Indoor Power-Line Communications: Channel Modelling

and Measurements
F. J ~ a ~ e t e - ~ o r r i pL.
i oDiez-del
', Rio, J. T. Entrambasaguas-Muiioz
Dpt. Ingenieria d e Comunicaciones. E.T.S.I.Telecomunicaci6n.
Universidad de Mhlaga.
Campus d e Teatinos. 2907 1 Mhlaga, Spain.
' Fax: +34 952132027. Phone: +34 952137121. E-mail: francis@ic.uma.es.

Abstract.- The subject of this paper is to analyse the low voltage distribution lines inside the consumer
premises, in order to attain a channel model useful to simulate its behaviour for high bit-rate digital
communications. Measurements have also been performed both in the laboratory and in residential buildings,
and results are presented and discussed. Finally, some estimations of the channel capacity have been carried out.
Key words.- Power line communication, high bit rate, channel modelling, impulse response, frequency
response, noise, measurements, channel capacity, OFDM, DMT.

1. Introduction
So far low voltage power-line communication systems have been mainly concerned for low data rate
transmission (e.g. standards like CEBus). In fact, there are regulations that restrict severely the frequency range,
in both Europe (EN 50065 [I]) and other parts of the world. Two kinds of applications are being developed in
this field: 'last mile' services, or access to large telecommunication networks; and indoor services, like the
installation of Local Area Networks (LAN) at ofices or homes. Due to limits of the channel capacity, an
extended frequency band should be released for transmission at higher rates.

Power lines were designed, obviously, to deliver electrical energy and not to transmit communication
signals, and it makes the conditions in the channel quite adverse. The channel parameters (i.e. amplitude and
phase responses, input and output impedances, or noise sources) depend on the particular location characteristics
under study: complex and often unknown topology (with many branches), kind of wires involved, heterogeneous
loads, etc. These parameters also exhibit variations with frequency, position and time (as electrical appliances are
switched onloff). However, changes are presumed to be much slower than the data rate, so the assumption of
time-invariance during several symbols is justified.

In the preceding literature on the topic, the hostile nature of this communication medium has been studied
[2]. Recently, attention has been turned to the high frequency range [3][4], highlighting different sources of
disturbances (broad band, like impulse noise, background noise, or the one produced by switching devices; and
narrow band: radio interferences, noise synchronous to the power frequency...), and signal degradation (from
impedance mismatch and discontinuities, branch cables...). All these difficulties can lead to model this channel in
a 'black-box' approach, extracting a few system parameters from extensive measurement results [5]. However,
in this work, a method to obtain channel frequency response from the physical characteristics of the network
(cables layout, loads connected, transmission distance, etc) is presented. This procedure, despite its complexity
and once its adequacy had been accomplished, would permit to produce channel models automatically without
the necessity of measuring. Although, in order to adjust the model performance, measurements of channels
responses, existing noise, and load impedances, have to be carried out.

2. Channel model
As mentioned above, in this paper a procedure to obtain the channel response from the power network
characteristics and topology, is proposed. It is based on considering power lines as a structure of transmission
lines, with multiple branched lines, which may be loaded or not, representing the different sockets placed in the
rooms. For infomation transmission purposes the link is usually established connecting the transmitter and
receiver subsystems between the line and neutral conductors of the power grid. Data from manufacturers of
typical cables used in indoor power wiring (with several sections for the different electrical circuits employed)
have been adopted to find the secondary parameters (R, L, C, G per unit length) of the elementary transmission
line considered: two parallel wires. The loads have been modelled as complex impedance vectors in frequency
(values estimated from measurements: TV7s,PC's, lamps. .. input impedance from the mains plug at frequencies
up to 30MHz). In case that no device were plugged, an open circuit is assumed.

It is necessary to use a coupling circuit to prevent the communication equipment from damage when
connecting it to the electrical wiring to launch or receive the information signal. In the model, the effect of the
introduction of this adaptation interface has to be added to the transmitter output impedance and receiver input
impedance, to reflect the mismatch losses.

time (us)

Figure 1: Impulse response of modelled channel (normalised).

From real electrical schemes of several residences, approximate models for channel simulations have been
created. In Fig. 1, impulse response for a model designed from an apartment power network is shown (two
sockets of the same electrical circuit in different rooms have been selected to form the link, and the distance
could be 15-20m approximately). Echoes appeared, due to the multiple reflections experimented by the signal as
it travels through the network (because of wave propagation phenomena produced by impedance misrpatch and
branches). The corresponding amplitude frequency response is presented in Fig. 2, where notches at certain
frequencies are observed, in which phase exhibits nonlinearity (see Fig. 3). Between them, there are bands with
nearly flat response that could be combined for reliable transmission. Although there is a high attenuation l e ~ e l
(present even in short-distance links). In general, the more branched and loaded the network, the longer the
impulse response.

10
modelled channel I
measured channel
: Y

-70
0 5 10 15 20 25 30 0 5 10 15 20 25
frequency (MHz) frequency (MHz)

Figure 2: Example of channel amplitude responses Figure 3: Example of channel phase responses obtained
obtained from the model and measurements. from the model and measurements.

The procedure validity is studied contrasting its results with actual measurements, as seen in Fig. 2-3
(measured impulse response has not been displayed in Fig. 1 to make clearer the plot, as it is similar to the
modelled one). It is not sensible to expect a perfect fit, since the model has been constructed without an exact
knowledge of the real layout. But certainly, the drafts bear rather resemblance. Anyway, the purpose of the
model is not to match with a certain channel configuration; it is to serve as a generator of feasible channel
models,usefil to prove diverse transmission techniques.

3. Channel measurements
In order to verify the simulations results and to get information about the medium, measurements of noise
and channel transfer functions have been performed for different indoor power line channels, under various
and loads. A network analyser plus two coupling circuits (essentially high-pass filters and
in the transmitter and receiver side respectively, have been employed to make the measurements.
The access points to the electrical wiring (distant sockets) were reached with the help of cbhial cables, whose
influence (and that of the coupling units) was contemplated in the measure procedure. The frequency range has
been extended up to 30MHz.

The channel response depends on the electrical path that signal follows, and especially on the location of the
branched lines. It is common to find shorter links with higher attenuation level than longer ones. For example
when transmitter and receiver are not on the same electrical circuit, because the main signal path includes the
distribution box, where all the circuits are connected in parallel, and important impedance mismatch (causing
reflections) and power distribution (and dissipation) among them occurs. According to the model results, the
transfer functions measured display some nulls in the amplitude responses at certain frequencies. In Fig. 4 the
attenuation of the frequency response for three measured channels in a residential building are plotted. The first
exhibit low, but not uniform, attenuation, despite the link was established over the same electrical circuit and the
distance was short (no more than 5-6m.). The second line correspond to another configuration over the same
circuit but with more distance (about 15-20m.), the attenuation registered is, evidently, higher, and more
frequency variant. Finally, a result for a little shorter distance, but between different circuits is drafted,
presenting a worse behaviour. On the right, the plot shows the measurement results for three arrangements of
plugged devices in the laboratory, but keeping the transmitter and receiver f ~ e d It . is clear the significant
influence in channel amplitude response both of location and, as observed in this figure, of loads (the
transmission distance in this case was approx. 15-20m).

-701
0 5 10 15 ZU 25
I
30
frequency (MHz) hequency (MHz)

Figure 4: Amplitude response for three different channels Figure 5: Amplitude response for a power line channel
measured in an apartment. Transmitter and receiver were measured in the laboratory with different configurations
located in different electrical circuits and positions. of load.

Noise power spectra have also been measured in different environments. he plots usually (but not always)
show a descending spectral power density as frequency increases. These facts suggest that the system should
avoid both the low frequency band, and the amplitude notches. In Fig. 6 an example of the recorded data in the
laboratory is depicted, whereas in the next figure, the results obtained from an apartment in the same socket but
in a repose state (config. I), and a noisy active state (config. 2) are presented. In Fig. 7 config. 1 no remarkable
appliances were connected to the power network (except the fridge, some incandescent lights...), and this is why
the noise level is quite low. Although some narrow band disturbances are observed, possibly caused by radio
interferences (e.g. peaks appeared when activating a cordless telephone). On the contrary in config. 2, additional
 devices were plugged in he vicinity (a microwave oven the most important), and the registered noise is
drastically increased. L

-120 1
0 5 10 15 a0 25 30
I -1201
0 5 10 15 20 25 30
frequency (MHz) frequency (MHz)

Figure 6: Noise measured in the laboratory Figure 7: Noise measured in an apartment in the same
situation, but in config.2 additional electrical appliances
were turn on.

4. Channel capacity analysis

Channel capacity is the upper bound in digital transmission, hence models and measurements results are
used to estimate the spectral efficiency (b1sfI-I~)attainable. The signal launched into the power network by the
transmitter is supposed to have a flat power spectral density of -20dBm/kHz (a not excessive value, but higher
than allowed by current regulations [6]). Three scenarios have been chosen to illustrate power line
communications capability; data in all cases are extracted from measurements.

0 5 10 15 20 25
frequency (MHz) ' frequency (MHz)

Figure 8: Channel signal to noise ratio for a laboratory Figure 9: Respective channel capacity estimation.
measurement.

The Shannon theorem has been applied to calculate the maximum ideal capacity, along with a more practical
approach, which is to use an Orthogonal Frequency Division Multiplexing (OFDM) scheme for transmission,
also known as Discrete Multitone (DMT). OFDM is a good technique over spectrally shaped channels [7]. It is
based on partitioning the data rate before digitally modulating (in general with QAM) several carriers in narrow
subbands (in which the channel characteristics can be considered uniform), which are then multiplexed to create
an OFDM symbol. A convenient number for the studied bandwidth is 512 subchannels. In order to avoid
intersymbol interference, a cyclic prefix (whose length depends on the channel impulse response duration)
 ~houldbe added, what reduces the overall capacity. An error probability of 10'~has been assumed for the
and the channel impulse response duration has been estimated to contain 99% of the total energy.
4
i
First results are depicted for the channel at the laboratory (the 3rdconfiguration in Fig. 5, and noise in Fig.
6). The signal to noise ratio measured is presented in Fig. 8, and the corresponding channel capacity in Fig. 9.
The decay of noise with increasing frequencies suggests a higher bit load in this part of the band. Moreover, for
s
efficient transmission, the frequency notches have to be removed. The averaged spectral efficiency with this
procedure is 2.2 b/s/Hz (with Shannon limit is 4.4b/s/Hz), resulting in a total capacity of more than 60MbIs.
From apartment measurements the channel capacity is analysed for the same configuration (Fig. 2), but with
r two different noise environments: best case (with noise of config. 1 in Fig. 7) and worst case (with noise of
config. 2 in Fig. 7). Both of them involved a transmission distance of about 15-20m. In the first channel
conditions, no appliances were connected to the power line, and that implies more favourable conditions for
signal transmission as observed in Fig. 10. Hence a higher spectral efficiency is registered (Fig. 11) and there are
no empty bands. The averaged value is 6.2 bls/Hz using OFDM (9.1 bIs1H.z in the ideal case), which means a
total capacity of almost 200Mbls. In contrast, the second channel studied has to face strong disturbances
produced by several loads, and this is why the signal to noise ratio measured is much lower, Fig. 12, (becoming
negative for some frequencies). Consequently, the channel capacity is hardly 45Mbls (or 1.5 b/s/Hz, and
Shannon limit equals to 3.5 b/s/Hz).

--------.--------
; - - . ; -* ---- ----:
5'1 --------:
; . -;.
- - - - - ---.: ----- ---

101 0
0 5 10 15 20 25 JO 0 5 10 15 20 25 '"30
hequency (MHz) frequency (MHz)

Figure 10: Channel signal to noise ratio for an apartment Figure 1 1 : Respective channel capacity estimation.
measurement. Best conditions of load

01 -
0 5 10 15 al 25 30
frequency (MHz) frequency (MHz)

Figure 12: Channel signal to noise ratio for an apartment Figure 13: Respective channel capacity estimation.
measurement. Worst conditions of load.
 These results c o n f m the good opportunity that power lines represents for broadband communications.
Anyway, the distances contemplated in the analysis are not too long, so it could be expected to find lower t
channel capacities when using larger power grids. 1 ,

5. Conclusions and further work

In this work, a method to achieve impulse responses for indoor power line channels has been proposed. The
explained procedure allows obtaining communication system parameters directly fiom the physical
characteristics in typical electrical wiring inside residential buildings. It has been used to model some real t

channels and verified with measurements. The results show remarkable characteristics of these channels:
spectrally shaped responses with high attenuation level even in short distances, great dependence on the
environment, time variation and important noise. The channel response simulations are being completed with
noise models, estimating the source features from measurements of several electrical appliances (impedance and 8

emission level). In spite of the chaotic appearance and strong disturbances in the channel, the signal to noise ratio
in some portions of the spectrum up to 30MHz would be enough to provide reliable high bit rate communication.

In the channel capacity analysis here presented, theoretical approaches and more practical ones (adoption of
discrete multitone with QAM) have been considered, demonstrating that transmission at tens of Mbls could be
feasible. However, this capacity may change seriously with time, due to the connection and disconnection of the
loads and noise sources (electrical devices mainly), what leads to think about the application of an adaptive
transmission technique, for example variable rate OFDM 171. I

6. References
[I] EN 50065, "Signalling on low-voltage electrical installations in the$equency range 3 kHz to 148.5 kHz".
CENELEC, 1991.
[2] Hooijen, O.G. "Aspectsof Residential Power Line Communications". Shaker Verlag, Aachen, 1998.
[3] Philipps, H. "Performancemeasurements of power line channels at high ffequencies ". Proceedings of the
1998 International Symposium on Power-Line Communications and its Applications (ISPLCA), Tokyo,
Japan, pp. 229-237.
[4] Cafiete-Corripio, F.J.; Diez, L.; Entrambasaguas, J. T. "Analysis of residential indoor power lines as a
communication medium". Proceedings of the 5* Workshop on Emerging Technologies in
Telecommunications, Baiona, Spain, pp. 202-206, Sep 1999. J

[5] Philipps, H. "Modelling of power line communication channels". Proceedings of the 3rd International
Symposium on Power-Line Communications and its Applications (ISPLC799),Lancaster UK, pp. 14-21.
[6] CISPR 22, "Limits and methock of measurement of radio disturbance characteristics of information
technology equipment", 1996.
[7] Aguayo-Torres, M.C.; Entrambasaguas, J. T.; Ruiz, F.; Baiios, J. "Variable rate DFE-QAM and OFDM
systems for maximum capacity in multipath ffequency selective fading channels". Proceedings of 5 0 ~
Vehicular Technology Conference, Amsterdam, The Netherlands, pp. 263-267, Sep 1999.