METHOD OF ESTIMATING A GEOGRAPHIC LOCATION OF A MOBILE DEVICE

Abstract

A method of estimating a geographic location of a mobile device configured to communicate using a telecommunications network comprising a plurality of cells is provided. The method comprises obtaining one or more data records from the mobile device, wherein each data record comprises a plurality of signal measurements and respective cell identifiers, wherein each signal measurement relates to a signal received by the mobile device from a cell of the plurality of cells attributed with the respective cell identifier. The method further comprises, for each data record, generating an ordered list of signal properties comprising the plurality of signal measurements, wherein each signal property in the ordered list of signal properties is correlated with a corresponding cell identifier assigned to the cell from which the signal was received, based on an index of the signal property in the ordered list of signal properties. The method further comprises, for each data record, estimating the geographic location of the mobile device based on theordered list of signal properties.

Claims

1. A method of estimating a geographic location of a mobile device configured to communicate using a telecommunications network comprising a plurality of cells, the method comprising: obtaining one or more data records from the mobile device, wherein each data record comprises a plurality of signal measurements and respective cell identifiers, wherein each signal measurement relates to a signal received by the mobile device from a cell of the plurality of cells attributed with the respective cell identifier; for each data record, generating an ordered list of signal properties comprising the plurality of signal measurements, wherein each signal property in the ordered list of signal properties is correlated with a corresponding cell identifier assigned to the cell from which the signal was received, based on an index of the signal property in the ordered list of signal properties; and for each data record, estimating the geographic location of the mobile device based on the ordered list of signal properties.

2. The method of claim 1, wherein the one or more data records comprise a first data record corresponding to a first time and a second data record corresponding to a second time different to the first time, wherein each data record further comprises an identifier linked to the mobile device, wherein the method further comprises: using the identifiers from the first and second data records to determine that the first and second data records have been obtained from the same mobile device.

3. The method of claim 2, wherein for the second data record, estimating the geographic location of the mobile device comprises estimating the geographic location of the mobile device based on the ordered list of signal properties and the estimated geographic location of the mobile device for the first data record.

4. A method of providing training data to a model for estimating a geographic location of a mobile device based on a plurality of signal measurements, the method comprising: obtaining one or more data records from a mobile device, wherein each data record comprises: a plurality of signal measurements and respective cell identifiers, and a corresponding measurement of the geographic location of the mobile device, wherein the mobile device is configured to communicate using a telecommunications network comprising a plurality of cells, wherein each signal measurement relates to a signal received by the mobile device from a cell of the plurality of cells attributed with the respective cell identifier; for each data record, generating an ordered list of signal properties comprising the plurality of signal measurements, wherein an index of each signal property in the ordered list of signal properties is correlated with a corresponding cell identifier assigned to the cell from which the signal was received, based on an index of the signal property in the ordered list of signal properties; and for each data record, providing the ordered list of signal properties and the corresponding measurement of the geographic location of the mobile device to the model.

5. The method of claim 4, wherein the mobile device is a first mobile device, wherein the ordered list is a first ordered list, wherein the model is configured to receive as an input an ordered list of signal properties obtained from a second mobile device configured to communicate using the telecommunications network, wherein an index of each signal property in the ordered list of signal properties from the second device is correlated with a corresponding cell identifier assigned to the cell from which the signal was received, based on an index of the signal property in the ordered list of signal properties, and wherein the model is configured to provide as an output an estimation of a geographic location of the second mobile device.

6. The method of claim 4, wherein generating the ordered list of signal properties comprises, for each signal property in the ordered list having an index corresponding to a cell identifier for which a respective signal measurement has not been obtained, assigning a value to the signal property in the ordered list indicating that no signal has been received by the mobile device.

7. The method of claim 4, wherein the mobile device is: a user equipment; an aerial vehicle; an unmanned aerial vehicle, UAV; a user equipment aboard a manned or unmanned aerial vehicle; or an airborne device.

8. The method of claim 7, wherein each data record further comprises an altitude measurement of the mobile device, wherein the model is configured for use with signal measurements relating to a predetermined range of altitudes and wherein the altitude measurement of the mobile device lies within the predetermined range of altitude measurements.

9. The method of claim 4,, wherein the plurality of signal measurements comprises: one or more signal strength measurements; one or more RSSI measurements; one or more RSRP measurements; one or more signal quality measurements one or more RSRQ measurements; and/or one or more location measurements.

10. The method of claim 4, wherein the method further comprises: receiving at the mobile device a plurality of signals from a subset of the plurality of cells, each signal comprising encoded data, wherein the encoded data comprises a cell identifier attributed to the cell from which the signal was received; for each of the plurality of received signals, determining one or more signal measurements; for each of the plurality of received signals, decoding the signal and determining the cell identifier encoded within the signal; and communicating a data record comprising the signal measurements and respective cell identifiers to a cell of the plurality of cells.

11. A method of correlating each signal property in an ordered list of signal properties to a corresponding cell identifier based on an index of the signal property in the ordered list of signal properties, wherein the ordered list of signal properties is suitable for input to a model for estimating a geographic location of a mobile device, the method comprising: reading one or more data records, wherein each data record comprises a plurality of signal data entries, wherein each signal data entry corresponds to a signal received by a mobile device and wherein each signal data entry comprises: a signal measurement of the signal, and a respective cell identifier assigned to a cell from which the signal was received; for each data record of the one or more data records and for each signal data entry of the plurality of signal data entries in the data record, reading the cell identifier of the signal data entry; and for each unique cell identifier read from the plurality of signal data entries and from the one or more data records, correlating a signal property in an ordered list of signal properties with the corresponding unique cell identifier, based on an index of the signal property in the ordered list of signal properties.

12. A method of generating an ordered list of signal properties, the method comprising: reading a data record comprising a plurality of signal data entries, wherein each signal data entry corresponds to a signal received by a mobile device and wherein each signal data entry comprises: a received signal of the signal, and a respective cell identifier assigned to a cell from which the signal was received; for each signal data entry of the plurality of signal data entries in the data record, populating a signal property in the ordered list of signal properties with the signal measurement of the signal data entry, wherein the updated signal property is correlated to the corresponding cell identifier of the signal data entry, based on an index of the signal property in the ordered list of signal properties.

13. The method of claim 12, further comprising initialising each signal property in the ordered list with a value indicating that no signal has been received.

14. The method of claim 12, wherein each cell identifier is: a Physical Cell ID, PCI; and/or Cell Global Identity, CGI.

15. (canceled)

16. A mobile equipment comprising: a processor; and a computer memory, wherein the computer memory comprises instructions that, when executed on the processor, case the mobile equipment to perform the method of claim 1, wherein the mobile equipment further comprises the mobile device from which the one or more data records are obtained.

17. A mobile equipment comprising: a processor; and a computer memory, wherein the computer memory comprises instructions that, when executed on the processor, case the mobile equipment to perform the method of claim 4, wherein the mobile equipment further comprises the mobile device from which the one or more data records are obtained.

18. A mobile equipment comprising: a processor; and a computer memory, wherein the computer memory comprises instructions that, when executed on the processor, case the mobile equipment to perform the method of claim 11, wherein the mobile equipment further comprises the mobile device from which the one or more data records are obtained.

Description

BRIEF DESCRIPTION OF THE FIGURES

[0086] The present invention will now be described in more detail with reference to a number of non-limiting examples, depicted in the following figures in which:

[0087] FIG. 1 illustrates a mobile device configured to communicate with a telecommunications network.

[0088] FIG. 2 illustrates a map showing example geographic locations of a mobile device and three cells.

[0089] FIG. 3 illustrates a system and data flow diagram.

[0090] FIG. 4 illustrates a flowchart of a method of estimating the geographic location of a mobile device.

[0091] FIG. 5 illustrates a flowchart of a method of providing training data to a model for estimating a geographic location of a mobile device based on a plurality of signal measurements.

[0092] FIG. 6 illustrates a flowchart of a method of correlating each signal property in an ordered list of signal properties to a corresponding cell identifier based on an index of the signal property in the ordered list of signal properties.

[0093] FIG. 7 illustrates a flowchart of a method of generating an ordered list of signal properties.

[0094] FIG. 8 illustrates a flowchart of a method of restructuring the data in a specific implementation.

DETAILED DESCRIPTION

[0095] Existing information representation format from an MR (Measurement Report) uses nominal numbering to identify neighbour cells. Systems for defining the numbering may be different depending on the mobile service provider. These Measurement Reports may be used for determining when to perform handover and to which neighbour cell handover should be performed. They may also be used for re-establishment of a connection after signal has been lost. To support these operations, the ordering of the neighbour cells in the MR may be from the neighbour with the strongest RSRP at position #1 to the neighbours with weaker RSRP values at higher numbers. Alternatively, the Reference Signal Receive Quality may be the criterion on which the list is sorted. Measurement reporting is defined in the 3GPP specification TS 36.331 for LTE (4G) and TS 38.331 for NR (5G), both of which are herein incorporated by reference.

[0096] It is already known to use Measurement Reports to estimate the geographic location of the mobile device. RPS techniques provide the MRs to Machine Learning ML models to train and estimate geographic location. However, the nominal numbering of the neighbouring cells in the measurement report may be misleading for Machine Learning (ML) models to learn patterns.

[0097] For a given sample MR, a mobile device may report multiple neighbour cells that have similar Reference Signal Receive Power (RSRP). From the measurement reports, it is not always clear which cell is the #1 neighbour cell in the MR and which is #2 and so on. Instead, the MR structuring systems can sometimes use arbitrary and nominal numbering for neighbour cells information representation. The nominal numbering systems in these MRs can mislead the ML model and prevent the model from learning useful features (e.g. neighbour PCls) and correlating each neighbour cell with its corresponding radio signal profile (e.g. RSRP).

[0098] Referring to FIG. 1, there is illustrated a mobile device 100 configured to communicate with a cellular telecommunications network. The telecommunications network comprises a plurality of cell transceivers 140, 142 and 144 and associated cells 150, 152 and 154. The mobile device in this specific non-limiting example is an unmanned aerial vehicle (UAV). The mobile device is configured to receive signals 160, 162 and 164 from cells 150, 152 and 154. For each received signal, the mobile device measures a property of the received signal (e.g. by measuring the received signal strength and/or quality) and determines a cell identifier indicating which cell transmitted the signal. The mobile device generates a measurement record from the signal measurement and cell identifier information and this measurement record (MR) is sent to one or more of the cells as a signal 168. Alternatively, the MR may be sent to another entity in the network.

[0099] In this Figure, each cell transceiver is shown as a discrete tower. However, in some cases a single tower may comprise cell transceivers serving different cells. The cell transceivers may be base stations. A base station may serve more than one cell. The reference signals are each associated with a cell so this application generally refers to signals from cells. Nevertheless, the invention could be implemented using base station identifiers, rather than cell identifiers.

[0100] Referring to FIG. 2, the measurement record may be used to estimate a geographic location of the mobile device 100, based on the signal measurements and cell identifiers in the measurement record that relate to cells 150, 152 and 154 in the geographic vicinity of the mobile device.

[0101] There may be a plurality of cells in the telecommunications network that are assigned with the same cell identifier. In other words, the cell identifiers may be reused at different geographic locations around the network. However, the reuse scheme should be planned so that neighbouring cells do not share a cell identifier. Moreover the reuse scheme should be planned so that no cell has more than one neighbour with the same cell identifier. The combination of signals received at a mobile device at any geographic location in the network, along with the respective cell identifiers and signal measurements of those signals, should be sufficient to provide a good estimate of the geographic location of the mobile device. This is because the planning of the reuse scheme, along with the combination of signals received from all neighbour cells in the geographic area should allow the model to determine which cell each signal relates to, even though the cell identifiers may not be unique.

[0102] Physical Cell ID (PCI) may be used as the cell identifier. There are 504 available PCls for use in an LTE telecommunications network. To avoid PCI collisions, neighbouring cells must not share a PCI. To avoid PCI confusion, no cell in the network can have two neighbours that share the same PCI.

[0103] To address the data issues that arise in prior art RPS techniques, an innovative feature engineering method is provided. This method may be used to combine and reformat neighbour cell data representation into a new format. In this new format, arbitrary and nominal numbering as neighbour cell features, which causes ML models confusion and misinterpretation of neighbour cells information, is removed from both training and testing data. Instead, in this new format, distinct neighbour cell PCls (for both training and testing data) are identified and set as new columns/features/elements in the data.

[0104] Per row/record/MR, if a neighbour cell identified by its PCI was reported and present, its row value (signal property in the ordered list of signal properties) is populated with its corresponding radio information (i.e. RSRP value). Otherwise, its row value is filled with zero. Alternatively, a different value may be used to signify that a cell identified by its PCI has not been reported (e.g. 999 or -999). In this way, a clear and explicit association between each neighbour cell and its radio signal profile is built for all MRs/records collected in a geo-area of interest. This approach may greatly improve input data quality and may also remove neighbour information ambiguity.

[0105] The newly regenerated sub-dataframe regarding neighbour cell information (the ordered list of signal properties) is merged back to the original dataframes (which got ‘old’ neighbour cell representation removed) to form new “modified” dataframes. In other words, the tuples in the measurement records providing neighbour cell information are replaced by an ordered list of signal properties. This may be performed for both training and testing data.

[0106] The reconstructed dataframes are used to train an existing ML model. This may be performed without modifying the other data in the measurement record or changing the ML model. In doing so, model prediction accuracy enhancement may be observed. This accuracy enhancement may exceed the performance of previous models.

[0107] In summary, the whole data set is transformed and reconstructed to provide complete and more accurate neighbouring information to ML models, which in turn can better detect & learn hidden patterns and radio FPs.

[0108] Table 1 illustrates an example of ten measurement records communicated to the network from a mobile device. Table 2 illustrates how the measurement records of Table 1 may be restructured before being provided to a model.

TABLE-US-00001 MR1 MR2 MR3 MR4 MR5 MR6 MR7 MR8 MR9 MR10 lat (degrees) 41.8089 41.8089 41.80889 41.80888 41.80888 41.80888 41.80888 41.80889 41.80891 41.809 Ion (degrees) 2.16326 2.16325 2.16322 2.16321 2.16318 2.16316 2.16314 2.16312 2.16309 2.16307 alt (m) 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 serving PCI 323 323 323 323 323 323 323 323 323 323 servingenodeb 81947-7 81947-7 81947-7 81947-7 81947-7 81947-7 81947-7 81947-7 81947-7 81947-7 serving rsrp (dBm) -97 -97 -102.6 -102 -103 -102.5 -101.5 -101 -98.7 -109 serving rsrq (dB) -12.8 -13 -15.4 -14.5 -15 -15.2 -14.2 -13.5 -13.3 -15 pci_neigh_1 290 322 322 45 45 45 45 45 45 251 rsrp_neigh_1 -106 -104 -107.8 -109 -111 -109 -109 -110 -110 -114 pci_neig_2 322 0 483 143 143 54 54 251 251 290 rsrp_neigh_2 (dBm) -105 0 -109.6 -109 -110 -109 -108 -107 -107 -110 pci_neigh_3 0 0 290 251 251 251 251 290 290 322 rsrp_neigh_3 (dBm) 0 0 -107.5 -109 -109 -105.5 -106.5 -108 -106.7 -116 pci_neigh_4 0 0 251 290 290 290 290 322 322 0 rsrp_neigh_4 (dBm) 0 0 -107.7 -107 -103.5 -103.5 -105.5 -108 -107 0 pci_neigh_5 0 0 45 322 322 322 322 405 405 0 rsrp_neigh_5 (dBm) 0 0 -109 -108 -109 -108.5 -107 -110 -110 0 pci_neigh_6 0 0 143 0 483 483 405 483 0 0 rsrp_neigh_6 (dBm) 0 0 -110 0 -109.5 -109 -109.5 -109 0 0 pci_neigh_7 0 0 0 0 0 0 0 0 0 0 rsrp_neigh_7 (dBm) 0 0 0 0 0 0 0 0 0 0 ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... pci_neigh_32 0 0 0 0 0 0 0 0 0 0 rsrp_neigh_32 (dBm) 0 0 0 0 0 0 0 0 0 0 num_neig_cells 2 1 6 5 6 6 6 6 5 3

TABLE-US-00002 MR1′ MR2′ MR3′ MR4′ MR5′ MR6′ MR7′ MR8′ MR9′ MR10′ lat (degrees) 41.80891 41.8089 41.80889 41.80888 41.80888 41.80888 41.80888 41.80889 41.80891 41.809 lon (degrees) 2.16326 2.16325 2.16322 2.16321 2.16318 2.16316 2.16314 2.16312 2.16309 2.16307 alt (m) 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 serving PCI 323 323 323 323 323 323 323 323 323 323 servinqenodeb 81947-7 81947-7 81947-7 81947-7 81947-7 81947-7 81947-7 81947-7 81947-7 81947-7 ta (T.sub.s) 85 85 85 85 84 85 84 85 85 84 serving rsrp (dBm) -97 -97 -102.6 -102 -103 -102.5 -101.5 -101 -98.7 -109 serving rsrq (dB) -12.8 -13 -15.4 -14.5 -15 -15.2 -14.2 -13.5 -13.3 -15 128 (rsrp dBm) 0 0 0 0 0 0 0 0 0 0 386 (rsrp dBm) 0 0 0 0 0 0 0 0 0 0 ... ... ... ... ... ... ... ... ... ... ... 143 (rsrp dBm) 0 0 -110 -109 -110 0 0 0 0 0 ... ... ... ... ... ... ... ... ... ... ... 405 (rsrp dBm) 0 0 0 0 0 0 -109 -110 -110 0 ... ... ... ... ... ... ... ... ... ... ... 290 (rsrp dBm) -106 0 -107 -107 -103 -103 -105 -108 -106 -110 ... ... ... ... ... ... ... ... ... ... ... 45 (rsrp dBm) 0 0 -109 -109 -111 -109 -109 -110 -110 0 ... ... ... ... ... ... ... ... ... ... ... 54 (rsrp dBm) 0 0 0 0 0 0 -109 -108 0 0 ... ... ... ... ... ... ... ... ... ... ... 322 (rsrp dBm) -105 -104 -107 -108 -109 -108 -107 -108 -107 -116 323 (rsrp dBm) 0 0 0 0 0 0 0 0 0 0 ... ... ... ... ... ... ... ... ... ... ... 483 (rsrp dBm) 0 0 -109 0 -109 -109 0 -109 0 0 ... ... ... ... ... ... ... ... ... ... ... 251 (rsrp dBm) 0 0 -107 -109 -109 -105 -106 -107 -107 -114 ... ... ... ... ... ... ... ... ... ... ...

[0109] Table 1 illustrates an example of ten measurement records (MR1 to MR10) communicated to the network from a mobile device. In this example, the Measurement Records (MRs) belong to a training dataset. Therefore, each MR comprises latitude and longitude information transmitted from the mobile device. Likewise, validation data may also contain location measurement data. If this were test data, rather than training data, the same principles would apply but the latitude and longitude measurements would be absent. As can be seen in Table 1, each MR further comprises a list of neighbour cell PCI and RSRP values numbered from 1 to 32. In this example, each MR has no data for neighbours 7 to 32. The rows of the Table corresponding to neighbours 8 to 31 are therefore omitted, for brevity. In this example, each MR also comprises some further information, such as the altitude of the drone, the PCI of the serving cell, the identifier of the Serving eNodeB (“servingenodeb”), RSRP and RSRQ values of the serving cell, and the like.

[0110] Timing Advance (TA) May be added to the measurement record by the eNB. Timing Advance is expressed in units of T.sub.s, which is the basic time unit defined in the 3GPP standard. Timing advance is a “negative” offset, at the UE, between the start of a received downlink subframe and a transmitted uplink subframe. This may allow the UE to synchronize uplink and downlink transmissions. The TA value may be continually measured by the eNB and can be dynamically adapted and signalled to the eNB. TA is a measurement of time so can be expressed in microseconds. However, it is more usual to express TA in multiples of a basic time unit (T.sub.s), which is defined in 3GPP standard 36.211. For example, T.sub.s=1/(subcarrier spacing × FFT-size block-by-block);T.sub.s=1/(15000 × 2048) seconds=0.0325 microseconds. This may be for an LTE network where the subcarrier spacing is 15kHZ and the FFT-size block-by-block is 2048. These numbers will likely be different for a 5G network so the basic time unit will be different (e.g. subcarrier spacing can be 30 kHz or 60 kHz depending on the spectrum frequency).

[0111] Table 2 illustrates how the measurement records of Table 1 (MR1 to MR10) may be restructured before being provided to a model. This particular example shows that each restructured measurement record (MR1′ to MR10′) is represented by a column in the table and the table comprises a row for each PCI in the data set. In this example, the rows corresponding to PCls for which no RSRP values are available for any of the MRs have been omitted. However, every available PCI may have a row in the table. The table could equally be configured so that each MR is represented by a row and the table comprises a column for each PCI in the data set. Importantly, the ordering of the signal properties for each measurement record is the same as the ordering for each other measurement record. Likewise, the ordering of the signal properties for the training and test data follow the same ordering system.

[0112] It should be noted that the signal strength values correlated to PCI 323 are all populated with “0” values. This is significant because 323 is the PCI of the serving cell for all the measurement records in this example. In the method used to modify the measurement records in this specific example, only data for neighbour signal strength measurements in the original MR is used to populate the ordered lists that form part of the updated measurement records (MR1′ to MR10′). In an alternative method, it may be beneficial to incorporate the signal strength measurement for the serving cell into the ordered list, as well as the signal strength measurements for the neighbour cells.

[0113] FIG. 3 illustrates a system and data flow diagram. Reference signals sent from one or more cells 150, 152 and 154 are received by a mobile device 100. The mobile device 100 determines a signal measurement of each signal and a cell identifier associated with the cell from which the signal was received. The mobile device 100 generates a measurement record comprising the signal properties and respective cell identifiers. The mobile device 100 sends the measurement record to a data processor 320. The measurement record may be sent to the data processor 320 by sending the measurement record to one of the cells 150, 152 and 154 from which a reference signal was received or a different cell. The measurement record may be sent to the data processor 320 by sending the measurement record to a base station associated with a cell. The cell/base station provides the measurement record to a data processor 320 in the network or comprises a data processor 320. The data processor 320 reads the signal measurements and cell identifiers in the measurement record and generates an ordered list of signal properties. The data processor 320 provides the ordered list of signal properties to a model 310, along with any other required information from the measurement record.

[0114] The measurement record may also comprise a geographic location of the mobile device 100. In this case, the model may use the location data and ordered list of signal properties (along with any other required information from the measurement record) as training data to improve the accuracy of future estimates.

[0115] The model may use the ordered list of signal properties to estimate a geographic location of the mobile device 100. This estimate of geographic location may be used elsewhere in the network. For example, the estimate of geographic location may be used to predict an arrival time of the drone at a specified location.

[0116] MRs may be collected and sent with a time interval of around 2-10 seconds. However, the methods can be performed with MRs sent with any time interval or single one-off records.

[0117] Whilst the model 310 and data processor 320 are described as being on the network side, these techniques could equally be used on the mobile device 100 itself. In this case, the step of transmitting the measurement record to a cell would not be required. The mobile device 100 could directly format/structure the signal measurements into an ordered list and use this with a model to estimate its own location. This may be advantageous in cases where the mobile device 100 is not equipped with other means by which to determine a location (e.g. GNSS). Moreover, this may save battery power compared to GNSS techniques.

[0118] FIG. 4 illustrates a flowchart of a method of estimating a geographic location of a mobile device. The method comprises the following steps: [0119] S401: Obtain one or more data records from the mobile device, wherein each data record comprises a plurality of signal measurements and respective cell identifiers from a mobile device; [0120] S403: For each data record, generate an ordered list of signal properties comprising the plurality of signal measurements; and [0121] S405: For each data record, estimate the geographic location of the mobile device based on the ordered list of signal properties.

[0122] FIG. 5 illustrates a flowchart of a method of providing training data to a model for estimating a geographic location of a mobile device based on a plurality of signal measurements. The method comprises the following steps: [0123] S501: Obtain one or more data records from a mobile device, wherein each data record comprises: a plurality of signal measurements and respective cell identifiers from a mobile device and a corresponding measurement of the geographic location of the mobile device; [0124] S503: For each data record, generate an ordered list of signal properties comprising the plurality of signal measurements; and [0125] S505: For each data record, provide the ordered list of signal properties and the corresponding measurement of the geographic location of the mobile device to the model.

[0126] FIG. 6 illustrates a flowchart of a method of correlating each signal property in an ordered list of signal properties to a corresponding cell identifier based on an index of the signal property in the ordered list of signal properties. The method comprises the following steps: [0127] S601: Read one or more data records, wherein each data record comprises a plurality of signal data entries; [0128] S603: For each data record and for each signal data entry in the data record, read the cell identifier of the data record; and [0129] S605: For each unique cell identifier, correlate a signal property in an ordered list with the corresponding unique cell identifier, based on an index of the signal property in the ordered list.

[0130] FIG. 7 illustrates a flowchart of a method of generating an ordered list of signal properties. The method comprises the following steps: [0131] S701: Read a data record comprising a plurality of signal data entries; and [0132] S703: For each signal data entry in the data record, populate a signal property in the ordered list of signal properties with the signal measurement of the signal data entry.

[0133] FIG. 8 illustrates a method performed to restructure the data in a specific implementation. The method comprises the following steps: [0134] S801: Collect and obtain all unique neighbour cell PCls (Physical Cell IDs) in both raw train and test sets; [0135] S803: Generate a list/collection of (PCI, RSRP, etc.) tuples for all neighbour cells reported per row/record; [0136] S805: Iteratively, generate such a list/collection of neighbour tuples for each record, and store all in a nested list/collection; [0137] S807: correlates every unique PCI in the data records with an index in an ordered list of signal properties. This may be achieved using an indexing scheme in the form of table headings so that each unique neighbour PCI is set as a new column/row/feature/element in the table; [0138] S809: Parse and convert the nested list/collection of neighbour tuples into ordered lists, whose value per row is populated with its associated tuple RSRP value. In other words, convert the tuples into an ordered list of signal properties; [0139] S811: Build the reconstructed neighbour cell dataframe by parsing and converting all entries in the nested list/collection; [0140] S813: Remove the original neighbour cell presentation from the raw dataframe, and merge the newly reconstructed neighbour cell dataframe with the remaining to form an updated train/test data set; and [0141] S815: Apply the updated data sets to train and test ML models.

[0142] After data transformation and reformatting, the arbitrary and nominal numbering as neighbour cell features, which may confuse and mislead ML models to learn from data, is removed from both dataframes (train and test).

[0143] The old neighbour cell information representation fails to associate each neighbour cell identified by PCI with its radio signal profile. Thanks to the novel method, a clear and explicit association is built and available for ML models to learn by filtering, combining and reconstructing neighbour cell information.

[0144] A new and novel approach is provided to exactly solve a data problem encountered in radio network measurement reporting and data collection. Rather than normalising data or performing a subset feature selection, which changes no data structure, this novel method is a ‘break and build’ breed to fundamentally change the neighbour cell representation structure, and regenerate it in a style unlocking more and accurate neighbour information for ML models to learn patterns and better presenting the data to the problem to solve. None of the prior art is similar or can achieve the same/similar results.

[0145] Existing prior art methods like OOB ML models, combinations, optimizations and others do not achieve sufficient accuracy in geo-location. This new method makes sense due in part to the fact that it breaks and reconstructs an improved neighbour data representation for ML models to better learn patterns and thus better address the problem.

[0146] The method replaces ‘nominal numbering’ with actual neighbour cell PCls as new features/columns, and associated radio signal profiles (RSRP) which may provide more relevant and useful information for ML models to learn patterns. That is, the proposed method may offer more accurate and complete neighbour information to ML models, and avoid misleading ML models with those ‘nominal numbering’ for neighbour cell information representation.

TABLE-US-00003 Original neighbour cell format Pci_neigh_1 rsrp_neigh_1 pci_neigh_2 rsrp_neigh_1 Sample 1 A va B vb Sample 2 B vb A va

[0147] Before neighbour data reformatting and transformation, ML models may have ‘thought’ these two samples contain different neighbour radio fingerprints, so may make mis-prediction.

[0148] For a human observer, these two MR records may appear equivalent. This may be because a human observer would perform subconscious ‘data re-organising and relationship reduction’. However, for a ML model, the nominal numbering (_1, _2) may cause the model to recognise neighbour _1 as being assigned with different values across samples from neighbour _2. Therefore, for the ML model, what is significant here is the feature/column name i.e. pci_neigh_1/pci_neigh_2. Most ML models do not have human-level intelligence to ‘realise’ the really meaningful neighbour cell representation should be the actual neigh cell PCls: A, B and their associated radio profiles (RSRP). What this novel method does is to make this information explicit to ‘help’ ML models to pick it up and learn right representation of neighbour radio fingerprints.

TABLE-US-00004 New neighbour cell format A B C D E Sample 1 va vb 0 0 0 Sample 2 va vb 0 0 0

[0149] After reformatting, the data provides a more explicit relationship between the PCI and the RSRP. ML models may therefore learn these records contain the same neighbour radio fingerprints, so make the right prediction.

[0150] The above are dummy examples to illustrate the situation. In real samples it can get much more complex and some patterns are implicit and hidden, hard for human users to comprehend. That is where Machine Learning models may be beneficial. A data-driven approach to make us if Al/machine learning techniques can assist with learning, understand and solving challenging problems.

[0151] In this application, the term “mobile device” is used to refer to the device that is configured to communicate with the telecommunications network. However, this device may equally be referred to as a “user device”, “subscriber device”, “mobile handset”, “cellular connected device”, “telecoms device”, and the like.

[0152] Whilst this application generally uses the term “cell” to refer to a source of a telecommunications signal, the skilled person would understand that this signal may come from a number of different elements in a telecommunications network. For example, the signal may originate from a NodeB, an eNodeB, a microcell, a picocell, a femtocell, and the like.

[0153] Specific embodiments of the present invention are described in this application in which the signal measurement is the Reference Signal Receive Power. However, in some cases, these methods may additionally or alternatively use other measures of signal strength. For example, these methods may use the Received Signal Strength Indicator RSSI or Received Channel Power Indicator RCPl. Moreover, in future iterations of the telecommunications standards, different measurements of signal strength may be used. Such measurements of signal strength may be utilised in methods described in the present application.

[0154] Likewise, specific embodiments of the invention described above use the Physical Cell Identifier PCI as the cell identifier. These methods may additionally or alternatively use other cell identifiers, such as the Cell Identification (Cl), Cell ID (CID), Enhanced Cell ID (E-CID), and the like. These cell identifiers may be used in combination with other identifiers, such as the Mobile Country Code MCC, Mobile Network Code MNC, Location Area Code LAC, Location Area Identity LAI, and the like. Moreover, global base station identifiers such as the Base Station Identification Code (BSIC), Cell Global Identity (CGI), and the like may be used in the methods described in this application, in place of (or as well as) the cell identifiers. Moreover, in future iterations of the telecommunications standards, different conventions for identifying cells may be used. Such cell identifiers may be utilised in methods described in the present application.

[0155] Whilst LTE (4G) and NR (5G) applications are described in specific examples above, these techniques could be used for 2G/3G/4G/5G and beyond. Moreover, WiFi signals could also be used in a similar manner to support location estimation techniques.

[0156] Although particular embodiments of the invention have been described, the skilled person will appreciate that various modifications and variations may be made without departing from the scope of the invention.