Analysis of narrowband power line communication channels for advanced metering infrastructure
 José Antonio Cortés^{1}Email author,
 Alfredo Sanz^{1},
 Pedro Estopiñán^{1} and
 José Ignacio García^{1}
https://doi.org/10.1186/s1363401502114
© Cortés et al.; licensee Springer. 2015
Received: 6 November 2014
Accepted: 25 February 2015
Published: 20 March 2015
Abstract
This paper analyzes the characteristics of narrowband power line communication (NBPLC) channels and assesses their performance when used for advanced metering infrastructure (AMI) communications. This medium has been traditionally considered too hostile. However, the research activities carried out in the last decade have shown that it is a suitable technology for a large number of applications. This work provides a statistical characterization of NBPLC channels in the CENELECA band. The presented results have been obtained from a set of 106 links measured in urban, suburban, and rural scenarios. The study covers the input impedance of the power line network, the channel response and the noise. The analysis of the channel response examines the delay spread, the coherence bandwidth, and the attenuation, while the assessment of the noise considers both its spectral and temporal characteristics. Since low voltage (LV) distribution networks consists of several conductors, they can be simultaneously used to set up multipleinput multipleoutput (MIMO) communication links. This paper investigates the correlation between the MIMO streams. The bit rates that can be attained both in the singleinput singleoutput (SISO) and in the MIMO cases are estimated and discussed.
Keywords
Power line communications Narrowband Smart Grid Advanced metering infrastructure1 Introduction
The conventional paradigm of electricity networks ‘generate what is consumed’ is shifting towards the new ‘consume what is produced’ [1,2]. This change is motivated by facts like the increased use of renewable sources, which have a much more decentralized structure than conventional ones and whose generating capacity is subject to unpredictable factors, and by new consumption patterns like electric vehicles charging, which complicate the demand forecasting.
The traditional electricity network must evolve into the socalled smart grid to support this change. Advanced metering infrastructure (AMI) is considered a constituent part of the Smart Grid. It enables applications such as automatic meter reading (AMR), demand side response, and distribution automation [1,3]. AMI requires bidirectional communication links between the medium voltage to low voltage (MV/LV) transformer stations and the costumers, usually known as the last mile. Recent studies suggest that power line communications (PLC) is the most costeffective technology for ARM [4]. In addition, it easily enables power quality measurements and distribution automation functions; it gives utilities full control of the communication network and seems to be the most appropriate technology for the communication between the onboard charging system of electric vehicles and the grid [5].
PLC technology can be classified in terms of the employed bandwidth into narrowband (NB) and broadband (BB) [5,6]. Data rates estimated for lastmile AMI applications suggest that they can be delivered by NBPLC in a more inexpensive way than with BBPLC [7,8]. Examples of suitable systems for this end are the ones defined in the ITUT Recommendations G.9902 (known as G.hnem), G.9903 and G.9904, and the IEEE P1901.2 [912]. ITUT G.9903 and G.9904 are based on the industry specifications G3PLC and Powerline Intelligent Metering Evolution (PRIME), respectively.
Nowadays, the CENELECA band (3 to 95 kHz) is the most widespread one in NBPLC [13]. However, the quantitative knowledge of the channel in this frequency range is still imprecise [14]. This is clearly reflected by the significant differences among the physical layer parameters of the latest NBPLC systems [5]. Recent noise measurements and models incorporated into the IEEE P1901.2 have provided much insight into the noise features in the frequency band above 100 kHz [15]. Nevertheless, their suitability for the CENELECA band has not been assessed. The same uncertainty applies to the ‘fading modeling method’ stated in the IEEE P1901.2 for the channel response. It results from a particularization of the wellknown model proposed in [16], but the appropriateness of the selected parameters to generate responses in the CENELECA band has not been evaluated. Similarly, measurements performed in selected scenarios have provided much information about the qualitative features of the channel response [15,17,18], but the absence of a statistical knowledge leads to a large uncertainty in the expected performance of NBPLC.
LV distribution networks consist of several conductors which can be simultaneously used to set up multipleinput multipleoutput (MIMO) communication links. This strategy is being successfully employed in wireless communications and in indoor BBPLC [19], where the phase, neutral, and protective earth conductors are used for the MIMO. However, there are almost no available works on NBPLC MIMO, and they are limited to explore indoor NBPLC channels [20] or to plain tests using existing single phase PRIME devices [21].

We provide a statistical analysis of the channel characteristics in the CENELECA band. These results are compared to the ones obtained with the more recent channel model for this band, proposed in the IEEE P1901.2 standard.

We analyze the correlation between the channels of the 3 × 3 MIMO links that could be established by injecting and receiving the communication signal between each of the phases and the neutral conductor.

We estimate the data rates that could be achieved both in singleinput singleoutput (SISO) and MIMO communications. These results will be useful to clarify whether there is a need for using BBPLC in the last mile of AMIs.
The rest of the paper is organized as follows. Section 2 provides a brief description of the employed measurement setup and signal processing algorithms. Sections 3 and 4 are devoted to the characterization of the input impedance of the power line network and its channel response, respectively. Section 5 is the noise counterpart of Section 4. Based on the presented characterization, Section 6 provides an estimate of the achievable performance. Finally, Section 7 summarizes the main conclusions.
2 Measurement methodology
2.1 Measurement setup
Estimates of the channel response are computed from the input and output signals v _{ T }(t) and v _{ R }(t) shown in Figure 1. Therefore, the attenuation due to both coupling circuits (about 1 dB in the passband) and the coupling loss between the SGS and the power line network are measured as part of the channel. The latter effect could be compensated, since the input impedance of the channels has also been measured. In fact, this must be done when the results are to be used in a channel emulator which separately models both magnitudes, channel response and input impedance [17]. However, the characterization accomplished in this work is intended for assessing the performance of actual communication systems. Hence, all the effects that are present in a real situation have to be taken into account, including impedance mismatch.
The characterization of the 3 × 3 MIMO links is done using the sequential measurements of the nine SISO channels accomplished with the aforementioned setup. Since the input phases not involved in the SISO channel that is being measured are left open circuit, this does not exactly model the actual situation in which these phases would be loaded with the impedance of the MIMO transmitter. However, this does not limit the validity of the results since the influence of the impedance connected to the unused phases has proven to be negligible. Simulations using multiconductor transmission line (MTL) theory have shown that the response of the MIMO streams and their spatial correlation is essentially determined by the cable characteristics, by the topology of the underlying LV line, and by the loads connected to the same phase.
Description of the measured scenarios
Site  Scenario  Cabling  Number of  Number of  Min.  Max.  Number of 

LV lines  customers  dist. (m)  dist. (m)  MIMO links  
1  Rural  OH  2  126  6  427  10 
2  Rural  OH  1  105  14  359  6 
3  Rural  OH  1  66  10  710  5 
4  Rural  OH  1  67  12  1063  6 
5  Semiurban  OH  2  124  16  615  10 
6  Semiurban  UG  4  132  5  163  6 
7  Semiurban  OH  2  183  34  423  13 
8  Urban  OH & UG  24  737  4  340  36 
9  Urban  UG  12  329  19  65  14 
2.2 Measurements processing
with 0≤n≤N _{ L }−1.
The employed value of L yields a frequency resolution of 100 Hz.
The selected value of L leads to a time resolution of 250 µs.
Channel response measurements are obtained by transmitting an orthogonal frequency division multiplexing (OFDM)like sounding signal generated using a 2048point DFT. Its lengths also equals C=26 European mains cycles. An estimate of the frequency response is then obtained by averaging the least squares (LS) estimations obtained from each symbol. The moderate length of the acquired signal has obliged to accomplish an asynchronous averaging (with respect to the mains) of the LS estimates in order to achieve a reasonable signaltonoise ratio (SNR). This provides an estimate of the average channel response in the frequency range 40 to 91 kHz, veiling possible periodic variations in the channel response [23].
The input impedance of the power line channel is estimated during the transmission of the OFDM signal used for channel sounding. The current and voltage signals i(t) and v(t) shown in Figure 1 have been employed for this purpose, following a similar approach to the one in [24]. To this end, the resistance R is fixed to a much larger value than the one of the PLC grid. For the sake of clarity, the circuits used for conditioning and digitizing i(t) and v(t) are not shown.
3 Impedance characterization
The input impedance of the power line network in the considered band is frequency selective. Its magnitude generally increases with frequency, reaching maximum values of tenths of Ω. Illustrative shapes can be found in [18,24].
The magnitudes shown in Figure 2 pose a twofold problem in the design of NBPLC systems. The first one is the difficulty of injecting signal levels of up to 5 V, as the ones allowed by the EN 500651 [13], into such low impedance values. The second one is the minimization of the coupling loss from the transmitter to the PLC network. To this end, the most desirable situation is to make the output impedance of the transmitter negligible with respect to the input impedance of the PLC grid. However, this obliges to make an output impedance on the order of a few m Ω, which is in the range of the resistance of some printed circuit board traces or the transformers wiring. On the other hand, conjugate impedance matching is also technologically difficult because of the aforementioned frequencyselective behavior of the impedance and its large variation between locations.
4 Channel response characterization
4.1 Statistical analysis
The coherence bandwidth and the delay spread are the most widespread parameters used to characterize the frequency selectivity of a channel response and its timedomain counterpart, the time dispersion. In this paper, the former is computed as the frequency separation for which the spacedfrequency correlation function falls down to 0.9 [26]. Both magnitudes have a plain relation to the parameters of OFDM communication systems, like the ones currently used in NBPLC. In particular, distortion in the OFDM signal is avoided if the cyclic prefix is larger than the length of the channel impulse response. Nevertheless, since increasing the cyclic prefix decreases the symbol rate, its optimum value is usually shorter than the channel impulse response length [27]. The delay spread is a root mean squared (rms) measure of the latter. Hence, the cyclic prefix length should generally be several times larger than the delay spread.
System parameters of NBPLC systems standardized by the ITUT
Parameter  G.9902  G.9903  G.9904 

(G.hnem)  (G3PLC)  (PRIME)  
Cyclic prefix length (µs)  60/120  55  120 
Intercarrier spacing (Hz)  1562.5  1562.5  480 
FFT window length (ms)  0.640  0.640  2048 
Max bit rate (kbit/s)  101.3  55.5  64.3 
The presented results indicate that the parameters of the ‘fading modeling method’ are unsuitable for the CENELECA band, although they might be appropriate for the 154.69 to 487.5 kHz range, where the channel response is known to have better transmission characteristics. This band is defined in the IEEE P1901.2 for transmissions in the Federal Communications Commission (FCC) and the Association of Radio Industries and Businesses (ARIB) bands.
where U and V are unitary matrices and D is a diagonal matrix whose values are the singular values, σ _{ i }, and (·)^{ H } denotes the Hermitian operator. The singular values are related to the eigenvalues of H H ^{ H }, λ _{ i }, as \( {\sigma}_i=\sqrt{\lambda_i} \).
Hence, the SVD decomposes the MIMO channel into a set of orthogonal SISO channels, or streams, with amplitude σ _{ i }. The ratio of the singular values can be used as a measure of the correlation between the constituent channels of the MIMO. If fact, the ratio of the maximum to the minimum singular value is the condition number of the matrix H, denoted by κ= max(σ _{ i })/ min(σ _{ i }). When the constituent channels of the MIMO are perfectly correlated, all singular values except one will be zero and the ratio is infinite. On the other side, when channels are absolutely uncorrelated, all the singular values have the same value and its ratio equals one.
5 Noise characterization
5.1 Statistical analysis
Values of the noise PSD fitting curves
Scenario  PSD _{ 0 }  Δ PSD 

Rural  41.412  371.863·10^{−3} 
Semiurban  29.879  233.782·10^{−3} 
Urban  33.939  230.935·10^{−3} 
Figure 10 also shows the PSD of the 12 noise patterns described in the IEEE P1901.2 model. In this case, the synchronous averaging used for the estimation of the PSD has taken into account that they correspond to a mains frequency of 60 Hz. Their background level is higher than that in the measured ones, especially at high frequencies, where differences can be up to 40 dB. In addition, the modeled noise shows no trace of the NBI. This is somehow surprising because these patterns do clearly reflect NBI in the frequency range above 100 kHz (not shown in the figure).
6 Performance assessment
The most straightforward measure of the performance that can be achieved in a given channel is the capacity. However, it requires precise knowledge of the noise statistics, which are still unknown in NBPLC. Moreover, simple closedform expressions are known only for certain distributions, like the Gaussian one. An alternative approach could be to assess the performance that can be attained by standardized systems. Results obtained in this way do not really inform of the potentiality of the channel, which might be under underutilized. Moreover, significant performance differences can be found depending on the algorithms employed at the receiver.
where j denotes the index of the MIMO stream and \( {\lambda}_k^j \), \( {N}_k^j \), and \( {P}_k^j \) are the eigenvalue, the noise, and the input power at the kth frequency index of the jth stream, respectively.
This is the approach used in this paper. For the sake of simplicity, the values of N _{ k } and \( {N}_k^j \) have been taken from the fitting PSDs whose parameters are shown in Table 3. Noise at the MIMO ports is assumed to be uncorrelated. However, to provide values as close to the stateoftheart technology as possible, the following practical constraints have been taken into account: a SNR gap of 5 dB has been included to model the SNR loss caused by the use of practical constellations; the backoff of the power amplifier at the transmitter has been assumed to be 8 dB, and the maximum number of bits per constellation symbol has been fixed to 6. In order to explore the performance limit of the channel, bit rate values achieved without the latter constraint have also been computed.
Limits for the transmitted level are defined in the EN 500651 [13]. It fixes both the signal level (134 dBµV) and a PSD mask (120 dBµV/200 Hz). When using MIMO communications, the maximum level that can be injected on any phase is 6 dB lower than that in the SISO case. The PSD constraint is the most restrictive one because transmitting at 120 dBµV/200 Hz in the 40 to 90 kHz band results in a signal level that exceeds 134 dBµV. Hence, in this paper, a flat PSD of 110 dBµV/200 Hz has been employed.
The comparison of the bit rates in Figure 13 and the ones achieved by the actual systems given in Table 2 reveals that their performance is largely limited by complexity (cost) constraints. In fact, they use constellations with at most 4 bits/symbol. In PRIME, all carriers must employ the same constellation. The same approach has been adopted in G3PLC, except for the possibility of avoiding transmitting in groups of adjacent carriers with very low SNR. Since the noise in NBPLC is strongly colored and the channel response is frequency selective, using the same constellation in all employed carriers causes a severe performance degradation. Additionally, to reduce the memory size, the efficiency of the physical layer frames (data symbols length/frame length) has been penalized. In G3PLC, for instance, the impossibility of fitting more than two ReedSolomon blocks per frame leads to an efficiency of 78% when D8PSK is employed.
Figure 13 also shows the bit rate that could be attained with a 3 × 3 MIMO system in which the maximum power level allowed by the EN 500651 is injected in all phases. As seen, MIMO can offer significant bit rate improvements when the complexity of the system is constrained. The minimum bit rate attained in 90% of the channels in the urban, semiurban, and rural environments improves to 73 kbit/s (2.61 ×), 241 kbit/s (1.57 ×), and 108 kbit/s (1.45 ×), respectively. Performance gains in 50% of the urban, semiurban, and rural channels are 2.89 ×, 2.75 ×, and 2.28 ×, respectively. It should be taken into account that despite the weak contribution of the third stream, part of the MIMO gain comes from the larger singular values of the first stream with respect to the SISO one. As expected from Figure 8, the lowest gains are obtained in the rural scenario. Removing the constraint on the number of bits per constellation symbol reduces the median values of the MIMO gain in the urban, semiurban, and rural channels to 2.34 ×, 2.35 ×, and 1.72 ×, respectively. As expected, it does affect the gains achieved in links with bad transmission characteristics.
7 Conclusions
This paper has presented a statistical analysis of the characteristics of NBPLC channels in the CENELECA band and has assessed their performance when used for advanced metering infrastructure (AMI) communications in rural, semiurban, and urban scenarios. The accomplished study has included the noise, the channel response, and the input impedance of the power line network. The analysis of the former has examined both its spectral and temporal characteristics. The channel response has been studied in terms of the delay spread, the coherence bandwidth, and the attenuation. An estimation of the data rates that could be achieved has been accomplished. The obtained results indicate that the highest performance is achieved in semiurban scenarios and the lowest in urban ones. In the former, 90% of the channels can deliver more than 153 kbit/s, while in the latter, this figure goes down to 28 kbit/s. Some of the reasons that prevent current NBPLC systems from achieving these performance have been highlighted. In addition, the use of the three phase conductors for MIMO communications has been explored. It has been shown that a practical 3 × 3 MIMO system can give performance improvements larger than 2.61 ×, 1.57 ×, and 1.45 × in 90% of the urban, semiurban, and rural channels, respectively.
8 Endnote
^{a}The low intercarrier spacing in PRIME is motivated by the use of a differentialinfrequency modulation.
Declarations
Acknowledgments
The authors thank Atmel corporation for the support provided to perform all the tests referred in the paper.
The authors would also like to thank the anonymous reviewers for their valuable comments and suggestions.
Authors’ Affiliations
References
 H Farhangi, The path of the Smart Grid. IEEE Power Energy Mag. 8(1), 18–28 (2010).View ArticleMathSciNetGoogle Scholar
 Kema Limited. GB Demand Response. Report 1: The Opportunities for Demand Response. Commissioned by the Energy Association. (Version 1.0) (Kema Limited,Arnhem, 2011).Google Scholar
 G Strbac, C Kim Gan, M Aunedi, V Stanojevic, P Djapic, J Dejvises, P Mancarella, A Hawkes, D Pudjianto, S Le Vine, J Polak, D Openshaw, S Burns, P West, D Brogden, A Creighton, A Claxton. Benefits of Advanced Smart Metering for Demand Response based Control of Distribution Networks. ENA/SEDG/Imperial College report on Benefits of Advanced Smart Metering (Version 2.0) (Energy Networks Association,London, 2010).Google Scholar
 H Edelmann, T Kästner. CostBenefit Analysis for the Comprehensive Use of Smart Metering. Technical report, Ernst & Young On Behalf of the Federal Ministry of Economics and Technology (Ernst & Young GmbH,Germany, 2013).Google Scholar
 S Galli, A Scaglione, Z Wang, For the grid and through the grid: the role of power line communications in the Smart Grid. Proc. IEEE. 99(6), 998–1027 (2011).View ArticleGoogle Scholar
 The OPEN meter Consortium, Description of current stateoftheart of technology and protocols description of stateoftheart of PLCbased access technology. European Union Project Deliverable FP7ICT2226369, d 2.1 Part 2, Version 2.3, March 2009. http://www.openmeter.com/files/deliverables/OPENMeter\%20WP2\%20D2.1\%20part2\%20v2.3.pdf.
 A Mengi, M Waechter, M Koch, in Proceedings of the IEEE International Symposium on Power Line Communications and Its Applications (ISPLC). 500 kHz G3PLC access technology for the rollouts in Germany (Glasgow, 2014).Google Scholar
 Engage Consulting Limited. Highlevel smart meter data traffic analysis. Technical report for Energy Networks Association (Energy Networks Association,London, 2010).Google Scholar
 ITU.T Recommendation G.9902. Narrowband orthogonal frequency division multiplexing power line communication transceivers for ITUT G.hnem networks (ITU,Geneva, 2012).Google Scholar
 ITUT Recommendation G.9903. Narrowband orthogonal frequency division multiplexing power line communication transceivers for G3PLC networks (ITU,Geneva, 2013).Google Scholar
 ITUT Recommendation G.9904. Narrowband orthogonal frequency division multiplexing power line communication transceivers for PRIME networks (ITU,Geneva, 2012).Google Scholar
 IEEE P1901.2. IEEE Standard for LowFrequency (less than 500 kHz) Narrowband Power Line Communications for Smart Grid Applications (IEEE,New York, 2013).Google Scholar
 European Committee for Electrotechnical Standardization (CENELEC). Signalling on lowvoltage electrical installations in the frequency range 3 kHz to 148,5 kHz  part 1: general requirements, frequency bands and electromagnetic disturbances. European Standard EN 500651 (CENELEC,Brussels, 2001).Google Scholar
 S Galli, Recent developments on the international standardization of narrowband PLC for smart grid applications. Keynote given at the IEEE Internationaly Symposium on Power Line Communications and its Applications (ISPLC) (2012). http://www.ieeeisplc.org/2012/program_44_1704874382.pdf.
 M Nassar, J Lin, Y Mortazavi, A Dabak, IH Kim, BL Evans, Local utiliy power line communications in the 3500 khz band: channel impairments, noise, and standards. IEEE Signal Process. Mag. 29(5), 116–127 (2012).View ArticleGoogle Scholar
 M Zimmermann, K Dostert, A multipath model for the powerline channel. IEEE Trans. Commun. 50(4), 553–559 (2002).View ArticleGoogle Scholar
 W Liu, M Sigle, K Dostert, Channel characterization and system verification for narrowband power line communication in Smart Grid applications. IEEE Commun. Mag. 49(12), 28–35 (2011).View ArticleGoogle Scholar
 G Chu, J Li, W Liu, in Proceedings of the, International Symposium on Power Line Communications and Its Applications (ISPLC). Narrow band power line channel characteristics for low voltage access network in China (Johannesburg, 2013).Google Scholar
 LT Berger, A Schwager, P Pagani, Scheneider DM (eds.), MIMO Power Line Communications: Narrow and Broadband Standards, EMC and Avanced Processing, 1st edn (CRC Press, Boca Raton, FL, 2014).Google Scholar
 Y Sugiura, T Yamazato, M Katayama, in Proceedings of the IEEE International Symposium on Power Line Communications and Its Applications. Measurement of narrowband channel characteristics in singlephase threewire indoor powerline channels (Jeju, 2008), pp. 18–23.Google Scholar
 A Sendin, A Llano, A Arzuaga, I Berganza, in Proceedings of the IEEE International Symposium on Power Line Communications and Its Applications. Strategies for PLC signal injection in electricity distribution grid transformers (Udine, 2011), pp. 346–351.Google Scholar
 AV Oppenheim, RW Schafer, J Buck, Discretetime Signal Processing (Prentice Hall, New Jersey, 1999).Google Scholar
 FJ Cañete, JA Cortés, L Díez, JT Entrambasaguas, Analysis of the cyclic shortterm variation of indoor power line channels. IEEE J. Selected Areas Commun. 24(7), 1327–1338 (2006).View ArticleGoogle Scholar
 M Sigle, W Liu, K Dostert, in Proceedings of the, IEEE International Symposium on Power Line Communications and Its Applications. On the impedance of the lowvoltage distribution grid at frequencies up to 500 kHz (Beijing, 2012), pp. 30–34.Google Scholar
 A Sendin, I Peña, P Angueira, Strategies for power line communications smart metering network deployment. Energies. 2014(7), 2377–2420 (2014).View ArticleGoogle Scholar
 JG Proakis, Digital Communications (McGrawHill, Massachusetts, 1995).Google Scholar
 JA Cortés, L Díez, FJ Cañete, JJ SánchezMartínez, JT Entrambasaguas, Performance analysis of OFDM modulation on indoor broadband PLC channels. EURASIP J. Adv. Signal Process. 2011(78), 1–12 (2011).Google Scholar
 A Goldsmith, S Chua, Variablerate variablepower MQAM for fading channels. IEEE Trans. Commun. 45(10), 1218–1230 (1997).View ArticleGoogle Scholar
 KF Nieman, J Lin, M Nassar, K Waheed, BL Evans, in Proceedings of the International Symposium on Power Line Communications and Its Applications (ISPLC). Cyclic spectral analysis of power line noise in the 3200 kHz band (Johannesburg, 2013).Google Scholar
 A Goldsmith, Wireless Communications (Cambridge University Press,New York, 2005).Google Scholar
Copyright
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.