System-oriented Measurement And Analysis Of Mimo

EUROPEAN COOPERATION IN THE FIELD OF SCIENTIFIC AND TECHNICAL RESEARCH

COST 273 TD(05) 063 Bologna, Italy 2005/Jan/19-21

————————————————— EURO-COST —————————————————

SOURCE:

MEDAV - TeWiSoft Germany

System-Oriented Measurement and Analysis of MIMO Channels

Uwe Trautwein1 Markus Landmann2 Gerd Sommerkorn2 Reiner Thomä2 1

MEDAV - TeWiSoft Ehrenbergstr. 11, D-98693 Ilmenau, GERMANY Phone: + 49-3677 668 433 Fax: + 49-3677 668 168 Email: [email protected]

2

Ilmenau University of Technology Institute of Communications and Measurement Engineering POB 100565, 98684 Ilmenau, GERMANY Phone: + 49-3677 69 1157 Fax: + 49-3677 69 1113 Email: [email protected]

System-Oriented Measurement and Analysis of MIMO Channels Uwe Trautwein1, Markus Landmann2, Gerd Sommerkorn2, Reiner Thomä2 1

2

MEDAV – TeWiSoft, Ilmenau, GERMANY, Email: [email protected] Ilmenau University of Technology, GERMANY, Email: [email protected]

Abstract: Realistic modeling of the radio wave propagation is an essential prerequisite for the evaluation of different concepts for future wireless communication systems. The German research project WIGWAM is targeted to achieve 1 Gbit/s overall throughput in a bandwidth of 100 MHz within the 5 GHz band, making the use of MIMO technology mandatory. This paper presents an overview of several measurement campaigns performed with a RUSK MIMO channel sounder in 2004 to gather data in representative deployment scenarios for such a system. High resolution antenna arrays have been used for both the mobile terminal (MT) side as well as the access point (AP) side. The RIMAX algorithm is applied to jointly estimate the spatial-temporal parameters of the discrete multipath components for both link ends, and to identify the contribution of the diffuse scattering. First results of the statistical analysis of 10 different measurement routes are presented, including distributions of Tx and Rx azimuth and delay parameters, CDF’s of their respective RMS spreads, the characterization of their joint properties as well as CDF’s for the parameters of the dense multipath components.

1

Motivation and Outline

The adoption of multiple-input multiple-output (MIMO) technologies for the air interface of new wireless communication systems promises to meet the increasing data rate demands of future applications within a reasonable radio bandwidth [11]. MIMO can provide an increased spectrum efficiency by exploiting the spatial dimension of the radio wave propagation. For MIMO techniques, the multipath propagation itself turns into a key component of the transmission system for separating multiple data streams transmitted at the same timeslot at the same frequency. Thus, MIMO transceivers depend on the joint spatial and temporal multipath structure at the transmitter (Tx) side as well as the receiver (Rx) side of the radio link. Hence, it must be accurately modeled. The conceptual considerations for the channel modeling approach, in particular the joint frequency-spatial-temporal correlation for one user and furthermore for multi-user MIMO channels, involves detailed insights into the physical nature of propagation [9] This can only be by conceived by performing multidimensional channel sounding campaigns with a broadband, real-time MIMO channel sounder such as the MEDAV RUSK MIMO [1][2][7] and subsequent estimation of the spatial-temporal characteristics of the various multipath components by means of a high resolution parameter estimation algorithm such as RIMAX [3][4]. A statistical analysis of these results is the foundation for the derivation of a stochastic-geometric model. Moreover, the statistical parameters can serve the purpose of “calibrating” more refined channel models and for deriving a classification of propagation environments with the respective parameter settings. At the beginning of 2004 the German national research project WIGWAM (Wireless Gigabit With Advanced Multimedia Support) [8] has been launched with the ambitious goal to develop the technological foundations for a new wireless communication system capable to provide 1 Gbit/s overall throughput. The project is funded by the German Ministry of Educa-

2

tion and Research (BMBF) and combines in total 10 partners, among them many big wireless equipment manufacturers active in Germany. MEDAV and TU Ilmenau contribute to this project by channel measurements and modelling. The WIGWAM concept puts high demands on the channel modeling aspects due to the high system bandwidth and the mandatory commitment to MIMO transmission techniques. The envisaged radio bandwidth of 100 MHz bandwidth leads to a very high resolution of multipath components in the delay domain. Presently there exists no suitable model for such a bandwidth. It is currently under investigation whether known concepts can be expanded straightforward, or whether an explicit introduction of a model for the diffuse scattering components is required [5]. This paper describes the system-oriented measurement and analysis of MIMO channels in the sense that the measurement environment, the measurement antennas, and the setup of the measurement procedure is arranged according to the needs of a specific system and deployment scenario and the intended use of the measurement data and the derivable models. (See [10] for a general discussion on this topic.) This should help to bridge the frequently observable gap between the “pure” propagation modeling aspects and the channel modeling needs of system designers. The paper is organized as follows: We present a brief introduction into the WIGWAM system concept and its influence on the measurements. Section 3 summarizes the basic data of the applied measurement technique. The overview of the measurement environments considered in this paper is given in Section 4. Section 5 presents details on the multipath parameter estimation procedure by means of the RIMAX algorithm. A complete example of one measurement setup and the respective data analysis in Section 6 familiarizes the reader with the concept and the computational steps leading to the comparative depiction of some statistical properties of 10 different channel measurements in Section 7.

2

System Aspects

This section summarizes some general information on the system concept that served as the reference for the specific channel investigations. The main design objective of the WIGWAM project [8] is to attain the required data rate of 1 GBit/s but at the same time not to exceed the limits given by the technology that can be expected to be available in 2007, when the project is finished. Several working groups have been established which are responsible for the system concept, the hardware platform, the physical layer, the link layer, and the network layer. Four different deployment scenarios are envisaged for the WIGWAM system: • Home Scenario: autonomous self-configuring network • Office Scenario: fixed network extension • Public Access Scenario: cellular network extension for hot spots • High Velocity Scenario: freeway and track information access Depending on the scenario, different system parameters and also different frequency bands are under consideration. Presently, it is most actively worked on an air interface for the 5 GHz band. Here, the bandwidth limitations are very tight. In order to achieve 1 GBit/s within the assumed channel bandwidth of 100 MHz a spectral efficiency of at least 10 bit/s/Hz is required which can only be reached by using multiple antenna techniques. MIMO-OFDM schemes are currently the candidates for the physical layer, combined with adaptive coding and modulation techniques. Currently, up to 4 x 4 MIMO schemes are considered.

3

The presentation in this paper focuses on MIMO channel characterization for the Public Access Scenario. Here, the system is aimed to provide high data rate access in urban and hot spot environments. It covers outdoor environments with a range of up to 500 m and medium mobility support as well as public indoor hotspots such as railway station or conference hall. The measurement locations (see Section 4) have been selected correspondingly. A measurement antenna with a horizontal field of view of 120° and a vertical field of view of 60° is well suited to resemble a potential AP antenna. The mounting sites are likewise realistically chosen, e.g., close to house walls or other suitable constructions and with a good sight on the scenario. The selected measurement antenna for the MT side provides omni-directional coverage in azimuth and about 90° in elevation.

3

Measurement Technique

Although channel measurement and modeling is a very active field of research, a general lack of true MIMO, i.e., double directional, measurements had to be stated at the beginning of the WIGWAM project. The rationale behind the setup of several measurement campaigns in 2004 was therefore to cover representative Public Access scenarios and to use multiple and / or flexible antenna configurations to allow the data to be used and analyzed for different investigations. This paper focuses exclusively on the measurements using well-calibrated high resolution antenna configurations. This is required for the extraction of the directions of arrival (Rx side) and the directions of departure (Tx side) of the individual multipath components by means of a high resolution parameter estimation algorithm. The following table summarizes information on the measurement parameters: Channel Sounder Carrier frequency / wavelength Measurement bandwidth Maximum multipath delay Measurement rate Tx/Rx synchronization Tx power at the antenna Number of attached Tx/Rx ports Access point antenna (Rx side) Mobile terminal antenna (Tx side)

RUSK ATM MIMO (Medav) [1][2][7] 5.2 GHz / λ = 5.77 cm 120 MHz 1.6 µs / 3.2 µs chosen according to the environment chosen according to desired mobile speed and / or hardware restrictions, 9 ms … 20 ms typ. Rubidium reference ca. 200 mW maximum 16 Tx / 16 Rx 8 element uniform linear patch array (8ULA), optionally dual-polarized, coverage ca. 120° sector view 16 element uniform circular array (16UCA), omnidirectional coverage

Table 1 Summary of measurement parameters

Figure 1 High resolution measurement antennas: left – 16UCA, right – 8ULA

4

4

Overview of the Measurement Scenarios

The following table lists the locations and some characterization on the 2004 MIMO measurement campaigns. The locations have been selected according to the specifications of a potential Public Access deployment scenario. The AP antenna array is usually mounted in a height of ca. 4 m. The AP mount locations are chosen such that it could be easily imagined to be the site of a system’s AP, e.g., by mounting at a house wall. The MT took either the role of a pedestrian or was mounted on a car. Results are presented for 10 selected measurement routes, each labeled by a route designator, which is a small subset of the totally available measurements. Location

Deployment

Route Misc. Information Designator

Ilmenau City

Microcell

street 1

street canyon (ca. 10 m width, pedestrian zone)

square 1

public square into street canyon

street 2

open street into street canyon

square 2

public square traversal, starting from adjacent street

4 AP locations

Munich Stachus

Microcell

square 3

large public square

Munich Siemens Tower

Macrocell

tower

AP mount height ca. 70 m with 30° down-tilt, i.e., significantly above surrounding buildings, route on factory premises, MT on the roof of a car

Munich Railway Station

Public indoor hotspot

hall dimensions 143m x 225m x 17m, heavily equipped with metal constructions station 1

route under NLOS along a train platform

station 2

route across the hall with LOS and partly NLOS

Ilmenau Con- Public indoor ference Hall hotspot Lobby

lobby

concrete/steel/glass construction, route mostly under NLOS, room dimensions 15m x 30m x 8m

Autobahn Bridge

bridge

MT on the roof of a car, AP mount height ca. 8 m with down-tilt, measured range ca.+80 m … –30 m

Freeway information access by car

Table 2 Overview of Channel Measurement Scenarios

5

High Resolution Multipath Parameter Estimation

For the characterization of MIMO channels the spatial properties of the radio wave propagation are especially important. The results presented in this paper are almost exclusively based on the identification of the parameters of the individual multipath components that can be observed for each position of the MT along its trajectory. The maximum likelihood parameter estimation algorithm RIMAX [3][4] is used for this purpose. The data model of the algorithm includes additionally the estimation of the parameters of the dense multipath components (DMC). Data Model The appropriate data model comprises two components which can be handled separately throughout the estimation procedure. The first part is deterministic and results from specularlike reflection. The second part represents the dense multipath components (DMC) which are 5

a consequence of the distributed diffuse scattering and the limited resolution of the measurement system. It typically occurs in complicated, multipath rich environment. This part is adequately modeled by a complex circular normal distribution. Its contribution varies depending on the complexity of the propagation environment. It can be almost negligible in macrocell LOS scenarios and can even dominate in complicated propagation environments such as factory halls. In the discrete angular-delay Doppler domain the specular part is described by a superposition of K R-dimensional Dirac deltas weighted by a 2x2 complex polarimetric path weight matrix with its components γ xy, k , where the indices x,y indicate horizontal and vertical polarization at Tx and Rx resp. The R dimensions are the DoD ϕT , ϑT (azimuth and elevation), TDoA τ, Doppler-shift α, and DoA ϕ R , ϑR : K γ HH , k γ VH , k  H (α , τ , ϕ R , ϑ R , ϕT , ϑT ) = ∑   δ (α − α k ) δ (τ − τ k ) δ ϕ R − ϕ R k δ ϑ R − ϑ R k δ ϕT − ϕT k δ ϑT − ϑT k k =1 γ HV , k γ VV , k 

(

) (

)(

)(

)

(1) The observable channel response s(θk) in the multidimensional aperture domain is defined by the limited observation time, finite bandwidth, and finite (effective) antenna apertures. θ k is the condensed propagation path parameter vector containing 14 real-valued unknowns. We arrange the sampled channel response in vectors as a(µ k ) = a(µ k( R ) ) ⊗ a(µ k( R−1) ) ⊗ … ⊗ a(µ k(1) ) , whereby the a µ k(i ) are complex exponentials resulting from Fourier transform of (1) and the (i )

( )

µ k are normalized path parameters [3]: s(θ k ) = γ HH ,k ⋅ G HH ⋅ a(µ k ) + γ HV ,k ⋅ G HV ⋅ a(µ k ) + γ VH ,k ⋅ GVH ⋅ a(µ k ) + γ VV ,k ⋅ GVV ⋅ a(µ k )

(2)

The linear projector matrices Gxy describe the measurement systems response which is composed by the Kronecker product of the frequency, Doppler and spatial responses, respectively. Resulting from many observations of measured channel responses an exponential decaying data model was defined to represent the dense multipath components (DMC) in the delay (correlation) domain ψ (τ ) with its corresponding frequency response Ψ ( f ) [5]. The parameter vector θ dds is composed of the parameters β d , τ d , α1 which are the normalized coherence bandwidth, base delay and maximum power respectively, see Figure 2. Note that due to the limited observation bandwidth, a distortion of this response will be observed in the delay domain:

{

ψ x (τ ) = E x(τ )

2

}

0 τ <τd     1 α1 ⋅ 2 τ = τd  = α ⋅ e − B d (τ −τ d ) τ > τ  d  1

−−−•

ψ x(f ) =

α1 βd + j 2 π f

⋅ e − j 2 π fτ d

(3)

6

power delay profile

α1 βd

0

α0

τd

time delay

Figure 2 Parameters characterizing the dense multipath components

Maximum Likelihood Parameter Estimation With the stationary measurement noise n and the dense multipath and specular components d and s resp. the total observed signal vector x is modeled as follows: K

x = n + d(θ dds ) + ∑ s(θ k ) = n + d(θ dds ) + s(θ sp )

(4)

k =1

having a conditional probability density of: pdf (x θ sp , θ dds ) =

1 −(x −s (θ sp ))H R (θ dds )−1⋅(x −s (θ sp )) e . π det (R (θ dds )) M

(5)

The related log-likelihood function is: L(x; θ sp , θ dds ) = − M ⋅ ln (π ) − ln (det (R (θ dds ))) − (x − s(θ sp )) ⋅ R (θ dds ) ⋅ (x − s(θ sp )) H

−1

(6)

Because of the Gaussian nature of the probability density, the maximization of (6) in essence is a nonlinear least squares problem. Since an exhaustive search in the multidimensional parameter space is not feasible, we are using an iterative search framework. This procedure proceeds snapshot by snapshot and takes advantage as much as possible from typical channel behavior which is known a-priori from propagation physics and from experimental experience. So the estimated parameter set of every snapshot is taken as the initial estimate for the next one. The global search for new paths (which has to be carried not only at the beginning of the sequence but continuously step by step) is carried out by a SAGE-like procedure. Rather than a random assumption for unknown parameters we use some kind of non-coherent combining of observations to reduce the parameter dimension. The problem of local search is completely different. We have found that in case of closely spaced coherent paths the coordinate-wise search strategy of SAGE [6] has serious disadvantages because of its slow convergence rate which is not only time-consuming but may also end in erroneous estimates when using a quantized parameter data base [4]. On the other hand, it is well known that the ML function is, under mild restrictions, quadratic at its maximum (in the local “attractor area”). Therefore a conjugate gradient search promises much better convergence performance when the parameters are coupled. From the variety of available procedures for nonlinear optimization we have decided to use Levenberg-Marquardt because of its robustness. To calculate the optimum step size and direction for parameter change these algorithms require the gradient,

7

the Jacobian and the Hessian of the log-likelihood function at the actual point in the parameter space. The approximation of the Hessian as its is used in the Gauss-Newton / LevenbergMarquardt algorithm is essentially an estimate of the Fisher information matrix (FIM). This provides us with information on the variance and on the interdependency of the parameter estimates. The variance estimate helps to evaluate the reliability of the parameters and is used to accept or drop estimated paths. The two major sources of excessive variance are line splitting and noise enhancement. As a result of the parameter estimation we get a propagation path parameter vector for each specular component plus one vector containing the parameters of the dense multipath components for each measured MIMO channel snapshot. These result vectors are the basis for the further analysis in this paper, aimed to achieve a better understanding of the wave propagation in the different scenarios using the model including dense multipath components. An important parameter about the scenario is the power ratio between all specular components (Pspecular) and the power of the dense multipath components (PDMC) which is strongly related to the estimated number of paths.

6

An Introductory Example to the Analysis Results

The purpose of this section is to portray the complete approach to the setup and the analysis of one specific measurement. We selected “square 1” from Table 2 for this example. The measurement location is in the city of Ilmenau. The AP antenna array (R2) is positioned in one corner of a small square (Apothekerbrunnen) with the looking direction diagonally across the square and into an adjacent pedestrian shopping zone, as depicted in Figure 3. The array is mounted approx. 6 m above the average surrounding street level and well below the roof tops, which yields a typical microcellular deployment scenario.

AP R2 50m

R2 16

3

MT

Figure 3 Map and photograph of the presented example scenario “square 1”

8

In the sequel, the measurement results of the route from point 16 to point 3 in Figure 3 are described, which lies in the continuation of the street to be seen in the photograph. At first, the MT traverses the square and enters then into an adjacent street. The MT covered a distance of ca. 70 m in a time of ca. 50 s with walking speed. A total number of ca. 3000 MIMO channel snapshots have been recorded during this time. On the square dominate line-of-sight (LOS) conditions, partly the LOS is obstructed by a van parking on the square. As soon as the MT leaves the square, the LOS disappears completely. This hard transition can be very well recognized from the time-variant power delay profile depicted in Figure 4.

0

10

0.2

0.4

0.6

0.8

[s]

sur em

20

-40

ent ti

30

-20

me

40

0

mea

average power [dB]

50

1

delay [µs]

Figure 4 Normalized magnitude of the time-variant power delay profile for the selected measurement run, averaged over all measured antenna channels

The first processing step towards the statistical characterization of the spatial-temporal multipath propagation is the parameter estimation by means of the RIMAX algorithm. Figure 5 and Figure 6 depict for each of the MIMO snapshots recorded along the measurement run the following parameters of the identified multipath components: the path delay, the Doppler frequency, the angle in azimuth at the MT side, the angle in azimuth at the AP side and the relative path weight in dB, which is coded in the color of the dots. 30

1 -10

delay [µs]

-20 0.6

-30 -40

0.4

-50 0.2

0 0

-60 10

20 30 measurement time [s]

40

50

-10

20 Doppler frequency [Hz]

0.8

-20

10

-30 0 -40 -10

-50

-20 -30

-60 0

10

20 30 measurement time [s]

40

50

Figure 5 Parameter estimation results for the selected measurement run – colors depict magnitudes of the path weights in dB (left: delay, right: Doppler)

9

150

-10

100

-20

50 -30 0 -40 -50 -50

-100

-60

-150 0

10

20 30 measurement time [s]

40

angle in azimuth @AP [deg]

angle in azimuth @MT [deg]

150

-10

100

-20

50 -30 0 -40 -50 -50

-100

-60

-150 0

50

10

20 30 measurement time [s]

40

50

Figure 6 Parameter estimation results for the selected measurement run – colors depict magnitudes of the path weights (left: angle of departure at the mobile terminal, right: angle of arrival at the access point)

A joint presentation of the multipath parameters can help to identify clusters of multipath components and can give indications on an eventual coupling of the path parameters. Figure 7 shows as an example the angle in azimuth versus the relative path delay for the selected measurement run in a scatter plot. It indicates for the very first multipath components a rather independent distribution with a tendency to smaller angles with increasing delays. For larger delays, a cluster-like appearance for certain delay/azimuth ranges can be stated.

angle in azimuth @AP [deg]

150

-10

100

-20

50

-30

0 -40

-50

-50

-100

-60

-150 0

0.2

0.4 rel. delay [ µs]

0.6

0.8

Figure 7 Joint representation of the parameter estimation result for the selected measurement run – relative delay vs. angle of arrival at the access point

Clustering effects can effectively be investigated by the inspection of the joint power density spectra over a limited segment of the measurement routes. Figure 8 shows this presentation for the relevant signal dimension combinations separately for the LOS and the NLOS part of the selected measurement run. For the LOS segment it can be observed that for short delays (ca. 200 ns), the MT and AP azimuth angles arrive via a large sector with two main clusters that are separable in the MT azimuth but not in the AP azimuth. For larger delays a few distinct scattering areas can be recognized. For the NLOS segment of the route there are also some clusters visible, but the cluster areas are close to another. They are especially hard to distinguish in the joint MT and AP azimuth spectrum, the delay component needs additionally to be considered.

10

-40

-20

-80

0.8

0.8

0.6

0.6

0.4

-40

-20

-100

0.4

0.2

0.2

0 -100

-100 0 100 angle in azimuth @M T [deg]

-60

-40

-20

-60 0.8

0.6

0.6 delay [ns]

0.8

0.4

-50 0 50 angle in azimuth @AP [deg]

-40

100

0.2

0.2

0

0 -100

-100 0 100 angle in azimuth @M T [deg]

-50 0 50 angle in azimuth @AP [deg]

100

-60

-40

-20

100 50 0 -50 -100 -150

-20

0.4

-80

150

-100

-50 0 50 angle in azimuth @AP [deg]

-60

angle in azimuth @MT [deg]

0

delay [ns]

-60

angle in azimuth @MT [deg]

-60

delay [ns]

delay [ns]

-80

-40

100

-20

150 100 50 0 -50 -100 -150 -100

-50 0 50 angle in azimuth @AP [deg]

100

Figure 8 Joint power density spectra for the LOS part of the selected measurement run (upper row) and for the NLOS part (lower row). The colors represent the normalized power in dB in relation to the total discrete multipath power of the respective segments.

A basic statistical description of the channel characteristics is obtained by computing the cumulative distribution functions (CDF’s) of individual channel parameters. The basis is the progression of the parameters over a measurement run within a specific environment. The first analysis item is the delay window duration, which characterizes the total delay time interval with significant multipath components present in the power delay profile. This parameter is important, e.g., for the definition of the guard interval in OFDM systems. The left part of Figure 9 shows the temporal progression over the measurement time, and the right part the CDF of this parameter. 1

800

Prob(delay window < Abscissa)

700

delay window [ns]

600 500 400 300 200

0.8

0.6

0.4

0.2

100 0 0

10

20 30 measurement time [s]

40

50

0 0

200

400 delay window [ns]

600

800

Figure 9 Estimated delay window duration for the selected measurement run (left: progression over the measurement time, right: CDF)

The rms delay spread of the channel is a very important basic channel parameter from a system design point of view. The progression of this quantity over the measurement run is depicted in the left part of Figure 10. Here, the transition from the LOS part of the route into the 11

NLOS part can be clearly identified by the step-like increase at around 20 s. When looking at the CDF of the rms delay spread in the right part of Figure 10, the presence of a clear distinction of the two “sub-scenarios” can be identified by a non-monotonic shape of the CDF curve. 1 Prob(rms delay spread < Abscissa)

rms delay spread [ns]

100 80 60 40 20

0 0

10

20 30 measurement time [s]

0.8

0.6

0.4

0.2

0 0

40

20

40 60 80 rms delay spread [ns]

100

Figure 10 Estimated rms delay spread for the selected measurement run (left: progression over the measurement time, right: CDF)

120 100 80 60 40 20 0 0

10

20 30 measurement time [s]

40

50

Prob(rms angular spread in azimuth @MT < Abscissa)

rms angular spread in azimuth @MT [deg]

MIMO transmission principles rely on the presence of multipath components distributed over a preferably large angular sector at the AP and the MT side of a radio link. This characteristics can be grasped by analyzing the rms angular spread of the angles of departure and the angles of arrival of the individual multipath components. The parameters depicted in Figure 5 are the basis for computing the rms angular spread values depicted in Figure 11 for the MT side and in Figure 12 for the AP side. For the first 20 seconds, where the MT traverses the square mostly under LOS, very large spread values for the MT side can be observed, which is probably a result of wall reflections from the buildings surrounding the square. The transition into the adjacent street goes along with a smooth decrease in the spread values. Finally, only small spread values are observed, indicating multipath propagation only along the street towards the square. According to these observations, the CDF of the rms azimuth spread values rises rather smooth for the selected measurement run. 1

0.8

0.6

0.4

0.2

0 0

20 40 60 80 100 rms angular spread in azimuth @M T [deg]

120

Figure 11 Estimated rms angular spread in azimuth at the MT for the selected measurement run (left: progression over the measurement time, right: CDF)

The rms angular spread values for the AP side depicted in Figure 12 are smaller than for the MT side. This can be expected, because the antenna covers only a sector. The change from the LOS area into the NLOS is likewise to be recognized. For the NLOS part the values are within rather tight limits, because the multipath components always arrive via propagation along the street and reflections at certain objects on the square at the AP antenna. The result of this characteristics is a steep shape of the CDF. 12

Prob(rms angular spread in azimuth @AP < Abscissa)

rms angular spread in azimuth @AP [deg]

120 100 80 60 40 20 0 0

10

20 30 measurement time [s]

40

50

1

0.8

0.6

0.4

0.2

0 0

20 40 60 80 100 rms angular spread in azimuth @AP [deg]

120

Figure 12 Estimated rms angular spread in azimuth at the AP for the selected measurement run (left: progression over the measurement time, right: CDF)

The depiction of the joint probability density functions (PDF’s) of the rms spread values of the multiple dimension in Figure 13 helps to discover mutual dependencies between the channel characteristics in the different signal dimensions. We can clearly see a cluster-like occurrence of certain delay/azimuth ranges plus a rather independently distributed fraction of the rms angular spread in azimuth at the MT. It is also interesting to see in the middle picture that there are two clusters with a correlation in the spread values. 3

4

1

250

250

200

200

150 100 50

0 0 50 100 rms angular spread in azimuth @M T [deg]

2

3

4

5

150 100 50 0 0 20 40 60 80 rms angular spread in azimuth @AP [deg]

1 rms angular spread in azimuth @MT [deg]

2

rms delay spread [ns]

rms delay spread [ns]

1

2

3

4

5

120 100 80 60 40 20 0 0 20 40 60 80 rms angular spread in azimuth @AP [deg]

Figure 13 Joint probability densities for rms delay spread, rms angular spread in azimuth at the AP and at the MT for the selected measurement run (the grey values represent the probability in %)

Especially the azimuth characteristics depends on the identification of a sufficiently high number of discrete multipath components. For informative purposes, the CDF for this number is depicted in the left part of Figure 14. Another important quantity for the evaluation of the significance of the estimated discrete multipath components is the ratio of the power contributed by the discrete or specular multipath components to the power of the dense multipath components. The larger this ratio, the better is the representation of the multipath scenario by the discrete components. Vice versa, a small ratio indicates, that the propagation situation is dominated by diffuse scattering and relying on discrete components only for modeling purposes might lead to erroneous results. The CDF of this parameter depicted in the right part of Figure 14 indicates potentially this situation for 65 % of the cases, where the power of the specular components is smaller than the power of the dense multipath components.

13

< Abscissa)

1

0.8

DMC

0.6

Prob(Power /Power

0.4

SC

Prob(number of paths < Abscissa)

1

0.2

0 0

10

20 number of paths

30

0.8

0.6

0.4

0.2

0 -10

40

-5

0 Power /Power SC

DMC

[dB]

5

10

Figure 14 Left: CDF of the number of estimated paths per snapshot for the selected measurement run, right: CDF for the ratio of the power of the specular components vs. dense multipath components

7

Analysis Results of Selected Measurements

In this section we present a comparative presentation of the characteristic channel properties introduced by the example in the previous section. This is important in order to classify propagation environments and to find out the range of the values for typical as well as for extreme cases. For the delay window as well as for the rms delay spread depicted in Figure 15, extremely large values are observed for the measurements at the railway station. At the other extreme we find the bridge scenario, were we have always LOS and no significant scatterers around. In all cases but the station, the delay window is found to be smaller than ca. 500 ns with 95 % probility, and the rms delay spread smaller than ca. 90 ns. 1

0.8

station 1 station 2 street 1 street 2 square 1

0.6

square 2 square 3 lobby tower

0.4

0.2

Prob(rms delay spread < Abscissa)

Prob(delay window < Abscissa)

1

0.8

0.6

0.4

0.2

bridge 0 0

500 1000 delay window [ns]

1500

0 0

50

100 150 rms delay spread [ns]

200

250

Figure 15 CDF’s for the estimated delay window (left) and the rms delay spread (right) for all scenarios

For the rms angular spreads in azimuth at the mobile terminal depicted in the left part of Figure 16 the smallest spread values are again for the bridge scenario, which is not really surprising. However, the largest values are here observed for the indoor scenario lobby and for the tower scenario. In both environments reflecting walls are close to MT antenna leading to multipath propagation into very different directions. The right part of Figure 16 shows in contrast the smallest spread values for the tower scenario, where the MT was relatively far away from the AP, and hence, all paths arrived from a rather small sector. Here, the two station

14

1

0.8

0.6

0.4

0.2

0 0

50 100 rms angular spread in azimuth @M T [deg]

150

Prob(rms angular spread in azimuth @AP < Abscissa)

Prob(rms angular spread in azimuth @MT < Abscissa)

measurements mark the upper bound of the spread values, reflections can arrive from a large sector via the metallic hall construction. 1

station 1

0.8

station 2 0.6

street 1 street 2 square 1

0.4

square 2 square 3 lobby tower

0.2

bridge 0 0

50 100 rms angular spread in azimuth @AP [deg]

150

Figure 16 CDF’s for the estimated rms angular spreads in azimuth at the mobile terminal (left) and the access point (right) for all scenarios

Figure 17 serves the purpose to illustrate the relative significance of discrete and dense multipath components. The path numbers are greatest for the tower scenario which is a very spacious environment with well-separated paths. These discrete paths contribute also to the highest power ratio of discrete vs. dense multipath components. The power of the diffuse scattering around the MT is too weak to reach the AP over the relatively large distance. On the other hand, the station scenario features an average number of estimated discrete paths and is nevertheless rather poorly represented by the discrete paths only. This can be recognized from the low values in the right part of Figure 17.

street 1 street 2 square 1

0.6

square 2 square 3 lobby tower

0.4

0.2

0 0

bridge

10

20

30 40 number of paths

50

60

DMC

station 2

SC

station 1

0.8

< Abscissa)

1

Prob(Power /Power

Prob(number of paths < Abscissa)

1

0.8

0.6

0.4

0.2

0 -30

-20

-10 0 Power /Power SC

DMC

10 [dB]

20

30

Figure 17 CDF’s for the number of estimated paths per snapshot (left) and CDF’s for the ratio of the power of the specular components vs. dense multipath components (right) for all scenarios

In order to keep the presentation clear, we limit the depiction of joint parameter representations to only 3 selected examples for the current paper. Interesting observations are for instance, that the MT and AP angular spread values for the scenario street 2 are rather independent. In contrast, the joint PDF of these parameters shows a clear dependency in the square 3 scenario. The rms delay spread and the rms angular spread at the AP in the tower scenario are pretty much focused to a very narrow area, whereas the joint PDF’s where the rms angular spread at the MT is involved shows a rather wide range of MT rms angular spread values within a narrow area of the other quantities.

15

250 200

200

150 100 50

0 0 50 100 rms angular spread in azimuth @M T [deg]

1

1.5

2

0 0 20 40 60 80 rms angular spread in azimuth @AP [deg]

2

200

200

50

0 0 50 100 rms angular spread in azimuth @M T [deg]

1

2

200

200

50

0 0 50 100 rms angular spread in azimuth @M T [deg]

4

6

8

100 50 0 0 20 40 60 80 rms angular spread in azimuth @AP [deg]

1.2

1.4

80 60 40 20 0 0 20 40 60 80 rms angular spread in azimuth @AP [deg]

2

3

120 100 80 60 40 20 0 0 20 40 60 80 rms angular spread in azimuth @AP [deg]

10

150

1

100

1

0 0 20 40 60 80 rms angular spread in azimuth @AP [deg]

2

0.2 0.4 0.6 0.8 120

8

50

250

100

6

100

250

150

4

150

3

rms delay spread [ns]

rms delay spread [ns]

50

250

100

4

100

250

150

3

150

2.5

rms delay spread [ns]

rms delay spread [ns]

0.5

2

rms angular spread in azimuth @MT [deg]

1 250

rms angular spread in azimuth @MT [deg]

1.5

2 rms angular spread in azimuth @MT [deg]

1

rms delay spread [ns]

rms delay spread [ns]

0.5

4

6

120 100 80 60 40 20 0 0 20 40 60 80 rms angular spread in azimuth @AP [deg]

Figure 18 Joint probability densities for rms delay spread, rms angular spread in azimuth at the AP and at the MT for 3 different measurements (the grey values represent the probability in %)

8

Summary and Outlook

New physical layer concepts need to be verified by using radio channel measurements and/or channel models validated for the appropriate system deployment scenarios. We have presented an overview on system-oriented MIMO channel measurements as well as first results of the statistical analysis. Much more effort will be necessary for the further evaluation and for drawing conclusions for a structural channel model and its parameters for the different deployment environments. It is expected that clustering properties of discrete components, coupling of the parameters in multiple dimensions and the explicit consideration of the dense multipath components will be required. Furthermore, the explicit introduction of sub-classes for LOS / NLOS situations for each of the scenarios seems advisable.

16

Acknowledgement This work was partly supported by the German Ministry of Education and Research (BMBF) within the project Wireless Gigabit with advanced multimedia support (WIGWAM) under grant 01BU375. A special thanks goes to the colleagues of the following institutions supporting the individual measurements: Siemens Communications, Munich, MEDAV GmbH, Uttenreuth, Ilmenau University of Technology, University of Ulm.

References [1]

R.S. Thomä, D. Hampicke, A. Richter, G. Sommerkorn, A. Schneider, U. Trautwein, W. Wirnitzer, "Identification of Time-Variant Directional Mobile Radio Channels," IEEE Trans. on Instr. and Measurement, pp. 357-364, April 2000.

[2]

R. S. Thomä, D. Hampicke, A. Richter, G. Sommerkorn, U. Trautwein, “MIMO Vector Channel Sounder Measurement for Smart Antenna System Evaluation,” European Trans. on Telecomm., ETT Vol. 12, No. 5, Sep./Oct. 2001.

[3]

A. Richter, M. Landmann, R. Thomä, “RIMAX - A Flexible Algorithm for Channel Parameter Estimation from Channel Sounding Measurements,” COST 273 TD(04)045, Athens, Greece, Jan. 2004.

[4]

A. Richter, M. Landmann, R. Thomä, “RIMAX - A Maximum Likelihood Framework for Parameter Estimation in Multidimensional Channel Sounding,” 2004 Intl. Symp. on Antennas and Propagation, Sendai, Japan, August 2004.

[5]

A. Richter, R. Thomä, “Parametric Modelling and Estimation of Distributed Diffuse Scattering Components of Radio Channels,” COST 273 TD(03)198, Prague, Czech Republic, Sep. 2003.

[6]

B. H. Fleury, M. Tschudin, R. Heddergott, D. Dahlhaus, and K. Ingeman Pedersen, “Channel Parameter Estimation in Mobile Radio Environment Using the SAGE Algorithm,” IEEE Journal on Selected Areas in Communications, Vol. 17, No. 3, pp. 434450, March 1999.

[7]

http://www.channelsounder.de

[8]

http://www.wigwam-project.com

[9]

Andreas F. Molisch, “A Generic Model for MIMO Wireless Propagation Channels in Macro- and Microcells,” IEEE Trans. Signal Proc., vol. 52, no. 1, pp. 61–71, Jan. 2004.

[10] Uwe Trautwein, Christian Schneider, Gerd Sommerkorn, Dirk Hampicke, Reiner Thomä, Walter Wirnitzer, “Measurement Data for Propagation Modeling and Wireless System Evaluation,” COST 273 TD(03) 021, Barcelona, Spain, Jan. 2003. [11] G. J. Foschini and M. J. Gans, “On limits of wireless personal communications in a fading environment when using multiple antennas,” Wireless Personal Communications, vol. 6, no. 3, pp. 311–335, March 1998.

17