WO2006067022A1

WO2006067022A1 - Method for estimating the displacement of a mobile phone customer

Info

Publication number: WO2006067022A1
Application number: PCT/EP2005/056262
Authority: WO
Inventors: Kurt Majewski; Johannes Nierwetberg
Original assignee: Siemens Aktiengesellschaft
Priority date: 2004-12-21
Filing date: 2005-11-28
Publication date: 2006-06-29
Also published as: TW200637381A; DE102004061538A1

Abstract

The invention essentially relates to a method for estimating the displacement of a mobile phone customer. According to said method, the respective speed of a mobile terminal is estimated using possibly all system and network data available. Statements regarding the reliability of these speed data can be established by calculating, except for a constant factor, the probability that a sequence of measuring data sets were acquired along a path, and calculating an estimated position of the customer during acquisition of a data set in the form of an averaged position vector and, based thereon, calculating an estimation variable for the speed of the user and an estimation variable for a covariance matrix of the speed. For this purpose, instead of a sum across all paths possible by n pixels, only approximately n times the sums across neighboring pixels are produced, thereby allowing to calculate the estimation variable for the speed vector and the covariance matrix of its mean variation for relevant values from n (e.g. n = 6) and number of pixels (e.g. 1000 pixels) using today's computers. The inventive method also makes use of the estimated speed and its estimated covariance matrix for establishing statements and probabilities for which these statements hold true.

Description

description

Procedure for estimating the movement of mobile subscribers

The invention relates to a method for estimating the movement of mobile radio customers, in which a respective speed of a mobile terminal is estimated on the basis of measured data of the mobile radio system.

For position-based services but also for system analysis in mobile networks, it is important to know the speed of the caller. Answering simple questions is often enough, such as: Is mobile communications moving? Fast or slow? North or south?

The technical problem is to use the available system metrics together with appropriate network data to provide estimates that help answer these questions. Statements about the movement of the customers are made difficult by the fact that the system measured values are typically only indirectly indicative of the speed since they actually serve other purposes. In order to be able to make statements about the dynamic behavior of the customers with system measured values, several observations are required. To make matters worse, the system measurements are usually subject to strong random interference.

System metrics that are particularly useful for estimating motion are transit time measurements of signals between mobile and base stations, and transit time differential measurements and receive strength measurements.

German published patent applications DE 100 56 222 and DE 102 32 177 (2000P20698, 2002P11357) disclose methods in which the system measured values of a customer obtained at different times and network data, for example B. Base station positions, transmit powers and receive strength forecasts, which is inferred to customers' movement.

From the website http: // ipcl. ee. queensu. ca / PAPERS / iswc98_bz_sdb .ps is a method for estimating the movement of mobile subscribers, in which a respective speed of a mobile terminal is estimated on the basis of measured data of the mobile radio system with the aid of a speed estimator.

The object underlying the invention is now to specify a method for estimating the movement of mobile radio customers in such a way that as far as possible all existing system and network data are used to obtain as meaningful an estimate of the speed as possible. Furthermore, the certainty with which the statements apply should be estimated.

This object is achieved by the features of claim 1. The further claims relate to preferred embodiments and uses of the method according to the invention.

The invention essentially consists in a method for estimating the movement of mobile radio subscribers, in which a respective speed of a mobile terminal is estimated on the basis of as far as possible existing system and network data and information on the security of this speed information is made by a probability that a Sequence of measurement records were taken along a path, is determined to a constant factor, an estimated position of the customer during the recording of a data set in the form of an average position vector and therefrom an estimator for the speed of the user and an estimator for a Covariance matrix of the velocity is determined. Here, instead of a sum of all possible paths through n pixels, only about n times sums are formed over adjacent pixels, whereby a calculation of the estimator for the velocity vector and the covariance matrix of its

Scattering for relevant values of n, z. B. n = 6, and number of pixels, e.g. B. 1000 pixels, only with today's computer systems feasible.

The inventive method will now be explained in more detail with reference to embodiments.

A search area is decomposed into a finite set of raster cells (pixels) P. We identify the pixels p in P with their position vectors. These vectors are typically two-dimensional when you want to estimate motion on a plane. However, they can also be selected three-dimensionally (or one-dimensionally) if the third spatial dimension can be relevant for the movement (eg in buildings) (or if the movement is limited to one spatial dimension (eg in a tunnel)) , We denote the dimension of the position vectors of the pixels with d.

For each screen cell p there is a set N (p) of screen cells from which this screen cell can be reached within the time that corresponds to two consecutive system measurements. Normally, this set N (p) consists of the pixel p itself and its neighboring pixels.

Let U (p, q) for pe P and qe N (p) be a weight that indicates how often a customer switches from pixel p to pixel q between two system measurements and U (p, off) the frequency that a customer is gets its last system measurement dataset from pixel p. Further, let U (off, p) be the frequency that a customer gets his first system measurement data set from pixel p. In the simplest case, all these values are one. A method for calculating these transition frequencies finds also in the above-mentioned German Offenlegungsschriften DE 100 56 222 and DE 102 32 177.

By comparing the measured values of a system measurement data set m with the expected measured values on the grid cell p, a score H (m, p) e [0, ∞] can be formed. This assessment should be advantageously designed so that

W (m, p): = exp (-H (m, p)) (1)

is approximately proportional to the probability that the measurement data set m from the pixel p comes. Such reviews can be found, for example. also in the above-mentioned German Offenlegungsschriften.

Now let Q (m) c P be the set of all pixels p from which the measurement data set m can be obtained, d. H . for which H (m, p) <∞ holds.

A sequence Hi ₁ , ..., m _k of measurement data records of a

Mobile subscriber is available. The probability that these measurement data sets were taken along the path p_7_, ..., p_e P is given (except for a constant factor) by

ripϊ; :: _. p _k ^mk == u ( _Po , _Pl ) π w (m _D , _P ) u (p _D , _{P] +1} ), <2)

D = I

where pg = Pk + 1 ^~ ° ff applies here and in the following. An estimated position of the customer during the recording of the data record m-j_ is thus by the averaged d-dimensional position vector

given. This can be the estimator

for the d-dimensional velocity vector, where t ^ - -tj_ represents the length of time between the recording of the first and last record. A corresponding estimator for which the d by d-dimensional covariance matrix of velocity is

From this covariance matrix estimates for the safety of the derived statements about the movement of the customer can be obtained.

The speed estimator and its covariance matrix are calculated particularly efficiently by the following method steps because the sum occurring in the above formulas does not have to be explicitly formed over all possible paths.

Complete the following steps:

1. (Initialization) Compute the values CLi (p) for p e Q (m ±)

GCi (p): = U (off, p) W (mi _f p). (6) This value represents, with the exception of a constant factor, the probability that the first measurement data set originates from the pixel p, whereby all other measurement data records remain unconsidered. By restricting the calculation to the relevant pixels in Q (m ^), computation time is saved. Let Aj be the reciprocal of the maximum of these weights. For each pe P, provide a d-dimensional zero vector θ _j (p): = (θι (p)) _pe p = 0 and a d times d-dimensional zero matrix Jl (P) ^{: =} 0. These symbols serve to formulate the following formulas in the case i = 2 in the same way as in the cases i> 2.

2. (iteration) For i = 2, ..., k, do the following: Compute the value QL ^ (q) for q G Q (m ±)

This value represents, with the exception of a constant factor, the probability that the i-th measurement data set originates from the pixel q, taking into account the measurement data sets obtained before, but not the following ones. Let Aj_ be the reciprocal of the maximum of these values. In the formula

(7) the factor A ^ _ _j serves the numerical stability, because the product of (often) small probabilities could result from calculation inaccuracies incorrect zeros.

Since the calculation determines the probabilities only "down to a constant factor" and the final formulas also quotients of these values are formed, they otherwise have no influence on the result.

Compute the d-dimensional vector Q ₁ (q) for every qe Q (m-jj)

(8th )

Q ₁ (q): = A ₁ -IW Om ₁ , q) L 1 (GC ₁ -I (p) (q p) -Q ₁ -I (p)) U (p, q) This vector represents an estimator for the position change between the first and ith measurement data set except for a constant factor, on the condition that the customer was in the pixel q when recording the ith measurement data set.

Finally, calculate the d by d-dimensional matrix γ ^ (q) for all q e Q (m ±)

YL (q): = Ai-i W (πii, q) L * ((qp) (qp) ^τ GC ₁ -I (p) (9) ^pε mr ^Q M

- (qp) © Vi (P) -i-i (p) (qp) ^τ + γ ± -i (P)) ^u (Pf q)

This matrix provides, except for a constant factor, an estimate of the covariance of the distribution of the estimated vector Θj. (q)

3. (completion) Compute the estimator for the velocity vector from the auxiliary vectors

and the estimator for its associated covariance matrix

and deliver them back.

The computational method described uses the intermediate values OCj _^ (q), Θi (q) and γ ^ (q) to form, instead of a sum over all possible paths through n pixels, only about n times sums over adjacent pixels. This will be a calculation the estimator for the velocity vector and the covariance matrix of its dispersion for relevant values of n (eg n = 6) and number of pixels (eg 1000 pixels) can only be performed with today's computer systems.

In the following it is explained how statements can be obtained from the estimators (10) and (11) concerning the movement of the mobile radio customer and with what certainty such statements apply. The following considerations serve this purpose:

Let X be a random d-dimensional vector representing the customer's speed. From the distribution of X we only know the vector of the mean values of its components v and its covariance matrix C. These can be z. B. are estimated by formulas (10) and (11).

An upper bound for the likelihood that the customer is not moving in the direction of a given vector u with u ^τ v> 0 is

P (U ^T X <θ) = P (u ^τ (x -v) <-u ^τ v) = p ((u ^τ (x -v)) ² > (u ^τ v) ² ) ₍₁₂₎

u ^τ Cu

In the estimation it is used that the random variable U ^ ^" (Xv) has the mean value 0 and the variance u ^ Cu Thus the customer moves at least with probability lu ^τ Cu / (u ^τ v) 2 somewhat in the direction of the velocity vector u.

In the special case u: = v, one obtains a statement about the probability of whether the customer is moving in the direction of the estimated speed. For the direction

( ₁₃ )

)

the barrier in (12) takes on its smallest value. In this case, the case of two-dimensional pixels has been assumed and the components have been designated by an index x and y. This smallest value is

cc - c ² ^x ' ^xy ' ^y -ϊ-i (15) ^v χ C _y , _y + v _y C _x , _x - 2 v _x v _y C _x , _y ^'

If this value is greater than 1/2, this means that no statement can be made about the direction of movement, d. H . the direction of movement can not be detected from the measured values.

For a given certainty, γ e] 0, l [

w: = ^C * ^{«+ C} ™, (16)

V l - Y

so applies

Thus, the probability that the actual speed is less than w of the expected

Speed v deviates at least γ. Two-dimensional pixels were assumed and C _xx and Cy y denote the diagonal entries of matrix C. If this w-value is small, the actual velocity deviates

Probability γ only a little from v. If w is greater than the amount of v, no γ-safe statement can be made as to whether the customer is moving at all. You put

\

so applies

^p jxfl ≦ ^w ') ≥ ¹ Y- (19)

So the actual speed is less than w 'with probability γ. If w 'is very large, it means that no statement can be made as to whether the customer is moving or not.

With the help of these considerations can thus be questions, whether a mobile radio customer moves, whether he moves quickly or slowly and in which direction he moves to be answered.

Together with a restriction of the search space on the set P, z. B. on street pixels, the method can be used to analyze the traffic flow. Measurements from selected customers can be used to determine if there are stagnation or congestion on the monitored roads.

Claims

claims

1. A method for estimating the movement of mobile subscribers, in which a respective speed of a mobile terminal and information on the security of these speed indications are thereby estimated,

that a probability that a sequence of measurement data records was recorded along a path is determined, with the exception of a constant factor,

- That an estimated position of the customer during the recording of a first and a last measurement data set in the form of an average position vector is determined, and

that from this with the help of the time period between the recording of the first and last measurement data set an estimator for the

Speed of the user and an estimator for a covariance matrix of the speed is determined.

2. The method according to claim 1,

in which a search area is decomposed into a finite quantity P of raster cells p and for each raster cell p there exists a set N (p) of raster cells from which this raster cell can be achieved within the time corresponding to two successive system measurements,

- in which π, the probability that a sequence of measurement data sets m _lr ..., m _k along a path p ± _r • • • P _r] ς_ e P were added until it is determined to a constant factor according to the following formula

where PQ = Pk + 2 = off, where U (p, q) for pe P and qe N (p) is a weighting of how often a customer switches from pixel p to pixel q between two system measurements, U (p, offj the The number of times a customer receives its last system measurement record from pixel p and U (off, p) indicates the number of times a customer gets his first system measurement data set from the pixel p and where W (m, p) represents a score that is approximately proportional to the probability that the measurement data set m comes from the pixel p .

- In which an estimated position of the customer during the recording of the record m ^ in the form of an average

Position vector p _{t is determined} according to the following formula

Σ -i rm ,,, m _k ._ Pi,, Pk '* - ± Pι,, Pk

P ₁ - = rm ,,, m _k

1 Σ-Ip _x ,, p _k TATXp " _u , p _k

- In which an estimator for the speed v of the user is determined according to the following formula

where t ^ -tj_ represents the time between the recording of the first and last record.

3. The method of claim 2, wherein an estimator for a covariance matrix C of

Speed is determined according to the following formula

4. The method according to claim 1,

in which a search area is decomposed into a finite set P of raster cells p and for each raster cell p there exists a set N (p) of raster cells from which this raster cell can be reached within the time corresponding to two successive system measurements, in which values OCj (p) are initially determined in an initialization step al), wherein the value OC _j (p) represents, with the exception of a constant factor, the probability that the first measurement data set πii originates from the pixel p, bl) a value Aj as Inverse of the maximum of these weights

OCj (p) is determined and cl) for each raster cell a d-dimensional zero vector θl (p): = (ΘΣ (P)) pe P = 0 and a d by d-dimensional zero matrix Ji (p): = 0 to be provided,

in which by subsequent iteration steps for i =

2, ..., k for qe Q (m _± ) ai) a temporary value OC ^ (q) is determined, this value OC ^ (q) except for a constant

Factor represents the probability that the i-th

Since the measured data set originates from pixel q, bi) the respective value Aj__j is determined as the reciprocal of the maximum of the values OL ± -i (p), ci) a respective d-dimensional auxiliary vector Θi (q) for each qe Q (m ^ ) and a respective d times d-dimensional auxiliary matrix J ± (q) for all qe Q (m ^) can be determined with the aid of the values Aj__j and QL ± -χ (p), - in which, in a final step, the estimator for the

Velocity vector can be determined using the auxiliary vectors Θi (q) and the estimator for the covariance matrix of the velocity using the auxiliary matrices y _± (q).

5. The method according to claim 4,

in which first in an initialization step the values QLi (p) for p e Q (m ±) are given by the formula

GCi (p): = U (off, p) W (mi _f p) where U (off, p) indicates the number of times a customer gets his first system measurement dataset from pixel p, and where W (m _lr p) represents a score which is approximately proportional to the probability that a first measurement data set m _± comes from the pixel p,

in which, by subsequent iteration steps for i = 2,..., k for q e Q (m ±), the respective value OC 1 (q) is given by the formula

where U (p, q) for pe P and qe N (p) represents a weighting of how often a customer switches from pixel p to pixel q between two system measurements, where W (m _{1 /} p) represents a score, which is approximately proportional to the probability that the respective measurement data set originates m _± from the pixel p, and

a respective d-dimensional auxiliary vector Q ₁ (q) for each qe Q (m- ^) by the formula

Q ₁ (q): = A ₁ -IW (ITi ^ q) 2- (GC ₁ -I (p) (qp) -Q ₁ -I (p)) U (p, q) and a single d times d-dimensional auxiliary matrix J ± (q) for all qe Q (m ±) by the formula

YL (q): = A ₁ -! W (ITi ₁ , q) L ^ ((qp) (qp) ^τ CC ₁ -I (p) ^pε HärGM

- (qp) © Vi (p) -Q ₁ -I (p) (qp) ^τ + Y ₁ -I (P)) u (p, q) are determined, and - in which, in a final step, the estimator for the

Velocity vector v from the auxiliary vectors Q ₁ (q) by the formula

and the estimator for the covariance matrix of the velocity from the auxiliary matrices J _± (q) by the formula

be determined, where u (p, off) the frequency that one

Customer receives its last system measurement dataset from pixel p and tj _ζ -t ^ represents the time between the first and last record being recorded.

6. Use of a velocity vector v of the terminal and an estimated covariance matrix C of its distribution according to claim 1, for determining the security of statements about the speed and direction of movement of mobile terminals.

7. Use according to claim 6, for determining a minimum probability that the terminal moves in a given direction, this

Minimum probability 1 - u ^τ C u / (u ^τ v) ² is given that the given direction satisfies the inequalities u ^τ v> 0 and u ^τ C u / (u ^τ v) ² <1, and otherwise this minimum probability is zero.

8. Use according to claim 6, for calculating the direction in which the mobile terminal moves with greatest certainty, this direction, in the case of a two-dimensional search space, being given by the following formulas: v _x c _{γ / γ} _{-v γ} C _{X / Y}

^V _x ^C _γ , _γ ^{+ V} _γ C _{X / X} - 2v _χ V _γ C _{X / γ} ^V _Y C _{X / X} - vu ": = ^v _χ C _{Y / Y} + v _γ C _{X / X} - 2 v _x v _γ C _{X / Y} and the probability for this direction of motion at least

1- is.

9. Use according to claim 6, wherein for a given probability γ a velocity value w is determined so that the difference of the correct velocity vector is estimated

Velocity vector v with probability γ has at most the amount w, where the difference w by

is determined.

10. The method of claim 7, wherein for a given probability γ a speed value w ^{f is} determined, such that the probability that the mobile terminal moves at a speed of less than w ^f is greater than γ, the speed value being calculated according to the following formula :