Progress Meeting of the PhD Thesis

Ahmad Mahbubul Alam

17.09.2014
Objective

Discussion based on the action items decided on the meeting held

on 22.07.14

2 / 47
Action Items for This Meeting

1. Improve the state of the art report and send.

2. Look at the power consumption model in Communication
magazine (Should be sent by Dr. Jean-Yves Baudais).
3. Express EE in terms of SE for the AWGN channel and draw it
for ratios of different parts of the power consumption model.
4. Rebuild the presentation slide (Latex beamer).

3 / 47
Outline

Power Consumption Models

Relation between Energy Efficiency (EE) and Spectral Efficiency

(SE) for AWGN Channel

Performance Analysis of Interference Limited Cellular Network

Applying Poisson Point Process (PPP)

Future directions

4 / 47
 Section 1

Power Consumption Models

5 / 47
Model One
Ï Macro BS
P = aPtx + b

Ï Micro BS
P = a(l)Ptx + b(l)

Ï P average consumed power per BS

Ï Ptx transmit power per BS
Ï a, a(l) coefficient for power consumption due to amplifier,
feeder loss, cooling (only macro) that scales with transmit
power
Ï b , b(l) Power that is independent of transmit power and
includes consumption due to signal processing, battery backup
and cooling (only macro)
Ï l traffic load ranging from 0 (no load) to 1 (maximum load)
A. J. Fehske, F. Richter, and G. P. Fettweis, “Energy efficiency improvements
through micro sites in cellular mobile radio networks,” in GLOBECOM, pp. 1–5,
IEEE, 2009 6 / 47
Model Two

Ï Macro BS

P = aPtx + b

Ï Micro BS

P = l(aPtx + b)

G. Koutitas and P. Demestichas, “A review of energy efficiency in telecommuni-

cation networks,” Telfor journal, vol. 2, no. 1, pp. 2–7, 2010
7 / 47
Model Three
Ï Macro BS

P = Nsec Nant (aPtxant + b)

a = 3.77, b = 68.73

Ï Micro BS

P = a(l)Ptx + b(l)
a(l) = 4.44 + 1.11l
b(l) = 16.65 + 15.26l

Ï Nsec number of sectors per site

Ï Nant number of antennas per sector
Ï Ptxant transmit power per antenna
F. Richter, A. J. Fehske, P. Marsch, and G. P. Fettweis, “Traffic demand and
energy efficiency in heterogeneous cellular mobile radio networks,” in Vehicular
Technology Conference (VTC 2010-Spring), 2010 IEEE 71st, pp. 1–6, IEEE, 2010
8 / 47
Model four
Ï Macro BS

P = Nsec Nant (laPtxant + b)

Ï Micro BS

P = laPtx + b + lc

Ï c Maximum value of the part of the BS power consumption

that is independent of the transmit power but dependent on
load

H. Klessig, A. J. Fehske, and G. P. Fettweis, “Energy efficiency gains in

interference-limited heterogeneous cellular mobile radio networks with random
micro site deployment,” in Sarnoff Symposium, 2011 34th IEEE, pp. 1–6, IEEE,
2011
9 / 47
Model in Communication Magazine

P = aPtx + Mb When,0 < Ptx <= Pmax

= MPsleep When,Ptx = 0

Ï M=Number of transceiver chains

Ï Psleep = BS power in sleep mode

G. Auer, V. Giannini, C. Desset, I. Godor, P. Skillermark, M. Olsson, M. A. Imran,

D. Sabella, M. J. Gonzalez, O. Blume, et al., “How much energy is needed to run
a wireless network?,” Wireless Communications, IEEE, vol. 18, no. 5, pp. 40–49,
2011
10 / 47
General model

P = a(l)Ptx + b(l)
a(l) = c + dl
b(l) = e + fl

Ï c , d , e , f =Constants

11 / 47
 Section 2

Relation between Energy Efficiency (EE) and

Spectral Efficiency (SE) for AWGN Channel

12 / 47
System Model

Channel bandwidth = w [Hz]

Transmit power = Pt [W ] £ ¤
Noise spectral density = No W /Hz
Pt £ ¤
Channel capacity, C = w log2 (1 + wN o
) b/s
C Pt £ ¤
Spectral Efficiency, SE = w = log2 (1 + wN o
) b/s/Hz
Total link power, Ptotal = aPt + b [W ]
a.wNo (2SE −1)+b
Energy per bit, Eb = Ptotal aPt +b £ ¤
C = C = w .SE J/b
1 w .SE w .SE
£ ¤
So, EE = Eb = a.wN (2SE −1)+b = ³ ´ b/J
o
a wNo (2SE −1)+ ba

13 / 47
 7 EE vs SE for W=5MHz, N =10−21 W/Hz (−180 dBm/Hz)
x 10 o
2.5
EE for a=14.5,b/a=49.10 (Macro)
EE for a=6.35,b/a=16.69 (Micro)
EE for a=8.4,b/a=1.77 (Pico)
EE for a=15,b/a=0.67 (Femto)
2

1.5
EE (b/J)

0.5

0
0 10 20 30 40 50 60
SE (b/s/Hz)

G. Auer, V. Giannini, M. Olsson, M. J. Gonzalez, and C. Desset, “Framework for

energy efficiency analysis of wireless networks,” in 2nd International Conference
on Wireless VITAE, pp. 1–5, 2011
14 / 47
 Section 3

Performance Analysis of Interference Limited

Cellular Network Applying Poisson Point Process
(PPP)

15 / 47
System Model I
Ï Large scale network with Base Stations (BSs) and users
forming 2 independent homogeneous Poisson Point Process
(PPP) Φb and Φu respectively
Ï Only downlink direction is considered
Ï Total Bandwidth available in downlink direction is same for all
the BSs and the bandwidth is equally shared among the users
in the cell
Ï All BSs have the same transmit power and equally shared
among the users in a cell
Ï Users are served by the nearest BS
Ï No intra-cell interference, but users receive interference from
the BSs having the same frequency band
Ï All the users in a cell receive signals at a particular instant
D. Tsilimantos, J.-M. Gorce, and E. Altman, “Stochastic analysis of energy sav-
ings with sleep mode in ofdma wireless networks,” in INFOCOM, 2013 Proceed-
ings IEEE, pp. 1097–1105, IEEE, 2013
16 / 47
System Model II

Similar to the first model except that only a single user receives
signal at a particular instant and all the power and bandwidth
available in the cell is provided to that user

J. G. Andrews, F. Baccelli, and R. K. Ganti, “A tractable approach to coverage

and rate in cellular networks,” Communications, IEEE Transactions on, vol. 59,
no. 11, pp. 3122–3134, 2011
17 / 47
Problem Formulation

Analysing the performance metrics of the systems such as coverage

probability and Shannon capacity (ergodic) of a user at a random
distance from the BS for different BS densities and frequency reuse
factors. Then other performance metrics such as Area Spectral
Efficiency (ASE) and Energy Efficiency (EE) are to be determined
and their relation is to be demonstrated1 .

1
Although the coverage probability and Shannon capacity are derived in the ref-
erences, there are small changes. For example, the Shannon capacity of a user
in model 1 is here upper bounded by that of a user in model 2. Besides, ASE
and EE provided here are not available in the references.
18 / 47
Definition I

Ï PPP A Point Process (PP) is a PPP if and only if the number

of points in a bounded set B ⊂ R d is a poisson random
variable, and the number of points in disjoint sets are
independent
Ï Independent thinning of PPP Independent thinning of a
PPP results another PPP. This is obtained by either keeping or
deleting some points independently with a probability from the
parent PPP. The intensity measure of the thinned PPP is
obtained by multiplying the intensity measure of the parent
PPP by the probability
Ï Frequency Reuse Factor System bandwidth is divided into
subbands and the cells are assigned one of the subbands.
Number of subdivisions of the system bandwidth is called the
frequency reuse factor

19 / 47
Definition II
Ï Coverage probability of a user at a random distance from
the BS is defined as the probability that the received Signal to
Interference plus Noise Ratio (SINR) is greater than some
threshold (here, ρ )
Ï Shannon capacity (ergodic) of a user at a random distance
from the BS is the average of maximum achievable rate
corresponding to all the fading states. However, here the
averaging is done also over the point process
Ï Outage capacity is the maximum achievable rate
corresponding to the SINR threshold
Ï Area Spectral Efficiency (ASE) is defined as the number of
successfully transmitted bits per sec per Hz per unit area
Ï Energy Efficiency (AEE) is defined as the number of
successfully transmitted bits per Joule

20 / 47
Coverage Probability
Ï First System Model Coverage probability of a user at a
random distance from the connected BS is,
Z ∞ ¡ λ
¢
−πx 1+ λ i I(α,ρ )
Pc1 = π e b dx
0

R∞ ρ
Where, I(α, ρ ) = 1 α dz
ρ+z 2
Ï Second System Model Coverage probability of a user at a
random distance from the connected BS is,

Pc 2 = Pc 1
Ï λb BS density
Ï λi Interferring BS density
Ï α Path loss exponent
Ï ρ SINR threshold for coverage
21 / 47
Shannon Capacity (ergodic)
Ï First System Model Shannon capacity (ergodic) of a user at
a random distance from the connected BS is,
³ ´
∞ ∞ λ
−πx 1+ λ i I(α,θ1 )
Z Z
Cu1 = π e b
dxd ²
²>0 x >0

² ln2
Where, θ1 = e w − 1
Ï Second System Model Shannon capacity (ergodic) of a user
at a random distance from the connected BS is,
³ ´
∞ ∞ λ
−πx 1+ λ i I(α,θ2 )
Z Z
Cu2 = π e b
dxd ²
²>0 x >0

² ln2
W
Where, θ2 = e δ −1
22 / 47
Ï W System bandwidth in downlink
Ï δ Frequency reuse factor
W /δ W /δ W
³ ´ ³ ´
Ï w Average bandwidth per user=min M , W δ =min ,
λu /λb δ
Here, M=Average number of users per cell=λu /λb
λu =User density

23 / 47
Proof of Coverage Probability I

Desired received signal by a user at a distance R from the nearest

BS, Pr (R) = Pt hs l(R);
Where,
Pt =Link transmit power
hs =Fading variable for the desired signal power
l(R)=Path loss at distance R from the connected BS
Hence, SINR of a user at a distance R from the connected
P (R) P hs l(R)
BS= Norw +I = Nt o w +I
Where,No =Noise spectral density

24 / 47
Proof of Coverage Probability II

Coverage probability of a user whose nearest BS is at a distance R

from it,
³ P h l(R) ´
t s
Pc1 |user is at distance R = P >ρ
No w + I
³ ρ (No w + I) ´
= P hs >
Pt l(R)
τρ
= e −So No w E [e −So I ] hs ∼ exp(τ);So =
Pt l(R)
−So No w
=e L I (So )

25 / 47
Proof of Coverage Probability III
Now,
h i
L I (So ) = E e −So I
h P i
= EΦI ,hI,i e −So i ∈ΦI Pt hI,i I (ri ) [ΦI is point process for I]
hY i
−So Pt hI,i I (ri )
= EΦI ,hI,i i ∈ΦI e [hI,i are fading variables for I]
hY £ −S P h l(r ) ¤i
= EΦ I i ∈ΦI E h I, i e
o t I,i i
[ΦI and hI,i are independent]
hY £ −S P h I (r ) ¤i
= EΦ I i ∈ΦI E h I e o t I i (Assuming all hI,i to be same)
³ £ ¤´
R∞
− R 1−EhI e −So Pt hI l(r ) λi 2πrdr
=e (Applying PGFL for PPP ΦI )
−πλi R 2 I(α,ρ )
=e (using l(r ) = r −α , l(R) = R −α
hI ∼ exp(τ))

26 / 47
Proof of Coverage Probability IV
If fR (R)=Pdf of the distance to the connected BS from the user to
be R, then
Z ∞
Pc 1 = (Pc1 |user is at distance R)fR (R)dR
Z0 ∞
= e −So No w LI (So )fR (R)dR
Z0 ∞
2
I(α,ρ ) −λb πR 2
= e −So No w e −πλi R e 2πλb RdR
0
Z ∞ τρ No w
¡ ¢ α2 ¡ λ
¢
x
− −πx 1+ λ i I(α,ρ )
=π e Pt λb
e b dx
0
(Putting the value of So and letting λb R 2 = x)
Z ∞ τρ No w
¡ ¢ α2 ¡ λ
¢
x
− −πx 1+ λ i I(α,ρ )
=π e Pt λb
e b dx
0

27 / 47
Proof of Coverage Probability V

Similarly,
Z ∞ τρ No W α ¡ λ
¢
δ
− ( λx ) 2 −πx 1+ λ i I(α,ρ )
Pc 2 = π e Ptx b e b dx
0
[Ptx = Total transmit power of a BS]
∞ τρ No w α ¡ λ
¢
W
Z
− ( λx ) 2 −πx 1+ λ i I(α,ρ )
=π e Pt b e b dx [When, Pt = Ptx , w = ]
0 δ
Z ∞ τρ No wM α ¡ λ
¢
− ( λx ) 2 −πx 1+ λ i I(α,ρ )
=π e Pt M b e b dx
0
W
[When, Ptx = Pt M , = wM]
δ
= Pc 1

28 / 47
Proof of Coverage Probability VI

In interference limited scenerio (Neglecting noise),

Pc 1 = Pc 2
Z ∞ ¡ λ
¢
−πx 1+ λ i I(α,ρ )
=π e b dx
0

29 / 47
Proof of Capacity I

Z
Cu1 = (Cu1 |user is at distance R)fR (R)dR
ZR >0
2
= EΦI ,hI [w log2 (1 + SINR)]e −λb πR 2πλb RdR
R >0 Z Z
2
= 2πλb P[w log2 (1 + SINR) > ²]d ²e −λb πR RdR
ZR >0 Z²>0
² 2
= 2πλb P[SINR > 2 w − 1]d ²e −λb πR RdR
ZR >0 Z²>0
² ln2 2
= 2πλb P[SINR > e w − 1]d ²e −λb πR RdR
ZR >0 Z²>0 τθ1 No w
2 2
e Pt l(R) e −πλi R I(α,θ1 ) d ²e −λb πR RdR
−
= 2πλb
R >0 ²>0
Z ∞ Z ∞ τθ No w x α ¡ λ ¢
− 1P ( λ ) 2 −πx 1+ λ i I(α,θ1 )
=π e t b e b d ²dx
x >0 ²>0

30 / 47
Proof of Capacity II
² ln2
Here, θ1 = e w −1

Z ∞ Z ∞ τθ2 No W α
δ
− ( λx ) 2
Cu2 = π e Ptx b
x >0 ²>0
³ ´
λ
−πx 1+ λ i I(α,θ2 ) ² ln2
e b
d ²dx Here, θ2 = e W /δ − 1

For interference limited scenerio,

³ ´
∞ ∞ λ
−πx 1+ λ i I(α,θ1 )
Z Z
Cu1 = π e b
dxd ²
²>0 x >0

31 / 47
Proof of Capacity III ³ ´
∞ ∞ λ
−πx 1+ λ i I(α,θ2 )
Z Z
Cu2 = π e b
dxd ²
²>0 x >0

[When all the BSs are active and interference comes from all the active BSs
except the BS with which an user is connected, active/interferring BSs can also
be modeled as PPP based on the Slivnyak’s theorem. This theorem states that
for a PPP, the reduced Palm distribution equals the distribution of the PPP
itself. Palm distribution is the distribution of a point process conditioned to
have a point at x ∈ R d (In this work, BS position can be thought as x). Reduced
palm distribution is the distribution of the point process without counting the
point at x ∈ R d . However, all the BSs might not be active as there might
be BSs without user and activity of the BSs is not independent of each other.
Hence, active/interferring BS density cannot be modelled as PPP as this is
not the independent thinning of the parent PPP. Moreover, if the frequency
reuse factor is more than one and the frequency subbands are not distributed
randomly, interferring BSs are not PPP. However, in the derivations as well as
in the simulations, interferring BSs are considered to be a thinned version of the
parent PPP]
32 / 47
Coverage Probability of a User when all BSs are Active

Coverage probability vs BS density for α=4, ρ=1, λi=λb/δ BS per m2

0.95
Pc for δ=1
1

Pc for δ=1
0.9 2

Pc for δ=3
1

Pc for δ=3
0.85 2

Pc for δ=5
1

0.8 Pc for δ=5

Coverage probability

0.75

0.7

0.65

0.6

0.55
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
BS density, λb (BS per m2) x 10
−3

33 / 47
Capacity of a User when all BSs are Active
Shannon
6 capacity (ergodic) vs BS density for α=4, W=5MHz, λu=10−3 user per m2, λi=λb/δ BS per m2
x 10
12
Cu for δ=1
1

Cu for δ=1
2
10 Cu for δ=3
1

Cu for δ=3
2

Cu for δ=5
Shannon ergodic capacity (b/s)

8 1

Cu for δ=5
2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
BS density, λb (BS per m2) x 10
−3

[In model 1, downlink bandwidth (BW) in a cell is divided by M. But, when M

is less than 1, BW available to a user exceeds the BW in the cell. Hence, I
restricted the BW of a user by BW available in the cell]
34 / 47
BS Activity Probability
BS Activity Probability,
λu
Pa = 1 − (1 + )−3.5
3.5λb
−3 2
BS activity probability vs BS density for λ =10 user per m
u
1

0.9

0.8
a
BS activity probability, P

0.7

0.6

0.5

0.4

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

BS density, λb (BS per m2) −3
x 10

C. Li, J. Zhang, and K. B. Letaief, “Throughput and energy efficiency analysis

of small cell networks with multi-antenna base stations,” IEEE Trans. Wireless
35 / 47
Coverage Probability When All BSs Are Not Active

Coverage probability vs BS density for α=4, ρ=1, λu=10−3 user per m2, λi=λbPa/δ BS per m2
0.95
Pc for δ=1
1

Pc for δ=1
0.9 2

Pc for δ=3
1

Pc for δ=3
0.85 2

Pc for δ=5
1

0.8 Pc for δ=5

Coverage probability

0.75

0.7

0.65

0.6

0.55
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
BS density, λb (BS per m2) x 10
−3

36 / 47
Capacity When All BSs Are Not Active
−3
Shannon
6 capacity (ergodic) vs BS density for α=4, λu=10 user per m2, W=5MHz, λi=λbPa/δ BS per m2
x 10
18
Cu for δ=1
1

16 Cu for δ=1
2

Cu for δ=3
1
14 Cu for δ=3
2

Cu for δ=5
Shannon ergodic capacity (b/s)

12 1

Cu for δ=5
2

10

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
BS density, λb (BS per m2) x 10
−3

[In model 1, downlink bandwidth (BW) in a cell is divided by M. But, when M

is less than 1, BW available to a user exceeds the BW in the cell. Hence, the
BW of a user is restricted by BW available in the cell]
37 / 47
Other Performance Metrics Analysis I
Power consumption model used is

P = aPtx + b

So, average power consumption in unit area,

Parea = λb (Pa aPtx + b)

SE1 (ergodic) = SE2 (ergodic)

³ ´
∞ ∞ λ
−πx 1+ λ i I(α,θ )
Z Z
=π e b
dxd ² [b/s/Hz]
²>0 x >0
²
Here, [θ = 2 − 1]

38 / 47
Other Performance Metrics Analysis II
Ï First Model
Ï Area Spectral Efficiency (ergodic),

ASE1 (ergodic) = λu SE1 (ergodic) [b/s/Hz/m2 ]

Ï Energy Efficiency (ergodic),

λu Cu1
EE1 (ergodic) = [b/J]
Parea
Ï Area Spectral Efficiency (outage),

ASE1 (outage) = λu Pc1 log2 (1 + ρ ) [b/s/Hz/m2 ]

Ï Energy Efficiency (outage),

λu Pc1 w log2 (1 + ρ )
EE1 (outage) = [b/J]
Parea

39 / 47
Other Performance Metrics Analysis III
Ï Second Model
Ï Area Spectral Efficiency (ergodic),

ASE2 (ergodic) = λb Pa SE2 (ergodic) [b/s/Hz/m2 ]

Ï Energy Efficiency (ergodic),

λb Pa Cu2
EE2 (ergodic) = [b/J]
Parea
Ï Area Spectral Efficiency (outage),

ASE2 (outage) = λb Pa Pc2 log2 (1 + ρ ) [b/s/Hz/m2 ]

Ï Energy Efficiency (outage),

λb Pa Pc2 W
δ log2 (1 + ρ )
EE2 (outage) = [b/J]
Parea

40 / 47
ASE (outage) vs BS density

−3 α=4, W=5MHz, λu=10−3 user per m2, λi=λbPa/δ BS per m2, a=6.35, b=106, Ptx=6.3W
x 10
1
ASE1 (outage) for δ=1

0.9 ASE2 (outage) for δ=1

ASE1 (outage) for δ=3
0.8 ASE2 (outage) for δ=3
ASE1 (outage) for δ=5
0.7 ASE2 (outage) for δ=5
ASE (outage) in b/s/Hz/m2

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
λb in per m2 x 10
−3

41 / 47
EE (outage) vs BS density

4 α=4, W=5MHz, λu=−3 user per m2, λi=λbPa/δ BS per m2, a=6.35, b=106, Ptx=6.3W
x 10
3
EE1 (outage) for δ=1
EE2 (outage) for δ=1
EE1 (outage) for δ=3
2.5
EE2 (outage) for δ=3
EE1 (outage) for δ=5
EE2 (outage) for δ=5
2
EE (outage) in b/J

1.5

0.5

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
λb in per m2 x 10
−3

42 / 47
EE (outage) vs ASE (outage)

α=4, W=5MHz,
4 λu=10−3 user per m2, λb=0:2x10−3 per m2 λi=λbPa/δ BS per m2, a=6.35, b=106, Ptx=6.3W
x 10
3
EE1 (outage) for δ=1
EE2 (outage) for δ=1

2.5 EE1 (outage) for δ=3

EE2 (outage) for δ=3
EE1 (outage) for δ=5
EE2 (outage) for δ=5
2
EE (outage) in b/J

1.5

0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ASE (outage) in b/s/Hz/m2 −3
x 10

43 / 47
EE (ergodic) vs ASE (ergodic)

α=4, W=5MHz,
4 λu=10−3 user per m2, λb=0:2x10−3 per m2 λi=λbPa/δ BS per m2, a=6.35, b=106, Ptx=6.3W
x 10
12
EE1 (ergodic) for δ=1

11 EE2 (ergodic) for δ=1

EE1 (ergodic) for δ=3
10 EE2 (ergodic) for δ=3
EE1 (ergodic) for δ=5
9 EE2 (ergodic) for δ=5
EE (ergodic) in b/J

2
0 1 2 3 4 5 6 7
ASE (ergodic) in b/s/Hz/m2 x 10
−3

44 / 47
 Section 4

Future directions

45 / 47
Future directions

Ï Adopting Point Processes (PP) that model the cellular

network more accurately. Hard Core Point Process (HCPP),
where there is a condition on the minimum distance between
the points might be a potential solution
Ï Including more practical system parameters in the system
model

46 / 47
References I
A. J. Fehske, F. Richter, and G. P. Fettweis, “Energy efficiency
improvements through micro sites in cellular mobile radio networks,” in
GLOBECOM, pp. 1–5, IEEE, 2009.

G. Koutitas and P. Demestichas, “A review of energy efficiency in

telecommunication networks,” Telfor journal, vol. 2, no. 1, pp. 2–7, 2010.

F. Richter, A. J. Fehske, P. Marsch, and G. P. Fettweis, “Traffic demand

and energy efficiency in heterogeneous cellular mobile radio networks,” in
Vehicular Technology Conference (VTC 2010-Spring), 2010 IEEE 71st,
pp. 1–6, IEEE, 2010.

H. Klessig, A. J. Fehske, and G. P. Fettweis, “Energy efficiency gains in

interference-limited heterogeneous cellular mobile radio networks with
random micro site deployment,” in Sarnoff Symposium, 2011 34th IEEE,
pp. 1–6, IEEE, 2011.

G. Auer, V. Giannini, C. Desset, I. Godor, P. Skillermark, M. Olsson,

M. A. Imran, D. Sabella, M. J. Gonzalez, O. Blume, et al., “How much
energy is needed to run a wireless network?,” Wireless Communications,
IEEE, vol. 18, no. 5, pp. 40–49, 2011.

47 / 47
References II

G. Auer, V. Giannini, M. Olsson, M. J. Gonzalez, and C. Desset,

“Framework for energy efficiency analysis of wireless networks,” in 2nd
International Conference on Wireless VITAE, pp. 1–5, 2011.

D. Tsilimantos, J.-M. Gorce, and E. Altman, “Stochastic analysis of energy

savings with sleep mode in ofdma wireless networks,” in INFOCOM, 2013
Proceedings IEEE, pp. 1097–1105, IEEE, 2013.

J. G. Andrews, F. Baccelli, and R. K. Ganti, “A tractable approach to

coverage and rate in cellular networks,” Communications, IEEE
Transactions on, vol. 59, no. 11, pp. 3122–3134, 2011.

C. Li, J. Zhang, and K. B. Letaief, “Throughput and energy efficiency

analysis of small cell networks with multi-antenna base stations,” IEEE
Trans. Wireless Commun, 2013.

48 / 47