Sie sind auf Seite 1von 4

Optimizing Physical Layer Energy Consumption for

Wireless Sensor Networks


Jennifer A. Hartwell, Geoffrey G. Messier and Robert J. Davies
Department of Electrical & Computer Engineering and TRLabs, University of Calgary,
2500 University Dr. NW, Calgary, Alberta, Canada, T2N 1N4
Email: gmessier@ucalgary.ca

Abstract— This paper investigates the use of physical layer tradeoff when deciding on the SER target. Raising the target
symbol error rate (SER) optimization to minimize wireless SER will lower the required signal to noise ratio (SNR) at
sensor network (WSN) energy consumption. Increasing the SER the receiver and reduce the energy spent on transmission.
maintained by the physical layer can save energy by lowering
transmit power. However, this also causes an increase in the However, retransmissions that occur due to a high target SER
amount of energy spent on frame retransmissions. A technique are an additional source of wasted energy that will offset this
for SER optimization that balances these two effects will be transmit energy reduction.
presented. The variation in average link delay that results when This effect has not been considered in previous research.
this technique is applied to a WSN will also be explored. The work in [6], [7], [8] has all been based on a fixed bit
error rate (BER) or SER target.
I. I NTRODUCTION
The contribution of this paper is to present a method
Wireless sensor networks (WSN’s) consist of a large number for including frame retransmissions in physical layer energy
of small, low data rate and inexpensive nodes that commu- consumption analysis. This method is then used to find an
nicate in order to sense or control a physical phenomenon. optimal target SER that minimizes overall physical layer
The number of new applications made possible using this new energy consumption. This optimization strikes a balance be-
network design approach is considerable [1]. tween reducing transmit energy by raising the SER target and
The designers of WSN nodes have been presented with new minimizing the energy spent on retransmissions. It will be
priorities. Rather than maximizing data rate, the primary goal shown that the optimal value of target SER is a function of
of WSN node design is typically to improve energy efficiency. the pathloss between the transmitter and receiver. As a result,
Most WSN nodes will be battery operated and replacing different links in a WSN may operate at different values of
these batteries will be difficult for many WSN applications. SER.
This may be due to deployment in remote locations, such as Setting physical layer SER solely based on energy efficiency
for forest monitoring, or in difficult environments, such as will have an impact on other WSN application quality of
livestock monitoring or the battlefield. Even when nodes are service (QoS) metrics. The primary one is link delay. The QoS
accessible, their low cost may make it more efficient to simply impact of having different links operate at different levels of
replace the entire node rather than just its battery. SER is that they will experience a different average number
As a result, considerable research has been performed on of retransmissions. The variation in average link delay that
reducing the energy WSN nodes use to communicate. Work results will also be investigated as part of this study.
has been done on saving energy by modifying the higher Section II presents a single input/single output (SISO) WSN
layers of the protocol stack [2], [3], [4] and through the use physical layer energy consumption model that includes frame
of cross-layer design [5]. It has also been shown that a signif- retransmissions. The optimization of SER to minimize the
icant amount of energy can be saved by optimizing physical energy consumption of this model is introduced in Section III.
layer operation [6], [7]. This physical layer research includes The effect of this optimization on the WSN physical layer
optimizing modulation order to achieve a balance between and the savings achieved relative to a system with fixed SER
circuit energy consumption and transmit energy consumption. is discussed in Section IV. Concluding remarks are made in
In [7], energy consumption is minimized by choosing different Section V.
multiple antenna schemes for different pathloss values. The
authors in [8] explore co-operative transmission among nodes II. P HYSICAL L AYER M ODEL
as a method for minimizing energy consumption. In this section, an expression for the energy consumed by a
One method that has so far been overlooked in the search for WSN physical layer link is derived that includes retransmis-
techniques to reduce WSN physical layer energy consumption sion energy. Similar to [6], [7], this link energy can be written
is the optimization of physical layer symbol error rate (SER). as
It is assumed that the QoS requirements of most WSN  
applications will include error free transmission such that any  PTx
ESym = + POH,Tx + POH,Rx TSym (1)
frames received in error will be retransmitted. This presents a η

1550-2252/$25.00 ©2007 IEEE 76


where PTx is the transmit power, η is the transmit amplifier Variation in the physical layer pSErr will change the number
efficiency and TSym is the duration of a communications of average retransmissions required to successfully deliver a
symbol. The transmit overhead power term, POH,Tx , is the frame. This implies that links with higher values of optimal
power consumed by all the transmit chain circuitry, except for pSErr will experience higher delays due to additional retrans-
the amplifier. The term POH,Rx is the power consumed by the missions. It is important to evaluate these delays in order to
receive circuitry. quantify the impact of the pSErr optimization process. The
For each transmitting node, it is assumed that closed-loop average time required to successfully transmit a frame can be
power control is used to adjust the value of PTx in order to written as
maintain a certain target SER at the receiving node. The SER
target can be mapped to a received SNR target that is denoted TF = TDat + (TDat + TNAck )pFErr
γT (pSErr ) to indicate dependence of SNR on the probability of +(TDat + TNAck )p2FErr + . . .
(5)
symbol error, pSErr . WSN node transmit power can be written TDat +TNAck
= (1−p N − TNAck
as PTx = γT (pSErr )APN F , where A is large scale channel SErr )
attenuation, PN is thermal noise at the receiver input and F where TDat is the time required to transmit a data frame and
is the receiver noise figure. This allows (1) to be expanded to TNAck is the time required to transmit a negative acknowledg-
  ment. Note that (5) does not include propagation delay since
 1 TDat and TNAck will be much larger due to the low data rates
ESym = γT (pSErr )APN F + POH,Tx + POH,Rx TSym
η used by most WSN nodes.
(2) Information from [6], [7] is used to estimate hardware
Frame retransmission is a random event with a probability energy consumption values of POH,Tx = 98.4 mW, POH,Rx
that depends on physical layer SER. The average energy = 110.7 mW, η = 0.35 and F = 10 dB. The value of TSym is
required to successfully transmit a single symbol must include 16 µs, as taken from the IEEE 802.15.4 WSN standard [3].
the overhead of any frame retransmissions. This average As such, TDat in (5) can be calculated according to TDat =
energy must be equal to (2) multiplied by a factor that accounts N TSym . The number of symbols in the acknowledgment frame
for the retransmission overhead. This factor can be determined is 22 [3] such that TNAck = 22TSym .
by writing the average energy per symbol with retransmissions
as III. O PTIMIZING SER
Finding the value of pSErr that minimizes (4) can be consid-
   ered a constrained optimization problem. The constraints are
ESym = ESym + (ESym )(pFErr ) + (ESym )(pFErr )2 + . . . (3)
achieved by putting bounds on pFSuc such that 0.01 ≤ pFSuc ≤
where pFErr is the probability of a frame error and ESym
is 0.99. This translates to the following bounds on pSErr
the energy term in (2). Since (3) is a geometric series, it can
be simplified to [9] 1 − 0.991/N ≤ pSErr ≤ 1 − 0.011/N (6)
It will be shown in this section that the objective function

ESym (4) is convex over the interval of interest defined in (6). This
ESym = 1−p
FErr guarantees that if a local minimum is found, it is also a global

ESym minimum [10]. A function is convex if its second derivative
=p is non-negative over the region of interest.
FSuc (4)
TSym Before the second derivative can be determined, it is neces-
= (1−p N
 SErr )  sary to define the function γT (pSErr ) in (4). This function is
· γT (pSErr )APN F η1 + POH,Tx + POH,Rx essentially the inverse of a standard SER versus SNR curve.
The physical layer simulated to determine this function is
where pFSuc = (1 − pSErr )N is the probability of a successful the 16-ary quasi-orthogonal modulation option defined in the
frame transmission and N is the number of symbols per frame. 802.15.4 standard [3]. A flat Rayleigh fading channel with
Note that (4) now captures the tradeoff a link experiences additive white Gaussian noise is assumed and the best fit
when pSErr is adjusted. When pSErr is increased, link energy equation to those simulation results is
is reduced by a reduction in γT (pSErr ) but it is increased by
1
a smaller (1 − pSErr )N factor in the denominator. As a result, γ(pSErr ) = (7)
it is necessary to optimize (4) with respect to pSErr in order 11.5pSErr
to ensure ESym is minimized. The Rayleigh fading used in the simulation could be due
The terms PN , F , POH,Tx , POH,Rx and TSym in (4) will to movement of mobile WSN nodes. For a static WSN, the
remain approximately the same for all wireless links in the simulation can also be interpreted as providing average BER
WSN. However, the path attenuation factor, A, in (4) will for different deployment scenarios where the position of each
change considerably for different links. This will cause a node is sufficiently altered to result in a different small scale
change in the optimal value of pSErr . fading value.

77
Substituting (7) into (4) results in comparison in Fig. 1. The flat portions of the SER curves
  at low pathloss values correspond to where the optimization
1 A
ESym = + B (8) routine encounters the lower SER bound. Using (5), the
(1 − pSErr )N pSErr variation in link delay caused by the changing SER values in
where Fig. 2 is calculated and shown in Fig. 3. In this figure, delay is
expressed as percent delay relative to the un-optimized system
PN F ATSym operating at pSErr = 10−3 .
A= η11.5 B = (POH,Tx + POH,Rx )TSym (9) The first observation made from Fig. 1 is that optimizing
Assuming f (x) denotes the dependence of ESym in (8) on pSErr can result in some significant savings. This is particularly
pSErr , the second derivative of (8) is given in (10) where only true for large pathloss values where the savings exceed 80%.
the first term in the numerator could possibly be negative The savings also fluctuate considerably with pathloss and go
depending on the value of N . The 802.15.4 WSN standard to zero at some point for each curve in Fig. 1. These locations
allows for 22 ≤ N ≤ 256 [3]. It was verified numerically that represent the pathloss values where a pSErr of 10−3 is optimal
the numerator in (10) does not go negative for these frame and they match where the optimal pSErr curves in Fig. 2
sizes and the range of pSErr in (6). As a result, it can be intersect the 10−3 line.
concluded that (8) is convex. It is important to note that if a different fixed pSErr was
 2 N +1
chosen for this comparison, the location of the zero percent
f  (x) = A x (1−x) (N +1)
x4 (1−x)2·N +2
savings points in Fig. 1 would move. However, the consider-
N +1 able energy savings seen above and below that point would
+ (1−x(N +1))(2x(1−x)
x4 (1−x)2·N +2
2
 (10) remain approximately the same.
+1)(1−x)N )
− x x(N
4 (1−x)2·N +2 The large variation in the optimal value of pSErr shown in
(N +1)(1−x)N Fig. 2 indicates that pSErr must be optimized separately for
+B · (1−x)2N +2 each individual link in the WSN. There clearly is not one
For this study, the minima of the cost function (8) is value of pSErr that will minimize energy consumption for the
determined by numerically solving the Kuhn-Tucker equations range of pathloss values encountered across a WSN. These
using the sequential programming method [10]. In a WSN, this optimal values of pSErr could be determined dynamically by
optimization would have to be performed for each wireless link the nodes themselves or they could be estimated during the
by the receiving node in order to determine its received SER planning stages of a static WSN.
target. That target could then be achieved using closed loop Fig. 2 also shows that the optimal value of pSErr is lower
power control. It is not expected this optimization would be a for longer frame lengths. Longer frames require more energy
burden to the node since many WSN links are static and the to transmit, which increases the penalty of retransmissions.
optimization will only be required infrequently. The lower value of pSErr is an attempt by the optimization
algorithm to offset this increased cost.
IV. O PTIMIZING A SISO L INK At low pathloss values, Fig. 3 shows that the optimization
This section evaluates the benefit of using pSErr optimiza- routine decreases link delay. This is supported by Fig. 2
tion in a SISO system. The energy per symbol consumed by which indicates that a pSErr of 10−3 is too high for low
the WSN physical layer after this optimization, ESym,Opt , can pathloss values. The result of using the higher fixed SER is
be determined by substituting the optimal pSErr value into unnecessary retransmissions and a higher link delay. However,
(4). To evaluate the energy saved by this method, ESym,Opt is delay increases sharply at higher pathloss values where the
compared with ESym,Fix , the energy consumed by a system optimal pSErr is much larger than 10−3 .
using a fixed pSErr value of 10−3 . While the selection of this Generally, the optimization routine raises pSErr for large
un-optimized pSErr is somewhat arbitrary, the value of 10−3 is pathloss values and lowers it for small pathloss values. At
similar to the fixed SER and BER targets used in [6], [7]. The large pathloss values where transmit power is high, reducing
percent energy saved by pSErr optimization is then determined the target SNR through a higher pSErr provides the most
relative to ESym,Fix . gain. At low pathloss values, the overhead hardware energy
It was pointed out in Section II that (4) is a function of consumption, represented by POH,Tx and POH,Rx in (4),
the pathloss factor, A. It is therefore reasonable to suggest exceeds the transmit energy. In this case, lowering pSErr to
that both the optimal value of pSErr and the savings obtained minimize retransmissions results in the lowest overall energy
by that optimization will change as a function of A. As a consumption.
result, a range of pathloss values from 50 dB to 125 dB are For large pathloss values, Fig. 2 shows that the optimal value
considered. For each pathloss value, the percent energy saved of SER appears to approach an asymptote. To determine this
by pSErr optimization is calculated. A plot of the result is asymptotic value, (8) can be written for large pathloss values
shown in Fig. 1. as
The optimal value of pSErr as a function of pathloss is
shown in Fig. 2 along with a straight line that indicates the A
ESym  (11)
fixed SER of 10−3 that is being used for the power savings pSErr (1 − pSErr )N

78
Setting the derivative of (11) to zero reveals the asymptotic
value of pSErr is
N =42
1 80 N =102
pSErr,Asym = . (12) N =162
N +1 70 N =222

% Over Fixed SER Case


This expression supports the conclusion that optimal SER 60

decreases with increasing frame length.


50

V. C ONCLUSIONS 40

This paper has shown that WSN physical layer energy


30
consumption can be reduced by optimizing the SER target
used by the physical layer. This optimization balances the 20

reduction in transmit power achieved by a higher SER with the 10


energy spent on additional retransmissions due to increased
symbol errors. Simulated results for a SISO link indicate 50 60 70 80 90
Path Loss [dB]
100 110 120

that a considerable amount of energy can be saved with this


approach.
Fig. 1. Energy savings for the SISO link.
The implication of setting the physical layer SER based
solely on energy consumption is that the WSN application
using the link may experience a change in average delay. How-
ever, simulated results show that this variation only becomes
significant at very large pathloss values. N = 42
−2 N = 102
10
N = 162
ACKNOWLEDGMENT N = 222

The authors would like to thank TRLabs, the Natural Sci-


ence and Engineering Research Council (NSERC) of Canada
and the Alberta Informatics Circle of Research Excellence
SER

−3
10
(iCORE) for supporting this research.
R EFERENCES
[1] K. Romer and F. Mattern, “The design space of wireless sensor
networks,” IEEE Wireless Communications, vol. 11, no. 6, pp. 54–61, −4
December 2004. 10

[2] W. Ye, J. Heidemann, and D. Estrin, “An energy-efficient MAC


protocol for wireless sensor networks,” in Proceedings of the 2002 50 60 70 80 90 100 110 120
Joint Conference of the IEEE Computer and Communications Societies Path Loss [dB]
(INFOCOM ’02), June 2002, pp. 1567–1576.
[3] IEEE Computer Society, “IEEE std 802.15.4 - 2003,” The Institute of
Electrical and Electronics Engineers, Inc., 2003, New York, USA. Fig. 2. Optimal pSErr for the SISO link.
[4] K. Jin and D. Cho, “A MAC algorithm for energy-limited ad-hoc
networks,” in Proceedings of the 2000 IEEE 52nd Vehicular Technology
Conference Fall (VTC 2000-Fall), Septmeber 2000, pp. 219–222.
[5] A. J. Goldsmith and S. B. Wicker, “Design challenges for energy-
constrained ad hoc wireless networks,” IEEE Transactions on Wireless
N = 42
Communications, vol. 9, no. 4, pp. 8–27, August 2002. 180
N = 102
[6] S. Cui, A. J. Goldsmith, and A. Bahai, “Energy-constrained modulation N = 162
160 N = 222
optimization,” IEEE Transactions on Wireless Communications, vol. 4,
no. 5, pp. 2349–2360, September 2005. 140
% Over Fixed SER Case

[7] S. Cui, A. J. Goldsmith, and A. Bahai, “Energy-efficiency of MIMO


120
and cooperative MIMO techniques in sensor networks,” IEEE Journal
on Selected Areas in Communication, vol. 22, no. 6, pp. 1089–1098, 100
August 2004.
80
[8] X. Li, M. Chen, and W. Liu, “Application of STBC-encoded cooperative
transmissions in wireless sen sor networks,” IEEE Signal Processing 60
Letters, vol. 12, no. 2, pp. 134–137, February 2005.
40
[9] I. N. Bronshtein and K. A. Semendyayev, Handbook of Mathematics,
Springer, 3 edition, 1957. 20
[10] S. S. Rao, Engineering Optimization: Theory and Practice, John Wiley
0
& Sons Inc., 3 edition, 1996.
−20
50 60 70 80 90 100 110 120
Path Loss [dB]

Fig. 3. Delay variation for the SISO link.

79

Das könnte Ihnen auch gefallen