EP1692909A1

EP1692909A1 - Method, computer program with program code means, and computer program product for the description of a propagation performance of a communication signal emitted by a base station in a communication network

Info

Publication number: EP1692909A1
Application number: EP04804596A
Authority: EP
Inventors: Anton Schwaighofer; Joachim Bamberger; Marian Grigoras; Clemens Hoffmann; Volker Tresp
Original assignee: Siemens AG
Current assignee: Siemens AG
Priority date: 2003-12-08
Filing date: 2004-11-29
Publication date: 2006-08-23
Also published as: DE10357213A1; WO2005055635A1; TWI260868B; TW200524310A

Abstract

Disclosed is a method for describing the propagation performance of a communication signal. According to said method, calibration measurements are taken at selected points in the communication network. A model for the propagation performance is determined using the calibration data, said model being generated using a Gaussian process which represents a measured physical property of the communication signal in accordance with a piece of location data or position data (forward model).

Description

Description ^'

Method and computer program with program code means and computer program product for describing a propagation behavior of a communication signal emitted by a base station in a communication network

The invention relates to a modeling of a propagation behavior of a communication signal emitted by a base station in a communication network.

Radio communication systems are based on e.g. B. ireless LAN, Bluetooth, GSM, UMTS or DECT are used in a wide variety of areas. They are omnipresent in industrial production plants and office environments, but also in healthcare.

Propagation properties of an electromagnetic field, which is generated by the communication system emitting communication signals, essentially determine the performance of the communication system in terms of area coverage, availability and transmission rate.

On the one hand, radio network operators are interested in the distribution of the field propagation properties or signal characteristics, such as. B. electromechanical field strength, phase, transit time, wave vector (wave vector), bit error rate, signal-to-noise ratio, etc. to be able to plan the radio network optimally, commissioned after installation of the network To be able to prove system properties within the framework of quality assurance or to be able to diagnose fault conditions during the operation of the system. On the other hand, the network service providers are interested in being able to offer their customers location-dependent services.

The position of the receiving device must be known for this. Since only data that is generated during normal network operation should be used for the position estimates, it is advisable to consider the signal characteristics here as well.

Approaches and procedures are known from the prior art which are concerned with the location of end devices, such as DECT

Handsets or wireless LAN-equipped PDAs and notebooks, in radio networks.

In some approaches, such as that known from [1], the localization is based solely on the network topology. The position of the end device is determined on the basis of the base station to which it is connected and its connection history.

However, the accuracy of such a method is low, since only a very large area around the base station to which the terminal is connected can be specified as a possible location.

Other known methods try to estimate the position based on the received field strengths of all available transmitters. In some cases, a detailed physical model for wave propagation is used. However, detailed information about the surroundings is required for this.

From [2] it is known to use knowledge of the electromagnetic properties of the various walls in the building. Such knowledge is generally not available. Therefore, mostly a field strength map based on of a wave propagation model, which will later be used for localization.

A point estimate of the receiver position is often made on the basis of the field strength map [3], [4].

[5] describes a recursive stochastic nonlinear filter method for position estimation of DECT mobile phones. Here, too, a dispersion model, in this case a statistical non-linear model, is created or used as the basis for the position estimation.

In most of these model-based procedures known from the prior art, the model for the propagation behavior is based on calibration measurements, in which a physical variable characterizing the propagation behavior, such as the field strength mentioned above, is measured at previously known positions (calibration positions).

Using the calibration positions and at them

The model for describing the propagation behavior is determined at positions of measured propagation quantities.

A disadvantage of these model-based procedures mentioned above is that a large number of calibration positions have to be measured in order to obtain a sufficiently precise model for the propagation behavior and thus for a sufficiently accurate position determination based on this.

The invention is therefore based on the object of specifying a procedure for model generation for a propagation behavior which requires fewer calibration positions to be measured.

This task is accomplished by the method as well as by the computer program with program code means and the computer program product for the description of a propagation behavior a communication signal emitted by a base station in a communication network with the features according to the respective independent patent claim.

In the method for describing the propagation behavior of the communication signal, a physical property of the communication signal associated with the respective selected position is measured at selected positions in the communication network.

The physical property characterizes the propagation behavior of the communication signal.

Using the selected positions or using corresponding position or location information of the selected positions and the associated measured physical properties of the communication signal, a model for the propagation behavior is determined, which model describes the propagation behavior.

The modeling is carried out using a Gaussian process, which represents the measured physical property as a function of the location information or the position information ("forward model").

To differentiate between models, it should be noted that in a so-called forward model, the physical property of the communication signal is described as a function of a position in the communication network or a distance. A backward or inverse model describes the position in the communication network as a function of the physical property of the communication signal.

A major advantage of the invention is that the modeling of the propagation behavior with Gaussian processes allows a drastic reduction of calibration measurements with only a slight loss of accuracy. These can be determined, for example, using a design process or "design process" [2].

The computer program with program code means is set up to carry out all steps according to the inventive method for describing the propagation behavior when the program is executed on a computer.

The computer program product with program code means stored on a machine-readable carrier is set up to carry out all steps according to the inventive method for describing the propagation behavior when the program is executed on a computer.

The computer program with program code means, set up to carry out all steps according to the inventive modeling process when the program is executed on a computer, and the computer program product with program code means stored on a machine-readable medium, set up all steps according to the inventive Carrying out modeling processes when the program is executed on a computer are particularly suitable for carrying out the method according to the invention or one of its further developments explained below. Preferred developments of the invention result from the dependent claims.

The further developments described below relate both to the method and to the software implementations.

The invention and the further developments described below can be implemented both in software and in hardware, for example using a special electrical circuit.

Furthermore, an implementation of the invention or a further development described below is possible by means of a computer-readable storage medium, on which the computer program with program code means, which carries out the invention or further developments, is stored.

The invention or any further development described below can also be implemented by a computer program product which has a storage medium on which the computer program with program code means which carries out the invention or further developments is stored.

In the case of communication in a communication network, such as a radio network, between a mobile communication device (mobile station), for example a mobile telephone, and a base station, for example an omnidirectional antenna or an omnidirectional antenna or one or more sectoral antennas, data that is Communication signals, in signal packets, so-called bursts, transmitted. Based on (measurable) physical properties of the transmitted or radiated communication signals or signal packets, various distance-relevant parameters can be determined, which in turn can be used as the basis for determining the radiation or signal characteristics of (signal) transmitters.

Such a distance-relevant, i.e. Distance-dependent parameters are, for example, a field strength of a communication signal or signal packet, a phase, a running time, a wave vector, a bit error rate or a signal-to-noise ratio.

The field strength of a transmitted communication signal has a natural dependence on the distance from a transmitter, the (conversational) base station, therefore provides information about the propagation behavior (propagation characteristic) of the transmitter and is particularly suitable for the modeling according to the invention using a Gaussian Process.

In a preferred development, calibration measurements are carried out at previously known calibration points at which the values of the physical property are measured at these points. A selection of the calibration points can be used by means of an optimal "design" method or lattice method, such as, for example, the method known from [2] for determining a hexagonal lattice.

Larger communication networks generally have several or a multiplicity of base stations, each of which emits a communication signal. Here it is useful to create a separate propagation model for each base station or for the communication signal of each base station.

The model or models created by the inventive procedure or in the case of several base stations can be the basis for numerous applications in communication networks.

The model or models can thus be used for planning and / or installation and / or commissioning and / or diagnosis of error states and / or quality assurance in the communication network.

The propagation model or propagation models created according to the invention can also be used for localization / position determination of at least one mobile communication device in the communication network, which is set up at least one mobile communication device for receiving the communication signal and / or for receiving the communication signals.

With such a localization or position determination of a mobile communication device in a communication network with a plurality of base stations, a likelihood of the physical property measured at the position to be determined, for example the field strength, can be determined in the Gaussian process model belonging to the respective base station. The position of the mobile device to be determined is determined by determining the point / position with a maximum likelihood. The key point in determining the position is "inverting" the Gaussian process model. When modeling the propagation behavior, the forward model is created as described above. When determining the position, the model is inverted such that - in the case of the inverted model - the position can be represented as a function of the likelihood of the physical property.

The invention or the inventive modeling of the propagation behavior of one or more base stations is particularly suitable for use in the environment of a digital, cellular mobile radio system, such as a GSM / UMTS network, and there, for example, for the localization of a GSM / UMTS telephone (mobile phone ).

When using the invention, only the data available to the mobile phone will be used, and there are no costly changes to be made to the GSM network or to mobile stations in the GSM network.

The invention is also suitable for use in the environment of further digital, cellular mobile radio systems, such as a WLAN, a network based on bluetooth or a DECT network, and there, for example, for localizing a DECT mobile telephone.

The invention is particularly suitable in the environment of unfavorable conditions for signal propagation, such as heavily noisy or reflected signals, shielded or switched-off base stations, interior scenarios. Under such conditions, physically exact models are not or only very difficult to create. An exemplary embodiment of the invention is shown in the figures and is explained in more detail below.

Show it

Figure 1 procedure for a position determination using a Gaussian process position determination system (GPPS) according to an embodiment;

FIGS. 2a and 2b are sketches which show a GPM (FIG. 2a) adapted with the original calibration data and the GMP (FIG. 2a) smoothed with the data obtainable from the GPM;

Figure 3 equations for determining derivatives according to the position t to be determined.

Exemplary embodiment: Gaussian process position determination system (GPPS) in a communication network (DECT network) with several base stations

Overview / procedure:

The position determination system (GPPS) described below for a mobile communication device in a communication network (here: for example a DECT network) based on Gaussian process models (GPM) is based on a calibration with calibration measurements in which signals from the base stations of the communication network and their field strengths are measured at previously known positions, ie calibration positions, in the communication network (FIGS. 1, 110). Gaussian process models (GPM) are adapted to the calibration measurements (see point 1) (Fig. 1, 100). The appropriate selection of kernel functions is important for the Gauss process models. Kernel functions of the master class are used here.

The adapted Gauss process models are then used to determine the position (see point 2). This is done by determining the likelihood of the Gaussian process models (Fig.l, 130) and their optimization (Fig.l, 140).

Furthermore, a procedure for the optimal selection or placement of calibration positions is described (see point 3) (Fig. 1, 100).

1. Determination of the Gaussian process models based on measured field strengths

The GPPS described is based on probability models or statistical models for the propagation characteristics of the communication signals or the signal strengths / field strengths of the individual base stations.

Models based on Gaussian models or a Gaussian process regression (GPR) are used as models, which are often used to solve nonlinear regression problems in Bayesian systems [12, 9].

The procedure or model creation is described below for the signal or the field strength of a (selected) base station. This procedure applies accordingly to all base stations.

This is a set of N calibration measurements in which the field strength γ ± (usually in dB) of the selected base station was measured at known positions XΪ, i = 1, ..., N in the communication network. GPR in general assume that targets are generated by an unknown function f: f 'via y _t = f {x _i ) + e _i with independent 2 Gaussian noise e ^ with a variance σ.

The basic model assumption here is that f (xi) is based on a Gaussian process. This means that the function values of the function f (xi) are distributed at the points x ^ Gauss, with an average value 0 and a covariance matrix K.

K itself is given by the kernel (covariance) function

The assumption of a Gaussian process (GP) based on only a few calibrations or calibration positions t requires a Gaussian distribution. Use the following relationships: v (t) = (k (t, x _x ), ..., k (t, x _N )) ^T (2.1)

y = (yi, ..., y _N ) ^τ (2.2)

Q = K + σ ² I (2.3)

the assumed mean value of the GPM results for some calibration positions t:

E (f (t) | D) = v (t) ^τ CT ¹ y (2.4)

with the variance: var (f (t) | ü) = k (t, t) - v (t) ^T Q ^_1 v (t). (2.5)

These relationships are described in introductory work on Gauss processes [11, 8, 12, 9].

2 The choice of the noise variance σ and the parameters θ of the kernel functions k is important. These are determined by maximizing the log likelihood of the training / calibration data according to the model parameters: σ ² , θ = arg max (- log det Q - y ^T Q J- (2.6) σ ² , θ

The matern kernel function

The appropriate choice of kernel (covariance) function is important when using GPM. Kernel functions describe the type of correlation between function values of two points.

A common choice is GP with squared kernels of the form: k (x, x ') = exp (- • w X X'

However, it is known from [10, 6] that this form of kernel functions is unnatural in the environment of stochastic processes if the example path is infinitely smooth, i.e. if the covariance function has an infinite number of derivatives in the origin.

The matern class of covariance / kernel functions are therefore used here [10], which allow a continuous parameterization of the smoothness of the example path by means of its parameter v.

Experimental examples have shown that GPM with kernel functions provides a realistic estimate of the assumed variance from Eq. (2.5) deliver.

The functional form of the Matern Kernel is:

k (x, x ') = M _v (z) = K _V (2Λ /), (2.7)

where T (v) is the gamma function, K _v (r) is the modified Bessel function of the second degree v and z = ∑ ^ _- ι WJ (XJ - x'jj with the input scale lengths WΪ. The parameters v determine the smoothness ("fractal dimension") of the example path and can be derived from Eq. (2.6) can be estimated.

Adaptation of the GP with Matern Kernel

For an efficient solution of Eq. (2.6) it is necessary to derive the matern kernel function Eq. (2.7) after all the parameters v, w.

Numerical gradients, the application of which is known, for example, from [10], require a large number of evaluations of the Bessel functions and therefore lead to enormous computational effort.

The analytical calculation of the derivatives used here is based on:

3κ _v (z) = -i (κ _v _! (Z) + K _{v +} i (z)), (2.9) dz 2

where Ψ (v) is the zero order polygamma function (called the psi function). Since no closed form of the derivatives of the Bessel function K _v (z) according to degrees v is known to date, this is approximated by

DK _v (z) = ^{Kv Z} ^ «e ^_1 (K _{v + e} (z) - K _v (z)). σv

The gradients of Eq. Determine (2.7) as follows: θM _Λv rZ. Yi = _v (z) [i + log (Vvl) - ψ (v) 3v

(κ _v _ ₁ (2Vvz) + K _{V +} I (2Λ / VZ)) + DK _V (2 / VZ). (2.10)

Based on Eq. the derivatives of Eq. (2.6) 2 are calculated according to the model parameters σ, v, w with standard matrix algebra.

The required relationships are described in introductory work on Gauss processes [11, 8, 12, 9].

Determination of the field strength model

The following shows how the GPM is created for the signal propagation of the selected base station.

It is assumed that the signal emitted by this base station has been measured for N calibration measurements at the calibration points x £, 1 = 1, ..., N.

Here points are to be considered in two dimensions; it should be noted that the procedure described should be applied accordingly to three-dimensional points.

Based on the calibration data D = {x ^, y} N for the i = l selected base station, the procedure is as follows:

1. An estimated position for the base station is obtained by selecting the three calibration points xi with the highest field strength values yi and from this the center of gravity is formed. The positions of base stations are usually not measured and known. In rare cases, however, such position information is available, which can then be used instead of the above estimate.

In order to obtain the mean value functions of the GPM, a linear model is adapted to the measured values, a logarithmic scale being selected as a function of the Euclidean distance from the base station.

The field strength values - if given in dB - can be used directly because they are already in logarithmic scale. The following propagation regularity is thereby modeled:

A signal of strength 1 at the base station is received with a strength exp (-d) at a distance d from the base station, the value of the mean function of the original measurement is subtracted.

From Eq. (2.6) you get the optimal model parameters, such as the variance σ, the matern smoothness parameters v and the input scaling lengths wi.

2a and 2b show an example of a GPM with the original calibration data and the smoothed data available from the GPM. The smoothed data reveal certain structures that were not apparent from the original measurement data, such as two corridors that extend to the left and right of the base station.

2. Position determination with GPM

Position determination based on the GPM described above is explained below.

Known from the calibration is assumed: C calibration measurements

- at the calibration points xi, i = 1, ..., C

- at B base stations

- With the received field strengths Ci _f j at location xi from base station j, j = 1, ..., B ci ^ j = 0, if the signal of base station j cannot be received at location xi.

With c the field strength vector is designated with all signals receivable at location xi.

In the test or application phase, a mobile user measures the field strengths of the base stations receivable at this location at an unknown location to be determined.

With s the vector of the field strengths receivable at the location to be determined is designated, with SJ components of the vector s as field strength s received from the base station Bj.

Position determination after the "nearest neighbor" (NNLoc)

In the NNLoc, the vector s of a position to be determined is compared with the calibration measurements ci, i = l, ..., i. Each "neighbor" or calibration point to the position to be determined is weighted depending on how well the measurement s matches the respective calibration measurement ci. These weights are taken into account for all receivable base stations and the current field strengths. The position t of the mobile user to be determined is estimated from the known and best matching calibration positions and the associated weights by means of an interpolation method.

Gaussian process position determination system (GPPS) The position determination using the GPPS is based on the GPM formation described above. For this purpose, using the GPM of the individual base stations, the likelihood of the field strengths of the base stations received at the location t to be determined is formed.

Using the calibration data i ^x i ^c i, jj> i ^e {l, ..., C}, each {l, ..., B}, the respective GMP Mj are formed for the individual base stations. Model Mj is based on data Dj from such calibration points i at which base station j can be received: Dj = ((xi, c _if j): c ^ j ≠ θ}.

In the application phase, only the models of those base stations that can be received at the location to be determined are taken into account here.

The likelihood of the field strengths receivable at the location t to be determined results from:

L (t) = π P (SJ | DJ, t). (3.1) j: sj ≠ 0

The assumed distribution of the GPM of the data Dj at point t is denoted by ps _j D _j , tj. This assumed distribution is a one-dimensional Gaussian distribution with the mean and the variance according to Eq. (2.4).

The point of the GPPS at which the shared likelihood of the received field strengths is at a maximum is now sought. The position t to be determined is obtained by optimizing or maximizing L (t) after t:

t = arg ax L (t) = arg max ∑ log p (s _j D _j t). (3.2) ttj; _S j ≠ O This optimization task is a forward solution of the gradients from L (t) to t.

Eq. (3.2) together with the gradient information can be solved by standard numerical optimization methods, such as "scaled conjugate gradient", and thereby the sought position t can be estimated. Equations for this are shown in FIG. 3.

Alternatively, L (t) can be solved by a grid method in which L (t) is calculated at grid points and the maximum is determined.

The GPPS can be clearly explained as follows: The reception of a high field strength of a signal from a specific base station indicates that the mobile user is approximately in a very small circle around this base station. The same applies to a very small field strength, which indicates that the user is in a very large circle or circle distance. The superimposition of these individual location estimates provides the estimated end position t.

Optimal choice of calibration points (Fig.l, 100)

A suitable position determination system, such as the GPPS, should manage with a minimum number of calibration points in order to keep the effort for the calibration as low as possible.

The calibration points should cover the area in question for localization as optimally as possible. Various approaches to this are known from the prior art [7]. Here the method for optimal selection of calibration points described in [7] is selected, which leads to a hexagonal grid of calibration points.

The following writings are cited in this document:

[1] Peyrard, F., Soutou, C, Mercier, JJ: Mobile Stations Localization in a WLAN, in: Proceedings of the 25th Annual IEEE Conference on Local Computer Networks (LCN'00), Tampa, Florida (2000) 136-142

[2] Hassan-Ali, M., Pahlavan, K.: A New Statistical Model for Site-Specific Indoor Radio Propagation Prediction Based on Geometry Optics and Geometry Probability. IEEE Transactions on Wireless Communications 1 (2002) 112-124

[3] Howard, A., Siddiqi, S., Sukhatme, G.S .: An Experimental Study of Localization Using Wireless Ethernet. In: Appears in: Proceedings of the 4th International Conference on Field and Service Robotics, Japan (2003)

[4] Bahl, P., Padmanabhan, V.N. : RADAR: An In-Building RF-based User Location and Tracking System. In: Proceedings of IEEE INFOCOM 2000. Volume 2., Tel Aviv, Israel (2000) 775-784

[5] Rauh, A., Briechle, K., Hanebeck, U.D., Bamberger, J., Hoffmann, C: Localization of DECT Mobile Phones Based on a New Nonlinear Filtering Technique. In: Proceedings of SPIE Vol. 5084, AeroSense Symposium, Orlando, Florida (2003)

[6] Gneiting, T. "Compactly supported correlation functions", Journal of Multivariate Analysis, 83 (2): 493-508, 2002

[7] Hamprecht, FA and Agrell, E. "Exploring a space of materials: Spatial sampling design and subset selection", in JN Cawse, ed., Experimental Design for Co bi- natorial and high throughput materials development. John Wiley & Sons, 2002

[8] MacKay, D.J. "Introduction to Gaussian processes", in CM. Bishop, ed., Neural Networks and machine Learning, vol. 168 of NATO Asi Series. Series F, Computer and Systems Sciences. Springer Verlag, 1998

[9] Rasmussen, C.E. "Evaluation of Gaussian Processes and other methods for non-linear regression", Ph.D. thesis, University of Toronto, 1996

[10] Stein, M. "Interpolation of Spatial Data. Some Theory for Kriging", Springer Verlag, 1999

[11] Williams, CK "Gaussian processes", in M. Arbib, ed., The Handbook of Brain Theory and Neural Networks. MIT Press, edn 2 ^nd., 2002

[12] Williams, C.K. and Rasmussen, C.E. "Gaussian processes for regression", in D. S. Touretzky, M.C. Mozer, and M.E. Hasselmo, eds. , Advances in Neural Information Processing Systems 8th MIT Press, 1996.

Claims

claims

1. A method for describing a propagation behavior of a communication signal transmitted by a base station in a communication network, in which a physical property of the communication signal associated with the respective selected position is measured at selected positions in the communication network, the physical property characterizing the propagation behavior of the communication signal , using the selected positions or using corresponding position or location information of the selected positions and the associated measured physical properties of the communication signal, a model for the propagation behavior is determined, which model describes the propagation behavior, characterized in that the modeling using a Gaussian process, which represents the measured physical property depending on the location information or the position information represents ("forward model").

2. The method according to claim 1, in which the model for the propagation behavior of the respective communication signal is determined in each case for a plurality of base stations, each with a communication signal in the communication network.

3. The method according to any one of the preceding claims, wherein the selected positions are determined using a design method and / or "design method".

4. The method according to any one of the preceding claims, wherein the communication network is a radio network, in particular a radio network based on wireless LAN or Bluetooth or GSM or DECT or UMTS, and / or the measured physical property of the communication signal is a propagation property of an electromagnetic field, in particular a Field strength, a phase, a transit time, a wave vector (wave vector), a bit error rate or signal-to-noise ratio.

5. The method according to any one of the preceding claims, in which the model or the models is or are used for planning and / or installation and / or commissioning and / or diagnosis of error states and / or quality assurance in the communication network.

6. The method as claimed in one of the preceding claims, in which the model or the models is used or are used for localization and / or position determination of at least one mobile communication device in the communication network, which is set up to receive the Communication signal and / or for receiving the communication signals.

7. The method according to the preceding claim, in which, for at least those Gaussian process models whose communication signals can be received at a position of the mobile communication device to be determined, a likelihood of the respective Gaussian process for the physical properties measured at the position to be determined Model is determined in which the position to be determined is determined by optimizing and / or maximizing the likelihood.

8. Computer program with program code means to carry out all steps according to claim 1 when the program is executed on a computer.

9. Computer program with program code means according to claim 8, which are stored on a computer-readable data carrier.

10. Computer program product with program code means stored on a machine-readable carrier to carry out all steps according to claim 1 when the program is executed on a computer.