DMT-based Power Line Communication for the CENELEC A-band Frederik PetrG* Marc Engels Bert Gyselinckx Hugo DeMant Interuniversity Micro Electronics Centre (IMEC). Kapeldreef 75,3001 Heverlee - Belgium tel.
+32-16-281406 fax. +32-16-281515 email :
[email protected]
Witlz the deregulation of the telecorn market, the power distribution network has become an interesting alternative for already existing access networks. However; recently proposed transmission schenzes for power lines are not designed for tlze high data rates required for multiirzedia transmission. We propose Discrete Multi Tone (DMT)as a means to obtain high data rates in the CENELEC A-band (9-95kHz). A key element of such a system is tlze bitloading strategy. Front the different loading algorithins we selected the rate-adaptive loading algorithin proposed by Leke et al. [ I I], which maxiinizes the bitratefor@ed bit error rate and a given power constraint, and extended it to derive a discrete bit distribution. This allowed us to derive upper boundsfor the bitrate pellformance of a DMT transnzissiort system. Siinulations indicate that bitrates above 1 Mb/s are feasible for distances below 1000 m using a transmit power of several tens of Watt. Pe$ori?zance degrades to several hundreds of kb/s when transmit power is reduced to 100 mW
1 Introduction Power line communications is a topic which has been studied for many years. Up till now, its most important applications were load management, remote meter reading, home automation, intelligent buildings and local area networks [I]. With the deregulation of the telecom market, the power distribution network can also be used as an access network besides already existing ones like the telephone access network or the CATV access network. Digital customer services like electronic banking, e-mail, internet access and digital atldio and video broadcast should become feasible in the near future, using the power network as a communication channel. However, recently proposed transmission schemes for power networks are not designed for the high data rates required for multimedia transmission. Dostert describes in [2] his results with frequency-hoppingspread-spectrum. For outdoor communications he reports a data rate of 60 bitsls and a frequency hop rate of 300 Hz using a spectral In [3] Waldeck proposes an improved range from 30 to 146 kHz. The bit error rate is in the order of frequency-hopping scheme with reduced constraints on the system clock. Hooijen deiigns in [4] a system that combines both frequency-hoppingand direct-sequencespread-spectrumtechniques. In [5] Tuite exploits the larger bandwidth (10-450 kHz), available in the United States and Japan, to report data rates of 19.2 kbls with a bit error rate of using direct-sequence spread-spectrum. In order to support higher data rates, other modulation schemes a e required. Based on Hooijen's channel model [6] we propose Discrete Multi Tone (DMT) [7] as a means to obtain high data rates in the CENELEC A-band (9-95 kHz). The concept of multitone transmission has attracted a lot of interest recently as a means to increase the data rate on a channel under given requirements such as fixed transmitter power budget and equal probability of error on all subchannels. DMT is the most common form of Multicarrier Modulation (MCM) and assigns a number of bits to each subchannel according to the subchannel signal-to-noise ratios. The scheme which assigns the energy and the bits to the different subchannels is called a loading algorithm. The organization of this paper is as follows. The next section discusses the major impairments of the power line channel for the CENELEC A-band and derives the channel model we used for simulations. Section 3 presents the structure of a DMT system and section 4 describes the loading algorithm we implemented. Simulation results are discussed in section 5. Finally, section 6 summarizes the most important conclusions and gives a view on future work. * This work was supported by a KUL-scholarship t Professor at the Katholieke Universiteit Leuven
2 Channel model Based on an'extensive measurement campaign in the city of Amsterdam, Hooijen [6] was able to derive a channel model for the residential power distribution network. Figure 1 shows the residential power circuit (RPC) which basically comprises everything attached to the secondary side of the distribution transformer. Hooijen measured the channel parameters of interest in the frequency-band from 9 to 95 kHz (CENELEC A-band), the frequencyband in which communications by the electricy providers is allowed. For our simulations we used a simplified version of Hooijen's channel model, since our primary aim was to derive upper bounds for the bitrate performance of a DMT system. In the following we give an overview of this model and we point out the gimplifications we made. HighlMedium Voltage Transformer
Figure 1: Residential Power Circuit The two major contributions to the attenuation on the RPC are the coupling loss and the line loss. The coupling loss is caused by the interaction between the coupling network and the RPC impedance. As a first approximation we neglect this coupling loss by assuming the transmitter's output impedance to be zero. The line loss is approximately frequency- and time-independant for a given location. The attenuation increases as the distance between the transmitter-receiver pair increases, with 40 to 100 dB1km. The previous conclusions are summarized by the following equation for the channel transfer function :
with a [m-l]varying between 0.004 (best case) and 0.01 (worst case). The noise on the RPC can be considered to be a summation of four noise types :
Background noise is the portion of the noise that remains when subtracting the other three noise types from the total noise measured at a certain location. The background noise power decreases with increasing frequehcies and is described by [6] :
where K is normally distributed with average -8.64 and standard deviation 0.5.
Single event impulse noise is primarily caused by switching transients and can be modeled as impulses, which last for a very small fraction of time (typically less than 100 ps). The impulse amplitudes typically lie more than 10 dB above the average background noise level with peaks of 40 dB. Noise syrzchronous to the power system frequency is mainly produced by silicon controlled rectifiers, found e.g. in light dimmers. In the time domain, this noise type shows itself as a train of noise impulses, arriving every l/(k. fnet) seconds, with k usually 1 or 2. The spectrum of this noise consists of a series of harmonics of the k . fnet fundamental component.
59
7
Filter
Noise Filter WGN 20
10
Figure 2: Simplified channel model for CENELEC Aband
30
40
50
80
70
80
80
I
Fre~uency(Wz)
Figure 3: Best and worst case noise psd
Narrowband noise is confined to a narrow portion of the frequency band. It can appear at any frequency within the CENELEC A-band, but most likely'at television related frequencies (i.e. 31, 47, 62, 78 and 94 kHz). Our model only contains the background noise, which is the most important noise source. Figure 2 summarizes our simplified channel model. The channel filter, described by equation 1, contains a distance dependant attenuation. The noise filter transforms white gaussian noise into colored noise with the desired power spectral density, described by equation 2. Figure 3 shows the noise psd for the best case, corresponding to K = ,u - 2 . a, and for the worst case, corresponding to I( = ,u 2 - u.
+
3 Discrete Multi Tone modulation In this section we provide some background on Discrete Multi Tone (DMT) and point out the advantages of adaptive bitloading. In DMT the total bandwidth is divided into N parallel subchannels. These subchannels are exactly independant and memoryless by using the basis vectors of the inverse fast Fourier transform (IFFT) as the subchannel carriers and adding a cyclic prefix to each symbol. This cyclic prefix is used to combat intersymbol interference and make the transmitted sequence look periodic. If the length of the channel impulse response is v + l , then the length of the cyclic prefix must be v. To eliminate intercarrier interference, the first v received samples in each DMT block are discarded. Because each transmitted block must have a cyclic prefix to be transmitted over any non-ideal channel, part of the available bandwidth and power budget is wasted by inclusion of the cyclic prefix. Therefore, one goal in the design of a multicarrier system is to minimize the percentage of each block wasted by transmitting the cyclic prefix. For any fixed channel impulse response, the percentage of each block lost to the cyclic prefix decreases as the FFTIIFFT size (which is 2N) is increased. For slowly time-varying channels, channel state information can be used in the transmitter to 'load' the different subchannels. Bits are assigned to subchannels in direct proportion to the subchannel signal-to-noise ratios. As a result, subchannels that suffer from little attenuation andlor little noise carry the most bits, while subchannels that are severely attenuated andlor very noisy might not carry any bits. The scheme which assigns the bits and the energy to the different subchannels is called a loading algorithm. Figure 4 shows a typical block diagram of a DMT transmitlreceive structure. The entire bandwidth is divided into N parallel subchannels. An input bit stream of R bitsls is buffered into blocks of b = RT bits, where T is the symbol period. The loading algorithm assigns a certain number of bits, b,, to each subchannel where
'
and N the number of subchannels. The encoder then translates the bits, b,,. into symbols, Xn, chosen from the appropriate constellation. The time symbols, x,, are obtained by inverse Fourier transformation, and the cyclic prefix is added to the beginning of the DMT symbol. Taking the Fourier transform of the received samples, after discarding the cyclic prefix, one obtains N independant parallel subchannels which can be individually decoded using a simple memoryless decoder for each subchannel. 'A 2N-point complex-to-red IFFT is required in the transmitter to ensure that the signd applied to the channel is real.
Input bit stream R bitsls
Xl ,k
I
. .
b bit buffer and encoder
.
. Xl,k
c
X2,k
IFFT
X~,k
N QAM
symbols
.
.
.
X2,k
'
'2N.k
,prefix
PIS and add cyclic
2N time domain samples
Lowpass Filter
DIA
i
:
Channel
.
Output bit
n
R hits/
