Capacity Analysis of a Wireless Backhaul forMetering in the Smart Grid

Babak Karimi and Vinod NamboodiriDepartment of Electrical Engineering and Computer Science

Wichita State UniversityWichita, Kansas

Email: {bxkarimi, vinod.namboodiri}@wichita.edu

Abstract—Many application scenarios envisioned as part of the‘Smarter’ Grid depend on an effective underlying communicationinfrastructure for power distribution. To this end, a linear chainwireless communication architecture is proposed in this work tomeet communication requirements at the distribution level of thepower grid. A theoretical study is done to understand the capacitylimitations of using such linear chain wireless technologies atthe distribution level. The application of advanced meteringinfrastructure (AMI) that encourages customer participation inenergy delivery is used as a case study to understand the implica-tions of any limitations imposed by the proposed communicationarchitecture.

I. INTRODUCTION

The need for improved communications at the power dis-tribution level takes on greater importance with the introduc-tion of the Smart Grid approach. Title XIII of the EnergyIndependent and Security Act 2007 [1] requires improvedoperation of distribution systems that includes developmentand incorporation of demand-side and energy-efficiency re-sources, deployment of real-time, automated, interactive tech-nologies that optimize the physical operation of appliancesand consumer devices, and provisions for timely informationand control options to consumers, to name a few. Thesedevelopments and deployments require additional capabilitiesof the grid, especially a better communication infrastructurebeyond the supervisory control and data acquisition (SCADA)level communication.

The critical requirement for advancement in the distributionsystem is real time information sharing and automation. Byimproving the communication infrastructure, a vital ingredientfor the Smart Grid, a more reliable approach could be taken tobetter manage assets. In addition to asset and outage manage-ment tasks, communication will also aid in better energy man-agement and tariff-related information. Deployment of smartdistribution systems necessitates proper identification of powersystem requirements and integrating suitable communicationand control infrastructure. The information communicationand control layer of the smart grid brings about numerousadvances, including the empowerment of customers to activelyparticipate in the maintenance of the supply-demand balancearound the clock and the resulting reliability improvement inelectricity service.

Advanced Metering Infrastructure (AMI) initiatives are apopular tool to incorporate the changes for modernizing the

electricity grid, reduce peak loads, and meet energy-efficiencytargets. With the introduction of AMI technology, two-waycommunication between a smart meter (SM) and the controlcenter, as well as between the smart meter and customer loadswould be facilitated for demand response, dynamic pricing,system monitoring, cold load pick-up, and greenhouse gas-emission mitigation [2]. AMI uses technology to capture andtransmit energy use to a concentration point on an hourly orsub-hourly basis in contrast to standard meters that providea daily energy usage total and a cumulative monthly bill.This application requires bidirectional communication: controlcommands from the control center of the utility to smartmeters, and load profiles and logs from smart meters to thecontrol center.

There are many communication technology and architectureoptions in deploying the AMI. As shown in Figure 1, theend-to-end communication infrastructure can be divided intotwo major phases: (i) the consumer level home area network(HAN) between a smart meter and various electrical equip-ment/appliances, (ii) the backhaul link that collects informa-tion from smart meters and carries it to the utility controlcenter. The backhaul can either be a direct link from individualsmart meters to the utility control center, or a link (or series oflinks) from a concentrator node, which aggregates data frommultiple smart meters through some form of a mesh network,to the control center. There is reasonable consensus on usingZigBee-based star topologies in HANs [3], [4], and ZigBeeor Wi-Fi based mesh topologies to collect and aggregate dataat concentrators [5], but the technologies and topologies to beused for the backhaul is still an open problem.

This work thus looks at the design of backhaul communi-cation for the AMI. An approach of deploying the backhaulcommunication infrastructure is to create a network throughthe existing feeder infrastructure. This approach has the ad-vantage of reducing costs by reusing existing infrastructureof electric poles and feeder lines if needed. Further, it canalso integrate well with sensors deployed on feeder polesto improve distribution automation through applications likeautomated fault location and self-healing feeders. In priorwork the authors compared power-line carrier (PLC), wiredtechnologies, and a wireless linear chain architecture based onWi-Fi [6]. The Wi-Fi chain architecture was found to be themost capable option in terms of features like communication

The 1st IEEE INFOCOM Workshop on Communications and Control for Sustainable Energy Systems: Green Networking and SmartGrids

978-1-4673-1017-8/12/$31.00 ©2012 IEEE 61

2

Fig. 1. Communications for AMI at the distribution level. The commu-nications infrastructure needed (indicated by links shown as clouds) canbe divided into the following: (i) backhaul infrastructure communicationspossibly through feeders, and (ii) consumer/customer premise home areanetworks.

data rate, flexibility, and resiliency. For example, a wirelesstechnology is less likely to suffer network outage when electricpoles are downed due to weather than other technologies thatrely on physical wires that can break.

It is not yet clear though what would be the communicationcapacity of a very long wireless linear chain network deployedfor the AMI application between concentrator nodes andthe utility control center. The AMI application presents thechallenge of collection and management of data from smartmeters. By current standards, each smart meter sends a fewkilobytes of data every 15 minutes to a smart meter [7],[8]. When this is scaled up to large numbers, many existingcommunication architectures will find it difficult to handlethe data traffic due to limited bandwidth. To this end, thiswork explores the feasibility of utilizing a wireless linearchain network to meet communication requirements for AMI,analyzes the limitations imposed by such an architecture onthe amount of data traffic that can be carried, and exploreshow grid operators could plan and operate their deployments.

II. CAPACITY ANALYSIS OF FEEDER LEVEL LINEARCHAIN NETWORK TOPOLOGY

In this section we theoretically analyze the capacity ofa linear chain topology likely to be used for feeder levelcommunications at the distribution level. Though a Wi-Fiarchitecture was proposed as the technology of choice forcommunications at the distribution level in our previous work[6], we keep the treatment in this section of the paper andfollowing sections to be more general allowing for the useof other wireless technologies as well in any communicationarchitecture that is used.

A. Problem Definition and Assumptions

Consider a linear chain communication network that ismulti-hop in nature with n + 1 nodes (n hops from sourceto destination) overall separated by a constant inter-nodedistance. Let C bps be the maximum single hop throughputor capacity, and r meters be the communication range of thetechnology used. The source node generates data at a rate ofλ bits per second. Let the sink node be L meters away fromthe data concentrator. We would like to determine what is theend-to-end data capacity, Xc, of this linear chain topology.

This problem is representative of the type of linear chainnetwork that will be used at the distribution level of thepower grid as described in our proposed Wi-Fi architecturein the previous section. The source node could represent aconcentrator or data aggregator where all data from manydistributed sources (e.g. smart meters of residences in the AMIapplication) could be collected and sent over the network.The sink node could represent a control center of the gridoperator where all data is collected. The scenario could alsobe the other way around, where the source node is the controlcenter sending commands to a sink node that could be anend-user’s smart meter. The smart meter to control centerdirection is more interesting in terms of network capacityanalysis as it is expected to have more data; the other directionis mainly envisioned for control commands. Figure 2 shows anillustration of the specific application scenario at the feeder-level. The assumption of constant inter-node distance is basedon the assumption that nodes are co-located with existingutility infrastructure (e.g. distribution feeders) that is deployedat some fixed density. Determining the maximum data trafficrate that can be supported by such a linear chain networkfor any given communication technology that is deployedis useful in network capacity and operational planning, andreliability analysis. Prior research on the capacity of ad hocwireless networks have studied similar problems, which alsoinclude single chain capacity analysis [9], [10]. However, theirfocus was solely on WiFi technology and considered manyother topologies (apart from linear chain), including randomnetworks; our goal is to analyze and study the capacity oflinear chain wireless topologies and how it applies to thedistribution level, and without being restricted to only onetechnology.

B. Single Chain Analysis

The shared broadcast nature of the wireless medium typ-ically1 dictates that when one node is transmitting, none ofthe nodes within the interfering range of the transmission cantransmit or receive. Let us assume that only one out of every mconsecutive links of a linear chain can be active at a time. So arough estimate of a single chain capacity is 1/m times the one-hop capacity. This estimate may however be an over-estimateas the technology used may not have perfect scheduling (true

1If techniques like code division multiple access are used, multiple nodeswithin interference range of each other can transmit at the same time. Theseare typically not used except in cellular systems, one reason being thecomplexity of the hardware involved.

62

3

Fig. 2. The backhaul linear chain architecture considered in this work forAMI.

for example in the case of WiFi at the MAC layer) and not allopportunities to transmit may be taken; it could happen thatif the first link is active, any of the links m+ 1 to 2m couldbe active simultaneously if they have data to forward (linkm + 1 being active is optimal in this case). We can improvethis estimate through the following analysis.

Assume we have linear chain with n links, L1 to Ln withn sufficiently large. Let f(n) denote the average number oflinks that are active in the chain given that the first link L1 isactive. Let g(n) denote the average number of active links inthe chain. We will use a recursive analysis to calculate f(n)and g(n).

If L1 is active, none of the links L2 to Lm can be active.Further exactly one of Lm+1 to L2m will be active as nodeson the chain always want to send data, and any link Li canbe interfered with only by links Li−(m−1) to Li+(m−1). IfLm+i is active, then the number of active links in the chainwill be 1 + f(n − (m + i) + 1). We will ignore the cornereffect of the end of the chain where there may not be as manylinks interfering; the assumption of n being large supports thisdecision. If pi, 1 ≤ i ≤ n is the probability that any link Li

is active, then we can compute the average number of activelinks given L1 is active as

f(n) = 1 +m∑j=1

(pj ∗ f(n− (m+ j)− 1))

Next we compute g(n) letting qi, 1 ≤ i ≤ n denote theprobability that Li is active. Then,

g(n) =

m∑i=1

(qi ∗ f(n− i+ 1)))

where we assume any link i from the first m links can beactive removing the conditionality imposed by f(n) on L1

being active. Continuing,

g(n) (1)

=

m∑i=1

qi

m∑j=1

(pj ∗ (1 + f(n− i+ 1− (m+ j) + 1)))

=m∑i=1

m∑j=1

qi ∗ pj +m∑j=1

pj

m∑i=1

(qj ∗ f(n− i+ 1− (m+ j) + 1))

= 1 +m∑j=1

pj ∗ g(n− i+ 1− (m+ j) + 1))

Assuming that g(n) is a linear function of n and using an+ bas a possible solution with b = 0, and pi = 1/m, 1 ≤ i ≤ m,we can solve and obtain a = 2

3m−1 . Thus, g(n) = an = 2n3m−1

is the average number of links active in linear chain at steadystate when n is large. This results in an overall chain capacity,Xc, of

Xc =g(n)

nC =

2C

3m− 1(2)

where C is the single hop capacity. For smaller values of n wecan utilize the recursion in Equation 1 to compute differentvalues of g(n) and compute end-to-end chain capacity.

Validation of analysis through simulationsTo verify our assumptions we conducted a series of simulationsand compared the results of our theoretical result above withsimulations of the WiFi technology.The simulations were donewith the ns2 simulator [11] whose parameters were set to a2 Mbps channel data rate2. For the theoretical results we setthe value of C in the Equation 1 to 1.7 Mbps as overheadreduces useful throughput from the 2 Mbps channel rate. Thecommunication range r was set constant to 100 m and theinterfering range of each node was set such that m = 4 in theabove analysis. The distance L was varied to vary the chainlength. The corresponding values of g(n) were computed usingEquation 1 and the resulting capacity values for each chainlength plotted.

From Figure 3 the results of our theoretical analysis andsimulations can be compared. For the largest data packetsize of 1500 bytes, which is expected to have the greatestthroughput, it can be observed that the difference between thetwo results is very small (< 6%) validating the analysis. Wecan further improve on our analytical result above (and getcloser to the simulation results) by adding a correction for theway the IEEE 802.11 standard medium access control (MAC)protocol works for WiFi. Under this protocol, there could beinstances where none of the m consecutive links could betransmitting as they might be counting down backoff slots inaccordance with the binary exponential backoff algorithm used[12]. We do not do account for this to keep our analytical resultpresented in this section more general.

2Though current technologies have higher data rates available, the actualvalue of data rate used does not change the overall result. Setting a data rateof 2Mbps further proves useful to compare against prior work done in thearea of capacity of ad-hoc networks in [9] where a similar analysis was donefor random networks.

63

4

(a) Theoretical chain capacity (b) End-to-end chain throughputthrough simulations

Fig. 3. Comparison of theoretical capacity and end-to-end throughput throughsimulations

C. Multiple Chain Analysis

If the rate of data generation is higher than what a linearchain communication topology can support, multiple chainscould be possibly deployed. These chains could be co-locatedusing the same physical path and the same (or different) nodes,but possibly different non-interfering frequency channels, oruse a different non-interfering path along an another set offeeders. It is thus useful to extend our single chain analysis tohow many chains would be needed for some given data gener-ation rate. The eventual wireless communication architecturecould thus be composed of multiple independent linear chainsfrom many concentrators to the control center. The presenceof multiple such chains, if planned and deployed properly,can add reliability to the communication network by providingalternate paths from common intersection points.

Let Xc be the capacity of single chain as found in Equation2. Let N be the ratio of data to be sent from the concentrator,λ, to the capacity of chain Xc. In order to estimate the numberof chains that we need to support the required λ, there is arelation which is always true and is:

λ ≤ NXc (3)

Thus to get the number of required chains, we define thefunction f as:

f

(N =

λ

Xc

)=

{1 if N ≤ 1

⌈N⌉ if N > 1(4)

III. DATA RATE REQUIREMENTS ON COMMUNICATIONINFRASTRUCTURE

The analysis in the previous section looked at how muchdata rate, Xc, a single linear chain communication networkcan support from a source to sink and how many chains, N ,will be needed to carry a specified amount of data traffic λ. Inthis section an analysis of data communication requirementswhen using the AMI application will be done so as to betterunderstand the magnitude of λ that will have to be supported.The application scenario under consideration was shown inFigure 2 where the smart meter aggregator or concentratorpoint gathers data from smart meters and sends it to the control

center.3 This section will also give insights on the amount ofdata and interval at which smart meters should send this datafor a given communication infrastructure.

A. AMI data output

Suppose there are z smart meters in a given area as partof the AMI application, all connected to a concentrator. Asthe rate of data generation by smart meters are not necessarilysynchronized, an average rate of data generation per second isa more useful value. Let each meter generate a data packet Pbytes long at an interval of t minutes. Thus, the average datatraffic rate reaching the concentrator from z smart meters willbe

λ =z · P60 · t

bytes/sec =z · P7.5 · t

bits/sec (5)

Current standards specify that each smart meter sends a 512byte packet every 5, 15, 30, or 60 minutes[7], but 15 minutesis most common [8]. Thus, henceforth we use the interval of15 minutes for evaluations. This translates to a data rate ofλ = 4.55z bps arriving at the concentrator.

This simple analysis can be further extended to include theconcept of smart meter density. For example, assume eachhouse in a residential neighborhood has a smart meter, and thesmart meter density (equivalent to housing density) per squaremeter, say ρH , is known. So, given the area of the locationof interest, A, and the housing density, ρH , the expected datatraffic arriving at the concentrator would be

λ =ρHA · P60 · t

bytes/sec =ρHA · P7.5 · t

bits/sec (6)

Figure 4(a) plots the data traffic at the concentrator forvarious values of ρH for an area of 1000 sq. m and packetsize P = 512 bytes. Three intervals at which meters couldsend data are considered. It can be seen that huge amountsof data can be easily generated by applications in dense AMIdeployments.

(a) Incoming data rate λ at concen-trator for various t and ρH

(b) Minimum smart meter sending in-terval t allowed for a given communi-cation infrastructure capacity Xc.

Fig. 4. Considerations on selecting an ideal sending interval for smart meters(a) based on housing density ρH (b) chain capacity Xc

3The technologies used in gathering data from each individual customer tothe smart meter and the capacity analysis of this network is not addressed inthis work and would be part of future work.

64

5

B. Data transfer interval for a given infrastructure

If the communication infrastructure is already in place, theparameter that decides the data rate that will need to besupported by this infrastructure is the interval t at which eachsmart meter sends data given ρH , A, P , N , and Xc. We canuse equations 3 and 6 to get the following inequality

ρH ·A · P7.5 · t

< N ·Xc

which simplifies to provide a lower bound for the sendinginterval as

t >ρH ·A · P7.5 ·N ·Xc

(7)

Figure 4(b) provides an idea of how the sending intervalfor each smart meter increases with a decrease in end-to-endcapacity of the communication infrastructure for m = 4, N =1, A = 1000 sq. m, and P = 512 bytes.

C. Data Handling Requirements for Large, Hybrid Topologies

The analysis so far was based on a homogeneous ar-chitecture where only one type of technology was used totransport data from concentrators to the control center. Herewe extend our analysis to study expected data traffic that needsto be handled over a large, wide-area network where possiblymultiple technologies might be needed from concentrators toreach the control center. For example one technology couldbe used within metropolitan areas due to a higher populationdensity, while another technology is used for suburban orrural users. We begin by developing tree-based model for thecommunication infrastructure and subsequently use this modelfor assessing data traffic handling requirements.

Figure 5 shows how a tree architecture can be used to modelh different levels of technologies used end-to-end. There is aroot node, which is the control center and the final destinationof all data collected from smart meters. All leaf nodes at thelowest level of the tree are smart meters generating data. Allother intermediate nodes of the tree that are neither leaves orthe root node, are the concentrators. We are interested in thedata traffic at each concentrator, and the overall data trafficreceived at the control center.

Let the height of the tree be h, starting from level 0 to levelh. Let there be nc(i) cluster(s) at each level i. Then there areN(i, k) aggregators (concentrators) in kth cluster of level i,where 0 < k < nc(i) and i > 0. This helps to identify thelower level data traffic senders to which a concentrator j ofcluster k at level i is connected to. Thus λijk is the rate ofdata collected at concentrator number j in cluster k at level iand defined as:

λijk =

N(i′,k′)∑j′=1

λi′j′k′ (8)

where i′ = i − 1 (lower level), j′ is the aggregator index incluster k′, and k′ is the cluster identifier in level i′ that sendsits data to aggregator with identifiers i,j and k.

Fig. 5. Tree architecture of communication hierarchy

To represent the total data traffic that is being collected atany specific level i, we use λi.. and define it as:

λi.. =

nc(i−1)∑j=1

λijk (9)

λ111 is a base case and is the data traffic by an aggregator atlevel 1 directly from smart meters (leaves at level 0) connectedto them and equals

λ111 =

N(0,1)∑j=1

P1

7.5 · tbps (10)

based on the data generation rate for each smart meter fromEquation 5 (without the multiplier z) for all concentrators jat level 1. Pk is the packet size used for data sent in clusterk.Finally the overall data traffic that is being transmitted fromsmart meters to the root control center over the distributiongrid equals to:

λTotal =

N(0,k)∑j=1

nc(0)∑k=1

λ0jk +

N(1,k)∑j=1

nc(1)∑k=1

λ1jk + · · ·+N(h,k)∑j=1

nc(h)∑k=1

λhjk

=h∑

i=0

N(i,k)∑j=1

nc(i)∑k=1

λijk (11)

IV. DISCUSSION AND FUTURE WORK

Though we have recommended one specific communicationarchitecture for the distribution level of the power grid in thispaper, an optimal architecture may vary widely. For example,at the distribution level of the power grid, engineers couldchoose to utilize a configuration in which short-range wirelesscommunication technology is used to send data from sensorsinstalled on electric poles to a master node on the pole (sothat no wire is needed between the sensors and router) and

65

6

use wired technologies like DSL, Fiber Options (FTTX), orPower Line Communication (PLC) to feed the data back tothe control center. Alternative configurations may allow thesmart meters to send the data directly back to concentratorwithout involving any wired technology. In response to such awide range of possible network and system configurations,the best practice begins with thorough and careful systemarchitecture and configuration design. After this process, sev-eral distinct configurations that involve wired or wirelesstechnologies at varying degrees may be derived. Therefore,apart from proposing a specific architecture based on somecommon applications in this paper, we also contribute towardsa step in the direction of providing a suite of communicationtopologies and an analysis of their technical feasibility, optimalconfiguration, and deployment considerations. Future workwould need to consider economic feasibility, reliability, andadditional communication metrics like end-to-end latency ofthese other technologies.

V. RELATED WORK

The work in [13] is closest to our work by consideringa wireless backhaul architecture. Their focus is however onenhancements to the air interface and network protocols in-stead of network capacity. The work in [14] considers mostlast-mile options for telecommunication in a Smart Grid,especially backhaul solutions for the distribution network.The authors of [15] considered PLC in low- and medium-voltage distribution grids to connect network nodes (e.g.,meters, actuators, sensors) through multi-hop transmission.The authors in [16] proposed a unified solution for AdvancedMetering Infrastructure (AMI) integration with a DistributionManagement System (DMS). They found that a challenge ofthe integration of AMI and DMS is that it entails differentcommunication protocols and requirements for handling vari-ous meter information models.

While the prior work above discusses how to design thecommunication network at the distribution level for AMI,some researchers have looked beyond. For example, [17]discussed one of the key components of a future Smart Gridcalled load leveling and elaborates how existing techniquesfrom computer networking research could be potentially ap-plied to solve these problems. The authors of [18] describedan approach to modeling wireless communications at thelink layer of the power grid, with emphasis on employing amedium access control/physical layer model to measure linkperformance in terms of reliability, delay, and throughput.Some researchers believe that wireless communication is notenough to meet the entire needs of Smart Grid communication.For example, the authors in [19] describe a hybrid Wireless-Broadband over Power Lines (W-BPL) technology. They be-lieve that this combination is suitable for rural and remoteareas.

REFERENCES

[1] “Title XIII of the Energy Independent Security Act,” available online athttp://energy.gov/sites/prod/files/oeprod/DocumentsandMedia/EISA Title XIII Smart Grid.pdf.

[2] US Department of Energy. (2008) What the Smart Grid Meansto Americans. by Consumer Advocates group in a series ofbooks of Smart Grid stakeholder groups. [Online]. Available:http://www.doe.gov/sites/prod/files/oeprod/DocumentsandMedia/ Con-sumerAdvocates.pdf

[3] C. Bennett and D. Highfill, “Networking ami smart meters,” in IEEEEnergy 2030 Conference, November 2008, pp. 1–8.

[4] V. Aravinthan, V. Namboodiri, S. Sunku, and W. Jewell, “Wirelessami application and security for controlled home area networks,” inProceedings of the Power and Energy Society (PES) General Meeting,ser. IEEE PES GM, 2011.

[5] P. Kulkarni, S. Gormus, Z. Fan, and B. Motz, “A self-organising meshnetworking solution based on enhanced rpl for smart metering com-munications,” in World of Wireless, Mobile and Multimedia Networks(WoWMoM), 2011 IEEE International Symposium on a, june 2011, pp.1–6.

[6] B. Karimi, V. Namboodiri, V. Aravinthan, and W. Jewell, “Feasibility,Challenges, and Performance of Wireless Multi-Hop Routing for FeederLevel Communication in a Smart Grid,” in In Proceedings of the 2nd In-ternational Conference on Energy-Efficient Computing and Networking(e-Energy), New York, May 2011.

[7] D. Bernaudo et al., “SmartGrid/AEIC AMI Interoperability StandardGuidelines for ANSI C12.19 / IEEE 1377 / MC12.19 End Device Com-munications and Supporting Enterprise Devices, Networks and RelatedAccessories,” The Association of Edison Illuminating Companies, Meterand Service Technical Committee report version 2, November 2010.

[8] E. E. Queen, “A Discussion of Smart Meters And RF Exposure Issues,”Edison Electric Institute (EEI), Washington, D.C, A Joint Project of theEEI and AEIC Meter Committees, March 2011.

[9] J. Li, C. Blake, D. S. D. Couto, H. I. Lee, and R. Morris, “Capacity of adhoc wireless networks,” in Proceedings of the 7th annual internationalconference on Mobile computing and networking, ser. MobiCom ’01.New York, NY, USA: ACM, 2001, pp. 61–69.

[10] A. Agarwal, C. Dutta, and D. Sanghi, “Capacity of ad hoc wirelessnetworks,” 2002.

[11] K. V. K. Fall, The NS Manual (formally known as NS Documentation),Information Science Institute(ISI), The University of Southern Califor-nia, May 2009.

[12] 802.11, IEEE Standard for Information technology Telecommunicationsand information exchange between systems Local and metropolitanarea networks Specific requirements Part 11: Wireless LAN MediumAccess Control (MAC)and Physical Layer (PHY) Specifications,LAN/MAN Standards Committee of The IEEE Computer SocietyStd., Rev. of IEEE Std 802.11-1999, June 2007. [Online]. Available:http://standards.ieee.org/about/get/802/802.11.html

[13] R. Mao and V. Julka, “Wireless broadband architecture supportingadvanced metering infrastructure,” in IEEE Vehicular Technology Con-ference (VTC Spring), May 2011, pp. 1–13.

[14] D. Laverty, D. Morrow, R. Best, and P. Crossley, “Telecommunicationsfor smart grid: Backhaul solutions for the distribution network,” in Powerand Energy Society General Meeting, 2010 IEEE, july 2010, pp. 1 –6.

[15] M. Biagi and L. Lampe, “Location assisted routing techniques for powerline communication in smart grids,” in Smart Grid Communications(SmartGridComm), 2010 First IEEE International Conference on, Oc-tober 2010, pp. 274 –278.

[16] L. Zhao, Z. W, J. Tournier, W. Peterson, L. Wenping, and W. Yi, “Aunified solution for advanced metering infrastructure integration witha distribution management system,” in Smart Grid Communications(SmartGridComm), 2010 First IEEE International Conference on, Oc-tober 2010.

[17] S. Gormus, P. Kulkarni, and F. Zhong, “The power of networking: Hownetworking can help power management,” in Smart Grid Communica-tions (SmartGridComm), 2010 First IEEE International Conference on,October 2010, pp. 561 –565.

[18] M. Souryal, C. Gentile, D. Griffith, D. Cypher, and N. Golmie, “Amethodology to evaluate wireless technologies for the smart grid,”in Smart Grid Communications (SmartGridComm), 2010 First IEEEInternational Conference on, October 2010, pp. 356 –361.

[19] G. Tsiropoulos, A. Sarafi, and P. Cottis, “Wireless-broadband over powerlines networks: A promising broadband solution in rural areas,” inPowerTech, 2009 IEEE Bucharest, 28 2009-july 2 2009, pp. 1 –6.

66