The energy consumed by ICT

Marco Ajmone Marsan Politecnico di Torino, Italy IMDEA Networks Institute, Spain

OUTLINE

§ Energy trends § The ICT role § Cellular networks

– Sleep modes – Renewable sources

§ Policy matters

Marco Ajmone Marsan 2

Towards Real Energy-efficient Network Design

TREND: The FP7 Network of Excellence on Green Networking

The Problem

§  Energy is becoming the issue of our future o Energy production with fossil fuels causes GHG

emissions which produce climate changes o Energy is becoming expensive

§  Energy efficiency is a goal in all sectors, ICT and networking included

Marco Ajmone Marsan 4

Fossil fuel prices

Marco Ajmone Marsan

Source: Trends to 2050, European Commission.

Energy price is growing

5

Electricity prices

Marco Ajmone Marsan

Source: Trends to 2050, European Commission.

Electricity price is also growing in the next 10 years

6

Electricity demand

Marco Ajmone Marsan

Source: Trends to 2050, European Commission.

Electricity demand is growing

7

Electricity generation

Marco Ajmone Marsan

Source: Trends to 2050, European Commission.

Energy from renewable sources will grow

8

§  Information and Communication Technologies (ICT) play a positive role for energy efficiency: –  moving bits instead of atoms

•  intelligent transport systems •  teleworking and telecommuting •  e-commerce •  electronic billing

–  new manufacturing systems –  sensors to monitor and manage our environment

•  smart buildings, neighborhoods, cities …

What about ICT?

Marco Ajmone Marsan 9

§  ICT will allow savings of the order of –  25-30% in manufacturing –  20-30% in transport –  5-15% in buildings

for a total of about 17-22% §  Moreover, ICT is expected to significantly improve the

energy generation, transport and utilization through the adoption of Smart Grids

Source: Ad-hoc Advisory Group “ICT for Energy Efficiency” of the European Commission DG INFSO, 2008.

Marco Ajmone Marsan

What about ICT?

10

However, ICT is also part of the problem §  ICT energy consumption is huge and increasing

Marco Ajmone Marsan

FP7-ICT-257740/ D1.6

Page 9 of 39

3.1.4 Overall trends and observations In Figure 2 we show the evolution of the worldwide electricity use of communication

networks, PCs and data centers. Growth trends — Over the last five years, the yearly growth of all three individual

ICT categories (10%, 5%, and 4% respectively) is higher than the growth of worldwide electricity consumption in the same time frame. The combined electricity consumption of communication networks, personal computers and data centers is growing at a rate of nearly 7% per year (i.e., doubling every 10 years), with the strongest growth observed in communication networks. All growth rates have decreased compared to what we predicted in a similar study five years ago. This can partly be attributed to a shift to more energy efficient technologies (such as from CRT to LCD monitors, and the introduction of server virtualization), and potentially to the effects of the global financial crisis in 2008.

Absolute power consumption — In 2012 each category accounts for roughly 1.5% of the worldwide electricity consumption. This highlights the need for energy-efficiency research across all three domains, rather than focusing on a single one. Taken together, the relative share of this subset of ICT products and services in the total worldwide electricity consumption has increased from about 3.9% in 2007 to 4.6% (or 900 TWh) in 2012.

0

5�000

10�000

15�000

20�000

25�000

0

50

100

150

200

250

300

350

400

2007 2008 2009 2010 2011 2012

Total�w

orldwide�electricity�use�(T

Wh/y)

ICT�category�electricity�use�(TWh/y)

Total�worldwide�electricity��useCAGR�2.9%

NetworksCAGR�10.4%

PCsCAGR�5.1%

Data�centersCAGR�4.4%

The�CAGR�value�is�the�compounded�annual�growth��rate�over�the�2007 to�2012�time� frame

Figure 2: Evolution of worldwide electricity use of networks, PCs and data centers (solid lines, left axis) and total worldwide electricity use (dotted line, right axis).

In the future, frequent estimates of the worldwide electricity use by ICT will be essential to provide timely feedback if indeed ICT electricity consumption remains relatively small, or instead continues to grow at an unsustainable rate.

Publications:

[1] W. Van Heddeghem, S. Lambert, B. Lannoo, D. Colle, M. Pickavet, P. Demeester, “Trends in worldwide ICT electricity consumption from 2007 to 2012”, submitted to Computer Communications.

[2] S. Lambert, W. Van Heddeghem, W. Vereecken, B. Lannoo, D. Colle, M. Pickavet, Worldwide electricity consumption of communication networks, Optics Express 20 (2012) B513–B524.

Source: TREND Final Deliverable on “Assessment of power consumption in ICT”, 2013 and S. Lambert et al., “Worldwide electricity consumption of communication networks,” Optics Express, Vol. 20, No. 26, 2012.

network consumption faster than

other sectors

1-1.5% of total

11

Traffic growth

Source: Cisco VNI, 2014.

number of devices grows (new markets)

new and more traffic intensive

services

Marco Ajmone Marsan Marco Ajmone Marsan 12

The case of mobile networks §  The mobile industry sums to 0.5% of global emissions §  Electricity cost is up to 70% of mobile operators OPEX!

Radio access networks are the prime target for

energy saving

Marco Ajmone Marsan

Source: S. Vadgama and M. Hunukumbure, Trends in Green Wireless Access Networks, IEEE ICC 2011.

13

Traffic growth will make it “worse”

Source: Cisco VNI Mobile, 2014.

Mobile network traffic is expected to grow by an order of

magnitude in 5 years!!!

n  Larger no. of devices (new markets)

n  More traffic intensive services

Marco Ajmone Marsan 14

Which segment of the network?

Terminals: Already energy-efficient by design

Core devices: •  Energy-hungry •  Not many •  Very critical •  Handle aggregate traffic

Marco Ajmone Marsan 15

Which segment of the network?

Access network: •  Energy-hungry •  Many devices •  Some redundancy •  Less critical than core devices •  Very close to the user, hence high traffic variability

Marco Ajmone Marsan 16

Mobile networks

3 billion x 0.1 W = 0.3 GW

3 million x 1.5 kW = 4.5 GW

10,000 x 10 kW = 0.1 GW

According to an estimate of Nokia Siemens Networks, worldwide…

Marco Ajmone Marsan 17

Energy efficient solutions §  Traditionally, little attention to consumption in the design

of the devices and of the network §  Many inefficiencies derive from the combination of

–  little load proportionality of consumption –  due to traffic variability, long periods under low to medium load

Marco Ajmone Marsan

15

00.20.40.60.810

10

20

30

40

50

60

70

80

90

100

Parameter L

Savin

g [%

]

Upper BoundDouble Switch Off (Max)1/2−2/31/2−3/42/3−3/43/4−8/9

Fig. 4. Double switch-off: network saving versus the parameter L for the synthetic traffic profile.

0

0.2

0.4

0.6

0.8

1

00:00 05:00 10:00 15:00 20:00 00:00

Traf

fic, f

(t)

Time, t [h]

WeekdayWeekend

Fig. 5. Business cell: weekday and weekend traffic profiles.

TABLE ICASE STUDY: SAVINGS WITH DIFFERENT SWITCH-OFF SCHEMES

Switch-off scheme S[%] - Business weekday S[%] - Consumer weekday

Single (8/9) 44.2 49.3Double (5/9)-(8/9) 47.5 51.0Triple (3/9)-(5/9)-(8/9) 48.7 51.3Maximum (Least-Loaded) 50.0 52.0

Inefficiency Pidle>>0

Psleep≈0

load

pow

er

18

Energy efficient solutions

§  Consume energy for capacity deployment, not for capacity usage

§  Many solutions try to make capacity adaptive to load –  Use of sleep modes for the BSs –  Network sharing –  Resource on demand architectures

§  Saving of up to 50% can be achieved,

Marco Ajmone Marsan 19

Planning and dimensioning: long time scales

night lunch

peaks

over-provisioning:“waste”ofcapacity

highcapacitytocarryallthetraffic

Daily traffic profile in a segment

Sleep modes of devices or portions of the network

•  Due to natural traffic variability, the network is over-dimensioned and wastes energy for long periods of time

Adapt capacity to actual traffic needs by putting to sleep mode

devices or portions of the network when traffic is low

Network planning with sleep modes

Sinceenergyconsump9ononlymarginallydependsonload,

over-provisioningàwasteofenergy

Adapt capacity to traffic by using sleep modes

Michela Meo – Politecnico di Torino

peaks define

planning

peak scheme

off-peak scheme

Network sharing

•  Severalcompe&ngmobileoperatorscoverthesameareawiththeirequipment

•  Networksaredimensionedoverthepeakhourtraffic•  Duringlowtrafficperiodstheresourcesofoneoperatoraresufficienttocarryallthetraffic

Make operators cooperate to reduce energy consumption

Michela Meo – Politecnico di Torino

n  Inturn,q  Switchoffthenetworkofoneoperator,whentrafficislowandtheac9veoperatorscancarryallthetraffic

q  Letusersroamtootheroperatorsq  Balancecosts

Networksharing

Michela Meo – Politecnico di Torino

Source: M. Meo, M. Ajmone Marsan. Energy efficient wireless Internet access with cooperative cellular networks. Computer Networks, Volume 55, Issue 2, February 2011, Pages 386-398.

Case study: Some European Countries

Country MNOs Market share [%] Subscr. [M]France 3 46 36 19 - 58.2

Germany 4 32 31 21 16 113.6Greece 3 51 28 21 - 15.4Italy 3 38 36 26 - 84.0

Netherlands 3 46 26 28 - 19.0Poland 4 29 29 28 14 47.5

Portugal 3 45 40 15 - 16.4Spain 3 44 34 22 - 51.4

Romania 3 41 32 26 - 24.2Russia 3 37 33 30 - 189.7

Ukraine 3 48 37 15 - 52.3U.K. 3 39 33 28 - 68.5

Table 1: Characteristics of the considered countries: Numberof MNOs offering both 2G and 3G services, market share foreach of the MNOs, total number of subscribers.

3. ENERGY BENEFITS IN EUROPEIn this section, we assess the effectiveness of network sharing interms of achievable energy saving by considering a number of Eu-ropean countries. In particular, we focus on the 12 countries indi-cated in Table ??, which are the countries whose total number ofsubscribers is larger than 15 Millions, according to publicly avail-able data.

For each country we collect approximate data about the number ofsubscribers for each of the active MNOs and the kind of providedservices. We then assume that network sharing is applicable onlyamong the MNOs that offer both 2G and 3G services. Indeed, aMNO offering access to 2G terminals cannot switch off its networkand make the users roam to a purely 3G network. In this case theoperator would probably switch off the 3G network leaving the 2Gaccess network on; however, since we only have aggregated dataabout the total number of subscribers and not the breakdown withrespect to the technology, we make the simplistic assumption thatnetwork sharing is implemented only among MNOs offering ser-vices to both 2G and 3G users.

Interestingly, the considered European countries present quite sim-ilar scenarios. As summarized in Table ??, except for two cases,namely Germany and Poland, all considered countries have 3 MNOsoffering both 2G and 3G services with relatively fair share of mar-ket. The smallest of the 3 MNOs has a share that is usually between20 and 30%, only in the case of Ukraine and Portugal the smallestof the three operators accounts for as low as 15% of the subscribers.Conversely, the largest of the 3 MNOs exceeds 50% of the marketshare only in Greece, where it is about 51%; otherwise, it is be-tween 37% and 48%. The case of Germany, with 4 MNOs, is in-teresting because it presents two dominant operators with about thesame number of subscribers, 36 millions, corresponding to 31% ofthe market, and other two smaller MNOs that share the remainingmarket. In Poland, three operators are about the same size, withalmost 30% of the share each, while the fourth operator accountsfor 14% of the market only. This substantial similarity of the situa-tions is probably due to historical reasons: in most of the Europeancountries similar network evolutions occurred roughly at the sametime.

We compute the energy saving achievable through network sharingfor each of the selected countries, and for both the consumer and

Figure 3: Saving achievable with network sharing in the Euro-pean countries with more than 15M subscribers; business andconsumer profiles, constant and variable cost models.

Figure 4: Utilization achievable with network sharing in theEuropean countries with more than 15M subscribers; businessand consumer profiles, week-days and week-ends.

the business traffic profiles shown in Fig. ?? (implicitly assumingthat the traffic profiles in Fig. ?? can be representative of trafficin all considered countries). Given a traffic profile and a country,we consider all the possible switch-off patterns, i.e., all the possi-ble orderings in which the MNOs of that country might switch off.Savings are obtained as described in the previous section, by de-riving switch-off and switch-on instants from (??), and by comput-ing saving from (??). Both the cases of variable and constant costmodels are evaluated. The saving achievable during week-days andweek-ends are properly weighted to get the average weekly saving.

Fig. ?? reports the maximum achievable energy saving, amongthose obtained from different switch-off patterns in a given sce-nario. The savings are really significant, typically larger than 40%:this confirms that network sharing, besides being a viable approach,already feasible with today technology, is very promising in termsof energy consumption reduction.

Observe also from the figure that the business traffic profile leadsto the largest saving. This is due to the profile having particu-larly steep transitions between peak and off-peak, and long peri-ods of very low traffic. Clearly, in reality, large service areas arecharacterized by a mixture of neighborhoods, some mainly withbusiness-like behavior of the users and others with consumer-liketraffic profiles. A switch-off scheme should then be applied byadapting, neighborhood by neighborhood, switching times to thespecific profiles. For example, a MNO that is going to switch-offits access network, might probably start from portions of the net-work in business areas, as soon as traffic drops below some thresh-old; some time later, when traffic drops also in the consumer ar-eas, other portions of the access network would be powered off. Interms of saving, this means that the achievable saving will be in be-tween what can be obtained from a business area and a consumerarea, with actual values depending on the traffic profiles and on theproportions of areas with business-like or consumer-like behavior.In case of some spare capacity, deployed to absorb medium termtraffic growth, some additional saving can be expected. With anoverprovisioning factor 1 + x = 1.2, for example, it is possible toreach savings between 50 and 59% for the consumer profile and be-tween 53 and 63% for the business profile under the constant costmodel. These values are even closer to the maximum theoreticalsaving that would be achieved when one network only has enoughcapacity to carry all the traffic; the maximum theoretical saving isequal to 66% for 3 MNOs, corresponsing to 1 network over threethat is carrying traffic, and it is equal to 75% for 4 MNOs.

A positive side-effect of network sharing is that active resources aremore effectively used than in traditional scenarios without sharing.Indeed, network sharing aims at reducing energy wastage that de-rives from daily periods of over-provisioning by making the avail-able capacity more closely follow the traffic profile. To evaluatethis effect, we compute the daily average utilization of the accessnetwork resources, by dividing the amount of generated traffic by

Case study: Some European Countries

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Fran

ce

Ger

man

y

Gre

ece

Italy

Net

herla

nds

Pola

nd

Portu

gal

Spai

n

Rom

ania

Rus

sia

Ukr

aine UK

Rel

ativ

e sa

ving

consumer, constconsumer, var

business, constbusiness, var

Energy efficient solutions

§  Consume energy for capacity deployment, not for capacity usage

§  Many solutions try to make capacity adaptive to load –  Use of sleep modes for the BSs –  Network sharing –  Resource on demand architectures

§  Saving of up to 50% can be achieved,

Marco Ajmone Marsan

is this enough?

27

Micro and macro effects of energy efficiency

Increase of energy efficiency to produce a good/service

reduces cost of the production and, hence, its price

increases the demand

increases the energy consumption

Jevons paradox:

Marco Ajmone Marsan 28

From energy efficient networking to sustainable networking

n  Energy efficiency is good since, through cost reduction, it leads to q  Higher production à better quality of life for more people q  Reduction of price increase and energy shortage q  Global environmental advantage if coupled with green taxes to

keep the price constant

n  but, for sustainability, it must be coupled with new energy generation principles (RES – Renewable Energy Sources)

Marco Ajmone Marsan 29

Powering BSs with renewables

n  Zero grid-Electricity Networking (ZEN) BSs rely purely on renewable energy sources and are not connected to a power grid q  Can acquire limited amounts of energy from (intermittent) local

generators exploiting renewable sources q  Any energy surplus is stored in a battery q  The BS can operate also in periods of low or no production, as

long as energy is available, but it is forced to switch off when the battery is depleted

n  Hybrid systems that, when battery is depleted, rely also on the power grid (or a secondary generator)

Marco Ajmone Marsan 30

n  Power BSs with Renewable Energy Sources q  To deploy networks where the power grid does not exist, is not

ubiquitous or is not reliable q  To achieve extremely low carbon footprints q  To reduce operational cost q  To survive to power grid outages or natural disasters which damage

the power grid

Possible scenarios: 1.  New opportunities for the deployment of networks in emerging regions

Marco Ajmone Marsan

Powering BSs with renewables

31

Marco Ajmone Marsan

Possible scenarios: Emerging areas

32

Possible scenarios: Emerging areas

§  Expected expansion of mobile networks in rural regions in Africa and Asia

§  Most of the BSs are powered with diesel generators §  Costly (especially due to fuel

transportation) §  Bad for the environment

Marco Ajmone Marsan

Source: “Green Power Design Approach and Feasibility Analysis,” Green power for Mobile Technical White Paper, Aug. 2014.

GPM Technical Paper, June 2014

6

However, the implementation of green power alternatives is far from reaching its true potential. There are 320,000 off-grid and over 700,000 unreliable-grid telecom sites in the world today (2014)1. The off-grid and bad-grid network globally is estimated to reach a total of approximately 1.2 million tower sites by 2020 from the current size of 1 million off-grid and bad-grid towers in 2014. Therefore, The MNOs and Tower Cos will deploy an additional 160,000 off-grid and bad-grid tower sites by 2020.

Figure 1: Global Off-grid and Unreliable-grid Mobile Network: Current Size and Future Growth

The major driver of the estimated growth in off-grid and bad-grid towers is the expected expansion of mobile networks into rural regions in Africa and Asia, large parts of which face limited access to reliable grid electricity and poor grid power infrastructure. Therefore, green alternatives for telecom power present a huge opportunity for MNOs and other stakeholders. This technical paper will focus on the aspects of analyzing the technical and economic feasibility of green power and the important parameters as well as clear approach to understanding the strategy and benefits of green power for telecoms.

1 GSMA GPM-Dalberg Research and Analysis, June 2014

bad grid

off-grid

33

Possible scenarios: Emerging areas

Marco Ajmone Marsan

Source: “Green Power Design Approach and Feasibility Analysis,” Green power for Mobile Technical White Paper, Aug. 2014.

GPM Technical Paper, June 2014

6

However, the implementation of green power alternatives is far from reaching its true potential. There are 320,000 off-grid and over 700,000 unreliable-grid telecom sites in the world today (2014)1. The off-grid and bad-grid network globally is estimated to reach a total of approximately 1.2 million tower sites by 2020 from the current size of 1 million off-grid and bad-grid towers in 2014. Therefore, The MNOs and Tower Cos will deploy an additional 160,000 off-grid and bad-grid tower sites by 2020.

Figure 1: Global Off-grid and Unreliable-grid Mobile Network: Current Size and Future Growth

The major driver of the estimated growth in off-grid and bad-grid towers is the expected expansion of mobile networks into rural regions in Africa and Asia, large parts of which face limited access to reliable grid electricity and poor grid power infrastructure. Therefore, green alternatives for telecom power present a huge opportunity for MNOs and other stakeholders. This technical paper will focus on the aspects of analyzing the technical and economic feasibility of green power and the important parameters as well as clear approach to understanding the strategy and benefits of green power for telecoms.

1 GSMA GPM-Dalberg Research and Analysis, June 2014

bad grid

off-grid In India, §  about 400,000 BSs §  75% of rural sites and 33% of urban

sites on RES by 2020

Source: K. Tweed, “Why Cellular Towers in Developing Nations are Making the Move to Solar Power,“ Scientific America, 15 Jan. 2013.

Renewables can be convenient n  Cost reduction n  Environmental concerns n  Political issues (related to

energy supply)

34

ZEN – Zero grid-Electricity Networking n  Power BSs with Renewable Energy Sources

q  To deploy networks where the power grid does not exist, is not ubiquitous or is not reliable

q  To achieve extremely low carbon footprints q  To reduce operational cost q  To survive to power grid outages or natural disasters which damage

the power grid

Possible scenarios: 1.  New opportunities for the deployment of networks in emerging regions 2.  New business models (high electricity price, green incentives and

environmental awareness) 3.  Dense urban areas where cabling is a problem 4.  Critical infrastructure protection 5.  …

Marco Ajmone Marsan 35

Pure PV (ZEN)

Marco Ajmone Marsan 36

Hybrid PV-Grid

Marco Ajmone Marsan

Connect to the Power Grid to reduce size of PV panel & number of batteries

37

Hybrid PV-diesel

Marco Ajmone Marsan

Use diesel instead of the grid as a backup

38

Two locations

!

H`!!

!Figura 28: Radiazione solare annuale in Europa (Fonte: PVGIS, Institute for Energy - Unione Europea,

2012)

!Figura 29: Radiazione solare annuale in Africa (Fonte: PVGIS, Institute for Energy - Unione Europea, 2012)

!

Torino(medium:4kWh/m2,

highseasonalvaria9ons)

Aswan(high:6.8kWh/m2,

lowseasonalvaria9ons)

!

H`!!

!Figura 28: Radiazione solare annuale in Europa (Fonte: PVGIS, Institute for Energy - Unione Europea,

2012)

!Figura 29: Radiazione solare annuale in Africa (Fonte: PVGIS, Institute for Energy - Unione Europea, 2012)

!

Solarradia9on

Solarradia9on

Marco Ajmone Marsan 39

Real traffic profile

Marco Ajmone Marsan

§  Traffic grows according to 50% Compound Average Growth Rate (CAGR) –  Focus on LTE, with initial low traffic load (0.02) due to start-up of the new

technology and maximum (0.75) in 10 years

Source: “Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 2013-2018,” http://www.cisco.com/

40

Real traffic profile

Marco Ajmone Marsan

§  Traffic grows according to 50% Compound Average Growth Rate (CAGR) –  Focus on LTE, with initial low traffic load (0.02) due to start-up of the new

technology and maximum (0.75) in 10 years

Source: “Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 2013-2018,” http://www.cisco.com/

15

00.20.40.60.810

10

20

30

40

50

60

70

80

90

100

Parameter L

Savin

g [%

]

Upper BoundDouble Switch Off (Max)1/2−2/31/2−3/42/3−3/43/4−8/9

Fig. 4. Double switch-off: network saving versus the parameter L for the synthetic traffic profile.

0

0.2

0.4

0.6

0.8

1

00:00 05:00 10:00 15:00 20:00 00:00

Traf

fic, f

(t)

Time, t [h]

WeekdayWeekend

Fig. 5. Business cell: weekday and weekend traffic profiles.

TABLE ICASE STUDY: SAVINGS WITH DIFFERENT SWITCH-OFF SCHEMES

Switch-off scheme S[%] - Business weekday S[%] - Consumer weekday

Single (8/9) 44.2 49.3Double (5/9)-(8/9) 47.5 51.0Triple (3/9)-(5/9)-(8/9) 48.7 51.3Maximum (Least-Loaded) 50.0 52.0

16

0

0.2

0.4

0.6

0.8

1

00:00 05:00 10:00 15:00 20:00 00:00

Traf

fic, f

(t)

Time, t [h]

WeekdayWeekend

Fig. 6. Consumer cell: weekday and weekend traffic profiles.

11:00 15:00 19:00 23:00 03:0011:00

15:00

19:00

23:00

03:00

τ1

τ 2

−5

0

5

10

15

20

25

30

35

40

1/2−2/31/2−3/42/3−3/43/4−8/9

Fig. 7. Business cell - weekday profile: network saving with double switch-off.

business residential

41

Total cost in 10 years

Marco Ajmone Marsan

Hybrid saves from the 8th year wrt the grid-only case

Torino

Hybrid is always cheaper than pure diesel

Pure PV saves from the 5th year wrt pure diesel

42

Fig. 6. Total cost for some values of PT , no energy selling

Observing the results in Fig. 6, we see that in the case withPT =100%, batteries need to be replaced only once, at the8-th year; instead, for all other cases, batteries are replacedtwice, as can be seen from the small steps in cost. In addition,batteries are added along the time axis, when necessary. Thepink line refers to the case of a BS powered from the grid only(i.e., with no PV panel and no battery). The results show thatthe hybrid cases, in which some energy can be bought fromthe grid, achieve significant reductions of the total cost withrespect to the fully renewable (off-grid) solution. The total costafter 10 years for the cases of PT =70% and PT =80% isabout 75% lower than the pure solar system. These solutionsare also cheaper than the grid-only case, starting from the 8-thyear on. The cost of using a diesel generator is much higherthan those of hybrid and pure grid solutions, and becomesquickly much higher than the pure solar solution.

We now consider the case in which the extra energyproduced by the PV panel, that cannot be stored in alreadyfully charged batteries, can be sold back to the power grid.Fig. 7 shows the cumulative total cost, which is initially equalto the CAPEX, and then decreases with time when there is apositive balance between the revenues obtained by selling theextra energy, and the cost of buying energy to power the BSwhen the batteries are depleted. In the cases with PT =70%and 80% (which are the most effective when energy cannotbe sold back, as in Fig. 6), the curves are almost flat, meaningthat OPEX and energy selling roughly balance.

When the system is dimensioned for PT =100%, instead,the size of the PV panel must be quite large, in order toguarantee that even in periods with low production, suchas winter, the produced energy is enough to power the BS;this implies a large extra production of renewable energyin other periods, which translates into significant revenues,so that the corresponding cost curve in the figure is steeplydecreasing. However, this case normally corresponds to an off-

Fig. 7. Total cost for some values of PT , with extra-energy selling

grid operation, so that the BS cannot inject the extra energyproduction into the power grid, and cannot obtain the revenuesassociated with the extra energy sellback, unless some specificpower distribution system is envisioned. For this reason, inthe case of PT =100% we report in Fig. 7 both curves withand without the extra energy sellback, the latter being morerealistic.

In all the considered cases, except the one with PT =100%and no energy sellback, the systems with renewable energystart being convenient with respect to the grid-only case, 5 or6 years after installation.

In the off-grid case, the pure solar solution is more ex-pensive than a diesel generator only in the first years, whenCAPEX dominate, but quickly becomes much less expensive,and at the end of the 10-year period the cost of the pure solarsolution is about two thirds of the cost with a diesel generator.

Similar results were derived for other cases, such as forthe business traffic profile, but are not shown for the sake ofconciseness. We also studied the case of other locations, wherethe variation of solar radiation over a year is much smaller thanin the considered case. Results for those cases indicate that theoverall system cost after 10 years is rather insensitive to thechosen value of the parameter PT , thus favoring an off-gridsolution. This suggests that hybrid solutions are particularlysuited to locations with seasonal variations of solar radiation.

IV. CONCLUSIONS

In this paper, we studied the dimensioning of a poweringsystem for a LTE macro BS composed of a PV panel, a set ofbatteries, and a connection to the power grid. We considereda scenario in which the amount of traffic in the BS and theelectricity cost increase with time according to well-knownpredictions, while the PV panel efficiency decreases due topanel degradation. We showed that the hybrid approach, thatuses both renewable energy and the power grid, can reduce thetotal cost of up to 75% with respect to a pure solar system,

Selling back energy §  Total cumulative cost, with the possibility to sell back

energy

Marco Ajmone Marsan

Save from the 6th year

43

Conclusion and discussion

Awareness about the huge energy consumption of communication networks has led to development of more energy efficient solutions and technologies However, for sustainability, energy efficiency alone is not enough

Marco Ajmone Marsan 44

Conclusion and discussion

Networking should be combined with new energy generation principles Powering BS with renewable sources

–  is cost effective –  allows service provisioning in disadvantaged areas –  is well suited to critical infrastructure protection –  allows communications after natural disasters –  avoids cabling difficulties

Marco Ajmone Marsan 45

Policy matters

§  Recommend energy reduction targets for European MNOs (and manufacturers) –  Identify adequate metrics

•  power •  power/throughput •  power/(throughput*area) •  power/(throughput*area*users)

§  Specify minimum fraction of RES §  Facilitate resource (network) sharing

–  Better energy efficiency in coverage-limited areas –  Better energy efficiency in capacity-limited areas

Marco Ajmone Marsan 46

Questions?

Marco Ajmone Marsan 47

Thank you!