IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 24, NO. 7, JULY 2006 1401
Power Line Enhanced Cooperative
Wireless Communications
Marc Kuhn, Stefan Berger, Ingmar Hammerström, and Armin Wittneben
Abstract—In this paper, we investigate the use of power line
communication (PLC) to assist cooperative wireless relaying. We
consider a communication scheme that uses the power line to
initialize and synchronize wireless amplify-and-forward relays
and to broadcast information between the relays. Starting from
an analysis of transfer functions and noise measurements of
PLC channels in office and residential environments, we propose a
power line transmission scheme for the inter-relay-communication
and assess the influence of this scheme on wireless relaying. This
scheme is based on linear precoded orthogonal frequency-division
multiplexing; it is designed to optimally exploit the frequency di-
versity available on PLC channels. The use of PLC leads to a very
flexible way of enhancing wireless communications by plugging
in additional relays where they are needed—without additional
wiring.
Index Terms—Cooperative wireless relaying, linear precoding,
multiple-input–multiple-output (MIMO) systems, power line com-
munication (PLC), precoded orthogonal frequency-division multi-
plexing (OFDM).
I. INTRODUCTION
P
OWER LINE COMMUNICATION (PLC) offers the pos-
sibility to use the well-developed infrastructure of the elec-
trical energy distribution grid for data transmission. For the time
being, there is no harmonized international standard for broad-
band PLC [1]. But IEEE started standardization of PLC phys-
ical and MAC layer in June 2005. In Europe, broadband PLC
is limited to frequencies between 1 and 30 MHz, because of re-
strictions regarding electromagnetic compatibility (EMC). Fu-
ture communication systems are expected to use much higher
data rates as today’s wireless local area networks (WLANs) [2].
In this paper, we study an approach to boost high data rate wire-
less communications by using existing power lines in a flexible
and cost-efficient way.
In wireless networks, spatial diversity and spatial mul-
tiplexing gains are achieved by multiple antennas at the
transmitter and at the receiver. Using
cooperative relaying
strategies [3]–[7] these gains are also possible for single-an-
tenna nodes. Spatial multiplexing is mandatory to achieve the
high bandwidth efficiency that is necessary for future Gigabit/s
wireless communication systems [2]. Practical cooperative
Manuscript received April, 18, 2005; revised December 15, 2005 and Feb-
ruary 20, 2006. This paper was presented in part at the Sixth IEEE Workshop
on Signal Processing Advances in Wireless Communications, New York, June
5–8, 2005.
The authors are with the Communication Technology Laboratory, Swiss
Federal Institute of Technology (ETH) Zurich, CH-8092 Zurich, Switzerland
(e-mail: kuhn@nari.ee.ethz.ch; berger@nari.ee.ethz.ch; hammerstroem@nari.
ee.ethz.ch; wittneben@nari.ee.ethz.ch).
Digital Object Identifier 10.1109/JSAC.2006.874407
Fig. 1. Scenario: wireless MU-ZFR assisted by PLC between the relays.
relaying schemes for spatial multiplexing gains usually need
the exchange of information between the relays.
In this paper, we consider a cooperative relaying scheme with
fixed infrastructure amplify-and-forward (AF) relays (support
nodes) connected to the power grid for power supply. We study
the possibility to use PLC for the considerably high signaling
overhead between the wireless relays in such scenarios; some
basic ideas to this were first published in [8]. In contrast to our
approach, the authors in [9] study the use of cooperative com-
munication for PLC.
The considered scenario is shown in Fig. 1: a room (e.g.,
a conference room) with nodes that need high-speed wireless
data communication; these nodes are mobile, but the velocity
is low (less than 5 km/h); an example for such a scenario is
a high-speed WLAN. Fixed infrastructure AF relays assist the
communication between the wireless nodes (
: number of re-
lays). The nodes are equipped with a single antenna. We con-
sider two different traffic patterns.
1) Ad Hoc WLAN: The wireless nodes form an ad hoc net-
work using orthogonal frequency-division multiplexing
(OFDM) under a two-hop relaying scheme. We assume
that the wireless nodes can be divided into
source/
destination pairs; the transmission from a source node
to its associated destination node includes two channel
uses: one for the uplink transmission from the source to
all relays and one for the downlink transmission when
each relay broadcasts an amplified (but not decoded)
version of its received signal to the destination nodes.
0733-8716/$20.00 © 2006 IEEE
1402 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 24, NO. 7, JULY 2006
2) Infrastructure-based WLAN: One relay is connected to a
high data rate backhaul [e.g., wide-area network (WAN),
LAN etc., see Fig. 1]. This relay acts as access point
(AP) for the wireless nodes. The transmission between
the AP and a particular wireless node is again done as de-
scribed in 1): The AP relay is considered as a new wire-
less source node and the other relays assist in a two-hop
relaying scheme.
In this paper, the AF relay gains are assigned such that the
interference between different source/destination links is nulled
by coherent combining the broadcasted signals; we refer to
this scheme as multiuser zero-forcing relaying (MU-ZFR) [7],
[10]. This essentially realizes a distributed spatial multiplexing
gain with single-antenna nodes, enabling high data rates. Using
this scheme,
source/destination pairs are orthogonalized
in the spatial dimension; if more wireless nodes are involved,
MU-ZFR can be combined with an additional multiple-ac-
cess scheme, e.g., time-division multiple-access (TDMA).
For MU-ZFR, all relays have to be synchronous and in an
initialization phase every relay has to broadcast its channel
state information (CSI) regarding all wireless nodes (uplink
and downlink CSI) to all other relays; in [10] is shown, how
this information can be acquired. The CSI has to be updated
from time to time, because otherwise, it becomes outdated. In
the following, we investigate if synchronization and initializa-
tion/updating of the wireless relays can be done using PLC.
We assume that every relay is supplied by a power outlet, and
therefore the power line can be used for communication between
the relays (PLC backbone). In addition, it is also possible to con-
nect one or more relays to the high data rate backhaul over the
power line (see Fig. 1); these additional APs do not have a fixed
infrastructure-based connection to the backhaul, i.e., high data
rate PLC has to be used. The wireless source and destination
nodes are not necessarily connected to the power line and com-
municate only over the wireless medium; they do not cooperate,
e.g., there is no joint decoding.
The analysis of the PLC backbone is based on extensive mea-
surements of the transfer function and the noise of the PLC
channels—for office and residential environments. The channel
capacity of these indoor PLC channels is determined at frequen-
cies between 1 and 30 MHz. A PLC transmission scheme is pro-
posed and the expected jitters are analyzed, because they deter-
mine the suitability of PLC for the relay synchronization.
This paper is outlined as follows. In Section II, the results
of the PLC measurement campaign and the investigation of the
power line channel capacity are given. In Section III, we de-
scribe MU-ZFR. In Section IV, a PLC backbone for wireless
multiuser zero-forcing is studied; we propose a PLC transmis-
sion scheme that is optimized regarding the properties of PLC
channels and address the issue of relay synchronization over
power line channels.
Notation: The operators
, , and denote the
Hadamard (element-wise) product, expectation with respect to
x, and conjugate complex transpose, respectively. The terms
, , and denote the element , the th
column, and the
th row of a matrix , respectively. The oper-
ator
has two meanings: When the argument is a matrix,
it takes the diagonal elements and puts them into a column
Fig. 2. Transfer functions of three measured PLC channels.
vector. When the argument is a vector, it puts the elements of
the vector into a diagonal matrix.
II. PLC M
EASUREMENTS
To determine the characteristic properties of indoor
low-voltage system PLC channels, measured transfer functions
and measured noise power density spectrum (PDS) of the chan-
nels are used. Transfer functions are measured by a network
analyzer, noise PDS by a spectrum analyzer. Couplers are used
to connect the measurement devices to the power line. Addi-
tional information about PLC measurements and the properties
of PLC channels can be found, e.g., in [1] and [12]–[18]. In the
following, we review some properties of indoor PLC channels
that are needed in the remaining part of this paper.
In our measurements, the transfer functions show high dif-
ferences regarding the frequency-selectivity and the average at-
tenuation—which are typical for indoor PLC channels (see the
above mentioned references). Fig. 2 shows three examples of
measured PLC transfer functions, roughly classified according
to the average attenuation in the categories “good,”“average,”
and “below average.” The PLC channels in one room (or in ad-
jacent rooms) usually belong to the categories “good” or “av-
erage.” In our measurement campaign, “below average” PLC
channels are typically found in case of connections between not
adjacent rooms; therefore, they do not match the scenario shown
in Fig. 1, and we will treat them as worst case situations for our
considerations. The characteristics of PLC transfer functions de-
pend on different factors: the cable length of the power line be-
tween the two power outlets that define entry and exit point of
the PLC channel, the included phase conductors and fuse cir-
cuits, and multipath propagation because of reflections (Fig. 3).
Such reflections are generated, e.g., at open-ended power out-
lets or at devices connected to the power line with their loads not
matched to the frequency-dependent impedance of the power
line network (cf. [17]). Fig. 4 shows an example of the varia-
tions within 8 h of the amplitude spectrum of a measured PLC
transfer functions. According to our measurements, the transfer
KUHN et al.: POWER LINE ENHANCED COOPERATIVE WIRELESS COMMUNICATIONS 1403
Fig. 3. Impulse response of a PLC channel (category “good”).
Fig. 4. Variations over time (8 h) of the amplitude spectrum of a PLC transfer
function.
functions of PLC channels vary only slowly over time except
for modifications of the power line topology next to the con-
sidered PLC channel, e.g., in case a device is plugged in. In
many cases, PLC channels seem even to be quasi-static, usu-
ally their frequency response is at least less variant than the
transfer function of a wireless channel between mobile nodes.
But in literature (e.g., [19]), cases are reported where PLC chan-
nels are highly time-variant, in particular, when devices with
time-variant impedances are in the neighborhood of the consid-
ered PLC channel.
In Figs. 5 and 6, noise measurements for PLC channels are
shown. The use of electrical devices is one reason for noise
at the power line (Fig. 5 shows the influence of a dimmer as
an example). Other reasons are narrowband interferers—e.g.,
(medium/short wave) radio transmitter and radio stations. Curve
2 in the lower part of Fig. 5 shows a typical noise PDS of a PLC
channel. Distinct narrowband interferences can be found.
In Fig. 6, the variations of the PDS within 24 h is shown.
All in all, it can be seen that transfer function and PDS of a
Fig. 5. Part of periodical time function of a dimmer and measured noise power
delay spectrum of a PLC channel with a dimmer (1) and without (2).
Fig. 6. Variations over time (24 h) of a measured noise PDS of a PLC channel.
PLC channel are frequency-selective, strongly depending on the
location and vary only slowly over time; in addition, impulsive
noise can be expected (upper part of Fig. 5). In the following, we
model the influence of impulsive noise by its spectral behavior.
A. Channel Capacity
In Europe, the maximum permitted radiation for unshielded
cables is restricted by standards (because of EMC reasons). In
this paper, a low transmit power density of
Hz (constant for the considered bandwidth ) is assumed,
that meets—according to [12]—the regulations of the NB 30
[22] specified by German national regulatory authorities and,
e.g., also applied in Switzerland to limit the interference of PLC
devices on other services [23]. This power density corresponds
to a transmit power of 8 mW at 50
and a bandwidth of
MHz (1 MHz 30 MHz).
To approximate the capacity of a channel, a measured transfer
function and a measured noise PDS are used. The channel is di-
vided into
narrowband flat-fading subchannels of bandwidth
1404 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 24, NO. 7, JULY 2006
Fig. 7. Measured PLC channels: cumulative distribution function of channel
capacity for the different categories; dashed-dotted: without water-filling, uni-
formly distributed transmit power; dashed: all PLC channels regardless of cat-
egory.
, where is the number of samples of the mea-
sured transfer function
and of the measured noise PDS
(1)
The noise of each subchannel
is approximated as additive
white Gaussian noise (AWGN) of variance
. The samples of the power density of the transmitted signal
are found using water-pouring.
Fig. 7 shows the cumulative distribution functions (CDFs)
of the channel capacity of 430 measured PLC channels. The
dashed curve shows the CDF for all measured channels, the
other three curves show the CDFs of the three categories for the
transfer functions. Dashed-dotted curves are showing the CDF
for the cases where water-pouring is not used but the transmit
power is uniformly distributed over frequency. A (small) differ-
ence can only be seen in the case of the “below average” PLC
channels, because of the low signal-to-noise ratio (SNR) for this
channel category. Obviously, channel capacities of more than
400 Mbit/s are not rare even at the low transmit power density
that is considered. More than 70% of all measured PLC chan-
nels have a capacity of more than 200 Mbit/s (but it should be
noted that a large share of the channels are characterized by short
distances because they are measured in the same room or in ad-
jacent rooms; this may explain the lower capacity values found
in some other published measurement campaigns); even in the
category “below average” there are only 3% of the channels with
less than 38 Mbit/s capacity.
III. W
IRELESS MULTIUSER ZERO-FORCING RELAYING
(MU-ZFR )
Shortly, we review the MU-ZFR scheme presented in [7] and
show its robustness with respect to noisy CSI and phase noise
at the local oscillators. To this end, consider Fig. 1. It depicts a
Fig. 8. MU-ZFR equivalent channel model.
scenario where source/destination pairs shall commu-
nicate with the assistance of
AF (also called “nonregen-
erative”) relays. All nodes in the network are assumed to utilize
a single-antenna only. The communication follows a two-hop
relay traffic pattern, i.e., each transmission cycle includes two
time slots: In the first time slot, all sources transmit their data
to the relays. There, the samples are scaled and rotated, i.e.,
multiplied with complex gain factors. In the second time slot,
the relays forward their signals to all destinations. The com-
plex gain factors at the relaying terminals are calculated such
that the transmit data of source node
is received by destina-
tion
without interference from other sources (MU-ZFR). Con-
sider the equivalent system model in Fig. 8. For one OFDM
subcarrier, the transmit symbols of all sources are stacked in
the vector
, and the received symbols at all desti-
nations in the vector
. The equivalent channel ma-
trix
is the concatenation of the source-relay
channel matrix
, the relay gain factors, and the relay-desti-
nation channel matrix
. Finally, the vector con-
tains the noise that is present at the destinations. It consists of
the noise at the relays, which is transmitted to the destinations,
as well as the noise contributions at the destinations themselves.
Let
denote the vector of channel coefficients from the
source
to all relays and the vector of channel coefficients
from all relays to destination
. The entries of the equivalent
channel matrix can be written as
(2)
where the vector
contains the complex-valued relay gain
factors. We define a matrix of interference coefficients with its
columns according to
(3)
The interference between different source/destination links is
nulled, if the relay gain vector
satisfies [7] . In this
case, the equivalent channel matrix
becomes diagonal. In
order to find such a gain vector, we choose an initial gain vector
which would orthogonalize the links
for asymptotic number of relays [7] and project it onto the null
space of
(4)
At least
relays are needed in order to perform
the nullspace projection (4), because the nullspace of
might
KUHN et al.: POWER LINE ENHANCED COOPERATIVE WIRELESS COMMUNICATIONS 1405
be empty for . We denote the case that
, which consequently is the minimum
number of relays needed to completely orthogonalize all links,
by minimum relay configuration. In a practical system, where
the number of relays is less than that, a scheduling algorithm
could orthogonalize sets of
source/destination pairs in time. The relays would then be able
to orthogonalize each of the sets at a time. Compared with the
case where enough relays are present to orthogonalize all links
simultaneously, the sum rate would decrease because more time
slots are needed for one transmission cycle.
In order to find their zero-forcing gain factors locally, each
relay needs to know
and perfectly, i.e., it needs to
have global channel knowledge. In addition to that, all relays
have to be phase synchronous in order to accomplish coherent
combining at the destinations.
MU-ZFR allows all source/destination pairs to communicate
in parallel over the same physical channel, which essentially
realizes a distributed spatial multiplexing gain. As a measure
of performance of the present system, we choose the average
sum rate, which is given by
(5)
where
denotes the instantaneous signal-to-interfer-
ence-noise ratio (SINR) of source/destination pair
for a given
channel realization. The factor 1/2 accounts for the fact that a
transmission occupies two channel uses as it needs two time
slots.
In the following, we show the performance of MU-ZFR in
terms of average achievable rates in bits per second per Hertz.
Further, we discuss the robustness of MU-ZFR with respect to
two main error sources: 1) noisy CSI due to channel estima-
tion and quantization of the channel estimates before feedback
and 2) the influence of errors in the phase of the local oscil-
lators of the distributed relays due to phase noise. From this,
we can derive requirements to the PLC backbone. Fig. 9 shows
the average sum rate
of a minimum relay configuration
versus the number of source/destination pairs. The parameter
of the curves is the average SNR. In order to have a defined
average SNR at each destination, we consider a reference sce-
nario with a single source/destination pair and only one relay
in between. From this, we determine the transmit power
of
source and relay which is needed to achieve this defined ref-
erence SNR. For the MU-ZFR scenario, we apply this transmit
power to the configuration that is to be evaluated. Thus, each
source has transmit power
, whereas all relays share the sum
transmit power of
. Further, it is assumed that relays and
destinations have same noise variances. For the simulations, we
assumed frequency flat Rayleigh fading on every single source-
relay and relay-destination link. All channel coefficients are sta-
tistically independent and drawn from a complex normal dis-
tribution with zero-mean and variance 1. The channel matrices
and are constant during each transmission cycle
Fig. 9. Average sum rate versus number of source/destination pairs for perfect
CSI and synchronization (solid lines) and estimated CSI with processing gain of
10 dB and Gaussian distributed phase jitter of 2
rms (dashed lines). Parameter
of the curve is the defined SNR.
(block fading) and temporally independent. All relays are as-
sumed to exhibit the same noise variance, as do all the destina-
tions.
The potential of MU-ZFR within a distributed network can
be easily seen in Fig. 9. For perfect CSI and synchronization
(solid lines), the average sum rate increases linearly with the
number of source/destination pairs which exhibit the achieved
distributed spatial multiplexing gain. However, imperfections
due to noisy CSI and synchronization have the effect that the
sum rate no longer increases linearly but saturates (dashed lines:
exemplary values for synchronization and channel estimation
imperfections) [11].
To visualize the effect of noisy CSI and synchronization er-
rors, we plot the average rate per link versus imperfections for
source/destination pairs. In Fig. 10, the average per link
rate versus
(excess SNR of the channel estimate rela-
tive to the defined reference SNR) is shown. Parameter of the
curves is the reference SNR. The error is modeled by a complex
Gaussian noise added to the exact channel matrices. It can be
seen that the rates are nearly unaffected by the noisy CSI for a
larger than 10 dB. Already for an excess SNR of 5 dB,
the loss in spectral efficiency is quite small.
Fig. 11 depicts the influence of an additive Gaussian error
of the reference phase at all relays on the average rate per link
for
source/destination pairs. It can be seen that for in-
creasing reference SNR the negative influence of the phase jitter
also increases. The phase jitter causes a residual noise variance
at the destinations which limits the average rates.
IV. PLC B
ACKBONE FOR WIRELESS RELAYING
A. Wireless MU-ZFR: Relay Initialization and Updates of CSI
In the initialization phase of wireless MU-ZFR, every relay
broadcasts its CSI regarding the wireless source/destination
nodes to all other relays using the power line. Each of the
relays has to estimate complex channel taps per OFDM
