Sie sind auf Seite 1von 4

IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 2, NO.

6, DECEMBER 2013 611

Energy Consumption Analysis of


Energy Harvesting Systems with Power Grid
Yuyi Mao, Guanding Yu, Senior Member, IEEE, and Caijun Zhong, Member, IEEE

Abstract—Energy Harvesting (EH) is an indispensable technol-


ogy for future wireless communication systems due to its potential
in saving nonrenewable energy and making the communication
environmentally friendly. However, due to the unpredictable and
intermittent nature of the external energy arrivals, commu-
nication systems solely powered by harvested energy can not
guarantee the delivery of reliable and uninterrupted services.
Consequently, having stable power supply from power grid is
still of great importance in EH communication systems. In this
letter, we consider a wireless communication system powered by (a) Hybrid-energy supply wireless communication system.
both power grid energy and harvested energy. Specifically, we + + + + +
( ( ( ( 1  ( 1 
propose two strategies depending on how the harvested energy
is used, and investigate the average energy consumed from the
ĂĂ ĂĂ
^( `
+
M
+DUYHVWHG
(QHUJ\

power grid in each case. The major contribution of this letter ^% ` %DWWHU\
M (QHUJ\/HYHO
is the derivation of recursive expressions for the grid energy % % % ĂĂ %1  % 1  ĂĂ

consumption in hybrid-energy supply communication systems,


which may provide an useful tool for transmission scheme (b) N-slot transmission frame.
designs.
Index Terms—Energy harvesting, hybrid-energy supply sys- Fig. 1. System model.
tem, power grid energy consumption.
investigate the average energy consumption from the power
I. I NTRODUCTION grid. Specifically, we study both the infinite and finite battery
NERGY Harvesting (EH) technology, which captures capacity scenarios, and derive recursive expressions for the
E energy from renewable energy sources such as solar
and wind, has received significant interests from both the
average grid energy consumption for both systems. Simulation
results are also presented to corroborate the analysis. These
industry and academia, because it can significantly reduce results will be beneficial to the transmission protocol design
the carbon dioxide emission and make the communication for future communication systems with hybrid energy supply.
environmentally friendly. As such, a great deal of works
have appeared recently, which investigated the performance of II. S YSTEM M ODEL
communication systems employing EH techniques, from both We consider a point-to-point wireless communication sys-
the theoretical aspects and practical implementation aspects tem with a hybrid-energy supply transmitter as depicted in
[1]-[3]. Fig. 1(a). The harvested energy is stored in a battery with
However, communication systems solely powered by har- capacity M . A battery energy level monitor is applied. For
vested energy can not guarantee to provide reliable services the realization of hybrid-energy supply, a switch is adopted
due to the randomness and intermittent nature of the external for harvested energy and grid energy selection.
energy arrivals. In [4], it has been demonstrated that the outage We will focus on an N -slot transmission and the length
probability of a communication link heavily depends on the of each slot is Ts as shown in Fig. 1(b). Before the i-th slot,
H
transmitter’s energy harvesting profile and energy storage energy packet with Ei−1 Joule arrives at the transmitter, where
H N −1
capability. Motivated by this, hybrid-energy supply systems {Ei }i=0 are modeled as identically and independently dis-
where the transmitter is powered by the energy from both EH tributed(i.i.d.) random variables (RVs) with probability density
and power grid have been proposed in literatures. In [5], the function (p.d.f.) f (x) , ∀x ≥ 0. The energy level in the battery
delay-aware control problems in hybrid-energy supply system right before i-th slot starts is denoted as Bi−1 . We consider a
were investigated, while later, [6] analyzed the outage prob- slow fading channel, such that the channel gain hi remains
ability and grid power consumption-outage tradeoff curve for the same for a total length of N slots, i.e., hi ≡ h. We
single cell systems with hybrid-energy supply baste-station. denote N0 and W as the background noise spectral density
In this letter, we propose two heuristic methods for the al- and system bandwidth, respectively. Without loss of generality,
location of the power obtained from the energy harvester, and we set N0 ≡ 1. Hence, to achieve a transmission rate of rg ,
the required energy
 rg in each time slot can be computed as
Manuscript received May 29, 2013. The associate editor coordinating the
review of this letter and approving it for publication was K. B. Huang. Eg = Ts Wh N0 · 2 Ts W − 1 .
This work was supported by the National Natural Science Foundation of Now, we present two intuitive power allocation schemes:
China (No. 61201229), the National Basic Research Program of China (No.
2012CB316006), and the Zhejiang Provincial Natural Science Foundation of Strategy I: When the energy stored in the battery before
China (No. LQ12F01006). the i-th slot is sufficient, i.e., Bi−1 ≥ Eg , i = 1, ..., N , then
The authors are with the Department of Information Science and Electronic the required energy Eg for information transmission will be
Engineering, Zhejiang University, China (e-mail: {maoyuyi, yuguanding,
caijunzhong}@zju.edu.cn). supplied solely by the battery; Otherwise, the power grid will
Digital Object Identifier 10.1109/WCL.2013.081913.130391 provide the total required energy of Eg .
2162-2337/13$31.00 
c 2013 IEEE
612 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 2, NO. 6, DECEMBER 2013

Strategy II: When the energy stored in the battery before


the i-th slot is sufficient, the operation will be the same as that Now, let us consider two separate cases. When y ∈
in Strategy I; Otherwise, the power grid will only supply the [0, min{x, Eg }], Eg comes from the power grid in the
shortage part of the energy, i.e., the grid energy consumption (i − 2)-th slot. Hence, the conditional p.d.f. becomes
is Eg − Bi−1 . fBi−1 |Bi−2 (x|y) = f (x − y). When y ∈ [Eg , Eg + x], Eg
Remark 1: It is simple to show that Strategy II always is from the battery in the (i − 2)-th slot, which implies
  that
consumes less power grid energy than Strategy I since when fBi−1 |Bi−2 (x|y) = f (x − y + Eg ). To this end, E E G can
Bi−1 < Eg , the power grid only needs to supply Eg − Bi−1 be computed recursively as follow:
in this strategy. However, in Strategy II, a single transmission N  Eg  min{x,Eg }
may be powered by the energy from two different sources,  G 
E E = Eg f (x − y) · fBi−2 (y) dy
which requires a switching operation, and hence incurs addi- i=1 0 0
tional implementation costs. In addition, the switching oper-  x+Eg

ation may result in extra delay and cause loss of energy. On + f (x + Eg − y) · fBi−2 (y) dy dx,
the other hand, in Strategy I, the transmission energy always Eg

comes from a single source, hence it is easier to implement. (5)


where fB−1 (x) = δ (x) and δ (x) is the dirac delta function.
III. I NFINITE BATTERY C ASE
In this section, we consider the infinite battery capacity case, B. Average Energy Consumption with Strategy II
i.e., M = +∞. Before going into details, we first introduce the In Strategy II, the energy level Bi can be expressed as,
following notations: we use fBi (x) to denote the p.d.f. of Bi ,
EiG and E G to denote the grid energy consumption in the i-th Bi−1 + EiH − Eg , Bi−1 ≥ Eg
Bi = , i = 1, 2, ...N − 1,
slot and total grid energy consumption in the N -slot frame, EiH , Bi−1 < Eg
respectively. In the remainder of this letter, we will analyze (6)
the average energy consumption of the strategies, namely, the and B0 = E0H . Hence, the average power consumption at the
grid energy consumption expectations. i-th slot EiG can be expressed as,
 Eg
 G
A. Average Energy Consumption with Strategy I E Ei = (Eg − x) · fBi−1 (x) dx, (7)
0
To start with, let us consider the variation of the energy
level in the battery. In Strategy I, the energy level Bi can be Similar as Strategy I, we consider two separate cases. When
expressed as y ∈ [0, Eg ), Eg is supplied by both the battery and power grid
 in the (i − 2)-th slot, hence fBi−1 |Bi−2 (x|y) = f (x). When
Bi−1 + EiH − Eg , Bi−1 ≥ Eg
Bi = , i = 1, 2, ...N − 1, (1) y ∈ [Eg , Eg + x], Eg is supplied by the battery only, which
Bi−1 + EiH , Bi−1 < Eg
implies fBi−1 |Bi−2 (x|y) = f (x − y + Eg ). Therefore, the
and B0 = E0H . Hence, fB0 (x) = f (x) , ∀x ≥ 0. average grid energy consumption in N slots can be computed
In Strategy I, the energy supplied by the power grid is by
binary, i.e., the power grid either spends no energy, or supplies N  Eg  Eg
the whole energy Eg . Hence, in the i-th slot, the average power  G 
E E = (Eg − x) f (x) · fBi−2 (y) dy
consumption can be shown as: i=1 0 0
 Eg  x+Eg

 G
E Ei = Eg · fBi−1 (x) dx. (2) + f (x + Eg − y) · fBi−2 (y) dy dx,
0 Eg
A close observation reveals that Bi−1 only depends on the (8)
power level at the previous slot, i.e., Bi−2 . Hence, we denote where fB0 (x) = f (x) and fB−1 (x) = δ (x) still hold.
the joint p.d.f. of Bi−1 and Bi−2 as fBi−1 ,Bi−2 (x, y), then We note that the expressions given in Eq. (5) and (8) are
fBi−1 (x) can be computed by averaging over Bi−2 , very general, and valid for EH systems with a continuous
 energy arrival distribution. In the following, we show that the
fBi−1 (x) = fBi−1 ,Bi−2 (x, y) dy, (3) analysis can also be extended to the discrete scenario. In the
y∈Ω(x)
discrete case, Eg is assumed to be a positive integer. For sim-
where Ω (x) is integral domain of y. In Strategy I, Ω (x) = plicity, we consider the important case when {EiH }N −1
follow
H i=0
[0, min{x, Eg }] [Eg , Eg + x], Eq.(3) can be rewritten as, the Poisson distribution, i.e., Pr Ei = k = k! , i =
−λ k
e λ

fBi−1 ,Bi−2 (x, y) dy 0, · · · , N − 1, k = 0, 1, · · · . Let us denote Eg − 1 as τ , then
y∈Ω(x) the discrete analogy of Eq.(5) and Eq.(8) can be written as,
 min{x,Eg }  Eg +x
N  τ min{k,τ
 } e−λ λ(k−j)
= fBi−1 ,Bi−2 (x, y) dy + fBi−1 ,Bi−2 (x, y) dy   
0 Eg E EG = Eg · Pr (Bi−2 = j)
 min{x,Eg } (k − j)!
i=1k=0 j=0
= fBi−1 |Bi−2 (x|y) fBi−2 (y) dy

k+Eg
0
  e−λ λ(k−j+Eg )
Eg +x
+ Pr (Bi−2 = j) ,
+ fBi−1 |Bi−2 (x|y) fBi−2 (y) dy. (k − j + Eg )!
Eg j=Eg
(4) (9)
MAO et al.: ENERGY CONSUMPTION ANALYSIS OF ENERGY HARVESTING SYSTEMS WITH POWER GRID 613

and where
N 
τ 
τ Pr (Bi = k) =
   e−λ λk ⎧ n (k)  −λ k−l 
E EG = Eg − k · · Pr (Bi−2 = j) ⎪ 1
Pr (B = l) · e λ
k! ⎪
⎪ l=0 i−1 (k−l)!
 
i=1 k=0 j=0 ⎪
⎪ n2 (k)


⎪ + q=Eg Pr (Bi−1 = q) · e(k+E
−λ k+Eg −q
λ
k+Eg
 e λ −λ (k−j+Eg ) ⎪
⎪ −q)! ,


g
+ Pr (Bi−2 = j) , k≤M
(k − j + Eg )! τ  M−l−1 e−λ λz 
j=Eg

⎪ Pr (B = l) · 1 −

⎪ l=0 i−1
 z=0 z!
M−q+τ e−λ λz 
(10) ⎪
⎪ n2 (k)

⎪ + q=Eg Pr (Bi−1 = q) · 1 − z=0 .
e−λ ·λk

⎪ z!
respectively. Also, Pr (B0 = k) = , ∀k = 0, 1, · · · , ⎩
k! k=M
Pr (B−1 = 0) = 1.
Proposition 1: In Strategy I, Strategy II,
  Δ  G    N 
τ
lim E EiG = E E∞ = (Eg − λ · Ts )+ , (11) E EG = (Eg − k) · Pr (Bi−1 = k) , (16)
i→+∞
i=1 k=0
where (x)+ returns the larger
 G value of x and 0. Also, Eq.(11) where
is an upper bound of E E∞ in Strategy II. Pr (Bi = k) =
Proposition 1 is conformed with Energy Conservation Law ⎧ τ
⎪ e−λ λk
and can be proved bythe Random Walk method similarly in ⎪
⎪ l=0 Pr (Bi−1 = l) · k!  
N
E [E G ]

⎪ n2 (k) e−λ λk+Eg −q
[7]. Here, we denote i=1N i as E [E G ]. ⎪
⎪ + Pr (B = q) · ,

⎪ q=E g i−1 (k+Eg −q)!


 k≤M
τ M−1 e−λ λz 
IV. F INITE -S IZE BATTERY C ASE ⎪
⎪ l=0 Pr (Bi−1 = l) · 1 −
⎪  z=0 z!


⎪ n2 (k) M−q+τ e−λ λz 

⎪ + Pr (B i−1 = q) · 1 − ,
In this section, we extend the results obtained in Section III ⎪

q=Eg z=0 z!

to the finite battery size case. We further assume the maximum k=M
battery level is larger than the required energy for information
transmission, i.e., M > Eg . where n1 (k) = min{k, τ } and n2 (k) = min{Eg + k, M }.
To this end, the remaining task is to evaluate the probability
With finite battery size, in both Strategy I and Strategy II,
that the battery level stays at a particular level. To address
the p.d.f. of B0 should be revised as,
this problem, we propose a Markov chain based approach.
  
M Specifically, we model the battery energy levels as the states
fB0 (x) = f (x) · U (x) + 1 − f (e) · de · δ (x − M ) . of a Markov Chain, i.e., Θ = {0, 1, 2, ..., M }. Then, the
0 transition probability matrix P = {pmn }(M+1)×(M+1) for
(12) Strategy I and Strategy II could be presented as follows:
If x ∈ [0, M ), U (x) = 1. Otherwise, U (x) = 0. Please note Strategy I,
that the appearance of δ (x − M ) in fBi (x) , i = 0, · · · , N −1 ⎧
is due to the fact that no more than M amount of energy can ⎪
⎪ Pr E H = n − m , m < Eg , n < M
⎪ H

be stored in the battery. Pr E = n − m + Eg , m ≥ Eg , n < M
Proposition 2: In finite battery size case, the average grid pmn = +∞ H

⎪ Pr E = k , m < Eg , n = M
energy consumptions are, in Strategy I, ⎪
⎩+∞
k=M−m H
k=M−m Pr E = k + Eg , m ≥ Eg , n = M
N   (17)
   Eg
E EG = Eg · fBi−1 ,Bi−2 (x, y) dydx, Strategy II,
i=1 0 y∈Ω(x) ⎧
(13) ⎪
⎪ Pr E H = n , m < Eg , n < M
 ⎪
⎨Pr E H = n − m + E , m ≥ Eg , n < M
where Ω (x) = [0, min{x, Eg }] [Eg , min{Eg + x, M }].
pmn = +∞ H g
And in Strategy II, ⎪
⎪ k=M Pr E = k , m < Eg , n = M

⎩+∞
H
N   Pr E = k + Eg , m ≥ Eg , n = M
   Eg k=M−m
(18)
E EG = (Eg − x) fBi−1 ,Bi−2 (x, y) dydx,
i=1 0 y∈Ω(x) Obviously, the Markov Chains for Strategies I and II are
 (14) ergodic. Hence the limiting distribution of the battery energy
where Ω (x) = [0, Eg ) [Eg , min{Eg + x, M }]. levels π = [π0 , π2 , ..., πM ] always exists which can be
We now look into the discrete case, and assume that obtained by solving the following linear equation set,
{EiH }N −1
i.e., Pr (B0 = k) =
i=0 follow the Poisson distribution,
 G
e−λ λk π·P=π
k! and Pr (B −1 = 0) = 1. Then E E can be computed M (19)
by, Strategy I, j=0 πj = 1.
G
N 
τ Denote Emc as the average power consumption computed
  
E EG = Eg · Pr (Bi−1 = k) , (15) by Markov Chain method. Then, substituting π into the
G
i=1 k=0 discrete versions of Eq.(2) and Eq.(7), we get E[Emc ] =
614 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 2, NO. 6, DECEMBER 2013

250 1.6
80 Strategy I, M=1.2E
g Strategy I, λ=4W
Strategy II, M=1.2Eg
60 1.4 Strategy II, λ=4W
Strategy I, M=infinite Strategy I, λ=6W
200 40
Strategy II, M=infinite 1.2 Strategy II, λ=6W
20 Strategy I, M=2.4Eg
Strategy II, M=2.4E
0 g 1

E [E G ](J)
E E G (J)

150 4 5 6
8
 

6 0.8

100 4
0.6
2
0.4
50 0
7 8 9 10
0.2

0 0
0 2 4 6 8 10 0 100 200 300 400 500
λ(W) N

Fig. 2. Relationship between E[E G ] and λ, with Eg = 5J and N = 50. Fig. 3. Relationship between E[E G ] and N (infinite capacity), Eg = 5J.

τ G

Eg · j=0 πj in Strategy I and E[Emc ] = j=0 (Eg − j) · πj
2.5
in Strategy II. Strategy I, λ=4W
Strategy II, λ=4W
2 Strategy I, λ=6W
V. N UMERICAL R ESULTS Strategy II, λ=6W
In this section, we present numerical results to validate our
analytical expressions. We only consider the discrete case, and (J) 1.5
assume {EiH }N −1
i=0 are Poisson RVs and Ts = 1s.
 G
E Emc

Fig. 2 illustrates the relationship


  between the average power 1
grid energy consumption E E G and the energy harvesting
rate λ. First of all, we see that the simulation curves over-
lap with the analytical results (highlighted by markers and 0.5
lines, respectively), which demonstrates the correctness of our
analytical results. Also, we see an intuitive result that the 0
energy consumption reduces when the energy harvesting rate 5 10 15 20 25 30 35 40
M(J)
increases, and the energy consumption of Strategy I is always
higher than that of Strategy II with the same battery size
Fig. 4. G ] and M (finite capacity), E = 5J.
Relationship between E[Emc
constraint. In addition, imposing
 a finite battery size M results g

in an increase of E E G . Finally, we see that if the battery


size constraint is different for the two strategies, it is possible required energy for information transmission and the average
that more energy will be consumed in Strategy II. harvested energy.
Fig. 3 shows the relationship between E[E G ] and N . It can
be observed that when N approaches infinity, E[E G ] goes to R EFERENCES
+
(Eg − λ · Ts ) , i.e., when λ = 6W, E[E G ] goes to zero and [1] O. Ozel and S. Ulukus, “Achieving AWGN capacity under stochastic
λ = 4W, E[E G ] approaches 1J as N increases to a value energy harvesting,” IEEE Trans. Inf. Theory, vol. 58, no. 10, pp. 6471–
sufficiently large which are in accordance with Proposition 1. 6483, Oct. 2012.
[2] J. Yang and S. Ulukus, “Optimal packet scheduling in an energy harvest-
Finally, Fig. 4 examines the impact of finite battery size ing communication system,” IEEE Trans. Commun., vol. 60, no. 1, pp.
M on the consumed energy. As we can readily observe, the 220–230, Jan. 2012.
average energy consumption reduces when the battery size [3] C. K. Ho and R. Zhang, “Optimal energy allocation for wireless com-
munications with energy harvesting constraints,” IEEE Trans. Signal
increases, and when M is sufficiently large, the average energy Process., vol. 60, no. 9, pp. 4808–4818, Sept. 2012.
consumption approaches the infinite battery capacity cases [4] B. Medepally, N. B. Mehta, and C. R. Murthy, “Implications of energy
G +
shown in Fig. 3, i.e., E Emc = (Eg − λ · Ts ) . profile and storage on energy harvesting sensor link performace,” in Proc.
2009 IEEE Global Telecommunications Conference, pp. 1–6.
[5] Y. Cui, V. K. Lau, and Y. Wu, “Delay-aware BS discontinuous trans-
VI. C ONCLUSION mission control and user scheduling for energy harvesting downlink
coordinated MIMO systems,” IEEE Trans. Signal Process., vol. 60, no.
In this letter, we have analyzed the average grid energy 7, pp. 3786–3795, Jul. 2012.
consumption in hybrid-energy supply wireless communication [6] J. Gong, S. Zhou, Z. Niu, and J. Thompson, “Energy-aware resource
allocation for energy harvesting wireless communication systems,” in
systems. For the two proposed energy allocation strategies, Proc. 2013 IEEE Vehicular Technology Conference – Spring, pp. 1–5.
recursive expressions for the average grid energy consumption [7] K. Huang, “Spatial throughput of mobile ad hoc networks with en-
were obtained. Our findings suggest that the finite battery size ergy harvesting,” IEEE Trans. Inf. Theory, to appear. Available: arXiv:
1111.5799.
constraint results in elevated energy consumption, and if the
communication period N is sufficiently large, the average grid
energy consumption per slot reduces to the gap between the

Das könnte Ihnen auch gefallen