You are on page 1of 4

Proceedings of the 14th European Radar Conference

Applying Antenna Synthesis Methods on a Path


Based MIMO Channel Model for Verification
Tobias Mahler, Thibault Deletoille, Johannes Frey, Jerzy Kowalewski, Thomas Zwick
Karlsruhe Institute of Technology (KIT), Institute of Radio Frequency Engineering and Electronics (IHE)
Engesserstraße 5, 76131 Karlsruhe, Germany
Email: tobias.mahler@kit.edu

Abstract—In this paper multiple input multiple output In this paper we present a simple channel model that is able
(MIMO) channel samples are generated by the use of a simple to generate MIMO channel samples of arbitrary MPCs. It
channel model that allows for arbitrary selection of the propa- resolves the DOD and the DOA and the propagation path
gation paths. The distributions of direction of departure (DOD),
direction of arrival (DOA) and phases and amplitudes can be phases and amplitudes that together yield the APS. Afterwards
selected for each path. The antenna array geometry and the we apply an eigenbeamformer synthesis method to the simu-
antenna element radiation patterns are also included and can be lated channels. Hereby we visualize and verify the matching
chosen arbitrarily. An antenna synthesis method that is based effect of the synthesized radiation patterns to the transmission
on eigenbeamforming is then applied to the simulated channels. channel under investigation.
It is the optimal solution of the considered antenna synthesis
problem that maximizes ergodic channel capacity. The resulting The remainder of this paper is organized as follows. Section II
fixed synthesized antenna radiation patterns are evaluated in introduces the channel model. Afterwards section III summa-
terms of their matching effect to the predominant directions rizes the facts about eigenbeamforming. Then we continue
of the transmission channel afterwards. The results show that with an overview of the simulation details in section IV and
the channel model is beneficial for the development of antenna the synthesis results in section V. Finally in section VI we
synthesis methods. It includes all parameters necessary for the
design and evaluation of antenna synthesis methods and still has conclude our work.
a low complexity. II. PATH BASED MIMO C HANNEL M ODEL
I. I NTRODUCTION The MIMO channel model used throughout this paper is
The knowledge about basic channel properties as for in- based on the generation of the MPCs that propagate between
stance predominant directions of transmission and reception the transmitter and the receiver antennas. A sketch of the
can help to achieve design guidelines for mobile multiple an- model is depicted in Fig. 1. The MIMO channel matrix
tenna systems. This is the goal of antenna synthesis methods. H ∈ CN ×M consists of the sum of Np propagation paths
The results of an antenna synthesis are fixed antenna radiation or MPCs:
Np
patterns that match the statistics of the expected channel as 
good as possible. The most important input information for H= ER (ΩR,p ) · ej·k·AR ·s(ΩR,p ) · αp · ej·ϕp
these methods is the sampled transmission channel. p=1 (1)
 T
The channel can be simulated by raytracing methods [1], j·k·AT ·
s(ΩT,p )
· e · ET (ΩT,p )
that model the wave propagation very well [2]. It is also
possible to get the angular power spectrum (APS) of the N and M are the numbers of transmitting and receiving
simulated channels that in turn helps to interpret the antenna antennas, (·)T denotes the transpose and k = 2π λ is the
synthesis results. But it is tedious to generate channel samples wavenumber. αp and ϕp stand for the amplitude and the phase
with arbitrary directions of transmission or reception because of the p-th propagation path between the antenna array centers.
the raytracing relies on three-dimensional scenario maps that ΩT,p and ΩR,p are the angles of departure and arrival of
determine the propagation paths and their directions. the p-th propagation path with the cartesian direction vectors
Another option is to measure the desired transmission chan- s(ΩT,p ) ∈ R3×1 and s(ΩR,p ) ∈ R3×1 of unit length. Ω =
nels with the help of channel sounders and to apply the (θ, ψ) is the merging of elevation angle θ and azimuth angle ψ.
antenna synthesis methods on the measured data [3]. But The array geometries are represented by the antenna elements
it is problematic or even unfeasible to verify the antenna position matrices AT ∈ RM ×3 and AR ∈ RN ×3 that contain
synthesis results afterwards because of the limited propagation the corresponding cartesian position vectors aT,m ∈ R3×1 and
path resolution that depends on the channel sounder. Most of aR,n ∈ R3×1 of each antenna element relative to the antenna
the channel measurements only deliver time and frequency array centers. The antenna elements radiation pattern matrices
dependent complex transmission coefficients of the channel are also included in the channel model and have a diagonal
that do not resolve each multipath component (MPC) as the structure:
raytracing simulations do. Generating channel samples with ET (Ω) = diag(eT,1 (Ω), . . . , eT,M (Ω)) ∈ CM ×M
(2)
arbitrary MPCs is also impractical with channel measurements. ER (Ω) = diag(eR,1 (Ω), . . . , eR,N (Ω)) ∈ CN ×N

978-2-87487-049-1 © 2017 EuMA 501 11–13 Oct 2017, Nuremberg, Germany


When including single beamforming, i.e. beamforming for a by RT = E[H† H] and RR = E[HH† ]. We have then the
single transmitter branch and for a single receiver branch, we following separation property for the complete correlation
get the transmission factor matrix:
hBF = bTR · H · bT ∈ C (3) Rh = E[vec(H)vec(H)† ] = RTT ⊗ RR (8)
of the beamformed channel. bT and bT are the beamforming where ⊗ denotes the Kronecker product. Such a model de-
vectors of the transmitter and of the receiver respectively. For scribes the fact that the fading from two transmit antennas j
multiple beamforming we get and k to the same receive antenna p is R∗Tjk and does not
depend on the receive antenna, and vice versa [7] [8].
HBF = BR · H · BT ∈ CNbf c ×Mbf c (4) Under such conditions, when the transmitter knows only
accordingly with the beamforming matrices BT at transmitter the statistics of the channel, i.e. RT and RR , [10] showed that
and BR at receiver side. Nbf c and Mbf c are the numbers of the optimal strategy is to transmit proper Gaussian signals x
beamformed branches at each side of the link. along the eigenvectors of RT , i.e. the optimal input covariance
matrix has the following shape:
III. E IGENBEAMFORMER S YNTHESIS M ETHOD
Qopt = UT ΛQ U†T (9)
Since the investigation of MIMO systems [4] [5], optimizing
the transmit strategy with respect to channel capacity (ergodic where ΛQ is diagonal and UT are the eigenvectors of RT :
or outage) has been an active field of research known as
eigenbeamforming, or precoding on specific channel state RT = UT ΛT U†T . (10)
information (CSI), for instance, precoding on mean or cor- In this work, we used this eigenbeamforming technique un-
relation CSI at the transmitter [6]. der statistical covariance information to perform the antenna
In this section we sum up the main results of eigenbeam- synthesis. The statistical channel information RT and RR
forming with correlation CSIT and apply them to the antenna can be obtained through measurements or simulations of the
synthesis problem. We consider here a Rayleigh-fading chan- channel, allowing us to find the optimal beamforming vectors
nel where the transmitter possesses only statistical information to be used for antenna synthesis. Indeed, when synthesizing
(CDIT, channel distribution information) and the receiver a fixed antenna radiation pattern for a specific channel type
has perfect CSIR. Following the introduction in [7] and [8] we can only rely on the statistics of the channel as the
of the Kronecker model which describes the transmit and radiation pattern cannot react to instantaneous changes of the
receive correlations as independent elements, [9] and [10] channel. Instantaneous optimization can only be performed
have derived the optimal transmit strategy which maximizes by an adaptive antenna system which changes its radiation
ergodic capacity. In the following, we will describe the adopted pattern(s) during operation. But in our approach we only focus
model and the corresponding optimal input, using the resulting on the synthesis of fixed radiation patterns, which decrease the
beamforming vectors for antenna synthesis. Let the MIMO complexity of the system.
channel be described by [11]: At the transmitter, we use the eigenvectors UT of the transmit
y = Hx + n (5) correlation matrix RT . Keeping the notation of (4), we get
BT = UT . At the receiver, we can consider the channel
where in (5), y is the N × 1 output vector, x is the M × 1 reciprocity principle to choose the eigenvectors UR of RR
input vector, H is the N × M channel matrix and n is a N × 1 as beamforming vectors: BR = U†R .
zero-mean Gaussian noise vector with unit variance. We want We can see that the optimal precoding strategy uses the
to find the optimal input shape which maximizes the ergodic statistically preferred eigendirections of the channel. It is
capacity. Let us define the input covariance matrix Q = E[xx† ] interesting to notice that the receive correlation RR does not
where (·)† denotes the complex conjugate transpose and E[·] affect the transmit directions UT (but it does affect the power
is the expectation operator. The maximum ergodic capacity is allocation if we were using it for a usual transmission scheme)
shown to be [5]: [6] [10]. This allows us to consider the optimal beamforming
   vectors at the transmitter and at the receiver independently for
C = max EH log det IN + HQH† (6)
Q;tr(Q)≤P 2 our antenna synthesis problem.
Additionally, it can also be shown that the antenna synthesis
where tr(Q) ≤ P denotes the power constraint (P being
method developed in [3] is in fact equivalent to the eigenbeam-
the total transmit power) and Q is hermitian non-negative
former method with statistical channel information described
semi-definite. We use the Kronecker model to describe the
in this paper.
correlated channel [9] [10] [12]:
1 1 IV. S IMULATION D ETAILS
H = RR2 Hw RT2 (7)
In this section we use the path based MIMO channel model
where Hw has complex Gaussian i.i.d. entries, and RT and that is described in section II to simulate a channel between
RR are the transmit and receive correlation matrices (which a transmitting base station and a receiving mobile station.
are hermitian non-negative semi-definite matrices) obtained Of course we cannot emulate all of the wave propagation

502
HBF

Path 1 with Į1 and ij1

ĺ ĺ
s(ȍT,p) s(ȍR,p)
1 1
Path p with Įp and ijp
BT BR
AT AR
ĺ ĺ
Mbfc aT,m aR,n Nbfc

eT,m(ȍ) ST Path Np SR eR,n(ȍ)


Locations of M Element Radiation Directions of Directions of Element Radiation Locations of N
Transmitting Pattern of M Departure (DOD) Arrival (DOA) Pattern of N Receiving
Antennas Tx Antennas of Np Paths of Np Paths Rx Antennas Antennas

Fig. 1. Path based MIMO channel model

effects of such a mobile channel in detail but we choose the gle ball shaped beam according to CT (θ, ψ) = sin(θ) sin(ψ)
available channel parameters of our channel model to resemble pointing at horizon (θ = 90◦ ,ψ = 0◦ ) with 7.7 dBi gain.
such a setup. The simulations are performed for a center The receiver antenna array is a uniform circular array (UCA)
frequency of 2.5 GHz and a total of Ncr = 3000 narrowband with an element distance of λ3 and an additional antenna
channel realizations. Each channel realization has a total of element in the middle of the UCA. It has seven antennas on
Np = 6 propagation paths divided in two path bundles. The the circle and the eighth antenna is in the middle to get N = 8
three paths within a path bundle have correlated DOD at the antenna elements in total. The antenna element radiation
transmitter and correlated DOA at the receiver and have a patterns at the receiver exhibit a doughnut shape according
standard deviation between 1◦ and 5◦ from the main direction to CR (θ, ψ) = sin(θ) with its nulls at zenith (θ = 0◦ ) and
of the bundle. towards ground (θ = 180◦ ) and a gain of 1.7 dBi. Coupling
The stationary base station is assumed to be mounted at an between antenna array elements is not considered.
elevated position and serves a sector of 120◦ in azimuth.
The mobile station is assumed to move through an urban-like V. S YNTHESIS R ESULTS
scenario that causes propagation paths to impinge from the The applied channel model allows direct access to the DOD
whole azimuth range of 360◦ . The scenarios are comparable ΩT and to the DOA ΩR of the propagation paths together
to the measured scenarios in [3]. For these assumptions we with the power transmission factor α2 of each path. Hence
set up a Monte Carlo simulation method with the parameter it is possible to calculate an angular power spectrum (APS).
distibution ranges from Table I. All parameter values are drawn Fig. 2 depicts the incoherent APS of the DOA at the receiver
for all Np paths and for the complete channel simulation of
TABLE I all Ncr channel realizations. It was calculated as the kernel
PARAMETER DISTIBUTION RANGES OF THE CHANNEL SIMULATION density estimation (KDE) with a Gaussian kernel, weighting
Description Parameter Min. value Max. value α2 , estimation bandwidth 0.02 and normalized afterwards.
Path Amplitude αp −119 dB −113 dB
Path Phase ϕp 0 2π
Azimuth DOD ψT −60◦ 60◦
Elevation DOD θT 90◦ 112.5◦
Azimuth DOA ψR 0◦ 360◦
Elevation DOA θR 67.5◦ 90◦

from a uniform distribution and filtered with a Gauss window


of 31 taps after generation to achieve a correlation of adjacent
parameter samples. A more sophisticated channel simulation
based on raytracing together with an application of the antenna
synthesis can be found in [2].
The transmitter antenna array is a uniform linear array (ULA)
with M = 8 elements and an element distance of 3λ. The an- Fig. 2. Incoherent angular power spectrum at the receiver for all simulated
tenna element radiation patterns at the transmitter exhibit a sin- propagation paths

503
The APS reveals that the paths are within the expected DOA can afterwards be applied to more complex channel models or
distribution ranges of Table I. The APS distribution exhibits channel measurements to achieve antenna synthesis results for
a global maximum at (θ = 78◦ , ψ = 153◦ ) and several specific MIMO antenna design problems.
local maxima in the azimuth range between ψ = 193◦ and The eigenbeamformer synthesis method was applied to the
ψ = 268◦ . simulated channels. It is the optimal solution of the con-
Fig. 3 shows the first two radiation patterns that were syn- sidered antenna synthesis problem that maximizes ergodic
thesized with the eigenbeamformer method described in sec- channel capacity. It matches fixed antenna radiation patterns
tion III. The first pattern C1 (θ, ψ) has its main beam direction to the channel when relying on the assumption of statistical
channel knowledge. The synthesis results proved to cover the
maximum regions of the APS very well with the resulting
fixed synthesized radiation patterns. That verifies the matching
of the synthesized radiation patterns to the channels under
investigation that omnidirectional radiation patterns of mobile
antennas do not possess.
ACKNOWLEDGMENT
The authors would like to thank the German Research
Foundation (DFG) for funding this work in the context of the
project OASE.
R EFERENCES
[1] T. Mahler, L. Reichardt, C. Heine, M. Pauli, and T. Zwick. Channel based
design of systems with multiple antennas. Progress In Electromagnetics
Research B, Vol. 64:pages 63–81, 2015.
[2] L. Reichardt, J. Maurer, T. Fugen and T. Zwick, ”Virtual Drive: A Com-
plete V2X Communication and Radar System Simulator for Optimization
of Multiple Antenna Systems,” in Proceedings of the IEEE, vol. 99, no.
7, pp. 1295-1310, July 2011.
[3] T. Mahler, J. Kowalewski, B. Nuss, C. Richt, J. Mayer and T. Zwick.
Channel Measurement Based Antenna Synthesis for Mobile Automotive
MIMO Communication Systems. Progress In Electromagnetics Research
Fig. 3. Directivities of synthesized radiation patterns C1 (θ, ψ) and C2 (θ, ψ)
B, Vol. 72, pages 1-16, 2016
[4] G. J. Foschini and M. J. Gans, On limits of wireless communication in
at (θ = 88◦ , ψ = 152◦ ) and 10.1 dBi directivity and the second a fading environment when using multiple antennas. Wireless Personal
pattern C2 (θ, ψ) has its main beam direction at (θ = 90◦ , Commun., vol. 6, no. 3:pages 311–335, Mar. 1998.
[5] I. E. Telatar, Capacity of Multi-antenna Gaussian Channels. Eur. Trans.
ψ = 218◦ ) and 9.8 dBi directivity. When comparing the two Telecomm. ETT, Vol. 10, no. 6:pages 585–596, 1999.
synthesized radiation patterns C1 (θ, ψ) and C2 (θ, ψ) we note a [6] E. Biglieri, R. Calderbank, A. Constantinides, A. Goldsmith, A. Paulraj,
complementary occurrence of main beam directions and nulls. H. V. Poor, MIMO Wireless Communications, 1st ed., Cambridge
University Press, 2007.
This is induced by the orthogonality of the eigendecomposition [7] D. Shiu, G. J. Foschini, M. J. Gans, and J. M. Kahn, Fading correlation
included in the eigenbeamformer. and its effect on the capacity of multi-element antenna systems. IEEE
Trans. Commun., pages 502–513, 2000.
[8] C. N. Chuah, D. N. C. Tse, and J. M. Kahn, Capacity scaling in MIMO
VI. C ONCLUSION wireless systems under correlated fading. IEEE Trans. Inform. Theory,
vol. 48:pages 637–650, 2002.
The proposed channel model allows to choose the MPCs [9] S. A. Jafar, S. Vishwanath, and A. Goldsmith, Channel capacity and
arbitrarily by feeding the path parameters with the corre- beamforming for multiple transmit and receive antennas with covariance
sponding distributions. That leads to channel simulations that feedback. IEEE International Conference on Communications, vol.
7:pages 2266–2270, 2001.
do not reproduce the reality of wave propagation as detailed [10] E. A. Jorswieck and H. Boche, Channel capacity and capacity-range
as e.g. a channel simulation based on raytracing, but at the of beamforming in MIMO wireless systems under correlated fading with
same time it keeps simulation complexity at a minimum and covariance feedback. IEEE Transactions on Wireless Communications,
vol. 3, no. 5:pages 1543–1553, Sept. 2004.
allows arbitrary DOD and DOA distributions for testing and [11] G. V. H. Tsoulos, Ed., MIMO system technology for wireless commu-
analyzing. This is beneficial for the development and design nications, ser. Electrical engineering & applied signal processing series.
of antenna synthesis methods. Furthermore it could be shown Boca Raton, Fla. [u.a.]: CRC Press, 2006.
[12] A. Forenza, D. J. Love and R. W. Heath, Simplified Spatial Correlation
that the proposed channel model is very well suited to visualize Models for Clustered MIMO Channels With Different Array Configura-
and analyze the spatial channel matching effects of the antenna tions. IEEE Transactions on Vehicular Technology, vol. 56, no. 4:pages
synthesis procedure. 1924–1934, July 2007.
The synthesis results presented in this paper depend on the
exemplary channel parameters used for simulation and are not
considered to match a mobile channel in general. The path
based MIMO channel model rather serves as a testbed for
the development of antenna synthesis methods. These methods

504

You might also like