Pattern-recognition-based geolocation

[0002] 1. Field of the Invention

[0003] The present invention relates to a technique for performing pattern recognition-based geolocation under various RF propagation conditions, and more particularly, to a technique for locating a mobile caller within a cellular service area under various RF propagation conditions.

[0004] 2. Description of the Related Art

[0005] A cellular telephone system must be able to locate a mobile caller within a cellular service area under various RF propagation conditions.

[0006] Conventional methods are based on either a triangulation technique, which requires signals from or to three or more base stations, or an angle of arrival technique, which requires at least two base stations. In many areas, the number of base stations that the mobile unit can detect or can be detected by, is less than two. Furthermore, both the triangulation and angle of arrival techniques suffer from inaccuracies and signal fading which result from multi-path propagation.

[0007] The present invention solves these problems by providing a method, apparatus, article of manufacture, and propagated signal which utilize pattern recognition-based geolocation to locate a mobile caller within a cellular service area under various RF propagation conditions.

[0008] The present invention matches the observed RF characteristics associated with the signal transmitted by a mobile unit (or signal received and reported back to a base station by a mobile unit) to a known set of RF characteristics and other information of a particular location. The location of the mobile unit is determined or estimated when a closely match pattern is found. Pattern recognition technology is utilized to find the matched pattern. In one embodiment, the pattern recognition technique is a Bayes formula for optimal estimation. The main advantages of the present invention are eliminating the need for detecting more than one base station and eliminating inaccuracies caused by multi-path propagation.

[0009] FIG. 1 illustrates a cellular service area divided into cells;

[0010] FIG. 2 illustrates the cells of FIG. 1 further divided into sub-cells;

[0011] FIG. 3 illustrates one embodiment of the present invention; and

[0012] FIG. 4 illustrates the structure and fields which make up the pattern database in one embodiment.

[0013] FIG. 1 shows a cellular (or PCS) service area 10 including a plurality of cells 12. FIG. 1 also illustrates a plurality of base stations BS₁. . . BS_Nand at least one mobile switching center (MSC) 16. FIG. 2 shows the cells 12 of service area 10 divided into sub-cells 20 represented by the squares formed by the grid lines. The numbers shown represent the sub-cell C₁, C₂, C₄, C₅, and C₆respectively. FIG. 3 shows the hardware architecture for determining the location of a mobile unit 30 in one embodiment.

[0014] In particular, FIG. 3 illustrates a mobile phone 30, three base stations BS₁, BS₂, BS₃, a mobile switching center 16, a geolocation server 32, and a data base 34.

[0015] As illustrated in FIGS. 1 and 2, the cellular service area 10 is partitioned into cells 12 and then sub-cells C₁. . . C_M. A set of detectable RF characteristics are defined for each sub-cell, which are referred to as the attributes/properties of the sub-cell. The mobile unit 30 measures the RF signals that are associated with the attributes/properties and reports the results to a primary base station BS₁which in turn, reports to the geolocation server 32 illustrated in FIG. 3 via the MSC 16. The geolocation server 32 statistically 25 compares the measured values with the known attribute values of all sub-cells in a predefined area. The sub-cell that has the best matched set of attribute values with the measured values is the one that the mobile unit 30 is reported to be in. The known set of attribute values in database 34 for each sub-cell account statistically for weather conditions, time of day, and other environmental variations that could affect the RF characteristics. As the mobile unit 30 moves, the comparison is periodically performed and the location of the mobile unit 30 can be determined at any given time. As a result, the problem of geolocation can be solved by pattern recognition.

[0016] In more detail, assume S is the area that the mobile unit 30 is known to be in. For example, S may be a sector that can be determined by the existing technology where the mobile unit 30 is. Further assume that {C₁, C₂, C₃. . . , C_m} is a set of sub-cells which is a partition of S, such that $S\cong \sum _{i-1}^m{C}_i$

[0017] and A={A₁, A₂, A₃, . . . , A_n} is the set of attributes of C_ifor I=1 to m. For example, A₁is the PN code that identifies a particular base station BS_x, A₂is the strength of the PN signal, and A₃is the phase shift. A₁. . . A_nare random variables. D_iis the domain of A_iwhich contains all possible values of A_iand a_jdenotes a property which is defined such that the attribute A_jhas a certain value or a value set that can be used to characterize a sub-cell. For example, a_jmay represent A_j=v or A_j>v where v is a value in D_j. If P(a_i) is the probability of the occurrence of a_iover all sub-cells, P(a_i|C_j) is the probability of the occurrence of a_Iin C_j, P(C_j) the probability that the mobile unit 30 is in C_jindependent of properties, and P(C_j|a_i) is the probability the mobile unit 30 is in C_jgiven an observed property of a_i. The goal of the present invention is to find the highest P(C_i|A*), namely the highest probability that the mobile unit 30 is in C_igiven a set of measured or observed values A*, where A*={a₁, a₂, a₃, . . . , a_n}. In one embodiment, P(C_i|A*) can be obtained using the following Bayes formula:

[0018] P(C_i|A*)=P(C_i)P(A*|C_i)/P(A*)

[0019] P(C_i)can be assumed to be 1/m initially, a uniform distribution, since there is no a priori knowledge where the mobile unit 30 might be. P(A*) can be obtained using:

[0020] P(A*)=ΣP(C_s)P(A*|C_s), S=1, . . . , m and

[0021] P(A*IC_i) can be obtained using:

[0022] P(A*|C_i)=P(a₁|C_i)P(a₂|C_i)P(a₃|C_i) . . . P(a_n|C_i).

[0023] As the mobile unit 30 moves, the measurement is taken at time t+Δt, and P(C_i) is updated with P(C_i|A*_t) where A*_tis the set of properties observed at t+Δt.

[0024] The database 34 for the sub-cells is defined by a set of attributes. The database 34 can be implemented as an adjunct server at the MSC 16 or any base station BS₁. . . BS_Nrunning any commercially available database management system. The database 34 is organized by cells and by sectors to provide efficient searching methods. The attribute values may be obtained by a survey of the coverage area and the probability distribution function for A* can be constructed based on the survey results.

[0025] Given the size of a regular cell is 2 km to 25 km in radius, the size of a sub-cell with 125 m radius will result in a reasonable size for the database. For example, if the cell has a radius of 25 km, the number of sub-cells will be approximately 40,000 in a cell. Hence, there will be 40,000 records per base station which is a relatively small database.

[0026] There are at least two methods that can be used to create the database 34.

[0027] One way is to use specially designed equipment to survey the service area 10 exhaustively. The equipment is used to gather RF characteristics for a location and record the data. The data records are then processed to create the database 34. This process is automated using a computer program to gather data and create the database 34. This method provides an accurate database.

[0028] The database 34 can also be created by the construction of a statistical model. The statistical model is constructed by using an RF propagation formula. The parameters in the model are verified by a limited number of observations of the RF characteristics in sample locations in the service area 10. The probability of certain observations can then be calculated and stored in the database 34.

[0029] FIG. 4 shows an example of the database 34 that contains the statistical data for C₁, C₂, C₃, C₄, C₅, and C₆as in the first column. The second column represents the attribute BS₁and its values, and the third column is the probability denoted by p1 of a particular attribute value could be observed in a sub-cell. For example, BS₁=161/232/16 has a probability of 1 in sub-cell C₁. In other words, let a₁be BS₁=161/232/16, then P(a₁/C₁) has the value 1. The numbers 161/232/16, for example, separate by “/” represent the base station ID, the pilot ID, and the average pilot strength for the base station and pilot, respectively. Similarly, the fourth column and fifth column represent the attribute values for BS₂and their probabilities denoted by p2, and so on for BS₃and p3. The four numbers separated by “/” in the latter cases represent the base station ID, the pilot ID, the chip offset, and the pilot strength, respectively.

[0030] Example

[0031] If an observation of interest is A*=(BS₁=178/416/24, BS₂=161/64/3/33, BS₃=172/208/3/25), the probability that this observation is from a particular cell can be calculated as follows:

[0032] P(A*/C₁)=0*0*0=0

[0033] P(A*/C₁)=0*0*0=0

[0034] P(A*/C₁)=(21.6/24*1)*(26/33*0.18)*(24/25*0.18)=0.022

[0035] P(A*/C₁)=(23.3/24*1)*(0)*(0)=0

[0036] P(A*/C₁)=(19.25/24*1)*(0)*(0)=0

[0037] P(A*/C₁)=(24.8/24*1)*(0)*(0)=0

[0038] Note that P(A*C_i)≧P(A*/C_j) for i≠j →P(C_i/A*)≧P(C_j/A*). Therefore, it is concluded that A* is an observation from C₃. Note also that the probability of P(a2/C₃) is adjusted by a value 0.79 which is derived from 26/33, since a2 does not match exactly the pattern of BS₂in C₃. In one embodiment, the present invention uses the ratio of the pilot strength stored in the database 34 to the observed pilot strength to define the measure of the closeness between the two values.

[0039] In the above embodiment, the RF attributes are discrete attributes, such as average pilot strength, chip offset, or pilot strength. However, in another embodiment, continuous attributes or features may be utilized, such as signature waveforms.

[0040] The set of attributes may be either replaced or enhanced by the signatures of the measured signal waveforms. Other pattern recognition techniques such as fuzzy logic can then be applied to analyze the continuous attributes.

[0041] Utilizing the pattern recognition-based technique or fuzzy logic technique described above may be performed utilizing only one base station or more than one base station.

[0042] The foregoing merely illustrates the principles of the invention. It will thus be appreciated that those skilled in the art will be able to devise various arrangements which, although not explicitly described or shown herein, embody the principles of the invention and are thus within its spirit and scope.