Professional Documents
Culture Documents
AbstractCooperative transmission can signicantly reduce clustering and routing problems are handled by upper layers.
the power consumption associated with long distance transmission The multi-hop transmission consists of several single-hop
in wireless sensor networks (WSNs). In this paper, we analyze cluster-to-cluster transmissions. Each single-hop transmission
the optimal power consumption of cluster-based multi-hop trans- consists of three phases, namely the Preparation Phase (PP),
mission for a cooperative WSN. With specic Quality of Service the Broadcast Phase (BP) and the Cooperation Phase (CP). In
(QoS) constraints on delay and channel capacity, we show that
the PP, the cluster head of the transmitter cluster calculates the
the power optimization problem of the whole network has no
closed-form solution in a slow at Rayleigh fading environment. transmission power to be used in the BP and CP. In the BP, the
Thus we propose a dynamic power control and optimization cluster head broadcasts the data along with the power control
(DPCO) scheme that can jointly determine the optimal number message in the transmitter cluster. In the CP, all nodes of the
of cooperative sensors and their transmission power. We further transmitter cluster that received the data successfully in the
propose a channel approximation algorithm that can signicantly BP cooperatively transmit the data to the cluster head of the
reduce the computational complexity of the DPCO scheme. receiver cluster through a multiple-input-single-output (MISO)
channel. For intra-cluster transmission in the BP, we assume
I. I NTRODUCTION that all nodes in the cluster have perfect transmit and receive
channel state information (CSI). For inter-cluster transmission
Energy optimization is an important issue in wireless in the CP, we assume that nodes in the receiver cluster know
sensor networks (WSNs) where the low-power sensors are the CSI and that nodes in both clusters know the distribution
expected to operate for many years without battery replace- of the CSI. Note that our MISO approach can be extended to a
ment. Furthermore, certain applications may have Quality of MIMO approach by utilizing selection diversity in the receiver
Service (QoS) requirements on delay and channel capacity. For cluster [5].
instance, it is important for the sink node to receive data in a
timely manner in an industrial control system [1]. Traditional To minimize the total energy consumption, we propose a
energy optimization techniques may not always guarantee the dynamic power control and optimization (DPCO) scheme that
QoS requirements in WSNs, especially in fading environments. can optimize the total energy consumption without violating
the QoS requirements. Given the number of cooperative n-
It has been proved that multi-antenna system use less ener- odes as an input condition, the scheme rst calculates the
gy for data transmission in fading channels compared to single- conditional optimal power consumption using the dynamic
antenna systems [2]. However, a multi-antenna system requires power control (DPC) algorithm. Then it determines the optimal
complex transceiver circuits that are not practical for low-cost number of cooperative sensors using the results of conditional
wireless sensors. Cooperative multiple-input-multiple-output optimal power consumption. We show that the DPC algorithm
(MIMO) [3] and virtual MIMO techniques [4] are proposed converges to the optimum in O (1) iterations. We further pro-
to enable MIMO techniques to be utilized in WSNs. For pose a channel approximation algorithm that can signicantly
the single-hop cooperative MIMO systems analyzed in [3], reduce the computational complexity of the DPCO scheme.
it is shown that for long-distance transmission, they are more
energy efcient compared to a single-antenna system. In [5], The rest of the paper is organized as follows: Section II
multi-hop cooperative MIMO channels are analyzed, where the describes the system model of the multi-hop cluster-based
design target is to minimize the outage probability given an cooperative WSN and expresses the QoS requirements. The
energy constraint and target outage channel capacity. However, DPCO scheme is proposed in Section III. We further propose
the minimum outage probability may not give the maximum an approximation algorithm to reduce the computational com-
average outage channel capacity [6]. Minimizing the outage plexity of the DPCO scheme. Section IV shows the simulation
probability may also lead to full-power operation which sig- results of the proposed scheme. Finally, we conclude the paper
nicantly decreases the lifetime of the WSN. in Section V.
es, namely the preparation phase (PP), the broadcast phase the STBC. Nm+1 N 0, 2 is the l 1 AWGN at the
(BP) and the cooperation phase (CP). The time duration of receiver. The transmission power of the transmitted space-time
the three phases are 1 T , 2 T and 3 T respectively where codeword is nPtrct
per node.
1 +2 +3 = 1 and T is the time duration of each single-hop Considering the nodes in the transmitter cluster only know
transmission. We assume that each cluster contains n sensors the distribution of the CSI, the outage capacity Cout [bps/Hz]
that are uniformly deployed in the cluster and the cluster head of the CP is dened as [6]:
is located near the center of the cluster.
Cout = 3 log2 (1 + out ) , (2)
1) Preparation Phase: In the PP, the cluster head in the
transmitter cluster calculates the optimal number of coopera- where out is the minimum signal-to-noise ratio (SNR) for
tive nodes (ntr ), the transmission power in the BP (Pbt ) and successful decoding at the receiver side. The outage probability
CP (Pct ) using the DPCO scheme. The value of ntr , Pbt and pout is dened as:
Pct are carried in the power control message and sent to other
cooperative nodes along with the data in BP. pout = Pr ( < out ) , (3)
2) Broadcast Phase: In the BP, the cluster head broadcasts where is the actual SNR at the receiver side. According to
the data along with the power control message to the coopera- our channel model in (1), is given by:
tive sensors within its cluster with power Pbt . The sensors that 2
received the data successfully are selected to transmit with the Pct HF
= , (4)
cluster head in the CP. Since we assume that the sensors in the ntr 2
cluster are uniformly deployed and the cluster head is located 2
in the center of the cluster, the number of the active sensors is where 2 is the noise power and HF is the squared
proportional to Pbt and the total number of sensors n in each Frobenius norm of the 1 ntr channel vector H. According to
2
cluster. By selecting an appropriate level of Pbt to broadcast the denition of H, HF 2 (2ntr ) is a chi-square variable
the data, we can have ntr nodes selected as the cooperative with 2ntr degrees of freedom.
nodes (including the cluster head itself) in the CP.
The average outage capacity is given by:
3) Cooperation Phase: In the CP, the cooperative nodes
jointly transmit the data to the cluster head in the receiver Cout = (1 pout ) Cout = 3 (1 pout ) log2 (1 + out ) . (5)
cluster using an orthogonal Space-Time Block Code (STBC).
We assume that the number of cooperative nodes is ntr , 1) Delay Requirement: We consider the delay of the multi-
including the cluster head itself. The total transmission power hop cluster-to-cluster communication using an automatic re-
Pct is equally divided among the cooperative nodes. The peat request (ARQ) protocol. We assume that the outage
transmission power used by each sensor is consequently nPtr ct
. probabilities are equal in each of the N 1 hops. Therefore,
for transmission through N 1 hops, the number of failures Proposition 2: For any transmission under the QoS re-
k follows the negative binomial distribution: quirement and D > (N 1) T , there exists pout that can
maximize Cout , which can be denoted as follows
k N B (N 1; pout ) . (6)
Cout = max {3 (1 pout ) log2 (1 + out )}
The probability mass function of the negative binomial distri- pout (13)
(N 1)T
bution is s.t. Cout T ,0
L
< pout 1
D
k+N 2 N 1 1
f (k; N 1, pout ) =
k
pkout (1 pout ) . where out = Pct F ntr(pout
2
|2ntr )
and F 1 (p |v ) is the inverse
chi-square cumulative distribution function (CDF) for a given
(7)
probability p and v degrees of freedom.
Proposition 1: Given the delay requirement D, the average
delay of the transmission through N 1 hops is given by: Proof: From (2), (3) and (5), we know that F 1 (p |v )
is directly proportional to p for any given v. Let =
(N 1) T
T = E [(k + N 1) T ] = . (8) F 1 1 NDT |2ntr . For pout 0, 1 (N 1)T
D , we have
1 pout
lim Cout = 0,
Thus we can have reliable transmission if pout 1 (N 1)T
D
pout 0
for any delay requirement D > (N 1) T . If D (N 1) T , Pct (14)
lim Cout = 3 (1 pout ) log2 1 + ntr 2
it is impossible to have reliable transmission under the delay pout 1 NDT
requirement.
Since lim Cout > 0 and (5) is continuous at pout
Proof: We rst prove that E [k] = (N1p 1)pout
. Since the pout 1 NDT
0, 1 (N 1)T
out
number of failures k follows the negative binomial distribution, , according to the extreme value theorem,
the mean value of k can be calculated as:
D
there exists pout 0, 1 (N 1)T that can maximize Cout .
D
E [k] = kf (k; N 1, pout )
k=0
C. Energy Consumption Analysis
k (k + N 2)! k N 1
= p (1 pout )
k! (N 2)! out For the transmission in each hop, the total power consump-
k=0
tion is given by:
(N 1) pout (k + N 2)! k1 N
= p (1 pout ) Pt = 1 PDP CO + 2 Pbt + 3 Pct + ntr Pcir , (15)
1 pout (k 1)! (N 1)! out
k=1
where PDP CO is the power consumption of the DPCO scheme
(N 1) pout
= f (k; N, pout ) (9) in the PP, Pcir is the circuit power consumption of the sensor
1 pout and ntr is the number of sensors used for transmission. The
k=0
(N 1) pout total energy consumption for the transmission through N 1
= . hops is give by:
1 pout
Note that in (8) N and T are constants. Thus the average delay (N 1) Pt T
Et = Pt T = . (16)
is given by: 1 pout
(N 1) T
T = E [(k + N 1) T ] = . (10) In the BP, we assume the channel is AWGN with free space
1 pout path loss (i.e., = 2). Thus given the transmission data rate
For any delay requirement D > (N 1) T , we must have Rbt , we have
Rbt
T = (N 1)T
1pout D and thus pout 1
(N 1)T
to satisfy the
D Pbt = r2 2 2 2 1 , (17)
reliable transmission requirement. Note that 0 < pout < 1. If
D (N 1) T , we should have pout 1 (N 1)T D 0 for where r is the intracluster transmission range. We assume that
reliable transmission, which is impossible to satisfy. the average cluster range is rc and n sensors are normally
2) Capacity Requirement: We assume that there are a deployed in each cluster. Therefore the approximate number
total of L bits for transmission. In order to have reliable of sensors that can successfully decode the broadcast message
transmission in the CP, we have is given by:
2
L r Pbt n
Rct Cout , (11) ntr = n= Rbt . (18)
T rc rc2 2 2 2 1
where Rct [bps/Hz] is the transmission data rate in the CP.
In the BP, the transmission data rate Rbt [bps/Hz] should In the CP, a slow at Rayleigh fading channel is assumed
be no less than the transmission data rate in the CP. Thus we in the channel model. From (4) and (5), we have
have Cout
Since there is no closed-form solution for cooperative We assume that n sensors are normally deployed in each
transmission power Pct|v , we use a dynamic power control cluster with the cluster range rc . Since the minimum value of
(DPC) algorithm to nd the conditional optimal Pct|v . We will Rbt is TL , given ntr = v, the optimal broadcast transmission
show that given any ntr = v, Pct|v converges to its optimum power Pbt|v is denoted as
Pct|v with a sufcient small change in O (1) iterations. L
vrc2 2 2 2 T 1
In Algorithm 1, we use Pct|v,I max as the approximation Pbt|v = . (24)
n
of Pct|v since Pct|v,Imax Pct|v where is sufcient
small. For energy saving purpose, there is no need to coop- After we have Pct|v and Pbt|v for all possible ntr = v, we
eratively transmit the data if the QoS requirement cannot be
can determine ntr according to (22) and thus in turn determine
fullled with the maximum transmission power Pct,max .
Pct|n and Pbt|n .
tr tr
B. Approximation Algorithm
Q
To reduce the computational complexity in the DPCO WU
&RRSHUDWLYH7UDQVPLVVLRQ3RZHU3 FWP:
QWU
scheme, we propose an approximation algorithm. Note that Q
Q
WU
WU
the computational complexity in DPCO scheme is mainly
2
Since HF in (4) follows a chi-square variable with 2ntr
degrees of freedom, we can calculate the mean value of as: ,WHUDWLRQ1XPEHU
Pct
2
2P
ct
= 2
E HF = . (25) Fig. 2. Performance of the dynamic power control algorithm with different
ntr 2 ntr
We also calculate the variance of as:
2 2
Pct 2 1 2Pct where p = 2pout 1. Compared with (13), we can see that
2 = var H = . (26)
ntr 2 F
ntr 2 (33) is a polynomial function on p that has a straightforward
If we expand (2) in Taylor series at we have solution for the optimization of Cout . If we substitute Cout
k for Cout and pout for pout respectively, the computational
3 complexity in the rst step of the DPC algorithm can be
Cout () = 3 log2 (1 + ) . (27)
k ln 2 1 + signicantly reduced.
k=1
According to (27) the second-order approximation for C is
given by IV. S IMULATION AND A NALYSIS
2
3 In this section, we give the simulation results to show the
C =E [Cout ] 3 log2 (1 + ) (28) efciency of DPCO scheme. We also show that the approxi-
2 ln 2 1 +
mation algorithm provides accurate channel estimations. The
where 2 is the variance of . Note that the approximation is parameters in the simulation are given in TABLE I.
2
only valid for | | < 1. If we expand Cout in a Taylor
2 TABLE I. PARAMETERS FOR SIMULATION
series, the second-order approximation for C is given by:
2
2 2 Symbol Description Value
C = E Cout (E [Cout ]) (29)
2 Path loss constant 1
2 4 Path loss exponent 2
3 n Number of sensors per cluster 10
2 4 .
ln 2 (1 + ) 4(1 + ) d Transmission distance 100 m
rc Cluster range 10 m
2 Noise power 5 W
We assume that the approximate channel capacity Cout Pct,max Maximum of Pct 200 mW
Pct,min Minimum of Pct 0 mW
follows a normal distribution with a mean C and a standard
0 Initial step in DPC algorithm 10 mW
2
deviation C , which is denoted as Cout N C , C . We Pcir Circuit power consumption 10 mW
L
Capacity Requirement 1 bps/Hz
further dene the outage probability pout for Cout , which is T
1 , 2 , 3 0.1, 0.15, 0.75
given by:
1 1 Cout C
pout = Pr C < Cout = + erf (30) First, we investigate the DPC algorithm. We set M = 20
2 2 2C and 0 = 10 mW. Given different ntr = v, the simulation
where erf () is the error function. From (30) we have results in Fig. 2 show that Pct|v converges to its optimal value
Pct|v with sufciently small difference in a limited number
Cout = C + 2C erf 1 (2pout 1) (31) of iterations (Imax = 30). Note that the optimal cooperative
1 transmission power Pct|n is inversely proportional to the
where erf () is the inverse error function, which can also tr
$YHUDJH2XWDJH&DSDFLW\ESV+]
3RZHU&RVWP:
2XWDJH3UREDELOLW\S RXW 1XPEHURI&RRSHUDWLYH1RGHVQ WU
Fig. 3. Average outage capacity with different pout for ntr = 6 and Fig. 5. Performance of the dynamic power control algorithm with different
Pct|ntr = 53 mW values of ntr
V. C ONCLUSION
5HDO&KDQQHO
$SSUR[LPDWH&KDQQHO In this paper, we investigated the power consumption and
optimization in a multi-hop QoS-constrained cooperative WSN
in a fading environment. We showed that there is no closed
2SWLPDO2XWDJH3UREDELOLW\