Characterisation of the SatellitetoIndoor Propagation Channel
F. PerezFontan
*
, V. Hovinen
†
, M. Schönhuber
¤
, R. PrietoCerdeira
§
, J. RiveraCastro
§
, P. Valtr
*
, J. Kyröläinen
†
, F. Teschl
¤
*
University of Vigo, Spain, email: fpfontan@tsc.uvigo.es, valtr@ieee.org
†
Centre for Wireless Communications, Oulu, Finland, email: veikko.hovinen@ee.oulu.fi, jukka.kyrolainen@ee.oulu.fi
¤
Joanneum Research, Graz, Austria, email: michael.schoenhuber@joanneum.at
§
European Space Agency, ESTEC, Noordwijk, The Netherlands, email: Roberto.Prieto.Cerdeira@esa.int, Juan.Rivera.Castro@esa.int
Abstract
— Reliable propagation models are required for the design of satellite broadcast to mobile or satellite navigation systems at Sband in the indoor environment. In particular, these models will support the development of hybrid satellite systems including complementary ground components. Apart from the effects of shadowing and multipath encountered in LMS (land mobile satellite) urban and suburban environments, the satellitetoindoor channel is affected by high building penetration losses, but also lower Doppler spread due to static or quasistatic situation of the user terminal. Results from a measurement campaign simulating a satellitetoindoor link are presented. Clusterbased wideband measurements were carried out with a channel sounder placed inside several buildings (airports, shopping mall, office buildings…), receiving the signal from the transmitter located in a helicopter hovering over the building at different (azimuth, elevation) combinations in igloo configuration. The measured data has been analysed with direct methods and iterative superresolution algorithm in order to derive building penetration losses, time spreading and angleofarrival information. Superresolution techniques make it possible to identify rays and clusters of rays, and thus give added value to the development of models. The superresolution information is effectively used for evaluation of intercluster and intracluster azimuth angle of arrival distributions. The availability of spatial information in MIMO and SIMO measurements helps developing and validating sitespecific models based on raytracing techniques given that the measured echoes are fully characterized in terms of amplitude, phase, delay and angle of arrival. Furthermore, the parameters corresponding to a directional statistical channel model have been extracted for each scenario thanks to the availability of this type of information. The results obtained contribute to a better characterisation of the satellitetoindoor propagation channel.
Satellitetoindoor, wideband statistical channel model, entry loss, angleofarrival, timespreading.
I.
I
NTRODUCTION
The development of new mobile satellite systems, including both satellite interactive communication systems and satellite broadcast systems to mobile users, intend to provide coverage from the satellite to users located on a wide range of environments, including indoor. ESA supports the development of precise prediction methods for satellite to indoor propagation environment, which lacks experimental data. As a first step, some statistical propagation models have been developed for narrowband and wideband signals transmitted from a geostationary satellite. Other prediction propagation models have to be developed for nongeo satellites or to assess the coverage performance obtained with satellites plus complementary ground components. The indoor environment presents strong attenuation values introduced by building structures. However, penetration through windows or the roof still can provide sufficient signal strength inside the building. The information of multipath spreading and angle of arrival is also of relevant importance. A well recognized channel model, due to Saleh and Valenzuela [1], has been used as the baseline model for describing time dispersion effects. A modification of the above model by Spencer et al. [2] has been used for describing the spatial scattering effects, i.e. angles of arrival. II.
T
HE SATELLITE

TO

INDOOR CHANNEL
Two main issues are of importance when modelling the satellitetoindoor channel, namely, (a) the entry loss and (b) the delay and angle spreading of the received energy. Entry loss is more critical of the two, given its very large values, while the delays are in the order of tens of ns. Echoes produced by interactions (reflections, diffractions) with other buildings will, in general, be negligible in comparison to the energy entering the building by direct illumination of its façade. Entry loss is the result of very complex processes including transmission through building materials and paths through different external and internal walls, the roof, the floor, through windows, glass façades, etc. where also window glass can be metal tinted, thus giving rise to strong attenuations. Moreover, satellite paths are very much dependent on the incidence angle.
9781424422043/08/$25.00 ©2008 IEEE
Figure 1. Assumed cluster structure in the SalehValenzuela model
Channel models can be classified as empirical, statistical or deterministic / analytical. Here an empirical approach has been selected for the entry loss while a statistical model has been used for the spreading effects. The developed model and associated simulator are based on a combination of two existing models. One adapted to satellitewireless LAN interference studies by GlazunovBerg [3] and based on earlier COST 231 work [4]. The other model is due to SalehValenzuela [1], including its directional version [2] (Modified
SalehValenzuela model due to Spencer et al.). The model's outcome, in its most general form, is a received echo structure (impulse response / scattering matrix) consisting of a maximum of 5 echo ensembles (four external walls and the roof). From this general result, other products: statistics, narrowband complex time series, tapped delayline complex timeseries for link level simulations, can be generated through the use in the postprocessing phase. The above assumption (up to five building entry paths) is a very general one, only rarely valid for the terrestrial case where the signal levels used are much larger than those in the satellite case. Thus, out of those five multipath ensembles, a maximum of two are expected to be observed above the measurement noise floor. Note that these ensembles are further composed of clusters of echoes and individual echoes or rays. The S&V model makes the assumption that echoes arrive in clusters, the overall impulse response being
)
ττ
(
δ
) jexp(
β
)
τ
(
kl l kl l k kl
T h
−−=
∑∑
φ
(1)
where index
l
indicates the cluster number and index
k
indicates the echo number within a cluster. The amplitude (representing the received voltage) of each echo is
β
kl
(Fig. 1). Each external wall and the roof gives rise to an echo ensemble consisting of one or several clusters of individual echoes spread out in time (delay axis). The overall received echo structure is given by the sum of the individual complex impulse responses,
h
(
τ
), for all external walls and the roof, i.e.,
)
τ
()
τ
()
τ
()
τ
()
τ
()
τ
(
Roof 4Wall3Wall2Wall1WallTot
hhhhhh
++++=
(2)
where one, or at most two terms are significant. The S&V wideband model was extended in [2] to include Angles of Arrival, AoA. From measurements carried out in indoor environments it was found how ray clusters are not only spread out in time/delay but in angle of arrival. Thus, the srcinal SalehValenzuela model was extended to include angles of arrival,
θ
, as indicated by the equation
)
θ
(
δ
)
τ
(
δ
) jexp(
β
)
θτ
,(
kl l kl l kl
l k kl
T h
θ τ φ
−Θ−−−=
∑∑
(3)
where, again, index
l
indicates cluster number and index
kl
indicates ray number
k
within a cluster
l
. The model for the angles of arrival,
θ
, is a zeromean Laplacian distribution with standard deviation
σ
, and where
Θ
l
is uniformly distributed between 0 and 2
π
. III.
D
ATA
A
NALYSIS
To clarify the data analysis process, the different data sets and preprocessing steps taken before the actual analysis are summarized. The data set consists of a number of "products". The Propsound channel sounder [5] from Elektrobit was used, providing an "instantaneous" data set consisting of one "complex" (I and Q) channel response v. delay per patch antenna and polarization. These instantaneous measurements are called cycles. The channel sounder transmits a pseudo random sequence where different code lengths are possible. Depending on the selected code length and code rate, the update rate of such set (full scan over all antennas and both polarizations) varies giving rates ranging from several tens of cycles per second to a few hundreds. The transmit antenna was circularly polarized (RHCP) while the receive antenna consisted of a set of patch antennas (SIMO) with two orthogonal linear polarizations covering a surface that approximates a semisphere. From each of the two linearly crosspolarized measurements, both the received copolar (RHCP) and crosspolar (LHCP) components can be calculated. Thus, one instantaneous measurement or cycle consists of an ensemble of instantaneous individual antenna complex channel responses obtained through crosscorrelation between a transmitted pseudorandom sequence and an identical, internally generated sequence in the receiver. The channel sounding process tries to measure the channel impulse response,
h
(
τ
). In channel modelling, it is usual to assume a sufficiently large bandwidth so that the impulse response is made up of complex deltas (ideal channel response). The band limiting effects are considered later when the impulse response is converted into a tapped delay line, TDL, fitting the bandwidth requirements of whatever system has to be simulated. In general, the results obtainable from the analyses reported should be valid for systems with RF bandwidths smaller than that of the channel sounder, i.e., 200 MHz (chip rate 100 Mcps).
50 0 50 100 15050494847464544434241Excess delay (ns)
N o r m a l i z e d p o w e r ( d B )
Figure 2. Measured and simulated / fitted power delay profile: convolution of the ideal impulse response and the response of a correlation receiver. Graz Airport, Gate 10, Elevation angle 60 degree. model TABLE I. S
UMMARY OF MEASUREMENTS AND RELATED
(
PRE
)
PROCESSED PRODUCTS
Instantaneous, single patch antenna, single polarization, complex channel response v. delay
Cycle
: ensemble of all (16 antennas × 2 polarizations) instantaneous, single patch antenna, single polarization, complex channel responses v. delay Accumulation of 5 cycles: input to superresolution processing Averaged power delay profile, APDP
Moreover, the impulse response is not constant in time but timevariant, i.e.,
h
(
t
,
τ
). This imposes some constraints on the sounding process (the duration of one full instantaneous measurement or cycle) whereby sounding has to be carried out at a rate consistent with the rate of chance of the channel (its coherence time). Assuming that the sounding process is fast enough, it is possible to consider the channel impulse response to be constant over the duration of one cycle. To sound the channel, pseudo random, noiselike sequences were used. Ideally, noise has a delta autocorrelation function. Pseudo random sequences show very narrow autocorrelation peaks: a base of the order of ±
T
chip
, the chip duration, and amplitude equal to the sequence length,
m
. The measurement of the channel impulse response is carried out by computing the crosscorrelation between the received signal and an identical sequence synchronized with the one at the transmitter. The measurement process does not exactly provide the ideal channel impulse response but is approximately the result of the convolution between the channel impulse response and the code autocorrelation pulse. Given the narrowness of the crosscorrelation pulse, this processing permits a good approximation of
h
(
τ
). The inphase and quadrature parts are obtained in this process. The so called power delay profile, PDP, is in fact the most frequently used characterization parameter for the wideband channel and is given by
2
)
τ
(*)
τ
(log10)
τ
(
wh P
≈
(4)
where
w
(
τ
) is the channel sounder correlation pulse. Further preprocessing of the row data includes the computation of socalled averaged power delay profiles, APDPs, encompassing a sufficient number of cycles (several tens of cycles). This is a standard procedure (ITUR Rec. P.1407) [8] for smoothing out fading effects throughout the delay range in the channel response due to possible inphase or outofphase combinations of echoes with similar delays. Table I summarizes the various measurement products. In superresolution approach instantaneous measurements or cycles are passed on to a SAGEbased (space alternating generalized expectationmaximization) [6] ISIS (Initialization and Search Improved SAGE) superresolution algorithm [7] for extraction of individual echoes or rays given in terms of amplitude, phase, delay, angle of arrival and departure,
)
ψδ
(
ψ
)
θδ
(
θ
)
ττ
(
δ
) j(exp
iiiii
a
−−−
φ
(5)
The ISIS algorithm needs at least a 15 dB dynamic range to operate correctly. For instantaneous channel data suffering from low dynamic range, a coherent combination of a few cycles before applying ISIS may be used to improve the signaltonoise ratio with respect to that of a single cycle. While ISIS still is producing good estimates for delay spread after moderate data combining, it however easily suffers from the phase distortion that most dramatically is reflected in the angle spread results. In this study, the relatively fast cycle rate allowed accumulation of data over five consecutive cycles while still preserving the phase information of the major echo contributions. Measurements were performed at fixed positions within selected buildings, with a helicopter carrying the transmitter attempting at being stationary during each measurement. At each measurement location, a sufficiently large number of instantaneous measurements (cycles) was recorded: about one minute. This allows the accumulation of sufficient data for the two types of cycle combinations pointed out above: every five cycles for superresolution ISIS analyses and several tens of cycles for computing APDPs. Two kinds of parameters were extracted from the measurements; the first group has to do with the entry loss and the other with the time and angle dispersion effects. As an example, Fig. 2 illustrates a measured averaged power delay profile and a simulated/fitted power delay profile resulting from the convolution of an ideal (deltas) impulse response and the response of a correlation receiver. Also in the figure, individual cluster responses are shown. For extracting the entry loss, the measured averaged power delay profiles, APDPs, were compared with a reference measurement carried out outside each building. In the beginning, to measure the entry loss, the sum of all samples above the noise floor was calculated both for the reference outdoor measurement,
P
Outdoor
, and the indoor
Figure 3. Measuring the building entry loss model
50 0 50 100 150 200 250 3005048464442403836343230Excess Delay (ns)
N o r m a l i z e d P o w e r ( d B )
Figure 4. Measured and simulated / fitted power delay profile: convolution of the ideal impulse response and the response of a correlation receiver. Graz Airport, Meeting Room, Elevation angle 15 degree.
measurements
P
Indoor
, i.e.,
∑∑
==
τ
10/)
τ
(Outdoor
τ
10/)
τ
(Indoor
Outdoor Indoor
10log10and10log10
P P
P P
(6, 7)
This criterion was later changed to take into account only the direct, LOS contribution as illustrated in Fig. 3. These parameters are not exactly the total received power, but rather some sort of power density (dB/ns), but this does not matter since the interest is in their difference (relative measurement). The entry loss is given by
( )
Indoor Indoor Outdoor Outdoor Entry
),(
P d d P L
−∆+=
(8)
where
)/log(20),(
Indoor Outdoor Indoor Outdoor
d d d d
=∆
(9)
is a distance correction factor needed when the reference measurement has been carried out at a different distance than that used in the indoor measurement.
P
Outdoor
(
τ
) and
P
Indoor
(
τ
) are the measured APDPs. Similarly, the averaged PDPs were normalized with respect the outdoor reference parameter
P
Outdoor
, i.e.,
( )
),()
τ
()
τ
(
Indoor Outdoor Outdoor Indoor Indoor
d d P P P
∆+−=
′
(10)
This implies that the direct, LOS contribution power outdoors is assumed to be 0 dB, thus the extracted model parameters (see next section) will be referred to the conditions just outside the building. Their transformation into absolute values is straightforward. IV.
M
ODEL PARAMETERS AND RESULTS
In this section, entry loss results for all buildings as a function of the enervation angle are provided, Table II, and Figs. 5 and 6. More details are given for two different rooms at Graz Airport, namely, Gate 10 and a conference room, are presented. Gate 10 is a room on the ground floor of the terminal building with external glass walls, with one floor above it. On the other hand, the conference room is on the first floor with a ceiling giving directly to the terminal's roof. The extracted parameter ranges (minmax) are provided in Tables III to VII. Fig. 4 illustrates a fitted APDP corresponding to the conference room. Here the time spreading is larger than at Gate 10, Fig. 2. V.
S
UMMARY
This paper presents results of a measurement campaign simulating the satellite to indoor channel at SBand where a helicopter has been used to carry a wideband, directive, SIMO channel sounder. A model initially based on the SalehValenzuela model has been developed. Preliminary wideband model parameters have also been reported. A
CKNOWLEDGMENT
This work has been performed under ESA / ESTEC Contract No. 19769/06/NL/GLC. R
EFERENCES
[1]
A. A. M. Saleh and R. A. Valenzuela, "A statistical model for indoor multipath propagation," IEEE Journal on selected areas in communications, vol. 5, no. 2, pp. 128137, Feb. 1987. [2]
Q. H. Spencer, B. D. Jeffs, M. A. Jensen and A. L. Swindlehurst, "Modeling the Statistical Time and Angle of Arrival Characteristics of an Indoor Multipath Channel," IEEE Journal on Selected Areas in Communications, vol. 18, no. 3, pp. 347360, March 2000. [3]
A. A. Glazunov and J. E. Berg, “Buildingshielding loss modelling”, Proc. Vehicular technology Conference 2000, vol. 3, pp. 18351839, 2000. [4]
European Cooperation in the Field of Scientific and Technical Research (COST) Telecommunications. Digital mobile radio toward future generation systems. COST231 Final Report, European Commission, 1999. [5]
Elektrobit. Propsopund. www.elektrobit.com [6]
B. H. Fleury, M. Tschudin, R. Heddergott, D. Dahlhaus, and K. L. Pedersen, "Channel parameter estimation in mobile radio environments using the SAGE algorithm," IEEE Journal on Selected Areas in Communications, vol. 17, pp. 434–450, Mar. 1999. [7]
A. Stucki and P. Jourdan, "MIMO Radio Channel Parameter Estimation Using the Initialization and Search Improved SAGE (ISIS) Algorithm," Proceedings of Virginia Tech's Thirteenth Symposium on Wireless Personal Communications, Blacksburg, VA, June 46, 2003. [8]
ITUR Rec. P.14073 (2007). Multipath propagation and parameterization.
Figure 5. Entry losses for all buildings Figure 6. Entry loss values for all buildings as a function of the elevation angle TABLE II. E
NTRY LOSSES FOR ALL BUILDINGS
Max Average Min Std Graz Airport. Conference room 21.5 14.5 7.3 3.9 Graz Airport. Gate10 23.3 19.5 15.6 2.1 Graz. Shopping Center 38.3 31.2 24.2 3.7 Graz. Residential Home 33.8 29.7 23.1 2.4 Vienna. International Airport 27.3 21.7 16.7 2.4 Vienna. Millenium Tower floor 22 33.1 28.0 22.1 2.6 Vienna. Millenium Tower floor 44 27.6 22.2 14.2 3.1 Vienna. Office building 28.3 22.2 15.4 3.2 TABLE III. S
UMMARY OF OTHER EXTRACTED PARAMETERS AT
G
RAZ AIRPORT
Elev. (deg)
T
max
(ns) # of clusters Delay spread (ns) Sp (dB/
µ
s)
µ
(ns) 15 100250 612 30.882.2 14.795.9 212 30 60170 48 24.644.7 80.4212.5 1025 45 90210 411 27.061.7 28.0144.6 210 Graz Airport Gate 10 (ground floor) 60 105140 412 27.568.8 60.1115.7 215 15 180360 917 55.6100.2 19.262.9 312 30 305400 1420 75.795.6 35.150.0 220 45 310350 1215 66.287.6 26.151.2 815 Graz Airport Meeting room (top floor) 60 350390 1316 74.589.1 32.344.2 815