated to said reference element. 9. Method (10) geolocation according to any one of paragraphs. 1-4, wherein the wireless link is uplink to the base station (81) of the wireless communication system (60). 10. The geolocation method (10) of claim 9, wherein determining (22, 42) the radio signature of the terminal (70, 72) comprises the steps of: transmit (220, 420) to the base stations (81) by means of the terminal (70, 72) a message (81), measure (221, 421) for each base station (81) a value representing the quality of the wireless link existing between said terminal (70, 72) and the considered base station (81), based on the message received from said terminal (70, 72 ), form (222, 422) by the server (82) connected to the base stations (81), radio signatures of the terminal (70, 72) based on the measured values. 11. The geolocation method (10) according to claim 10, wherein the calibration terminal (72) is equipped with a positioning system and determining (22) the radio signature of said calibration terminal (72) further comprises the steps of: enter in the message transmitted by the said calibration terminal (72) the current geographical position of the said calibration terminal (72) measured by the positioning system, retrieve (223) by means of the server (82) the geographic location contained in the said message. 12. Method (10) geolocation according to any one of paragraphs. 1-4, wherein the value representing the quality of the wireless link existing between the terminal (70, 72) and the base station (81) is the received power level of the radio signal exchanged between said base station (81) and said terminal (70, 72). 13. Method (10) geolocation according to any one of paragraphs. 1-4, in which the wireless link is an ultra-narrowband communication channel. 14. Method (10) geolocation according to any one of paragraphs. 1-4, in which the estimation (44) of the geographic location of the terminal of interest (70) based on the radio signature of said terminal of interest (70) and the modified reference element set (36) is performed using a machine learning classification algorithm based on a decision tree method. 15. A wireless communication system (60) comprising terminals (70, 72) and an access network (80) containing base stations (81) and a server (82) connected to said base stations (81), wherein said system (60) is characterized in that it contains a database storing a set (26) of reference elements formed on the basis of messages transmitted by terminals, called "calibration terminals (72)", and received by base stations (81), each reference element contains a radio signature associated with the geographical position of the calibration terminal at the time when the message is transmitted, said radio signature contains a value representing the quality of the wireless links existing between the said calibration terminal (72) and the base stations (81), and in which the access network (80) is configured to: measuring values representing the quality of wireless links existing between base stations (81) and a terminal (70) whose geographic location is to be estimated, called "terminal of interest (70)", generating a radio signature corresponding to all values measured for said terminal of interest (70), determining certain geographic locations, called "points of interest", based on a set (26) of reference elements, modification of the set (26) of reference elements based on the identified points of interest, estimating the geographic location of the terminal of interest (70) based on the radio signature of the terminal of interest (70) and the modified set (36) of reference elements.

Images