GB2563825A - Localization of mobile devices - Google Patents

Abstract

A method of estimating the location of a device (Figure 1, P), comprising receiving multiple signals, relating to multiple cells (Figure 1, A, B, C), then for each signal; determining information which may be an indication of signal strength or a cell identifier, and using that information to ascertain a plurality of parameters characterising a probability distribution for the location of the device given the determined information; for each signal calculating one or more refined parameters characterising a joint probability distribution for device location given the determined information from all signals; thirdly obtaining a location estimate based on the refined parameters and joint distribution. Ascertaining parameters may involve querying a database using the cell identifier as a key, and they may be an estimated location for the cell centroid and a confidence that the subscriber device is there, and calculating a joint probability distribution may involve calculating a weighted average of the centroids weighted by the confidence value.

Description

Localization of mobile devices

Background [0001] Localization of a mobile device is a term for determining the current location of a mobile device. There are a number of reasons that a device’s location may be required. Possible motivations include providing navigations instructions to a user via their smartphone and remotely tracking the location of a fleet of vehicles. It may be that the device needs to know its own location (as in the former case) or that another entity needs to know the location of one or more mobile devices in the network.

[0002]This application is concerned with mobile devices that are configured to communicate using a cellular network. These devices may also be subscriber devices, endpoint devices or Mobile Subscriber (MS) devices. These may include user devices such as smartphones also include so called “Machine to Machine” (M2M) devices or other connected devices in the internet of things (loT). The devices may be moveable or may alternatively be stationary when installed in certain pieces of equipment, such as vending machines or gas meters.

[0003] Many techniques currently exist for determining the location of mobile devices. An accurate location of the device can be acquired using a global navigation satellite system (GNSS). Examples of such systems include the United States’ Global Positioning System (GPS), the Russian Global'naya Navigatsionnaya Sputnikovaya Sistema (GLONASS) and the European Union and European Space Agency’s Galileo Positioning System. However, any such system requires direct line of sight from the mobile device to the satellites, which may not always be available. Moreover, the receiver uses a considerable amount of power and so is a drain on the power supply of the mobile device. This can significantly reduce the battery life of the mobile device in question.

[0004] Localisation may also be performed using information about the serving cell of the mobile device. Basic techniques for localization of mobile devices simply report the location of the serving base station or the centroid of the serving cell of the cellular network. However, such techniques only provide a rough location of a device with limited accuracy. This accuracy may be insufficient for some applications.

[0005] A more advanced technique obtains an estimate of the location of the mobile device through the use of “fingerprinting”. This technique compares the network conditions for the serving base station to a database of network conditions for that base station at various locations. The network conditions may include Cell Identification (Cell ID), Received Signal Strength (RSS), Angle of Arrival (AoA), Time of Arrival (ToA) and Time Difference of Arrival (TDoA). These techniques are discussed in the chapter “Localization in Real GSM Network with Fingerprinting Utilization” from the book “Mobile Lightweight Wireless Systems” by Jozef Benikovsky, Peter Brida and Juraj Machaj.

[0006] Fingerprints require a large effort to collect, at least with the current capabilities. Moreover, they are only valid for a relatively small timeframe, due to changing conditions in the network. Current estimates show that approximately 20% of network and cell id locations in a fingerprint database are required to be updated each month. In other words, 1 in 5 of the data points contained in the database for a particular cell tower becomes incorrect within a month of being entered into the database and more data is required for that tower in order to improve accuracy.

[0007] An alternative approach is for determining the location of a mobile device is through the use of a Serving Mobile Location Centre (SMLC). The SMLC is a standardized network localization solution that allows for a more accurate localization of terminals using the control plane. The control plane is the part of the network responsible for maintaining a routing architecture and carrying signalling traffic, based on the network topology.

[0008] If a device is to be located, the SMLC is queried with a Mobile Station International Subscriber Directory Number (MSISDN). The SMLC triggers the mobile device to provide all available information (including cell information, signal strength, neighbouring cells, Timing Advance (TA) and the like) and uses this information to determine a location. The algorithm behind this localization can be implemented in a number of ways and is likely vendor specific.

[0009] One problem with SMLC is scalability of the system in terms of coverage of available networks and in different countries. To use a SMLC in a country, the SMLC needs to be integrated with the network. Hence it requires integration with all service provider markets individually, if this functionality is to be used on a large scale across many countries which have different available service providers. In countries without a network owned by the service provider, agreements need to be made with other service providers in order to obtain the required information to operate SMLC. Considerable integration effort for each region (for example, each country) is therefore required.

[0010]One particular implementation of SMLC provides an Adaptive Enhanced Cell ID (AECID) localization technique, which combines fingerprint data techniques mentioned above with real time reported radio measurements. However, the network conditions change over time and so these techniques require the database of fingerprints to be continually updated.

[0011] Other solutions are provided through the use of a cell database API. This API can be queried by providing the serving cell information and information of available neighbouring cells. In some implementations W-LAN data is also submitted to improve accuracy. The device also provides parameters relating to the network conditions. For example, the device may provide the signal strength and the timing advance (TA) to the API. One such implementation is provided by Google’s Geolocation API.

[0012] However, this solution is only based on the information of available cells and does not take into account the network conditions. The accuracy of many evaluated cell database APIs (such as, Google, Combain and TCS) do not provide sufficient accuracy for many asset tracking applications. One such application developed as part of a Vodafone product is the Mobile Asset Tracking (MAT) application. Experiments showed that the best of these systems was only able to provide average accuracy in urban areas in the region of 500 m.

[0013] Another localization solution (termed “CompAcc”) uses a compass and accelerometer to match the movement of a mobile device over time to a map. However, errors can grow over time if new location information is not available and constant monitoring of the sensors is required to determine a position.

[0014] One further problem with the CompAcc technique is the requirement for a compass and accelerometer to function access to such sensors is not always available. Additionally detailed maps are required. Moreover, this method would have a high power consumption. This localization method is therefore more useful for navigation than for localization.

[0015] A further technique (called “Calibree”) uses many other devices in the network to locate a single device. This is achieved by using devices equipped with a GNSS (which provides an exact location) in the vicinity. The distance between these devices and the target device is calculated and this information is used to perform localization fairly accurately. However, this technique requires there to be nearby devices with a GNSS signal and also requires the devices with a GNSS to determine the distance to the target device.

[0016] 5. This technique requires devices, which support this method of localization. Also a lot of devices are required in its vicinity in order for this method to function accurately.

[0017] It is an object of the present invention to overcome these issues.

Summary [0018]The present invention provides a method for estimating the geographic location of a subscriber device. The subscriber device is configured to communicate using a cellular communications network. The method comprises receiving a plurality of signals at the subscriber device. Each of the plurality of received signals relates to a respective cell. The method further comprises determining information from the received signal for each of the plurality of signals. The method further comprises using the determined information, for each of the plurality of signals, to ascertain respective probability distribution for the geographic location of the subscriber device given the determined information. Each respective probability distribution is characterised by a plurality of parameters. The method further comprises calculating one or more refined parameters characterising a joint probability distribution of the location of the subscriber device given the determined information from all of the plurality signals. The method further comprises obtaining an estimate of the geographic location of the subscriber device from the refined parameters characterising the joint probability distribution.

[0019]This method provides improvements over prior art methods because it makes use of the information obtained from a plurality of received signals relating to a plurality of cells. In other words, not only the signal received from the serving cell but also the signals received from neighbouring and/or available cells may be used for location estimation. These signals may come from a plurality of geographically separate base stations. These additional measurements allow the formulation of a refined joint probability distribution. By combining the information from a plurality of probability distributions, the accuracy of the estimate is improved and the confidence in the estimate is increased.

[0020] Moreover, the system of the present invention is able to provide location data, without the need for additional sensors (such as a GNSS) or specific parameters obtained from the base stations that are not ordinarily available in some networks (such as the round trip time).

[0021] The system of the present invention uses a probability distribution, rather than a matching algorithm. Therefore, the system of the present invention is less prone to errors than a best-guess fingerprinting technique.

[0022]The information determined from each received signal may comprise an identifier for the cell and/or an indication of signal strength. The identifier for the cell may be used to query further characteristics of the cell, such as a geographic location (either of the base station serving the cell or of the centroid of the cell itself), transmit direction and/or a characteristic loss profile. This additional information is useful in estimating location.

[0023] The identifier for the cell can take a number of different forms. The particular form of the identifier will depend on the network and will be different in 3G and 4G networks, for example. In particular, the identifier may comprise a global identifier for the cell, for example the Global Cell ID (GCID). Alternatively or additionally, the identifier may comprise one or more of the components that make up the global identifier. For example, the identifier may comprise one or more of the MCC, MNC, Cell ID, LAC, RNCID and/or TAC. One of these identifiers may be sufficient to identify a cell. However, for scenarios that are globally deployed, all of the components of the global identifier are preferably available to identify the cell uniquely. This can help to differentiate between cells if a particular identifier maps to two different cells.

[0024] Ascertaining a plurality of parameters may comprise reading values from a database using the identifier for the cell as a key and using the values from the database to arrive at the parameters characterising the probability distribution for the geographic location of the subscriber device. The database may contain the location/direction mentioned above. Additionally, the database can be updated with data observed by other devices. This will be discussed in more detail below.

[0025] The parameters characterising each probability distribution for the geographic location of the subscriber device may comprise an estimated geographic location of a centroid of the cell and a confidence that the subscriber device is at the estimated location of the centroid of the cell. Calculating one or more refined parameters characterising a joint probability distribution of the location of the subscriber device may comprise calculating a weighted average of the estimated locations of the centroids, weighted by the confidence in each centroid.

[0026] The combination of readings from multiple cells allows a broad estimation of the location of the device, based on the stations from which signals can be received.

[0027]The information determined from each received signal may comprise an indication of signal strength. The confidence may be based on the indication of signal strength for the respective cell.

[0028] The weighted centroid technique assumes a symmetrical attenuation profile around the centroid of a cell. The relative strengths of the signals from the base station provide an indication of the likely distance from the centroid of the cell and the weights of the centroids may be set accordingly when estimating the location of the device.

[0029] The weighted centroid technique is fairly simple to compute and can be performed quickly to obtain a good estimate with no prior knowledge of the location of the device.

[0030] The values read from the database may comprise the location of the centroid of the cell served by the base station. By providing knowledge of the identity of the base station, knowledge of specific characteristics of the base stations may be taken into account.

[0031] Ascertaining a plurality of parameters may comprise reading values from a database using the identifier for the cell as a key. The values may comprise a plurality of locations provided by devices served by the base station. The method may further comprise using the plurality of locations to calculate the estimated location of the centroid of the cell.

[0032] The estimated location of the centroid of the cell may be ascertained by reading a plurality of reported locations provided by devices served by the cell from a database, using the identifier for the cell as a key. These reported locations may be combined to estimate the location of the centroid of the cell.

[0033] The location of the cell area (and hence the centroid) may not be a fixed. Instead, the area and/or centroid may be calculated using location data from devices that have previously been served by the base station. This allows for a system that is continually changing and does not require continual manual updates to the databases whenever a tower is reconfigured. By using “real” data (in other words, data measured on a device using a technique, which is assumed to be accurate) to determine cell locations, the system of the present invention provides a system that is self-updating and continually being tuned to provide data with improved accuracy.

[0034] Alternatively, the centroid can be queried from the database, using the cell identifier as the key.

[0035]The values for each cell may further comprise a signal strength threshold value.

The worse quality the signal becomes, the weaker it is. Signals from the plurality of signals that are weaker than the threshold value may be disregarded when estimating the geographic location of the subscriber device.

[0036] The values for each cell may further comprise a ratio of an average received signal quality to an area served by the cell. The confidence that the subscriber device is at the estimated location of the centroid of the cell may be determined using the ratio. This ratio is a measure of how quickly the signal quality decreases with distance.

[0037] In one example for the weighted centroid localization, the signal strength value may be used to apply weight to the observed cells, rather than to determine a location. The model refinement component may then refine the datapoints. For example, the algorithm may populate a grid of many possible centroid locations. If many datapoints are received and variations are determined within one block in the grid, an average value may then be calculated and a variance defined for the block in the grid. Once the grid is completed, a mathematical function may be determined to best describe the signal strength distribution through the grid. The same method may be used to determine a function for the variance. Together they may be used for the sector analysis to later build the probability distribution in the weighted centroid technique.

[0038] The values for each cell may further comprise an indication of the probability that the subscriber device is located within the estimated geographic area of the cell, given that the subscriber device is being served by the cell. The confidence that the subscriber device is at the estimated location of the centroid of the cell may be determined using the probability.

[0039] The radius of the centroid may determine the most probable area of the device to be located. This is preferably set with the assumption of a standard deviation, meaning approximately a 68% probability. To get to a higher probability (>98%) the radius may be extended to three times the original size, which increases the search area by a factor of 9.

[0040] Because the weighted centroid method operates using confidence values for each centroid, a number of different factors (such as observed signal strength and centroid accuracy) may all be taken into account fairly simply. Each factor affects the weight of the respective centroid and the location estimate is carried out using the centroids and weights.

[0041]The base station attenuation profile may be used to estimate or refine the current estimate of the location of the centroid of the cell. This estimation may be performed using signal strength data and location data from a device reading and estimating the centroid location using the attenuation profile. A confidence of the centroid location may also be determined based on these readings. This can be combined with a pre-existing centroid location estimate and/or estimates from a number of other device readings to provide a refined estimate and refined confidence. This is particularly useful where an accurate centroid location is not available (for example, if the cell belongs to a different service provider).

[0042] The method may further comprise determining a geographic search area for the subscriber device comprising a plurality of sub areas. The parameters characterising each probability distribution for the geographic location of the subscriber device may comprise a value for each sub area within the search area, the value for each sub area being related to a probability that the subscriber device is located within that sub area.

[0043] The sector analysis localization method is more sophisticated than the weighted centroid method and requires more computation. However, the estimated location is provided with a good accuracy because non-symmetrical attenuation patterns and short distance fluctuations cause by reflections and the like may be taken into account.

[0044] The refined parameters characterising the joint probability distribution of the location of the subscriber device may comprise a refined value for each sub area within the search area. The refined value for each sub area may be calculated by combining the individual values for that sub area from each probability distribution. The values may be combined using a weighted average.

[0045] The parameters characterising each probability distribution for the geographic location of the subscriber device may further comprise a weight based on the indication of signal strength from the respective cell. The individual values may then be combined to obtain the refined values for each sub area, according to their respective weights.

[0046] In other words, each signal provides a discrete probability distribution defined by the probability values for each section in the search area. The joint probability distribution is provided by combining the individual probability distributions and taking into account the confidence we have in each distribution. The weighting of the average may be done according to the signal strength from each respective signal. In other words, base stations that provide weak signals only influence the joint probability distribution a small amount, whereas base stations that have a strong received signal will influence the joint probability distribution more. This is because a distribution that comes from a base station with a weak signal is likely to be less accurate than one from a base station with a strong signal. Therefore, the information contained in the distribution is of less significance and is therefore assigned a lower weight.

[0047] Alternatively, the values from each cell may be combined by summing the average values together to obtain non-normalised likelihood values for each sub area.

[0048]The information determined from each received signal may comprises an identifier for the cell and an indication of signal strength. Ascertaining a plurality of parameters for each of the plurality of signals may comprises reading values from a database using the identifier for the respective cell as a key. A probability of the device being in each of the plurality of sub areas may then be calculated using the indication of signal strength and the values read from the database.

[0049] The database may contain for each cell and for each of a plurality of sub areas in the search area, an expected signal strength for the sub area and a variance for the sub area. The database may contain an expected signal strength for each of the plurality of sub areas in the sub area. Preferably however, the database need only contain data relating to a subset of the plurality of sub areas. The areas for which no data is available can be assumed to have a vanishingly small expected value (in other words, assume the unknown values are zero). Alternatively, if data is available for neighbouring sub areas then interpolation may be used to fill in the gaps in the data, as will be discussed below.

[0050] Calculating the probability of the device being in each of the plurality of sub areas may use the expected signal strength, the variance and the indication of signal strength determined from the received signal to determine the probability that the subscriber device is located within that sub area. The method may assume that the values of signal strength are normally distributed. Other distributions of signal strength may also be appropriate.

[0051] The probabilities may characterise a respective probability distribution for the geographic location of the mobile device, given the determined information.

[0052] In other words, the present invention may calculate probability distributions of the signal strength (rather than calculating signal strength distributions, which may or may not be calculated as an intermediate step).

[0053]The expected signal strength and variance for each sub area (for example, in the intermediate step) may be calculated from a plurality of observed signal strength readings from devices located in the subarea. Preferably, many data points are available from each sub area, the data coming from a number of different devices.

[0054] In other words, the system may make use of historical data that is continually updated. This (as with the weighted centroid technique) permits the use of values for the sub areas of the search area that are not fixed. Instead, the expected signal strength may be calculated using readings and location data from devices that were previously in that sub area. Not only does this allow for a system that is continually changing and does not require continual manual updates whenever a tower is reconfigured. But also, the system of the present invention can respond to fluctuations in signal strength that occur over time and can provide a self-updating and self-tuning system to improve the accuracy of location estimates.

[0055]The database may contain, for each cell, for a plurality of signal strengths, and for a plurality of sub areas in the search area, a probability of the device being in each of the plurality of sub areas, given the signal strength is the indication of signal strength of the signal received from the respective cell.

[0056] A probability of the device being in each of the plurality of sub areas may be calculated by reading the probability of the device being in each of the plurality of sub areas from the database for a signal strength that is closest to the indication of signal strength.

[0057] Alternatively, a probability of the device being in each of the plurality of sub areas may be calculated by interpolating the probability of the device being in each of the plurality of sub areas from the database between two or more signal strengths that are closest to the indication of signal strength.

[0058] The method may further comprise calculating interpolated values for the expected signal strength, variance and or probability of the device being in the sub area, for each of the sub areas in the search area for which an expected signal strength and variance is not available.

[0059]The probabilities of the device being in each of the plurality of sub areas, given the signal strength is the indication of signal strength of the signal received from the respective cell may be calculated by using a model of expected signal strength.

[0060] Interpolated values for the expected signal strength may be calculated by using a model of expected signal strength.

[0061]The model of expected signal strength may decrease logarithmically with distance from the base station serving the respective cell.

[0062] Variance of the expected signal strength may increase with distance from the sub areas for which the values of expected signal strength and variance are available in the database.

[0063] The model of expected signal strength may decrease logarithmically with distance from the base station serving the cell and also include an additional decrease that increases with distance from the base station. This model is useful when the base station serves a smaller cell, for example in an urban area. The additional decrease helps to provide adjustment for decreases in signal strength/quality due to obstacles such as buildings.

[0064] The model used may be subject to refinement. It may be assumed to be logarithmic at the beginning, but it is not limited to this. Additionally it may not be appropriate to assume that the signal quality decreases with distance. This is true for some cases but it is not always the case. This is especially true in urban areas, where the signal quality can behave very irregularly due to reflections from buildings.

[0065] Other models may therefore be appropriate for the present invention, depending on the environment of the cell. In some circumstances, the signal strength can increase over distance, due to beneficial reflections or reduced interference. Elevation may also influence the signal strength. For example, a cell contain a mountain. The area of the cell between the base station and the mountain might have a low signal quality, as there may be many obstacles between the subscriber device and the base station. However at a higher altitude up the mountain (but further away from the base station), the signal quality may improve due to better line of sight. The most appropriate model may therefore implement any mathematical function. The method of the present invention may use any mathematical model that approximates the datapoints that have been gathered from other devices.

[0066] Using the determined information to ascertain a plurality of parameters characterising a respective probability distribution for the geographic location of the subscriber device, given the determined information may comprise the steps of: using a model to calculate the value related to the probability of the subscriber device being in each sub area, given the indication of signal strength for the respective cell.

[0067] Alternatively, using the determined information to ascertain a plurality of parameters characterising a respective probability distribution for the geographic location of the subscriber device, given the determined information may comprise the steps of: using a model to calculate an expected signal strength for each sub area (as an intermediate step); and calculating the value related to probability for each sub area by comparing the expected signal strength to the indication of signal strength determined from the received signal.

[0068] Using the model used to calculate the value related to probability of the subscriber being in each sub area (or the expected signal strength as an intermediate step) may comprise estimating the attenuation of the signal from a base station serving the cell to the sub area by calculating a distance between the base station and the sub area and/or an angle between a primary transmit direction of the base station and a direction from the base station to the sub area. The model can be used when no real data is available.

[0069] The present invention also provides a method for estimating the geographic location of a subscriber device by first estimating an approximate geographic location of the subscriber device by a using a weighted centroid method as described above to determine a search area and by second performing a sub area search in the search area to determine a more accurate estimate of the geographic location of the subscriber device. This allows for a combined method that is both computationally efficient and accurate.

[0070] The method may further comprise receiving an accurate indication of the location of the subscriber device. This may come from a GNSS, for example. The method may then further comprise calculating updated database values that would have resulted in a more accurate estimate of the geographic location of the subscriber device. For example, the device may update the database with new (up-to-date) signal strength data for a particular sub area. Alternatively, the device may provide information that indicates that the current location estimate for the centroid of a particular serving cell is not very accurate. The system can then reduce the confidence weighting for that serving cell accordingly. This update to the database values allows the system to more accurately predict location based on the available parameters and the historical data in the future. The method may then also comprise updating the database with the updated values.

[0071]The present invention also provides a method for updating and maintaining such a database. Such a method does not need to include all of the steps of the previously mentioned method for estimating the geographic location of a mobile device. It is not necessary for the devices whose data contributes to the data to themselves estimate a location using their signal data. If the device is enabled with GNSS (as most of the devices that provide useful training data for the system are) then they do not need to actually calculate their location using the algorithm mentioned above. It may sufficient only to store signal data from these devices (such as the signal data discussed above) and location data (obtained from a GNSS, for example).

[0072] The present invention also provides a method for estimating the geographic location of a subscriber device that is configured to communicate using a cellular communications network. The method comprises receiving a plurality of signals at the subscriber device, the plurality of signals relating to a respective plurality of cells. The method further comprises, for each of the plurality of signals, determining an estimated location of a centroid of the cell and a confidence that the subscriber device is at the estimated location of the centroid of the cell. The method further comprises calculating a weighted average of the estimated locations of the centroids, weighted by the confidence for each centroid to determine an averaged location and an associated confidence. The method further comprises using the averaged location and associated confidence to determine a geographic search area for the subscriber device, the geographic search area comprising a plurality of sub-areas. The method further comprises, for each of the plurality of signals, determining a value for each sub-area within the search area, the value for each sub-area indicating a probability that the subscriber device is located within that sub-area. The method further comprises, for each sub-area, combining the values from each signal to obtain a joint value for each sub-area indicating the joint probability that the subscriber device is located in that sub-area. The method further comprises obtaining the estimate of the geographic location of the subscriber device from the joint values. This method may be combined with any of the adaptations described above in relation to the earlier method.

[0073] The present invention also provides a cellular telecommunications network comprising a plurality of base stations and a mobile device configured to communicate using the cellular communications network. The cellular network is configured to estimate the geographic location of the mobile device using a method as described above. The method may be performed at the mobile device. Alternatively, the mobile device may send the information derived from the received signals to another part of the network and location estimation may be performed there.

Brief Description of the Drawings [0074] The present invention may be put into practice in a number of ways, and some specific embodiments will now be described by way of example only and with reference to the following drawings.

[0075] Figure 1 illustrates the process of trilateration.

[0076] Figure 2 shows the results of analysis to determine from which angles and distances a cell could be seen by a MAT device.

[0077] Figure 3 shows the relationship between the distance of a measurement location from a cell tower against the observed signal strength.

[0078] Figure 4 shows an illustration of the weighted centroid localization technique.

[0079] Figure 5 shows a representation of the sector analysis localization method.

[0080] Figure 6 shows an illustration of a superposition of the probability distributions from a plurality of base station.

[0081 ] Figure 7 shows evaluation of a joint probability grid to determine the highest probable location.

[0082] Fig ure 8 shows the Localization use case.

[0083] Figure 9 shows the Model Refinement use case.

[0084] Figure 10 shows the component decomposition of the Enhanced Cell-based Localization system.

[0085] Figure 11 shows the interface description of the ECbL system.

[0086] Figure 12 represents the basic classes required for the localization component.

[0087] Figure 13 represents the basic classes required for the model refinement component.

[0088] Figure 14 shows the state diagram for the Localization and Model Refinement components.

[0089] Figure 15 shows the sequence diagram for Localization.

[0090] Figure 16 shows the sequence diagram for Model Refinement.

[0091] Figure 17 shows how data may be consolidated.

Detailed Description [0092] The solution of the present invention is based on a trilateration algorithm. Such an algorithm requires the distances from a mobile device to be located (a handset or M2M device, for example) to at least three fixed points (such as cell towers) and the positions of the fixed points. Figure 1 illustrates the process of trilateration.

[0093] As the information provided by the mobile device is not suited to calculate a distance, the proposed solution uses a stochastic model and a heuristic model to determine the most probable location. In short these models do not determine a single probable distance from the point P to the points A, B and C in Figure 1. In contrast, the proposed solution determines many probable distances and works out the most probable distance by overlying all probable distances from all points around P. This mesh forms a grid which is used to determine the most probable location.

The enhanced cell-based localization algorithm [0094] The object of the proposed localization algorithm described in the present invention is to improve the current cell based localization capabilities provided by prior art systems.

One characteristic that the present invention aims to improve is the average cell-based localization accuracy. Specifically, an accuracy of <500m in urban and suburban environments was desired.

[0095] One additional object of the present invention is to function in any country in which a network is provided by the service provider implementing the solution. It is also desired that the algorithm may still be used in countries without a network belonging to the service provider in question. However, in such networks, different/additional network learning algorithms may be required. This is in order to gather the required geographical information about the cell network. Details on these cell-learning algorithms may be appreciated by the skilled person and do not form part of the present invention.

[0096] General best practices are also considered for the design and implementation of the solution. This is to permit for future scenarios. A high flexibility of the deployment options of the localization functionality is therefore required. Therefore, a low coupling and high cohesion design has been followed.

Prerequisites [0097] A basic knowledge about the radio environment is required where a device is to be located. Countries with networks operated by the service provider implementing the solution, or partners of the service provider, can make use of the new solution more easily if cell information of the given country is available. For countries without a network operated by the service provider deploying the solution, external data sources can be used to calculate the required information.

[0098] A paper from Thomas Kuhner “Provider-based Localization of GSM-Enabled Devices” describes that external sources, with the right refinement mechanisms, can provide cell propagation information with equivalent or improved accuracies.

[0099]The information that is required to perform the improved localization algorithm is: • Latitude and longitude of a cell centroid and of a cell tower to the corresponding cell (If only the cell tower position can be provided, additional information regarding the cell is required. This information may include cell direction and opening angle of the cell antenna. This additional information allows for a calculation of the centroid of the cell. • Cell information, including: o an identifier for the cell; o an indication of the location of the centroid of the cell or the antenna direction; and o an indication of the size of the cell.

[0100] The indication of the centroid or cell direction may be determined using one or more of Location Area Code (LAC), Radio Network Controller ID (RNCID) or Tracking Area Code TAC (the TAC is used in LTE and is comparable to the LAC in GSM), for example. If more than one of these parameters or other additional parameters indicating the location/direction of the cell are provided to the algorithm, the estimate of the location/direction of the cell may be improved.

[0101] The size of the cell can be the probable signal range or a radius for a given cell centroid. This indicator should specify the probable signal propagation area of the cell.

[0102] Other prerequisites for the algorithm are not required but optional. The achieved accuracy may be significantly improved by providing the following functionality.

[0103] To improve accuracy, a learning functionality or a refinement of the calculation model may be feed with data from the network. The source of this data can vary, but must include information about a cell at a known location. For the asset tracking scenario for example, all measurements are provided to the enhanced cell-based localization algorithm, or more specifically, the model refinement function.

[0104] In one asset tracking scenario, the cell-based localization may be a fallback mechanism with the devices mainly determining their location via a built-in GNSS module. These measurements may be provided to the model refinement component of the enhanced cell-based localization algorithm. In this way, the cell signal propagation may be learned. This data may be used for improving the localization estimation for the cell-based method.

Localization enhancement algorithm [0105] For the algorithm, two types of cells may be considered: 1. Umbrella cells: and 2. Sector cells.

[0106] Umbrella cells are created by transmitting the signal 360° around the positon of the cell tower. This forms a radial signal propagation around the cell tower. Sector cells are created by transmitting the signal with an angle usually <180°. In most instances the antennas send signals with a 120° opening angle. Of course in the real world the signal is not limited to the 120° vector. The signal can be reflected, refracted, diffused, diffracted or absorbed.

[0107] For the asset tracking proof of concept, a single sector cell was analysed and the results were used to determine from which angles and distances the cell could be seen by an asset tracking device. Figure 2 shows the results of one such experiment. The illustration shows the 120° sector cell (white line) and various positions (coloured dots) around it from where the asset tracking device could measure the cell. The colours indicate the signal strength (with green dots corresponding to readings having a good signal and red dots corresponding to readings having a bad signal).

[0108] As can be seen from Figure 2, many of the data points are where one would expect them to be (in front or within the sector). However, there are also measurements where they were not expected (behind or much further away from the given cell range).

[0109] Red dots can also be seen quite close to the cell tower position (blue dot) and light green dots further away. In an ideal setup, one would expect that closer to the cell towers all the dots are green and the further away the measurement is taken, the redder they become.

[0110] Figure 3 shows the distance from the location of the measurement to the cell tower (y axis) and the signal strength (x axis). The signal strength is considered good at the far left (low level of absorption) and worse at the far right (high level of absorption).

[0111] As it can be seen for this sector cell, if the signal quality is good, it is likely that the reading has been taken close to the cell tower. However, if the signal is bad, the reading may have been taken close or far away. The deviation in the range of distances increases with loss of signal quality.

[0112] These characteristic curves can look quite different from cell to cell. For this reason, improved accuracy can be achieved if the signal propagation is learned for each cell. One other way to improve the reading that can be learned from these observations is to only consider the better half of the signal quality measurements for refining the signal propagation model. In this way, a more accurate model may be provided.

[0113]The solution provided by the present invention consists of two components: 1. The localization component. This component performs the actual localization. It takes the following input: a. Serving cell information (MCC, MNC and Cell ID) and (LAC, RNCID or TAC), and the RSSI (signal strength) value to this cell b. Neighbouring cell information, including the same parameters as mentioned above. The amount of neighbours can vary between 1 and many. Note: The RTT (round trip time) of the signal is not required. It could not be provided for the asset tracking proof of concept. However, it would likely improve the localization accuracy even further. The “localization” component returns the latitude and longitude of the probable location. Additionally, it returns an accuracy indicator. This value is usually a distance value which represents a radius around the given location. The meaning of the radius is that with a 68% probability the tracker is located within this circle. Depending on how many values were used to determine the position and what the quality of the measurements were, the accuracy is higher (smaller circle) or lower (larger circle). 2. The model refinement component. This component refines the mathematical model which is used to calculate the most probable location. The source of the input can vary. It can be collected from network probes, log files or loT devices that submit data. The parameters that are required are: a. The exact position of the device whilst the measurement was taken (latitude and longitude). The accuracy of this position should be <20 meters (most likely a GNSS determined position) b. Cell information, including (MCC, MNC and Cell ID) and (LAC, RNCID or TAC) and the RSSI (signal strength) value to this cell at the given location.

The component outputs parameters which are required for the localization component to locate a device accurately. The more information is fed into the model refinement component, the more accurate the localization component will become. Note: Both components can work independently from one another. More details on both components are provided below.

Localization Component [0114] The proposed solution uses two different localization techniques to determine the most probable location. One method simpler and therefore requires less computational expenditure to calculate a rough location. The result of this localization may then be used to search for the device in a smaller area with the second method, which is more complex and requires more computational power.

[0115] The first method uses a weighted centroid localization technique, where the locations of the centroids from the given cells are combined using a weighted average.

The “weighted” part of this algorithm is based on the RSSI value.

[0116] Figure 4 shows an illustration of the weighted centroid localization technique. Let us assume the asset tracking device reported 3 cells it can see (A, B and C). To each of these cells it also broadcasted the signal strength level. This is the signal quality at the location of the device we are searching for the signal strength for “Cell A” is better than for “Cell B” and “Cell C”, the red circle is pushed towards “Cell A”, as the device is more likely closer to “Cell A”.

[0117] This technique on its own only gives a rough indication of where the device is located. As mentioned above, the signal strength is not an optimal indicator for determining the distance from the cell tower. Therefore, the purpose of this step is to reduce the search area for the second method. This allows for a higher performance and to minimize errors, which the second method is more susceptible to.

[0118] The second method is termed sector analysis localization. This method uses a stochastic distribution function, which may be determined by the “model refinement” component. Using this function, a predefined search area is evaluated. The search area may be determined using the Weighted Centroid Localization technique described above.

[0119] Figure 5 shows a representation of the Sector Analysis Localization Method. The circle represents the search area. The cell tower and the arrow indicate the cell signal propagation direction. The values within the grid represent the probability that the device is within that square of the grid (higher numbers correspond to higher probability). This is an abstract representation of what the algorithm does. The actual algorithm may use the full rational number range. Moreover, the probabilities may be represented in a number of different ways, as will be apparent to the skilled person (such as fractions, decimals, percentages and the like). The numbers in the grid do not necessarily need to sum to a probability of 100% (there is likely to be a non-zero probability that the device is located outside the grid).

[0120] The same action will be performed for each cell tower that was observed by the device. The probabilities from each cell tower are weighted according to their signal strength level. This is necessary because the worse the signal strength becomes the bigger the statistical error (due to the variance) becomes. An illustration of a superposition of the probability distributions from each base station is shown in Figure 6.

[0121]The cell signal propagation directions that are used in the sector analysis localization are determined the antenna transmit direction As mentioned above, there are often surprising points of high signal strength (coloured areas of Figure 2) outside of what we would consider the normal beam direction and scope (white lines).

[0122] Afterwards, the grid is evaluated and the highest probable location is returned to requester, as shown in Figure 7. If more than one grid cell has a high probability, the algorithm searches for the highest concentration of high probabilities. A centroid is formed (green circle) including the adjacent grid cells and the centre of the centroid is returned as the probable location of the device, which is to be located.

[0123]The performance of this algorithm varies greatly on the search area size and the number of observed cells within the device vicinity.

Model Refinement Component [0124] In order for the two localization methods (described in the previous section) to function accurately, both algorithms may also have one or more parameters that can be used to tune the localization function.

[0125] Preferably, a set of variable parameters that can be adjusted for many different geographical areas is defined for these techniques. Algorithms in the prior art approximate the signal loss function for an ideal scenario and try to accommodate its reliability in the real world by identifying influencing factors (like trees, mountains, buildings, etc.) [0126] The proposed solution uses a “Model Refinement Component” which continuously evaluates data which provides information for each individual cell and its signal loss curve. This is possible today due to Big Data. As each cell covers a different environment with different influencing factors, it is much easier to deduct a propagation model for each cell rather than to find the one model that fits all.

[0127] The data which is required for the Model Refinement Component can come from network probes, crowd data or loT devices with GNSS, as long as the input parameters as described in the section “Components” can be submitted.

[0128] For the Weighted Centroid Localization the variable parameter is incorporated into the weighting function. An indicator is calculated which represents how often the device is actually within the area where the given cell is understood to be. The centroid that we are using for this cell being the geometric centre of the cell area. The indicator is then used to determine a degree of validity of a cell centroid (in other words, how accurate the data for that cell is), so valid cell centroids can be given greater weight than less accurate cell centroids.

[0129] For example a Cell A and Cell B have been observed 5 times by different devices. For Cell A: every time this cell was a serving cell of the device, the device was within the cell area of Cell A. Therefore the indicator would be 100%. For Cell B only 3 of the 5 measurements were within the cell area of Cell B (because the information that is being used for the cell area of Cell B is inaccurate). Therefore the indicator would be 60%. For the Weighted Centroid Localization algorithm this would mean that if a device receives a signal from Cell A and B with a similar signal strength value, the weight of the location estimate would shift towards Cell A, as the centroid of cell A is known to be more precise that the centroid for Cell B.

[0130] For the Sector Analysis Localization, two additional statistics are created for each cell: 1. Angle deviation from signal propagation direction 2. Signal loss over distance [0131] The outcome results in a set of values. For each signal strength value, a discrete average value and a standard deviation value is calculated. Putting the two together allows for a computation of a Gaussian curve for each signal strength value. These values are required by the localization algorithm in the localization component to determine the probable location.

[0132] Data for each cell is required to calculate these functions. To allow the localization to function with cells where there is not enough data available, a default function is provided. This default function will be adjusted periodically when new data becomes available.

[0133] In conclusion the “Model Refinement Component” uses available data to calculate the necessary functions per cell and includes this information in a mathematical model. This model is than shared with the “Localization Component” which uses the model to calculate the probable location of the device.

[0134] In contrast to the Serving Mobile Location Centre (SMLC) methods shown in the prior art, the present invention only requires serving and neighbouring cell information, including the Cell ID, Mobile Network Code (MNC), Mobile Country Code (MCC), Location Area Cod (LAC) and Received Signal Strength Indication (RSSI). The Timing Advance (TA) is not required for the method of the present invention. Additionally, the functionality does not require local market integration. The data necessary for the calculation is owned by the network provider or can be queried from external sources.

[0135] Prior art systems described above use fingerprints collected for localization. The present invention uses the cell tower position and a centroid of the cell for the localization. These parameters may be calculated in an alternative method using the fingerprints available from prior art systems. This may be required for areas where the service provider has no own network coverage or an agreement to get this information from a partner network is in place.

[0136] In contrast to the CompAcc system described above, compass and accelerometer data are not required in the present invention. The solution of the present invention can locate a device without having to constantly monitor movement of a device to determine the location. This has advantages in terms of the battery consumption of the device.

[0137] In contrast to the Calibree system described in the prior art, the present invention does not require data obtained from other nearby devices to determine its location.

Instead, knowledge about the network in which it is operating is used to determine a location.

[0138]The proposed solution makes use of modern technologies that have not been available in the past (Big Data). By making use of other data which is also collected, a much more precise localization model can be created.

[0139] Compared to prior art solutions, the system of the present invention makes use of unique knowledge of how the service provider’s cellular network is setup. All other known solutions therefore rely on a fingerprinting technique for localization. Fingerprinting has the disadvantage of costing much effort and a long training phase to start functioning and the fingerprints become obsolete fairly quickly, as the mobile network changes. That is why prior art systems are more focused on WiFi localization.

[0140] In the past, if services wanted to locate a device they made use of GNSS, as this still provides the best accuracy. That is why there was no demand in the past to further improve localization via cells. However GNSS is not always available and in the future with loT we have learned that many future devices will need to focus on preserving battery life. This is very difficult with GNSS as it uses much computation power and therefore energy to get an accurate location.

[0141] The proposed solution allows for localization without any additional network components or integration (SMLC), it does not require a training phase and improves its localization accuracy over time. The device battery power is conserved as no localization computation needs to be carried out on the device. The Localization Component can be deployed centrally or locally. The proposed solution can locate devices which broadcast the necessary information much more precisely than any other commercial service we found on the market today.

Examples [0142] As mentioned earlier, this application focuses on the Mobile Asset Tracking (MAT) Scenario (though this is also applicable to any asset tracking scenario). The following two examples will highlight two different use cases for the MAT scenario, when interacting with the enhanced cell based localization capability. The skilled person will appreciate that applications of the present invention are not limited to MAT but may be used as a localization method for any mobile device.

Localization use case [0143] In the localization use case the MAT device needs to be located. This device is called the Mobile MAT Device. The actor in this use case is the MAT operator. This is a person who has access to the MAT Server and can view the current locations of all MAT devices for his level of authorisation.

[0144] The MAT Server is a central MAT component (and not part of the enhanced cell based localization algorithm). This component acts as a central hub for all MAT devices with which they can communicate. The MAT server receives data from the MAT devices and stores it in a database. The Enhanced Cell-based Localization Server hosts the localization algorithm.

[0145] As can be seen in Figure 8, the level of interaction between the components is distinct. The communication is as follows: • Mobile MAT Device —> MAT Server • MAT Server <-> Enhanced Cell-based Localization Server • MAT Operator <-> MAT Server

Model refinement use case [0146] In the model refinement use case the Mobile MAT device transmits cell information but also a GNSS location. This data packet is used to improve the enhanced cell-based localization capability.

[0147] The actor and the systems are the same as in the previous use case. However, the MAT Server and the Enhanced Cell-based Localization Server behave differently.

[0148]The MAT Operator is still querying the MAT Server for the location. This time the MAT Device is transmitting a GNSS location which is already accurate enough for the operator, so the Enhanced Cell-based Localization Server is not required for the actual localization of the Mobile MAT Device.

[0149] To improve the Enhanced Cell-based Localization Server capabilities, the data gathered by the Mobile MAT Device is “donated” to the Enhanced Cell-based Localization Server. This is of benefit to MAT, as other device which might not determine a location using a GNSS, but can observe the same cell as another device did before, the Enhanced Cell-based Localization Server can determine its location much more accurate.

[0150] Figure 9 shows the model refinement use case illustration. As can be seen in Figure 9, the connectivity is identical to the previous use case: • Mobile MAT Device MAT Server • MAT Server —> Enhanced Cell-based Localization Server • MAT Operator <-> MAT Server [0151] The only difference in this use case is that the Enhanced Cell-based Localization Server is not communicating any information back to the MAT Server.

SW architecture [0152] This section relates to the Enhanced Cell-based Localization algorithm. Firstly, various object models are described. This should provide an overview, about the required components, their level of connectivity and a method of access.

[0153]Then, dynamic models will be illustrated. These are important to understand the call flows and level of interaction between the various components described in the object models.

[0154] Finally, a detailed description on how the localization and model refinement components actually work will be given under the heading “Pseudo algorithmic description”.

Object models [0155] This section starts out with the component decomposition to give an overview of the various components and their composition. This illustrates how the various parts are connected to each other. Next is the interface description. This is the specification on how the Enhanced Cell-based Localization algorithm can be made usable for other systems. Finally, the last subsection of the object models contains the main class descriptions which are required for the localization and model refinement. In short: We look at how the Enhanced Cell-based Localization system is structured, then look at the interfaces and afterwards go into depth on how it works.

Component decomposition [0156] As mentioned above, the Localization and Model Refinement components make up the essence of the Enhanced Call-based Localization algorithm.

[0157] Figure 10 shows the component decomposition of the Enhanced Cell-based Localization system. As can be seen, there are two additional components, which have only been mentioned implicitly.

[0158] The “ECbL Fagade” is the interface to the Enhanced Call-based Localization System from the “outside”. For the MAT scenario this would be the MAT Server. Using this interface a location query can be initiated on given cell information, or the model refinement model can be feed with new data for the cell signal propagation models. For details on this interface, please see the following chapter “Interface description”.

[0159] The “Location Data Proxy” acts as a mediator between the data layer and ECbL components. Its purpose is to translate and relay cell information. The connection towards the Cell Location DB is used to get the cell information prerequisites (as described above). This information is currently accessible through the GLC (Group Location Capability) Cell API 2.0. The ECbL Data Model DB contains the required parameters for determining the cell signal propagation model. This information is calculated by the Model Refinement component and later used by the Localization component.

Interface description [0160] To access the functionality of the Enhanced Call-based Localization system the ECbL Fagade is provided. It has two public functions which allow for the localization of a device or the submission of cell information for the model refinement component. Figure 11 shows the interface description of the ECbL system.

[0161]The object Cell contains: • MCC : String • MNC : String • LAC, RNCID or TAC : String • Cell Id : String [0162] Strings are chosen instead of numeric values as these are identifiers and not values used for computation.

[0163]The object Coordinates contains: • Latitude : double • Longitude : double • Accuracy : integer [0164] The accuracy value is used to form a circle around the given coordinate. It represents the radius in meters. The circle indicates the area in which the device is likely to be located with a 68% probability (standard deviation).

Class decomposition [0165] Two different class diagrams are discussed below. Both diagrams share classes.

For readability purposes both diagrams are incomplete. Only the main functions and variables are given. Also not all classes are visualized, due to readability purposes.

Abstract classes for command and strategy patterns are excluded. The last negative note is that the class names are incorrect by adding spaces. This is also due to readability.

[0166] Figure 12 represents the basic classes required for the localization component. At the very top you find the Fapade interface, which was described in the previous chapter.

[0167] The Localization Controller manages all call flows. It is called by the ECbL Fagade with the locateDevice() method. For a detailed call flow, please view the sequence diagrams in the following chapter.

[0168] A simple description would be that the cell information passed to the ECbL algorithm will first need to be enriched with the cell locations and model parameters available for the individual cells. This is managed by the Cell Data Crawler.

[0169] Next the enriched cell information is passed to localization functions. The Localization Executer manages to pass the cell information through the Weighted Cell Localization first. This returns a search area (represented by coordinates with a radius) which is required for the Sector Analysis Localization. This function in turn determines the probable location with the given cell information.

[0170] Figure 13 represents the basic classes required for the model refinement component. At first glance the Model Refinement architecture looks similar. Only three classes are not displayed in the previous diagram. All new classes are in the Model Refinement package. All other classes are the same (missing methods in the other diagram are due to readability).

[0171]The call flow is again similar in the beginning. The cell information is enriched with the cell location information and the model parameters. Afterwards the cell information is passed to the Refinement Executer. This class manages the analysis of the model with the given coordinates obtained from a GNSS to for example determining the correct signal propagation for the Sector Analysis.

Dynamic models [0172] This section contains the dynamic models for the Enhanced Cell-based Localization algorithm. The first subsection illustrates the coupling between the Localization component and the Model Refinement component by using a state diagram. In the second subsection, a sequence diagram is used to detail the call flows through the class diagrams illustrated in the previous chapter.

State diagram [0173]The state diagram describes the various states a system can be in. As the Localization component and the Model Refinement component operate independently from one another there are two start points. Again, due to readability purposes the endpoint was discarded from this diagram.

[0174] Figure 14 shows the state diagram for the Localization and Model Refinement components. The state diagram for Localization is rather simple and is therefore not further detailed. However, note the “Accessing DBs” state marked in orange.

[0175] The Model Refinement state diagram illustrates a periodic accessing, processing and updating of the DB, as compared to a continuous workflow. This is mainly because of performance reasons. The changed to the cellular propagation model are minimal with only a single data point. Also updating the ECbL Model DB requires locking of the DB for the time of the update. This blockage, illustrated by the orange colouring in both diagrams, is to be kept to a minimum.

Sequence diagram [0176]This sub-section describes two sequence diagrams. These diagrams relate to the class diagrams of the previous chapter.

Please note: Both diagrams were reduced due to readability reasons. The Cell Location DB Connector and the ECbL Data Model DB Connector are only mentioned by a grey label.

[0177] Figure 15 shows the sequence diagram for Localization. As already mentioned before, the Localization Controller is the main class in the Localization component. It managed the call flows to the various other classes.

[0178] Figure 16 shows the sequence diagram for Model Refinement. Compared to the Localization sequence, the Model Refinement sequence requires a loop to go through all cells in the Model Refinement package. This is mainly because each cell needs to be refined individually, where as in the Localization sequence the cluster of cells determines the location.

Pseudo algorithmic description [0179]This section focuses on the Enhanced Call-based Localization algorithm, specifically the localization algorithms and the model refinement algorithms.

Weighted Centroid Localization def _executeLocalization(cell_list, Loc_Source): lat, Ion, rad = 0.0, 0.0, 0.0 cell_count = len(cell_list) total_weight = 0 total_radius = calculateTotalRadius(cell_list, cell_count) for i in range(0, cell_count): weight = calculateWeight(cell_list[i], total_radius) lat += cell_list[i][MYConstants.LATITUDE_CELL] * weight Ion += cell_list[i][MYConstants.LONGITUDE_CELL] * weight total_weight += weight lat = lat/total_weight Ion = lon/total_weight return lat, Ion def _calculateTotalRadius(cells, count): total = 0 for i in range (0, count) : total += cells [i] [MYConstants. RADIUS] return total def _calculateWeight(cell, total_radius): weight = 0.1 if cell[MYConstants.RSSI] > (-83): weight = (83 + cell[MYConstants.RSSI]) * (total_radius / cell[MYConstants.RADIUS] )** (-0.6) return weight [0180] The Weighted Centroid Localization is rather simple. The interesting method is the _calculateWeight() method. This can be varied easily to get very different results. The particular values used can vary, depending on the individual cell.

[0181] During the proof of concept we noticed that the worse the signal quality becomes, the less useful the measurement is. Therefore, there is a hard border after which the measurement is nearly not considered at all anymore. In this instance it is -83 dBm.

[0182] Another variable is the relation between the weighting of the signal quality compared to the cell size. Intuitively a smaller cell size means that the cell can only be observed in a relatively small area. This decreases the potential search area significantly.

Weighted Centroid Model Refinement [0183] As mentioned previously, two variables are the signal quality barrier and the signal quality radius ratio. Both can be optimised by running simulations for a specific cell. The optimisation potential is however rather slim with these two parameters.

[0184] A better optimization can be achieved by determining a more accurate cell centroid, than the given cell centroid from the service provider. The service provider determines the cell centroid from the direction of the antenna and the power output of it. The centroid should represent the area where the cell should serve as a serving cell.

[0185] Using the data collected by the Mobile MAT Device a more accurate cell centroid can be determined and stored in the Weighted Centroid Model Refinement.

Sector Analysis Localization def _executeLocalization(e_cell_list, search_area): if len(e_cell_list) == 1: raise SingleCell('ServingSectorAnalysisLocalization; _findWeightedSector: Only one Cell found!') grid =_createRadialSectorGrid(search_area, e_cell_list) probability_grid =_addProbabilitiesToGrid(grid, e_cell_list) lat, Ion, rad, max_probability_block = _identifyProbableLocation(probability_grid) return lat, Ion, rad def _addProbabilitiesToGrid(grid, e_cell_list): for sector in e_cell_list: tower_lat, tower_lon = sector.getTowerPosition() cell_direction = sector.getCellDirection() cell_rssi = sector .getRSSI() for block in grid: block_lat, block_lon = block .getLatCenter(), block.getLonCenter() deviation_angle = _calculateDeviationAngle(tower_lat, tower_lon, block_lat, block_lon, cell_direction) distance_tower_to_block =

CoordinateCalculator .distanceBetweenPoints(tower_lat, tower_lon, block_lat, block_lon) * 1000 probability_angle = distribution .getAngleProbability(cell_rssi, deviation_angle) probability_distance = distribution . getTowerDistanceProbability(cell_rssi, distance_tower_to_block) block.addLocationProbability(probability_angle) block.addLocationProbability(probability_distance) return grid [0186] The Sector Analysis is the second Localization algorithm which is executed after the Weighted Centroid Localization. As described above, this algorithm uses a grid that is filled with numbers. Each number represents a probability that the Mobile MAT device is located within that particular grid square. The most interesting part is how the probabilities are determined. For this the cell tower position is required. The direction of the signal is determined by calculating the baring between the cell tower and the cell centroid.

[0187] The class distribution can be one of two types. Either it contains a function to determine the signal propagation depending on distance from the cell tower or the angle of deviation from the centre axis. Or the distribution class contains a prefilled grid of signal strength distributions that were collected from other measurements with coordinates obtained from a GNSS.

[0188] As these prefilled grids are never 100% complete, the gaps between the different grid blocks are interpolated. For this again two different methods are used. For large cells, which are likely in rural areas, the assumption is that the signal between two points decreased logarithmically. For smaller cells that are more likely in urban areas, the assumption is that the signal quality also decreased logarithmically, however the signal quality is additionally much lower between the two points (as obstructed by buildings for example).

Analysis Model Refinement [0189] As mentioned above, the key component of the sector analysis is the distribution model of the cell signal propagation. The refinement component makes use of the donated data to calculate functions which represent the signal loss curve for the specific cells. Once a sufficient amount of data was donated it needs to be consolidated.

[0190] Figure 17 shows how data may be consolidated. Measurements close to others are grouped together into one measurement with a weight value. The weight value represents the amount of measurements that contributed to the consolidated measurement. Also for each consolidated point a variance is stored. This allows a mapping also to fluctuating signal strengths.

[0191] Once the function calculation mechanism does not find a matching function anymore, or exceeds a function degree higher than 7, the prefilled grid will be used for determining the probabilities. For cells where no data was donated, a generic signal loss curve is used.

[0192] The deployment design described above can vary greatly, depending on the scenario the algorithm is to be used for. This document has focused on the MAT scenario, where a device transmits the cells in range to the MAT server. This server connects, to a localization server API where it can query the location of the device by inputting the cell information, transmitted by the device. In this scenario, the localization server API may perform the localization algorithm described and perform the actual localization. Other scenarios outside of the MAT implication are possible and will be appreciated by the skilled person.

[0193] In one example of an alternative deployment, the localization mechanism may be performed locally on the device. In this scenario, the model refinement parameters described above may need to be calculated on a remote server and transmitted to the device periodically.

[0194] The general architectural design for the present invention described above may be easily adapted to take into account alternative implementations of the algorithm according to the present invention. The invention is implemented in accordance with recommended coding practices for object oriented design (such as low coupling and high cohesion) in order to facilitate adaptation for future requirements.

Claims (28)

1. CLAIMS:

   1. A method for estimating the geographic location of a subscriber device configured to communicate using a cellular communications network, the method comprising: receiving a plurality of signals at the subscriber device, the plurality of signals relating to a respective plurality of cells; for each of the plurality of signals, determining information from the received signal; using the determined information to ascertain a plurality of parameters characterising a respective probability distribution for the geographic location of the subscriber device given the determined information, for each of the plurality of signals; calculating one or more refined parameters characterising a joint probability distribution of the location of the subscriber device given the determined information from all of the plurality signals; and obtaining an estimate of the geographic location of the subscriber device from the refined parameters characterising the joint probability distribution.
2. 2. The method of claim 1, wherein the information determined from each received signal comprises an identifier for the cell and an indication of signal strength.
3. 3. The method of claim 2, wherein the identifier for the cell comprises one or more of GCID, MCC, MNC, CelllD, LAC, RNCID and TAC.
4. 4. The method of claim 2 or claim 3, wherein ascertaining a plurality of parameters comprises reading values from a database using the identifier for the cell as a key and using the values from the database to arrive at the parameters characterising the probability distribution for the geographic location of the subscriber device.
5. 5. The method of any preceding claim, wherein the parameters characterising each probability distribution for the geographic location of the subscriber device comprise: an estimated geographic location of a centroid of the cell; and a confidence that the subscriber device is at the estimated location of the centroid of the cell, and wherein calculating one or more refined parameters characterising a joint probability distribution of the location of the subscriber device comprises calculating a weighted average of the estimated locations of the centroids, weighted by the confidence in each centroid.
6. 6. The method of claim 5, wherein the information determined from each received signal comprises an indication of signal strength, and wherein the confidence is based on the indication of signal strength for the respective cell.
7. 7. The method of claim 5 or claim 6, wherein ascertaining a plurality of parameters comprises reading values from a database using the identifier for the cell as a key, wherein the values comprise the estimated location of the centroid of the cell.
8. 8. The method of claim 5 or claim 6, wherein the estimated location of the centroid of the cell is ascertained by: reading a plurality of reported locations provided by devices served by the cell from a database, using the identifier for the cell as a key; and combining the plurality of reported locations to estimate the location of the centroid of the cell.
9. 9. The method of claim 7 or claim 8, wherein the values for each cell further comprise a signal strength threshold value, wherein signals from the plurality of signals that are weaker than the threshold value are disregarded when estimating the geographic location of the subscriber device.
10. 10. The method of any of claims 7 to 9, wherein the values for each cell further comprise a ratio of an average received signal quality to an area served by the cell, wherein the confidence that the subscriber device is at the estimated location of the centroid of the cell is determined using the ratio.
11. 11. The method of any of claims 7 to 10, wherein the values for each cell further comprise an indication of the probability that the subscriber device is located within the estimated geographic area of the cell, given that the subscriber device is being served by the cell, wherein the confidence that the subscriber device is at the estimated location of the centroid of the cell is determined using the probability.
12. 12. The method of any of claims 1 to 4, further comprising determining a geographic search area for the subscriber device comprising a plurality of sub areas, and wherein the parameters characterising each probability distribution for the geographic location of the subscriber device comprise a value for each sub area within the search area, the value for each sub area being related to a probability that the subscriber device is located within that sub area.
13. 13. The method of claim 12, wherein the information determined from each received signal comprises an indication of signal strength, and wherein the refined parameters characterising the joint probability distribution of the location of the subscriber device comprise a refined value for each sub area within the search area, the refined value for each sub area being calculated by combining the individual values for that sub area from each probability distribution.
14. 14. The method of claim 13, wherein the parameters characterising each probability distribution for the geographic location of the subscriber device further comprise a weight based on the indication of signal strength from the respective cell, and wherein the individual values are combined to obtain the refined values for each sub area, according to their respective weights.
15. 15. The method of any of claims 12 to 14, wherein the information determined from each received signal comprises an identifier for the cell and an indication of signal strength, and wherein ascertaining a plurality of parameters for each of the plurality of signals comprises: reading values from a database using the identifier for the respective cell as a key; and calculating a probability of the device being in each of the plurality of sub areas using the indication of signal strength and the values read from the database.
16. 16. The method of claim 15, wherein the database contains, for each cell, and for each of a plurality of sub areas in the search area: an expected signal strength for the sub area; and a variance for the sub area, and wherein calculating a probability of the device being in each of the plurality of sub areas uses the expected signal strength, the variance and the indication of signal strength determined from the received signal.
17. 17. The method of claim 16, wherein the expected signal strength and variance for each sub area are calculated from a plurality of observed signal strength readings from devices located in the subarea.
18. 18. The method of claim 15, wherein the database contains, for each cell, for a plurality of signal strengths, and for a plurality of sub areas in the search area, a probability of the device being in each of the plurality of sub areas, given the signal strength is the indication of signal strength of the signal received from the respective cell, and wherein calculating a probability of the device being in each of the plurality of sub areas comprises either: reading the probability of the device being in each of the plurality of sub areas from the database for a signal strength that is closest to the indication of signal strength; or interpolating the probability of the device being in each of the plurality of sub areas from the database between two or more signal strengths that are closest to the indication of signal strength.
19. 19. The method of any of claims claim 16 to 18, further comprising: for each of the sub areas in the search area for which either: an expected signal strength and variance; or a probability of the device being in the sub area, given the signal strength is the indication of signal strength for the respective cell is not available, calculating interpolated values for the expected signal strength, variance and/or probability using available values in the search area.
20. 20. The method of claim 19, wherein the interpolated values for the expected signal strength are calculated using a model of expected signal strength.
21. 21. The method of claim 15, wherein the probabilities of the device being in each of the plurality of sub areas, given the indication of signal strength for the respective cell, are calculated using a model of expected signal strength.
22. 22. The method of claim 20 or claim 21 wherein the model of expected signal strength decreases logarithmically with distance from a base station serving the cell and variance of the expected signal strength increasing with distance from the sub areas for which the values of expected signal strength and variance are available in the database.
23. 23. The method of claim 20 or claim 21, wherein the model of expected signal strength decreases logarithmically with distance from a base station serving the cell with an additional decrease to provide adjustment for obstacles and variance of the expected signal strength increasing with distance from the sub areas for which the values of expected signal strength and variance are available in the database.
24. 24. The method of any of claims 12 to 14, wherein the information determined from the received signal comprises an identifier for the cell and an indication of signal strength, and wherein using the determined information to ascertain a plurality of parameters characterising a respective probability distribution for the geographic location of the subscriber device given the determined information comprises: using a model to calculate the value related to probability of the subscriber device being in each sub area, given the indication of signal strength for the respective cell.
25. 25. The method of claim 24, wherein using the model used to calculate the value related to probability of the subscriber device being in each sub area comprises estimating the attenuation of the signal from a base station serving the cell to the sub area by calculating: a distance between the base station and the sub area; and/or an angle between a primary transmit direction of the base station and a direction from the base station to the sub area.
26. 26. A method for estimating a geographic location of a subscriber device configured to communicate using a cellular communications network, the method comprising: estimating an approximate geographic location of the device by the method according to any of claims 5 to 11; and estimating a geographic location of a subscriber device by the method according to any of claims 12 to 25 wherein the geographic search area is determined using the approximate geographic location of the subscriber device. 2.Ί. The method of any preceding claim, wherein the information determined from the received signal comprises an identifier for the cell and an indication of signal strength, wherein ascertaining a plurality of parameters comprises reading values from a database using the identifier for the cell as a key and using the values from the database to arrive at the parameters characterising the probability distribution for the geographic location of the subscriber device, the method further comprising: receiving an accurate indication of the location of the subscriber device; calculating updated database values that would result in a more accurate estimate of the geographic location of the subscriber device; and updating the database with the updated values.
27. 28. A method for estimating the geographic location of a subscriber device configured to communicate using a cellular communications network, the method comprising: receiving a plurality of signals at the subscriber device, the plurality of signals relating to a respective plurality of cells; for each of the plurality of signals, determining: an estimated location of a centroid of the cell; and a confidence that the subscriber device is at the estimated location of the centroid of the cell; calculating a weighted average of the estimated locations of the centroids, weighted by the confidence for each centroid to determine an averaged location and an associated confidence; using the averaged location and associated confidence to determine a geographic search area for the subscriber device, the geographic search area comprising a plurality of sub-areas; for each of the plurality of signals, determining a value for each sub-area within the search area, the value for each sub-area indicating a probability that the subscriber device is located within that sub-area; for each sub-area, combining the values from each signal to obtain a joint value for each sub-area indicating the joint probability that the subscriber device is located in that sub-area; and obtaining the estimate of the geographic location of the subscriber device from the joint values.
28. 29. A cellular telecommunications network comprising a plurality of cells and a subscriber device configured to communicate using the cellular communications network, wherein the cellular is configured to estimate the geographic location of the subscriber device using a method according to any preceding claim.