Analysis of Scattering in Mobile Radio Channels Based on Clustered Multipath Estimates

Transcription

1 Int J Wireless Inf Networks (2008) 15: DOI /s Analysis of Scattering in Mobile Radio Channels Based on Clustered Multipath Estimates Maurice R. J. A. E. Kwakkernaat Æ Matti H. A. J. Herben Published online: 24 September 2008 Ó Springer Science+Business Media, LLC 2008 Abstract This paper presents the results of angle and delay measurements in physically nonstationary radio channels obtained in an outdoor urban environment. The multidimensional estimation data are obtained using a recently developed 3-D high-resolution channel sounder. The estimation results are compared with results obtained from a 3-D deterministic propagation prediction tool. For a better analysis, a hierarchical clustering method is presented that can separate and group the multidimensional estimation data into clusters. Measurements performed at a fixed position as well as along a trajectory are used to characterize the angular dispersion in both azimuth and elevation. The angular dispersion in terms of the rms cluster angular spread in both elevation and azimuth of the different clusters is analyzed over space and time and related to its physical scattering sources. Compared to the measurements, a large number of multipath clusters are missing in the predictions. Furthermore, it is observed from the measurements that different objects cause different angular spread values in azimuth and elevation. The results can be very helpful for the identification, improvement and calibration of deterministic propagation prediction models. Keywords Channel sounder Direction of arrival estimation Land mobile radio Mobile communications Radiowave propagation M. R. J. A. E. Kwakkernaat (&) M. H. A. J. Herben Eindhoven University of Technology, 5600 MB Eindhoven, The Netherlands m.r.j.a.e.kwakkernaat@tue.nl 1 Introduction The efficient utilization of smart-antenna and MIMO based systems in cellular radio networks demands accurate knowledge of the propagation effects. More importantly, this knowledge should be available in propagation prediction models for the planning and optimization of next generation cellular networks [1]. Current ray-optical based deterministic propagation prediction models use principles of geometrical optics and the uniform geometrical theory of diffraction, which generate specular-like contributions [2]. In reality, however, the energy is not directed into the specular direction only, but it is also scattered in other directions close to the specular direction in a stochastic way [3]. These scattered contributions contain a significant amount of energy and the performance of smart-antenna and MIMO-based systems in cellular radio networks is, to a large extent, dependent on the angular distribution of multipath components. It is important to characterize these effects using accurate measurements to obtain new knowledge regarding propagation. Previous measurement-based studies were mainly limited to the azimuth plane and only a few were performed under mobile conditions [4, 5]. Recently, a 3-D highresolution channel sounder was developed based on a 3-D antenna array in combination with a multi-dimensional version of the Unitary ESPRIT algorithm [6, 7]. The measurement system is capable of accurately measuring the angle-delay characteristics in both azimuth and elevation under mobile conditions. The results from this measurement system are of high importance in the field of propagation research, but additional processing is required to be able to correctly interpret the multidimensional estimation results. Since the multipath estimates obtained in physically nonstationary radio channels tend to appear in

2 108 Int J Wireless Inf Networks (2008) 15: elongated cluster-like structures within the multi-dimensional space, clustering of the multi-dimensional data can significantly improve the data representation and support further data analysis. Since clustering by visual inspection of the data becomes difficult due to the size and the multidimensional nature of the data, automatic clustering becomes necessary. Once clusters are obtained, they can be related to physical objects that interact with the propagating waves. The results are important for the understanding of the propagation mechanisms and the modelling of scattering, as well as the verification and calibration of deterministic propagation prediction models. To validate and improve the results from deterministic propagation prediction models the clustered high-resolution angle-of-arrival (AoA) estimates presented here are compared to the results obtained from a deterministic propagation prediction tool [8]. Furthermore, the angular dispersion in both the azimuth and elevation domain is determined from the clustered measurement data under static as well as mobile conditions to characterize the small-scale and large-scale angular fluctuations caused by movements of the environment [9]. The results are important for the understanding of the propagation mechanisms and the modeling of scattering, as well as the verification and calibration of deterministic propagation prediction models. This paper is organized as follows. In Sect. 2 the measurement system and the measurement scenario are presented. Next, in Sect. 3 the clustering method used to cluster the estimation results is discussed. Section 4 describes the simulations and in Sect. 5 the results of the measurements, the clustering method and the simulations are presented and compared. Finally, Sect. 6 gives conclusions. 2 Measurement Setup The data used in this paper is obtained from outdoor experiments performed at the campus area of the Technische Universiteit Eindhoven (TU/e) in Eindhoven, The Netherlands. Measurements are performed using a recently developed 3-D high-resolution channel sounder [6], the receiver of which is shown in Fig. 1. The system is based on a 3-D tilted-cross switched antenna array that consists of 31 monopole antennas and uses an improved version of the Unitary ESPRIT algorithm to obtain signal parameter estimates at mobile conditions [7, 10]. The measurement system operates at a center frequency of 2.25 GHz in a band of 100 MHz and is able to produce complex impulse responses (CIRs) with a range of 5.1 ls in the delay domain and a multipath power sensitivity ratio of 35 db, which is defined as the difference in db between the 3-D Switched antenna array Omni-directional video camera Power generator Receiver, Data collection, GPS, Gyroscopes, Accelerometers, Compass Fig. 1 The receiving part of the 3-D high-resolution channel sounding system that consists of a measurement vehicle, the 3-D switched antenna array, omni-directional video camera, receiver and auxiliary equipment maximum and minimum power value that can still be detected in the power-delay profile obtained from a backto-back measurement. The resolution in the delay domain is 20 ns and samples are available at 10 ns instances. Using multiple snapshots, high-resolution angle-of-arrival (AoA) estimates can be obtained under mobile conditions at each delay instant in both azimuth and elevation with a typical resolution better than 5 and an accuracy of 0.1. A single measurement snapshot is taken every 6.5 ls and sets of 10 consecutive snapshots are used as input for the Unitary ESPRIT algorithm. As a result, the angle and delay characteristics of the channel are estimated every 65 ls. In each of the measurement scenarios presented herein a total number of 1,000 snapshot-sets of 10 snapshots each are used, resulting in a measurement duration of 65 s. Prior to the measurement campaign, a back-to-back measurement is performed to calibrate the system. The measurement scenario is presented in Fig. 2, where Tx (1) and (2) mark the transmitter positions. In the case of the moving receiver Rx (1) marks the beginning of the first measurement trajectory in combination with Tx (1) and Rx (2a) marks the beginning of the second measurement trajectory in combination with Tx (2). Rx (2b) marks the position of the receiver in the case of the static receiver setup. The measurement trajectories are represented by the red dotted lines and the numbers along the trajectory correspond to the snapshot-set numbers. The yellow lines represent the building database information that is used in the simulations. The environment is characterized by several high buildings and scattered vegetation. The transmitting antenna (Tx) consisted of a 8-dBi waveguide horn antenna with an azimuthal half-power-beam-width of 55. In scenario (1), the Tx antenna was positioned outside, at 57 m above ground level and pointed southwards at a tilting

3 Int J Wireless Inf Networks (2008) 15: Fig. 2 Layout of the measurement site. The red dotted lines represent the measurement trajectories of scenario 1 and 2. The values along the trajectory correspond to the snapshot-set numbers. Copyright: Google Earth, angle of 20 downward. In scenario (2), the antenna was positioned behind glass, inside an elevated walkway made of steel at 5 m above ground level and pointed eastwards at a tilting angle of 20 downward. The receiving antenna (Rx) consisted of the 3-D antenna array [10], mounted on top of a measurement vehicle at 3.5 m above ground level, as shown in Fig. 1, and was moved at a constant speed of about 13.5 km/h along a trajectory of approximately 245 m. This effectively means that a single angle-delay estimate of the channel is available every 0.25 m corresponding to wavelengths. During the measurement campaign omni-directional video data was captured from the receiver perspective using a omnidirectional video camera that was mounted directly underneath the antenna array. The azimuth angles are in accordance with the orientation of the measurement vehicle, where 0 corresponds to the back of the vehicle and 180/-180 corresponds to the front. Due to the obstruction of the vehicle below and the antenna array above, the video data are effectively limited to an elevation range of -35 to Clustering Method The automatic clustering of measurement channel data obtained in physically nonstationary radio channels was recently addressed by the authors of [11, 12]. In their work a method was proposed that is based on the sequential clustering of windowed multipath estimates using a K-means clustering algorithm and the tracking of cluster centroids in the consecutive data windows. Although the algorithm is fast and useful for analyzing large measurement records, some difficulties may occur. Because K-means is a heuristic, hill-climbing algorithm, the algorithm is guaranteed to converge on a local, but not necessarily global optimum. This means that the clusters found may not be optimal and the choices of the initial clusters affect the quality of the results. Although this problem can be mitigated by doing multiple restarts, the shapes of the clusters may vary between multiple runs. The pairing of the clusters between two consecutive windows can be difficult because it requires the size of the sliding window to be correctly selected to prevent merging of multiple clusters to one, or too many clusters to be created. Instead of clustering the measurement data using sliding windows we propose in this paper to use a hierarchical whole clustering method to cluster the measurement data. Although this method is not fast, the data pool is clustered as a whole, which guarantees the conversion on a global optimum and prevents any difficulties with pairing clusters from consecutive windows. The problem of scaling data with different dimensions is also circumvented. The data to be clustered consist out of N number of estimated multipath components (MPCs) each of which are described by a position in a four-dimensional space, i.e. time-delay (s), AoA in azimuth (/), AoA in elevation (h) and measurement snapshot-set number (k). Additionally, the estimated power (p) is also available for each estimate. The task of the clustering algorithm is to group estimates that have similar angles and delays that slowly evolve over time, i.e. that are close together in the four dimensional space. Since the acquired data are obtained with a moving receiver, multipath components will evolve over time and tend to appear in strings or elongated clusters within the four-dimensional space. Single linkage clustering, also called nearest neighbour method, is an agglomerative hierarchical clustering method and known to be well suited to detect chains or elongated data structures [13]. The single linkage algorithm used here is given in Fig. 3. Here, the angle ðj ij Þ between MPC i and MPC j is defined as the angle between the two three-dimensional vectors pointing in the direction of the estimates i and j, defined by their azimuth and elevation angles, in the following manner 80 < J ij ¼ cos : cos / i cos h i sin / i cos h i sin h i 1 0 cos / j cos h j sin / j cos h j sin h j 19 = A ; : ð1þ The distances in delay and snapshot-sets are obtained as Ds ij ¼js i s j j and Dk ij ¼jk i k j j; respectively. Furthermore, t J ; t s and t k are the thresholds for each domain specified by the user. The values for t J and t s are set according to the resolution of the measurement system, i.e. t J ¼ 5 and t s = 20 ns. The value for t k was set to 10 snapshot-sets, which corresponds to a distance of 18 wavelengths and was found to produce satisfying clustering

4 110 Int J Wireless Inf Networks (2008) 15: of ray-launching and uses reflection, penetration and diffraction as the most dominant effects. For the building walls a thickness of 0.5 m was assumed and a relative permittivity e r ¼ 3 0:4j: In accordance with the measurements, the noise floor was set to -100 dbm. Except for the minimum power level, no other limitations to the number of interactions were set. 5 Results and Comparison This section presents the results from the measurements and the simulations and the application of the clustering method presented in Sect. 3. Unreliable estimates were removed by using the reliability indicator presented in [6, 10]. From the remaining dataset MPCs with very low power (\-100 dbm, in accordance with the simulations) are removed since they do not contribute significantly. Note that values for the intra-cluster delay-spread are not determined, because the delay resolution of 10 ns is too low to give an accurate estimate. 5.1 Measurements Compared to Predictions Fig. 3 Single linkage clustering algorithm results. The algorithm will proceed in clustering all data into clusters according to the thresholds set for each domain, meanwhile reducing the data set that needs to be searched for nearest neighbours. 4 Simulations To obtain propagation prediction results in the area used in the measurements, a commercial deterministic propagation prediction tool was used. Simulations were performed using information of a 3-D building database, the top view of which is shown by yellow lines in Fig. 2. A trajectory of 245 m, similar to the measurement trajectory of scenario (1) was defined and a step size of 1.8 wavelengths was used in accordance to the measurements. The height and orientation of the receiver and transmitter were correctly set and the actual antenna patterns were used. Values for the signal parameters and cable losses were in accordance with the measurements. The simulation is based on the principle The results in Fig. 4a present a set of N = 20,288 MPC estimates in the angle-delay domain obtained by applying 3-D Unitary ESPRIT to the measurement data. The colors indicate the absolute received power levels of the individual MPCs in dbm. Although the estimates in the three domains are interrelated, it is difficult to present them in a single multi-dimensional figure. The three domains are shown here independently and as a result the connection between the different domains is not explicitly visible. The evolution of several strong MPCs, the strong-est being the line-of-sight (LOS) component, can clearly be observed in all domains. The change in delay and angle of the LOS component is in agreement with the scenario, i.e. the time-delay decreases, the elevation angle increases and the azimuth angle changes with respect to the orientation of the receiver as the receiver moves towards the transmitter. Close to snapshot-set k = 400 there is a short and sudden drop in signal level of the LOS component. This is caused by passing underneath an elevated walkway made of steel. After taking the turn (k = 700) the elevation angle increases up to 50 and the level of the direct component drops again since LOS is obstructed. Beside several strong MPCs, many weaker and scattered components are also visible throughout the measurement and can be related to trees and other objects using the captured video data. An increase of these scattered components is observed when the receiver is surrounded by vegetation (starting close to k = 200). In Fig. 4b the results of applying the clustering algorithm to the measurement data are presented. The colors

5 Int J Wireless Inf Networks (2008) 15: Fig. 4 Angle and delay results of (a) the estimated MPCs from the measurements, (b) the clustered MPCs, (c) the MPCs obtained from a ray-launching simulation tool and (d) the clustered simulation results indicate the different MPC clusters (MPCCs) and connect them across the different domains. The 30 largest clusters in terms of total cluster size and total cluster power are presented. The results show that the clustering method can successfully distinguish many different clusters of different size and life-time within the data. These clusters can be related directly to the physical environment by comparing them with the captured omnidirectional video data. For example, between the interval 100 k 500 a series of 5 clusters appear and disappear in sequence (,,,, ). They are arc shaped in both angular domains and have a linear increasing delay. These clusters are caused by reflections from the vertical elevated sections of building Wh in the center of Fig. 2. A scattering effect caused by through-building propagation of the same building is visible between 0 k 300 ( ). This cluster has a large delay and angular spread in azimuth with the center of the cluster moving from / = 160 to 50. Although the lifetimes of the all the MPCCs are different, they generally tend to evolve rather slowly along the trajectory. Figure 5 presents an example of a single video frame of snapshot-set k = 280 combined with (a) the angular estimation results and (b) the results of clustering. In Fig. 5a the size and the color of the markers represent the signal intensity. In Fig. 5b the color of the markers represents the different clusters in accordance to Fig. 4b. This

6 112 Int J Wireless Inf Networks (2008) 15: Fig. 5 Results of (a) MPC angles-of-arrival and (b) clusters of MPCs superimposed on omnidirectional video data at snapshotset k = 280. The colours of the markers correspond to the clusters in Fig. 4b ELEVATION ANGLE (DEG) (a) AZIMUTH ANGLE (DEG) POWER (dbm) ELEVATION ANGLE (DEG) (b) AZIMUTH ANGLE (DEG) representation directly relates the estimation results and the clusters to the actual environment. When all the frames of the entire trajectory are combined an illustrative video is created in which the MPCs and MPCCs can be observed and tracked. The results in Fig. 4c are obtained from a deterministic propagation prediction tool in which the measurement scenario was accurately modelled and simulated. The results show that there is a good agreement between measurements and simulations for the LOS component in both power as well as angle-delay domain. Some of the other weaker components seem to have lower predicted power, are less spread in the angle-delay domain or seem to be missing completely in the simulation results. Possible reasons for the mismatch between the measurements and the predictions are an inaccurate or incomplete building database, incorrect building material properties ðe r Þ; incorrect modelling of irregular surfaces, vegetation and small objects. The effects of errors in the predictions, especially in the angular domain, can have a large impact on the design of multi-antenna systems for cellular networks and addresses the need for improving propagation predictions. The results from clustered high-resolution AoA measurements, as presented here, can be very helpful for the identification, improvement and calibration of deterministic propagation prediction models. The propagation predictions can possibly be improved by incorporating statistical information about the angle-delay spreads of MPCCs. The intra-cluster properties from clusters obtained from measurements can then be used to calibrate simulation parameters related to corresponding clusters from the simulations. Furthermore, the fact that the different MPCCs tend to evolve rather slowly along the trajectory may allow for interpolation techniques to reduce the number of positions used in the simulation and as a result speed up the simulation time. 5.2 Cluster Angular Spread The results in Fig. 6a present the thirty strongest MPC clusters that contribute to 98% of the total power available in the MPCs using Tx (2) and Rx (2a). At the beginning of the trajectory building obstructions cause only little power to be received. After snapshot set k = 300, changes in the channel caused by moving the receiver along the trajectory are immediately visible in the data. The change of several strong MPC clusters, the strongest being LOS component, ( ) can clearly be observed. The change in delay and angle of the LOS component is in agreement with the scenario, i.e. the time-delay first decreases and increases again after taking the turn, and the azimuth angle changes with respect to the orientation of the receiver as the receiver takes the turn and moves away from the transmitter. Beside several strong MPC clusters, many weaker and scattered components, caused by trees and other objects, are also visible throughout the measurement. At snapshot set k = 650 the receiver is closest to the position of static receiver setup Rx (2b), presented in the next section. The clustered estimation data provide high-resolution azimuth as well as elevation information, which makes it possible to analyze the cluster angular spread in both domains, i.e. rms cluster azimuth spread (CAS) and rms cluster elevation spread (CES). Reliable estimates of the distribution functions for the angular spreads can not be obtained directly from the estimation results, because the MPC estimates are based on the specular wave model. The mean AoA in azimuth for MPCC c is estimated as / ¼ P I i¼1 P N ðiþ ðiþ n¼1 ^/ n PðiÞ / n PI P N ð2þ ðiþ i¼1 n¼1 PðiÞ / n and the rms CAS is calculated as

7 Int J Wireless Inf Networks (2008) 15: Fig. 6 Angle and delay results of the MPCCs for (a), receiver setup Rx (2a) and (b), receiver setup Rx (2b). In (a) the thirty most dominant clusters are presented, each in their own colour. The size of the markers corresponds to the power of the MPCs, where the largest marker corresponds to the highest received power. In (b) the eight most dominant clusters, each in their own colour, are presented. Note that in (a) the elevation values for each cluster are offset by multiples of 40 degrees vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P I P N ðiþ ðiþ u i¼1 n¼1 j^/ n ^r / ¼ /j2 P ðiþ t / n : ð3þ PI i¼1 P N ðiþ n¼1 PðiÞ / n Here, I represents the total number of delay-bins and N the total number of estimated MPCs that belong to the corresponding MPCC. The estimated azimuth angle of the nth ^/ ðiþ n MPC in the ith delay-bin is represented by : The value for P ðiþ ðiþ / n corresponds to the power of ^/ n : Values for the CES are determined in a similar way. In order to characterise the dispersive effects of the individual clusters, mean values for CAS and CES of different clusters that can be related to specific buildings and objects, are determined and summarized in Table 1. The CAS and CES values presented here serve merely as an indication and show that they can be characterised using measurements obtained while moving trough an urban environment. Although the values for the angular spread are small, differences between clusters are observed that can be related to their origin. More details on the cluster origin and the effect of the movement of the transmitter on the angular spread are discussed in the next section, where measurements using a static receiver are presented. 5.3 Effects of Receiver Movement The results in Fig. 6b present the eight strongest MPC clusters in the angle-delay domain that contribute to 97% of the total power available in the MPCs in the case of Rx Table 1 Mean values for CAS and CES for different objects determined from the clustered MPCs Mean CAS (deg) Receiver setup: Rx (2a) LOS Building Wh Building PT Building WL Building LG Tree trunks Receiver setup: Rx (2b) LOS Building Wh Building PT Building HG Moving truck Moving truck Moving car Moving bicycle Mean CES (deg) (2b). Each MPC cluster is represented by its own color. The size of the marker indicates the absolute received power levels of the individual MPCs in dbm. Note that in Fig. 6b the elevation angles are offset by multiples of 40 degrees for each cluster for a better representation. Although the measurement is performed under static conditions channel fluctuations are visible in the

8 114 Int J Wireless Inf Networks (2008) 15: ELEVATION ANGLE (DEG) HG PT LG Truck 2 WL Wh AZIMUTH ANGLE (DEG) Fig. 7 Angles of arrival of the MPCC superimposed on omnidirectional video data for receiver setup Rx (2b) at time instance 45 s. The colours of the markers correspond to the clusters in Fig. 6b, and the measurements. The line-of-sight (LOS) component, ( ), visible near / ¼ 180= 180; has the shortest delay and the highest received power. By using the captured video data, shown in Fig. 7, the other MPCCs can be related to the physical environment to find their physical scattering sources. The MPC clusters result from interactions with building Wh, (, ), building PT ( ) and building HG ( ). The two clusters that are visible as slopes in the time-delay domain (, ), are caused by trucks moving towards and away from the transmitter. Although the effects of trucks were most dominant, cars ( ), and even bicycles, traffic signs and lamp posts can be identified as physical scattering sources using the captured video data. In all clusters, rapid variations of the spread, caused by small subtle movements of the environment are visible as well as larger variations caused by moving objects such as cars and trucks. Mean values for CAS and CES of different buildings and objects in the measurements are summarized in Table 1. Apart from the larger fluctuations that are caused by the changing environment, small variations are caused by the variance of the estimator due to the limited accuracy and resolution. These effects are minimised by averaging the CAS and CES values over a large number of snapshots and makes the identification of the mean CAS and CES values acceptable. It can be observed that the mean CAS and CES values can be different within a cluster and vary between different clusters. The LOS component, which ideally should have a very low angular spread, has a mean CAS of 0.3 and a mean CES of 0.7. Although quite small, the observed angular spread can be explained by the fact that the Tx antenna is positioned inside a elevated walkway with a steel floor and ceiling which causes additional reflections, more dominantly in elevation. The second cluster, related to building Wh ( ), shows a lower CES but an increased CAS. The spread is likely caused by the irregularities of the building, consisting of vertical steel tubes. The cluster related to building PT, ( ), shows only a small angular spread in both domains. This can be explained by the fact that the face of the building is almost plane metal at the reflection point, which creates a more specular reflection. received power is represented by the size of the markers, where the largest marker corresponds to the highest received power The cluster related to building HG ( ), which can also be identified due to its larger propagation delay, has much more spread in both elevation and azimuth. This can be explained by the irregular surface. In general, the mean values of the CAS and CES are similar in both the mobile and static measurement. It was found that the values for the CAS range between 0.3 and 1.7, and for the CES between 0.3 and 2.4. Although larger values are observed in other measurements, the mean values presented here are generally smaller than the results in [5], where similar analyses in the azimuth domain were performed. Here, values for the azimuth spread vary between 2 and Summary and Conclusions In this paper, the results of angle and delay measurements in physically nonstationary radio channels obtained with a 3-D high-resolution channel sounder are presented. A method to cluster these multi-dimensional estimation data is presented and it was shown that it can successfully separate clusters in multi-dimensional MPC estimation data obtained in physically nonstationary radio channels. From the clusters the scattering effects of specific objects can be isolated and the angular dispersion of these objects in azimuth as well as in elevation can be analysed. The results of the measurements and the clustered data are compared with the results obtained from a 3-D deterministic propagation prediction tool and it was found that in the simulations a significant amount of MPCs was missing. Furthermore, the spreads in delay and angle are much larger in the measurements. It was also observed that the MPCCs tend to evolve rather slow along the trajectory, which may allow for interpolation techniques to reduce simulation time. It was shown that in a static scenario, subtle changes in the channel cause rapid variations of the angular spread, whereas the movement of larger objects cause more gentle variations. The amount of angular spread in azimuth or elevation can be quite different and values for the CAS range between 0.3 and 1.7, and for the CES between 0.3 and 2.4.

9 Int J Wireless Inf Networks (2008) 15: The effects of errors in the predictions, especially in the angular domain, can have a large impact on the design of multi-antenna systems for cellular networks and addresses the need for improving propagation predictions. The results from clustered high-resolution AoA measurements, as presented here, can be very helpful for the identification, improvement and calibration of deterministic propagation prediction models. The propagation predictions can possibly be improved by incorporating statistical information about the angle-delay spreads of MPCCs. The intra-cluster properties from clusters obtained from measurements can then be used to calibrate simulation parameters related to corresponding clusters from the simulations. Furthermore, the fact that the different MPCCs tend to evolve rather slowly along the trajectory may allow for interpolation techniques to reduce the number of positions used in the simulation and as a result speed up the simulation time. Both of the afore mentioned subjects require additional research and are currently under study. Acknowledgement The authors acknowledge the support of TNO Information and Communication Technology and KPN. In particular, the first author is financially supported within the research framework Dutch Research Delta, a research co-operation between KPN, TNO and a large number of Dutch universities. The framework comprises three research themes: Liquid Bandwidth, Transsectoral Innovation and Social Innovation. References 1. L. M. Correia, Mobile Broadband Multimedia Networks, Elsevier Science & Technology Books, T. K. Sarkar, T. K. Sarkar, Z. Ji, K. Kim, A. Medouri, and M. Salazar-Palma, A survey of various propagation models for mobile communication, IEEE Antennas Propagation Magazine, Vol. 45, No. 3, pp , M. Bengtsson and B. Volcker, On the estimation of azimuth distributions and azimuth spectra, in Proc. IEEE 54th Veh. Technol. Conf., Vol. 3, pp , M. Ghoraishi, J. Takada, and T. Imai, Identification of scattering objects in microcell urban mobile propagation channel, IEEE Transactions on Antennas and Propagation, Vol. 54, No. 11, pp , L. Vuokko, P. Vainikainen, and J. Takada, Clusters extracted from measured propagation channels in macrocellular environments, IEEE Transactions on Antennas and Propagation, Vol. 53, No. 12, pp , M. R. J. A. E. Kwakkernaat, Y. L. C. de Jong, R. J. C. Bultitude, and M. H. A. J. Herben, High-resolution angle-of-arrival measurements on physically-nonstationary mobile radio channels, IEEE Transactions on Antennas and Propagation, Vol. 56, No. 8, pp , M. R. J. A. E. Kwakkernaat, Y. L. C. de Jong, R. J. C. Bultitude, and M. H. A. J. Herben, Improved structured least squares for the application of Unitary ESPRIT to cross arrays, IEEE Signal Processing Letters, Vol. 13, No. 6, pp , M. R. J. A. E. Kwakkernaat and M. H. A. J. Herben, Analysis of clustered multipath estimates in physically nonstationary radio channels, in Proc. IEEE 18th International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC 2007), Athens, September 2007, pp M. R. J. A. E. Kwakkernaat and M. H. A. J. Herben, Analysis of scattering in physically nonstationary mobile radio channels, in 2nd European Conf. on Antennas and Propagation (EuCAP 2007), Edinburgh, UK., M. R. J. A. E. Kwakkernaat, Y. L. C. de Jong, R. J. C. Bultitude, and M. H. A. J. Herben, 3-D Switched antenna array for angle-ofarrival measurements, in Nice, France, 1st European Conf. on Antennas and Propagation (EuCAP 2006), J. Salo, J. Salmi, N. Czink, and P. Vainikainen, Automatic clustering of nonstationary MIMO channel parameter estimates, in Conference Proceedings of the 12th International Conference on Telecommunications, Cape Town, Südafrika, N. Czink, G. Del Galdo, and C. Mecklenbräuker, A novel automatic cluster tracking algorithm, in Proc. IEEE 17th Int. Symp. on Personal, Indoor and Mobile Radio Communications (PIM- RC2006), Helsinki, Finland, A. K. Jain and R. C. Dubes, Algorithms for Clustering Data, Prentice Hall, Author Biographies Maurice R. J. A. E. Kwakkernaat was born in Venray, The Netherlands, in He received the M.Sc. degree in electrical engineering from Eindhoven University of Technology (TU/e), Eindhoven, The Netherlands, in In September 2004, he started as a Ph.D. student in the Radiocommunications group at TU/e. His research interests include wireless communication systems, radio wave propagation modelling and experimentation and signal processing for multiple antenna systems. He is a member of the IEEE, the Royal Institute of Engineers in The Netherlands (KIvI) and the Dutch Electronics and Radio Society (NERG). Since December 2006, he is a member of the Management Committee of the COST Action 2100 Pervasive Mobile Ambient Wireless Communications. In 2007, he received the Best Paper Award of PIMRC 2007 with his paper titled: Analysis of Clustered Multipath Estimates in Physically Nonstationary Radio Channels. Matti H. A. J. Herben was born in Klundert, The Netherlands, in He received the M.Sc. degree (cum laude) in electrical engineering and the Ph.D. degree in technical sciences from Eindhoven University of Technology (TU/ e), Eindhoven, The Netherlands, in 1978 and 1984, respectively. He has been with the Radiocommunications group at TU/e since 1978, where he currently is as an Associate Professor. He was Associate Editor of Radio Science from 1993 to 1996 and is since 2007 Associate Editor of the IEEE Transactions on Antennas and

10 116 Int J Wireless Inf Networks (2008) 15: Propagation. His research interests and publications are in the areas of antennas, radio wave propagation, channel modelling for wireless communications, and atmospheric remote sensing. Dr. Herben is a member of the IEEE, the Royal Institute of Engineers (KIvI), the Dutch Electronics and Radio Society (NERG), the Dutch URSI Committee, treasurer of the IEEE Benelux joint Chapter on Communications and Vehicular Technology, and a member of the Management Committee of the COST Action 2100 Pervasive Mobile Ambient Wireless Communications.