CA2442950A1 - Method and system for indoor geolocation using an impulse response fingerprinting technique - Google Patents

Abstract

The location of people, mobile terminals and equipment in a mine is of great current interest. In an indoor environment such as a mine (may be generalised to any indoor environment), the multipath caused by reflection from walls, ceiling, floor and objects, and the non-line of sight (NLOS) due to the blockage of the shortest direct path between the transmitter and receiver are the main sources of range measurement errors. The unreliable measurements of location metrics such as RSS, AOA and TOA/TDOA yield to the deterioration of the positioning performance.
Hence, alternatives to the traditional parametric geolocation techniques have to be considered.

In this paper, we present a novel method for mobile station location using wideband channel measurements' results applied to an artificial neural network (ANN).
The proposed system (WBNN-Locate) learns off-line the location 'signatures' from the extracted location dependent features of the measured channel impulse responses data (any relevant location dependent parameter may also be used) for LOS and NLOS situations. It then matches on-line the observation received from a mobile station against the learned set of 'signatures' to accurately locate its position.

With this approach, multipath becomes useful since it improves location 'signatures' characterisation, and the presence of a line of sight (LOS) is no longer essential, unlike in other geolocation techniques. The proposed location system has the characteristic to operate with a single fixed station and may be applied to a three-dimensional environment.

The error median distance and the 90th percentile of the error distance found with this technique were 1.2 m and 2 m, respectively.

Description

TITLE OF THE INVENTION
METHOD AND SYSTEM FOR INDOOR GEOLOCATION USING AN IMPULSE
RESPONSE FINGERPRINTING TECHNIQUE
FIELD OF THE INVENTION
The present invention relates to a method and system 'For indoor geolocation using an impulse response fingerprinting technique. In particular, the present invention relates to a method and system for locating a mobile station using a fingerprinting technique based on wideband channel measurements' results in conjunction with an artificial neural network.
BACKGROUND OF THE INVENTION
One problem of growing importance in indoor environments is the location of people, mobile terminals and equipment. Indoor radio channels suffer from extremely serious multipath and non-line of sight (NLOS) conditions that: have to be modelled and analysed to enable the design of radio equipment for geolocation applications.
Since telecommunication and geolocation applications have different objectives (see for example K. Pahlavan, P. Krishnamurthy and J.E3eneat, "Wideband Radio Propagation Modeling for Indoor Geolocation ~4pplications", IEEE
Communications magazine, April 1998, which is incorporated herein by reference) existing radio channel models are not appropriate, and different models and techniques have to be applied to provide adequate location accuracy.
In traditional wireless geolocation applications, the basic function of the location system is to gather parametric information (received signal strengths RSS, angles of arrival AOA, times of arrival TOA or time differences of arrival TDOA) about the position of a mobile station (MS) and process that information to form a location estimate (see, for example, K.Pahlavan, X.Li, et al., "An Overviev~ of Wireless indoor Geolocafion Techniques and Systems", Proceeding of M~fIICN 2000, Paris, France, May 2000 which is incorporated herein by reference).

' 2 In indoor environments where conditions of signal propagation are severe (multipath, NLOS), the traditional parametric geolocation techniques (RSS, AOA, TOA, TDOA) or their combinations (TDOA with AOA or RSS) fail to provide adequate location accuracy. For these techniques, all the paths used for triangulation have a LOS to ensure an acceptable accuracy, a condition that is not met in an indoor environment.
Geolocation based on the received signals' fingerprint (fingerprinting technique) performs better in such an environment (see, for example, C.Nerguizian, C.Despins and S.Affes, "A Framework for lndonr Geolocation using an Intelligent System", 3'd IEEE Workshop on WLANs, Boston, September 2001 [hereinafter Nerguizian et al.]
which is incorporated herein by reference).
Most existing indoor geofocation applications use a network-based system architecture in which base stations (BS) or access points (AP) extract location dependent parameters or metrics (RSS, AOA, TOA or TDOA) from the received radio signals transmitted by the mobile station (MS) and relay the information to a control station (CS). Then the position of the user (MS) is estimated and displayed at the CS.
In the parametric geolocation technique, the concept of the line of position (LOP), with at least two observations, is used in order to obtain a two-dimensional position fix.
Most of the geolocation system architectures and techniques developed for cellular systems are applicable for indoor geolocation systems with special considerations (see, for example, K.Pahlavan, X. Li and J.P.Makela, "Indoor Geolocation Science and Technolog,~', IEEE Communications magazine, F=ebruary 2002 [hereinafter Pahlavan et al.] which is incorporated herein by reference).
For the parametric geolocation technique using the received signal strength (RSS), a mathematical model describing the path loss {PL) attenuation with distance (d) is required. If the power transmitted by a mobile station is H;nown, measurement of the received signal strength provides a distance estimate between the mobile terminal (MS) and the fixed station (BS or AP). The estimated distance will determine a circle, centred at the receiver, on which the mobile transmitter lies. Three RSS
measurements will provide the two-dimensional location ~of the mobile.
For the parametric geolocation technique using the angle of arrival (AOA), antenna arrays at fixed stations are required for the direction finding of the signal of interest (see, for example, J.C.Liberty and T.S.Rappaport, Smarl Antennas f~r ~1/ireless Communications:IS-95 and Third Generati~n ODMA Applfcatior7s, Prentice Hali PTR, 1999 which is incorporated herein by reference). Two fixed stations measure the arrival angles of the signal that is transmitted from a mok~ile user. Sased on the AOA
estimates and the known positions of the fixed ;~fations, simple geometric relationships are used to form the two-dimensional location estimate.
In the parametric geolocation technique using the tinne of arrival (TOA), if the estimate of the propagation time (TOA) of the signal transmitted by a mobile station and received by a fixed station is known, the estimated distance between the mobile user and the fixed station can be determined providing geometrically a circle, centred at the receiver, on which the mobile transmitter lies. Three TOA measurements will provide the two-dimensional location of the mobile.
In the parametric geolocation technique using the time difference of arrival (TDOA), the locus of the estimated constant TDOA of a pair of fixed stations (receivers) defines a hyperbola, with foci at the receivers, on which the mobile transmitter lies.
Three TDOA measurements will provide the two-dimensional location of the mobile.
The accuracy of these traditional parametric geolocation techniques (RSS, AOA, TOA and TDOA) or their combinations (TDOA with AOA or RSS) depends on several factors, including the:
~ indoor environment (multipath, non-line of sight and local shadowing);
~ path loss model used in the RSS technique to estirnate the ranges;
~ plane wave/near field propagation model, the angular resolution of the antenna array and the direction of arrival's estimation algorithm used in the AOA technique;

~ time or time difference of arrival's estimation algorithm used in TOArfDOA
techniques;
~ number of fixed stations involved in the intersection of the lines of position LOP), ~ geographical location of the mobile station relative to the fixed stations;
and ~ positioning algorithm used to estimate the user's location.
Tree main measurement errors, introduced during the extraction of the location dependent metrics, are due to the indoor environment. 'The lines of position (LOP), due to these errors, do not intersect at a point resulting in large estimation errors.
Moreover, each of these techniques has its limitations.
Although the RSS indoor geolocation is easy to realise, variations in the F2SS
can be as great as 30 dB over distances on the order of a half wa~,elength due to small scale fading and shadowing effects.
Due to the limited angular resolution of the antenna array, large location errors occur in the AOA indoor geolocation because the scatterers are located around both the transmitter and the receiver. Moreover, a mobile unit situated between two FSs, placed face to face, cannot be localised.
For the time-based (TOA and TDOA) indoor geolocation techniques, LOS
propagation conditions are necessary to achieve high Ic~cation accuracy.
Moreover, the TOA technique requires strict time synchronisation k>etween the transmitter and the receivers, whereas only time synchronisation among ail the receivers is needed for the TDOA technique.
In general, the time-based TDC)A technique is the most popular one, and may be combined with other techniques to improve the location accuracy (http:l/www.comm-nav.com/e911.htm).
The location accuracy reported by companies, which use the time-based indoor geolocation technique with proprietary infrastructures (i.e. 3D-iD), is in the range of 3 meters. For the parametric indoor geolocation, Kalman filtering and fusion of multiple metrics may be used to improve positioning performance (see, for example, E. D. Kaplan, Understanding GPS: Principles and Applicati~ns, Artech House, which is incorporated herein by reference). However, in a non-line of sight indoor environment, alternatives to the parametric geolocation techniques have to be considered.
To improve the accuracy of the user's location in a harsh environment (multipath and non-line of sight), the effect of multipath has to be miticlated or multipath has to be used as constructive information.
A radio frequency signal transmitted from a given geographical IViS location has a distinct pattern by the time it reaches a receiver. Interference caused by natural or man-made objects causes the signal to break up into a number of different paths (multipath). Hence, each location produces a uniquE; 'signature' pattern called fingerprint (see, for example, httpa/www.uswcorp.com).
The process of geolocation based on the received signals' fingerprint (location fingerprinting or premeasurement-based location pattern recognition technique) is composed of two phases, a phase of data collection (off-line phase) and a phase of locating a user in real-time (real-time phase, see, for example, s~lergcaizian et al.). The first phase consists of recording a set of fingerprints (in a database) as a function of the user's location, covering the entire zone of interest. During' the second phase, a fingerprint is measured by a receiver and compared with the recorded fingerprints of the database. A pattern matching algorithm (positioning algorithm) is then used to identify the closest recorded fingerprint to the measured one and hence to infer the corresponding user's location (Figure 1).
To constitute a "signature" pattern or a fingerprint, several types of information (referring to Figure 2) can be used such as, received signal strength (RSS), received angular power profile (APP) and received power delay profile (PDP) or channel impulse response (CIP) On the other hand, several types of pattern-matching algorithms may be used in the fingerprinting technique, which have the objective to give the position of the mobile station with the lowest location error. The most popular algorithms are based on the:
~ nearest neighbours) in signal space (location estimate defined as the lowest Euclidean, Box-Cox or statistical metric in signal space);
cross-correlation between signal vectors (location estimate defined as the highest correlation coefficient between signal vectors); and ~ use of an artificial neural network (location estimate defined as the closest ANN's output to the training set's outputs).
It has to be noted that the accuracy of the method is primarily a function of the reproducibility and uniqueness of the estimated set of fingerprint information.
Reproducibility means, the achieving of almost the samE; estimated set of fingerprint information in one location, for different observation times. lJniqueness means that the set of fingerprint information in one location is relatively different from the one in another location (no aliasing in the signature patterns).
Several geolocation systems, using fingerprinting technique, have been deployed in outdoor and indoor environments. The main differences between these systems are the types of fingerprint information and the types of pattern matching algorithms. An overview of these systems is given next.
RADAR (designed by IVticrosoft Corporation see http://www.microsoft.com) is an RF
network-based system for locating and tracking users inside buildings (see, for example, P.Eahl and V.N.Padmanabhan, "RA~AR : An ln-Building R'F-based User L~cation and Tracking Systerrr"', Proceedings of IEEE INF~C~Nl 2000, Tel Aviv, Israel, IVlarch 2000 which is incorporated herein by reference). It uses RSS
(narrowband measurements) fingerprint information gathered at multiple receiver locations to determine the user's co-ordinates. The system, operating with technology, has a minimum of three access points (fixE:d stations) and covers the entire zone of interest.

The pattern-matching positioning algorithm consists of the nearest neighbours) in signal space. The minimum Euclidean distance (in signal space), between the observed RSS measurements and the recorded set of RSS measurements, computed at a fixed set of locations, gives the estimated user's location.
~CM (concept designed by \1T'T Information Technology/, see http:llwww.vtt.fi) is an RF handset-based system for locating and tracking users in a, metropolitan outdoor environment (see, for example, H.Laitinen, T.Nordstrc~m and J.Lahteenmaki, "Dafabase Correlation Method for GSM Location", IIEEE ~fehicular Technology Conference, Rhodes, Greece, May 2001 which is incorporated herein by reference).
The mobile terminal that needs to be located performs measurements of signal strength (narrowband measurements) received from the serving cell and six strongest neighbours. The gathered information is then sent to a location server, where the location of the user is estimated and this estimate is sent back to the application server. Other types of signal information (cell i~, propagation time delay) can also be used within the network (see, for example, I-I.Laitinen, T.Nordstr~m and J.Lahteenmaki, "Location of GSM Terminals using a G~atabase of Signal Strength Measurements", URSI XXV National Convention on Radio Science, Helsinki, Finland, September 2000 which is incorporated herein by reference). The system, operating with the GSM Cellular technology, has several fixed stations and covers the entire zone of interest.
A simple correlation algorithm is used to estimate the user's location. A best match search, between the observed RSS measurements by the mobile station and the recorded set of RSS measurements in the location server, is computed at a fixed set of locations and the MS's location is estimated.
It has to be noted that, since 17CM is a handset-I~ased location system, its implementation involves some software modifications of the mobile terminal in order to enable the retrieval of received signal characteristics.
In the framework of the VVILMA project (see http:ll .wilmaproject.org) RSS
fingerprinting technique is used to locate users in a building with a iNLAN
infrastructure (see, for example, R.Sattiti, T.L.Nhat and A.!/iliani, "Location-Aware Computing: A Neural Network Model for Determining Location in VIlireless LANs", Technical Report # DIT-02-0083, University of Trento, 'rrento, Italy, February which is incorporated herein by reference). The pattern-matching algorithm involved is an artificial neural network, which consists of a multi-layer perceptron (MLP) architecture with 3, 8 and 2 neurones in the input, hidden and output layers respectively to achieve the generalisation needed when confronted with new data, not present in the training set.
RadioCamera (designed by US Wireless Corporation, see http:llwww.uswcorp.com) is an RF network-based system for locating and tracking users in a metropolitan outdoor environment. It uses multipath angular power profile (APP) information gathered at one receiver to locate the user's co-ordinates. The system, operating with cellular technology, has one-antenna array per cell ('fixed station) and covers the entire zone of interest. The pattern-matching algorithm, used to estimate the user's location, consists of the nearest neighbours) in signal space. The minimum statistical (Kullback-Liebler) distance (in signal space), between the observed APP
measurements and the recorded set of APP measurements, computed at a fixed set of locations, gives the estimated user's location (see, for example, U.S.
Patent 6,112,095 for Signature Matching for Location Determination in Wireless Communication Systems which is incorporated herein by reference).
DCM, operating with UMTS technology and using CIR as fingerprint information, is the second RF handset-based system conceived by VTT Information Technology for locating and tracking users in a metropolitan outdoor environment (see, for example, S.Ahonen, J.Lahteenmaki, H.Laitinen and S.Horsmar~heimo, '6Usage of Mobile Location Techniques for UMTS Network Planning in Urban Env~ironmenY', IST
Mobile and Wireless Telecommunications Summit 2002, Thesaaloniki, Greece, June 2002 which is incorporated herein by reference). It has several fixed stations and covers the entire zone of interest. To form the database, a set of fingerprints is modeled by computing the radio channel impulse responses (CIR) with a ray-tracing tool.
The magnitudes of these impulse responses or the power delay profiles (PDP) are calculated (after setting a threshold value in order to reduce contributions of noise power and interference from other codes) from each fixed station to each receiving point corresponding to the user's location. The mobile terminal that needs to be located performs measurements of channel's impulse responses (wideband measurements).
The magnitude of the impulse response from the strongest fixed station is correlated with the content of its database (pattern-matching algorithm) at the location server.
The receiving point with the highest correlation coefficient is taken to represent the co-ordinates of the mobile station.
Referring to T.Nypan, K.Gade and T.Maseng, "Location using Estimated Impulse Responses in a IVlobile Communication System", 4t" Nordic Signal Processing Symposium (NORSIG 2001), Trondheim, Norway, October 2001 Chereinafter iilypan et al.) which is incorporated herein by reference, a measured channel's impulse responses are used for database collection and for location estimation algorithm.
The system performs an outdoor geolocation using GSM and UMTS technologies.
The pattern-matching algorithm involved is based on the nearest neighbour in signal space. The minimum Box-Cox distance (see, for example, l-.Nypan, K.Gade and O.Hallingstad, "Vehicle Positioning by Database Comparison using the fox-Cox iVletric and Kalman Filtering", IEEE i/ehicular Technology Conference, Vol.
55, No. 4, Birmingham, USA, February 2002 which is incorporated herein by reference) between the observed CIR measurement and the CIR .measurements contained in the database gives the estimated user's location.
The accuracy and coverage of the geolocation systerns, using the fingerprinting technique, depend on the resolution and the size of the database. Calibration measurement and database maintenance are essential in the operation of these systems. Moreover, the search methodology, involved in the pattern-matching algorithm should be efficient to minimise the time needed for the localisation.
Systems, using RSS fingerprinting technique (RADAR and WILMA for indoor, DCM
for outdoor), require the involvement of several fixed stcitions to compute the user's location. Moreover, RSS yield a great amount of variation (due 'to fading effects) for a specific location implying a reproducibility concern, which may degrade the location accuracy.

The system, using APP fingerprinting technique (RadioCamera for outdoor), requires the use of an antenna array with high angular resolution for indoor geolocation since the scatterers are around both the transmitter and the receiver.

Systems, using CIR or PDP fingerprinting technique (DCiVI and Nypan et al. for outdoor), have the advantage to be reproducible and it respects the uniqueness property, especially when the localisation is done ~on a continuous basis (user's tracking).
As a conclusion, it seems that the signature based on t:he impulse response of the channel gives the best location accuracy for an indoor ge:olocation, see, for example, Nerguizian et al.. However, its implantation involves the use of a wideband receiver.
On the other hand, the pattern-matching algorithm used ire RADAR, DCM and RadioCamera systems may show a lack of generalization (an algorithm that gives an incorrect output for an unseen input), a lack of robustness against noise and interference, a lack of pattern match in some situations (i.e. the Euclidean distance can be minimized without having the match of the two patterns) and a long search time needed for the localization (done during the real-time phase) especially when the size of the environment or the database is large. H~:nce, the use of an artificial neural network (ANN), as the pattern-matching or positioning algorithm, is essential to the enhancement of the geolocation system.
As a measure of performance, the median resolution of the location estimation for indoor and outdoor geolocation systems, using fingerprinting techniques, is reported to be in the range of 2 to 3 meters and 20 to 150 meters respectively.
Referring to Table 1, the different geolocation techniques are presented in order to compare their features, strengths and weaknesses.

SUMML~RY OF THE INVENTION
In order to address the above and other drawbacks the present invention implements a fingerprinting technique using the channel's impulse response information as a novel approach for geolocation in mines, which has a bE:fiter reproducibility property, compared to the other two fingerprint information (RSS and angular power profile).
The use of an artificial neural network as a pattern-matching algorithm for the proposed system is a new approach that has the advantage to give a robust response with a generalisation property (the location fingerprint does not have to be in the fingerprint database). Moreover, since the training of the ANN is off-line, there is no convergence and stability problems that some control (real-time) applications encounter. Finally, the transposition of the system from two to three dimensions is easy (addition of a third neuron in the ANN's output layer corresponding to the z position of the user).
~n the other hand, the fingerprinting technique needs the digital map of the environment and is not well suited for dynamic areas. Preliminary measurements in mine showed that the influence of low human activity is negligible on the wideband measurement results at the specific frequency of operation. hlowever, a heavy machinery or vehicle may considerably change the properties of the channel, obliging an update of the database's information (a new training of the neural network). This channel variation issue can be addressed, for example, by using a master neural network. After detecting the changes in the channel's properties, the system identifies the specific situation (channel state) via a scanning process and activates the trained neural network corresponding to this specific situation.
Finally, the method may also be applicable to many other indoor applications (shopping centres, campuses, office buildings). In addition, some advanced simulation programs may be used to generate impulse responses as a function of user location (for the training set of data of the neural network) instead of deriving these impulse responses from wideband measurements. This approach will reduce the database generation time for the proposed geolocation system and will act in favour of the proposed system's implementation.

BRIEF DESCRIPTION OF THE DRAWINGS
Figure 1 illustrates the process of geolocation using received signals' fingerprint, a) off-line phase, b) real-time phase;
Figure 2 provides an overview of the types of fingerprint information;
Figure 3 is a map of the underground gallery with wideband measurement positions;
Figure 4 is a digital photograph showing a part of the underground gallery, showing that the walls have some roughness, the floor is not flat and that it contains some plaques of water;
Figure 5 illustrates the operation of the proposed systern, a) learning phase (off-line phase), b) recalling phase (real-time phase);
Figure 6 illustrates the proposed pattern-matching ANN;
Figure 7 provides estimated and true position locations in x and y, with inputs corresponding to the training set of data defined by the number of positions of the mobile station;
Figure 8 provides location errors in x, y and Euclidean distance (c~, with inputs corresponding to the training set of data defined by the number of positions of the mobile station;
Figure 9 provides cumulative distribution functions (CDI=s) of location errors in x, y and Euclidean distance (~, with inputs corresponding to the training set of data defined by the number of positions of the mobile station;
Figure 10 provides estimated and true position locations in x and y, with inputs corresponding to the untrained set of data defined by the number of positions of the mobile station;

Figure 11 provides location errors in x, y and Euclidean distance (ark, with inputs corresponding to the untrained set of data defined by the number of positions of the mobile station;
Figure 12 provides cumulative distribution functions (CC~Fs) of location errors in x, y and Euclidean distance (c~, with inputs corresponding to the untrained set of data defined by the number of positions of the mobile station; and Figure 13 provides cumulative distribution functions (C~Fs) of location errors in Euclidean distance (cn, with inputs corresponding to the untrained set of data and with three positioning algorithms (Euclidean metric, Box-Cox metric and artificial neural network).
DETAILED DESCRIPTION OF AN ILLUSTRATIVE EMBODIMIIIIENT
The proposed geolocation system is an RF network-based system for locating and tracking users in an indoor mime. It uses a channel's multipath power delay profile or impulse response (obtained from wideband measurements) information (see, for example, Nerguizian et al.), gathered at one receiver, to locate the user's co-ordinates (uplink network based approach). The system, can be operated with different radio access technologies, has one fixed station (a second one can be used as redundancy) and covers the entire zone of interest.
Measurements were conducted in an underground gallery of an abandoned gold mine. Located at a 40 m underground level, the gallery stretches over a length of 75 meters with a width and height both of approximately 5 meters. Figure 3 illustrates the plan of the gallery with all its under adjacent galleries.
A central frequency of 2.4 GFIz has been used throughout the measurements in order to have a compatibility with WLAN systems, which may be used far data, voice and video communications as well as for radio location purposes. ~ue to the curvature of the gallery, the existence of a non- line of sight NLc~S case is visible.

The digital photograph, given in figure 4, shows a part of the underground gallery. It can be seen from the photo that the walls have some roughness, the floor is not flat and it contains some plaques of water.
The complex impulse response of the channel (wideband measurements) has been obtained using the frequency channel sounding technique. During the measurements, the vector network analyser has performed the transmission and the reception of the RF signal. The inverse Fourier transform (IFT) has been applied to the measured complex transfer function of the channel iin order to obtain its impulse response.
The chosen frequency band was centred at 2.4 GHz with a span of 200 MHz corresponding to a theoretical time resolution of 5 nanoseconds (in practice, due to the use of windowing, the time resolution is estimated to be around 8 nanoseconds).
The sweep time of the network analyser has been decreased to validate the quasi-static assumption of the channel. Each sweep consisted of 201 complex samples spaced of 1 MHz from each other giving an unambiguous delay time of 1 microsecond, which was far beyond the sum of the maximum excess delay for the studied mining environment and the propagation delay of the cable.
The wideband experimental procedures (see, for exampie, C.Nerguiziian, C.Despins, S.Aifes and M.Djadel, "Narrornrband and I~llideband Radio Channel Measc~rements in an Underground Mine with Narrov~ lAeins at z.4 sJHz", paper submitted to the IEEE
Transactions on Wireless Communications, September 2003 which is incorporated herein by reference) were defined to characterise the relevant parameters of the channel and to utilise these parameters in order to perform a radio location of workers in the underground gallery. Hence, for thne radio-location purposes (fingerprinting technique), the experimental procedures given in this article are different from those encountered in previous works.
The network analyser and the PC were stationed with the receiving antenna and the other receiver components at the predefined referential. The equipment was tested for flat response in the measurement band and calibrated in the presence of the RF
cable. The transmitting antenna and the other transmitter components were moved to different locations within the underground gallery by varying their position of 0.5 metres in width (6 positions having a distance of 0.5 meter for the gallery width of 5 metres) and 1.0 metres in length (70 positions having a distance of 1 metre for the gallery length of 70 metres). Some other extra intermediate positions have also been 5 used for the LOS and NLOS cases giving a total of 490 location measurements (Fig.
3). During the measurements, the transmitting and receiving antennas were both mounted on carts at a height of 1.9 meters (simulating, for example, an antenna placed on the helmet of a miner).
10 The complex transfer function was obtained at all 490 measurement locations. For each location, a temporal average has been performed on a set of ten (10) measurements of different observation times (a local spatial average may also be performed). The time domain magnitude of the complex impulse response has been obtained, from the measured samples of the frequency domain response, using the 15 inverse fast Fourier transform (IFFT).
From the magnitude of the complex impulse response, the mean excess delay (am), the RMS delay spread (zns), the maximum excess delay (zmax), the total received power (P), the number of multipath components (ll~, the power of the first path (P~) and the arrival time (delay) of the first path (z?) of the channel have been computed at alf 490 measurement locations by using a predefined threshold of 20 dB for the multipath noise floor. The first five (5) parameters characterized the time-spread nature of the indoor channel and the last two (2) parameters gave an emphasis about the difference between LOS and NLOS situations. Then, these seven relevant parameters (instead of the magnitude of the impulse response), defining the location-dependent features, have been used as the input for the artificial neural network (positioning algorithm).
The choice of these parameters was based on the necessity to have a good reflection of the user's location 'signature' (good location-dependent features of the channel impulse measurements) without having excessive ANN input vector size to avoid the over-fitting of the ANN during its training phase.

A trained artificial neural network can perform complex tasks, such as classification, optimisation, control and function approximation. The pattern-matching algorithm of the proposed geolocation system can be viewed as a function approximation problem (non-linear regression) consisting of a non-linear mapping from a set of input variables containing information about the relevant parameters of the channel's impulse response (zm, z~ms, Z'max~ P, N, l'T, z~) onto a two output variables representing the two dimensional location of the mobile station (x, y).
The feed-f~rward artificial neural networks that c:an be used as function approximation are of two types, Multi-Layer herceptron (MLP) networks and Radial Basis Function (RBF) networks. Either type of the two networks can approximate any nonlinear mapping to an arbitrary degree of precision provided the right network complexity is selected (see, for example, S.Haykin, Neural Network, a C~mprehensive Foundation, MacMillan, 1094 which is incorporated herein by reference). Specific learning algorithm is associated for each type of the two networks, which has the role of adjusting the internal weights and biases of the network based on the minimization of an error function, and defines the training of the network.
The MLP networks enable to reach globally any non-linear continuous function due to the sigmoid basis functions present in the network, which are nonzero over an infinitely large region of the input space. Accordingly, they are capable of doing a generalisation in regions where n~ training data are available (generalisation property). On the other hand, the RBF networks can reach the given non-linear continuous function only locally because the basis functions involved cover only small, localised regions. However, the design of an RBf= network is easier, and the learning is faster compared to the MLP network.
A generalised regression neural network (GRNN), which is an RBF type network with a slightly different output layer, and an MLP type network have been tested for the proposed geolocation system. The MLP network showed a higher location error, compared to the GRNN, during the memorisation of the data set. However, it showed a lower location error during the generalisation phase of the network.
Since the generalisation property of the system was of greater importance, the MLP
type network has been chosen for the pattern-matching algorithm used in the proposed geolocation system.
The ANN, used in the proposed system, consisted of two phases, a supervised learning phase (training of the network) and a recalling (testing or functional) phase.
During the off-line phase, the MLP network was trained to form a set of fingerprints as a function of user's location and acted as a function approximator (non-linear regression). Each fingerprint was applied to the input of the network and corresponded to the channel's relevant parameters (mean excess delay, RMS
delay spread, maximum excess delay, tote! received power, number of multipath components, power and arrival time of the first path) extracted from the impulse response data received by the fixed station. This phase, where the weights and biases are adjusted in iterations to minimise the network performance function, is equivalent to the formation of the database (recording of the set of fingerprints as a function of user's location) seen with other fingerprinting systems.
During the real-time phase, the mentioned relevant parameters from a specific mobile station (obtained from the measured channel's impulse response) were applied to the input of the artificial neural network (acting as a pattern-matching algorithm). The output of the ANN gave the estimated value of the user's location (see Figure 5).
It has to be noted that when the size of an ANN is increased, the number of the internal parameters (weights and biases) increase inducing more local and global minima in the error surface, and making the finding of a global or a nearly global minimum, by the local minimisation algorithm, easier (see, for example, Y.Shang and B.IN.iNah, "Global Optimization for Neural Netnrork Training", COMPUTER, pp 45-56, March 1996 which is incorporated herein by reference).
However, when the size of the ANN is large or equivalently, wl-ien the number of the weights and biases is large for the selected training set, an over-fitting problem occurs. This means that although the error on the training set is driven to a very small value, when new data is presented to the network the error is typically large.

This is a case where the network has memorised (for example, using a look up table) the training set, but it has not learned to generalise to new situations (see Fi.Demuth and M. Beale, Neural Network Toolbox for use with Matiab (User's Guide), The MathWorks Inc., 1998 [hereinafter Matlab User's GuideJ).
Hence, to have a network with a good generalisation property, the size of the network should be chosen just large enough to provide an adequate fit. A way of improving the generalisation property is the use of a regularisation method (modification of the performance function by adding to the mean sum of squares of the network errors a term that consists of the mean of the sum of squares of the network weights and biases). Moreover, to have an automated regularisation (determination of the optimal regularisation parameters in an automated fashion), i3ayesian regularisation in combination with Levenberg-Marquardt algorithm may be used (see, for example, F.~.Foresee and M.T.~lagan, "Gaussian-Newton Approximation to Bayesian Regularization", Proceedings of the 1997 International Joint Conference on Neural Networks, pp 1930-1935, 1997 which is incorporated herein by reference).
Hence, property trained MLP networks tend to give reasonable answers when presented with inputs that they have never seen (generalisation property, see Matlab User's Guide). Typically, a new input will lead to an output similar to the correct output (target) for input vectors used in training that are similar to the new input being presented (no need to train the network on all possible inputloutput pairs).
In order to have a good generalisation property, the MLP architecture used consisted of seven (7) inputs corresponding to the channel's relevant parameters, one hidden layer and an output layer with two (2) neurons corresponding to (x, y) location of the user (see Figure 6). A differentiable tan-sigmoid type of transfer function has been associated for neurons in the hidden Payers and a linear one for the output layer.
A simulation was carried out using the Neural Network Toolbox of Matlab (see Matlab User's Guide) with the results showing that ten neurons corresponding to the hidden layer are adequate to do the required regression. Special attention has been given to the over-fitting problem to respect the generalisation property (the trainbr.m function of Matlab has been used, which applies the Ba~yesian regularisation with the Levenberg-Marquardt algorithm). Hence, the designed network was robust to perturbations at its input (i.e. errors in the measurement data) and was able to do a generalisation rather than a memorisation (giving the right location for an unseen and non trained input). Moreover, since MLP has an inherent low pass filter property, it may remove the high frequency components present ire the location error signal.
4Nith seven (7) inputs, two (2) output neurons and ten (10) hidden neurons, the total adjustable number of weights and biases was equal to 102 ([7*10]+[10*2] for the weights, +[10]+[2] for the biases). This is almost four times smaller than the total number of the training set, which is equal to 367 and corresponds to the 75%
of the measured wideband data. As a rule of thumb, to have a good generalisation property and to avoid the memorisation of the network, the number of the patterns in the training set has to be around four times the number of the internal adjustable ANN
parameters. Hence the use of ten (10) hidden neurons was justified.
It has to be noted that, before training, the inputs and the targets have been scaled or normalised using the premnmx.m function of Matlab so that they fall in the range [_ 1, +1]. The outputs of a trained network, having scaled inputs and targets, will fall in the range of [-1, +1]. To convert these outputs back into the same units, which were used for the original targets, the postmnmx.m function of MatBab has been used. The normalisation of the inputs and targets is essential for the performance improvement of the ANN optimisation process.
Moreover, typical data sets often contain redundant information, or measured values (computed relevant channel's impulse response parameters), which are highly correlated. It is useful in this situation to reduce the dimension of the input vectors by transforming the full set of training examples into a compressed set that contains only essential information. The prepca.m function.of Matlab has been used to do this operation based on the principal component analysis which performs three effects: it renders orthogonal the components of the input vectors (the vectors become uncorrelated with each other), it orders the resulting orthogonal components (principal components) so that those with the largest variation come first, and it eliminates those components which contribute the least to the variation in the data set (see Maflab User's Guide).
Using the Neural Network Toolbox of Matlab, the proposed neural network 5 architecture has been designed. For the learning phase, the seven relevant parameters of the channel's impulse response and the measured true mobile station positions have been used as the input and as the target of the ANN
respectiveiy.
From the 490 measured data, 367 patterns have been employed to train the network.
For the recalling phase, as a first step, the same 367 patterns have been applied to the pattern-matching neural network to obtain the location of the mobile station (validation of the memorisation property). The estimated and true position locations, the location errors as well as their cumulative density functions (CDF) have been computed for analysis purposes. The plots of the corresponding position locations, location errors and CDFs of location errors are given in figures 7, 8 and 9.
It has to be noted that the localisation error has been calculated as the difference between the exact position of the user and the winning position estimate given by the localisation algorithm, and hence represents the f~MS position location error.
Moreover, CDF of the location error has been used as the performance measure of the system.
For the training set of data, it can be seen (see Figure 8) that the location error in x varies between -3.2 meters and 3.7 meters, the location error in y varies between -1.9 meters and 1.8 meters and the maximum error in Euclidean distance, between the estimated and the true positions, is equal to 3.9 meters.
Moreover, it can be seen, from Figure 8, that a distance location accuracy of meters is found for 90~/a of the trained patterns. An improvement of the location accuracy is feasible at the cost of the generalisation property.
As a second step, the remaining 123 non-trained patterns have been applied to the network to verify the generalisation property of the proposed geolocation system.

The estimated and true position locations, the location errors as well as their cumulative density functions (CDFs) have been computed and plotted (Figures 10, 11 and 12).
For the untrained set of data, it can be seen (see Figure 11 ) that the location error in x varies between -3.7 meters and 5.8 meters, the location error in y varies between -2 meters and 2,9 meters and the maximum error in Euclidean distance, between the estimated and the true positions, is equal to 5.8 meters.
llAoreover, the accuracy of the position estimate depends on the resolution of the map, which in turn depends on the distance threshold used in the map building process. After localisation has been achieved, the theoretical error between the actual and estimated position (localisation error) should therefore vary between zero and the distance threshold. Since the size of the grid used in the indoor wideband measurements was 0.5 meter wide and 1 meter long, the geolocation accuracy that one may expect with the proposed fingerprinting technique, should be between 0 and 1.12 meters (distance threshold) in terms of the Euclidean distance.
It can be seen, from Figure 12, that the location accuracy corresponding to the distance threshold is achieved for 40% of all the untrained patterns. ~ience, the proposed fingerprinting technique used for the geolocation of the studied mine, gave an accurate mobile station location. The results showed that a distance location accuracy of 2 meters has been found for 90% and 80% of the trained and untrained patterns respectively. This location accuracy, which may be improved at the cost of the generalisation property, is much smaller than the one found in the literature for indoor geolocation using fingerprinting techniques.
In order to see the advantage of using an ANN in an indoor geofocation system using the fingerprinting technique, three different pattern-matching algorithms (Nearest neighbour minimising the Euclidean distance, nearest neighbour minimising the Box-Cox metric, see for example R.V.D. Heiden and F.C.A. Groen, "The Box-Cox Metric for Nearest Neighbour Classificafion Improv~emeni", Pattern Recognition Society, Vol. 30, No. 2, 1997 which is incorporated herein by reference, and artificial neural network) has been used with the same empirical data set (untrained patterns).

The three curves of Figure 13 give the CDFs of location errors in Euclidean distance (d} for the involved three pattern-matching algorithms.
Only the CDF of location errors using the AIVN with the trained patterns is added on the figure since the associated curves for the two other algorithms are not necessary (their location errors tend to zero due to the memorisation of the two algorithms).
It can be seen that, for the generalisation property (the most important property for the fingerprinting technique), the artificial neural netvrork works the best giving an error less than 2 meters for 80%, for all the untrained patterns, compared to 68%
and 72% for the Euclidean and Cox-Box metrics respectively.
In indoor environments, the largest excess delay corresponding to the detectable multipath component may be on the order of 500 nanoseconds (see, for example, H.Hashemi, "Impulse Response Modelling of Indoor Radio propagation Channels", IEEE Journal on Selected Areas in Communications, Vol. 11, No. 7, September 1993 which is incorporated herein by reference). ~n the other hand, to characterise the discrete-time impulse response model or equivalently the multipath power delay profile, a reasonable bin (small time interval) resolution is needed. The value chosen for a bin depends on the indoor environment of interest.
The resolution of the measured channel impulse response depends on the system bandwidth. The effect of a limited bandwidth is that multiple reflections may end up in the same time bin on the delay axis, implying the vector combination of the reflections and yielding a resultant signal large or small depending on the distribution of phases among the component waves. This will give rise to a reproducibility problem of the measured channel impulse responses.
For efficient operation of the proposed system, it would be advantageous to be able to resolve all multipath components to obtain the power' delay profile or the impulse response as a function of the user's location. Hence the radio access technology used for an effective implementation of the system should satisfy this requirement (resolution of the multipath differential delays in the nanoseconds range).
Several existing technologies, with some modifications, are good candidates for such an application. The most promising technologies found iri practice are, the mobile radio system, the impulse radio system and the WLAN system. In this section, an overview about these three technologies will be given with their advantages and disadvantages. The choice of the wideband receiver technology and its implementation depend on the specific application and is still an open area of research.
The popular standards defined in digital mobile radio systems are the CDMA2000, the GSM and the IJMTS. The CDMA uses a code division multiple access (CDMA) with a direct sequence (DSSS) spread spectrum (spreading codes to separate signals). The chip duration of a CDMA-DSSS is about 815 nanoseconds (a pseudo noise sequence chip rate of 1.2288 Mbps). A CDMA RAKE receiver is able to resolve the closely spaced multipath components with delays greater than chip duration apart. Hence the choice of this technology without any modification appears inadequate for the proposed system. A super-resolution method for multipath delay estimation (see, for example, G.Morrison and M.Fattouche, "Super-Resolution Modeling of the Indoor Radio Propagation ChanneP', IEEE Transactions on Vehicular Technology, Vol. 47, No. 2, May 1998 and F.Bouchereau, D.Brady and C.Lanzl, "Multipath Delay Estimation using a Super-resolution PIlI-Correlation Method", IEEE
Transactions on Signal Processing, Vol. 49, No. 5, May 2001 which are incorporated herein by reference) or an over sampling method (limited by the hardware's clock rate) can be used to improve the time resolution during the process of power delay profile collection.
The GSM uses a time division multiple access (TDMA) with a modulation data rate of about 270.833 Kbps (time resolution of 3700 nanoseconds). Hence the choice of this technology without any modification would also appear inadequate for the proposed system.
The WCDMA (Wide CDMA) accepted for the UMTS standard utilizes a chip rate of 3.86 Mcps and higher (time resolution of about 260 nanoseconds and lower).
When the standard for receivers in WCDMA will be specified, a WCDMA RAKE receiver (measuring the delay and the signet strength of all fingers, see Pahlavan et al.) with a high-resolution algorithm as discussed above, may be an acceptable choice for the proposed geolocation system.
The Ultra Wide Band (UWB) signalling used in an impulse radio system is a viable candidate for indoor geolocation applications since it has fine time resolution properties (see, for example, M.Z.Win and R.A.Scholtz, "On f?obustness of Ultra Wide Band Signals in ~lultipath Environments", IEEE Communication Letters, February 1998 which is incorporated herein by reference). It is characterised by very low power transmission and by wide bandwidths (greater than a gigahertz). UWB
signalling uses pulses of very short duration (on the order of a nanosecond) with a certain repetition period (on the order of 100 nanoseconds). The data modulation of UWB is accomplished by a pulse position modulation at the rate of many pulses per data bit. Pseudo-random time hopping (time hopping codes) is used to eliminate catastrophic collisions in multiple accessing. Properly designed UWB receivers are capable of resolving multipath components with differential delays of a fraction of a nanosecond (see, for example, K.Siwiak, P.Withington and S.Phelan, "Ulfra Wide Band Radio: The emergence of an important New Technology", IEEE Vehicular Technology Conference, Rhodes, Greece, May 2001 which is incorporated herein by reference). However, the received UWB signal acquisition is more rigorous than a CDMA signal. In the near future, when the Federal Communications Commission (FCC) in the United States accepts the use of the UWB technology without any restriction, it can be a good choice for the proposed geolocation system. In a Wireless Local Area Network (WLAfV), access points (AP) represent the fixed stations and PC cards installed in terminals represent the mobile stations.
The standard IEEE 802.11 defines the Medium Access Control (MAC) sublayer, MAG
management protocols and services, and the Physical (PHY) layers (see B.O'Nara and A Petrick, The IEEE 802. ? ? Handbook, IEEE Press, 1999).
The Medium Access Control of IEEE 802.11 supplies the functionality required to provide a reliable delivery mechanism for user data over noisy and unreliable wireless media.
The Physical layer of IEEE 802.11 is the interface between the MAC and wireless media, which transmits and receives data frames over a shared wireless media.

The standard has four types of physical layers, IEEE 802.11-DSSS (Direct Sequence Spread Spectrum), IEEE 802.'11b-HR/DSSS (High Rate DSSS), IEEE
802.11-FHSS (Frequency Hopping Spread Spectrum) and IEEE 802.11x-OFDM
5 (Orthogonal Frequency Division Multiplexing).
The IEEE 802.11-DSSS works at 2.4 GHz with a rate of 1 to 2 Mbps (approximate resolution time of 500 nanoseconds).
10 The IEEE 802.11-FHSS works at 2.4 GHz with a rate of 2 Mbps (approximate resolution time of 500 nanoseconds).
The IEEE 802.11b-HRIDSSS works at 2.4 GHz with a rate of 11 Mbps (approximate resolution time of 90 nanoseconds).
The IEEE 802.11a-OFDM works at 5.8 GHz with a rate of 54 Mbps (a potential of acceptable resolution time may be possible) and its zone of coverage is smaller than the HRIDSSS.
Another physical layer IEEE 802.11g-OFDM will be available soon. It will work at 2.4 GHz with a rate of 54 Mbps. The IEEE 802.11x, fEEE 802.11b and IEEE 802.11g WLAN systems may be acceptable choices for the proposed geolocation system.
Moreover, since the frequency of operation of the IEEE 802.11b and 802.11g systems is lower than the one found in the former system, the reproducibility and uniqueness of the estimated set of fingerprint information is more easily obtained in the studied mining environment where rough wall surfaces induce a scattering of the incident signals.
It has to be noted that if the geolocation coverage area exceeds the range of the access point (AP), several APs may be used, each performing a geolocation and covering a part of the zone of interest.
In summary, the choice of the radio access implementation technology depends on several parameters such as, time resolution, power consumption, distance coverage, data rate, users capacity, signal to noise ratio, tolerance to interference, availability of the product and modification of the available products (hardware and/or software).
For the proposed geolocation system, the time resolution is a key factor.
Moreover, an important issue in the proposed geolocation system is the multiple-access factor (localisation of several mobile stations at the same moment).
Each of the implementation technologies described above has a certain type of multiple access technique (Time Division Multiple Access - TDMA, Code Division Multiple Access -CDMA and Carrier Sense Multiple Access - CSMA). CDMA and TDMA
systems are typically easier to implement than CSM.A (found in WLAN systems).
However, the cost of a WLAN system is lower than sy stems operated with CDMA
or TDMA.
A possible alternative may be the use of Air5 WL.AN systern, which operates at a frequency around 5 GHz and implements collision avoidance through its TDMA-based synchronous MAC subiayer (see, for example, P. Fowler, "5 GHz Goes the Qistance for Home Networdsing", IEEE Microwave Magazine, Vol. 3, No. 3, September 2002 which is incorporated herein by reference). Another alternative will be to design a set of signals from different mobile stations in such a manner that the access point can distinguish the signals from different mobile units.
Hence, the multiple-access issue is another important factor to consider for the proposed geolocation system.
It has to be noted that in practice, it is convenient to have a system able to support both telecommunication and geofocation applications. Hence, impacts on economic and technical issues should be considered during the chosen technology's implementation.
In general, it is the belief of the authors that, the WCDMA system may be focused for outdoar applications while the WLAN system may be used for indoor applications.
Although the present invention has been described hereinabove by way of an illustrative embodiment thereof, this embodiment can be~ modified at will, within the scope of the present invention, without departing from the spirit and nature of the subject of the present invention.

Claims (2)

1. A method for determining the location of a transmitter transmitting in a zone of interest, comprising the steps of:
(a) receiving a signal from the transmitter;
(b) deriving a fingerprint from said signal's multipath power delay profile or channel impulse response;
(c) comparing said derived fingerprint to a set of fingerprints stored in a database, each of said stored fingerprints corresponding to a location in the zone of interest;
(d) matching said derived fingerprint to the stored fingerprint using an artificial neural network to determine a closest match to said derived fingerprint, thereby inferring the location of the transmitter.

2. A system for determining the location of a transmitter in a zone of interest, the system comprising:
a data bank comprising a plurality of stored fingerprints, each of said stored fingerprints corresponding to a location in the zone of interest;
a receiver for receiving a signal from the transmitter;
a processor for deriving a fingerprint from said signal's multipath power delay profile or channel impulse response, said processor comprising a pattern matching algorithm using an artificial neural network for identifying the closest of said stored fingerprints to said derived fingerprint and thereby inferring the transmitter location.