Minimizing Wireless Connection BER through the Dynamic Distribution of Budgeted Power
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
1
Minimizing Wireless Connection BER through the
Dynamic Distribution of Budgeted Power
Bilal Khan∗ Ghassen Ben Brahim† Ala Al-Fuqaha† Mohsen Guizani†
Abstract— We develop a new dynamic scheme which contin- In this paper, we will not consider mobility-related issues.
uously redistributes a fixed power budget among the wireless Although our investigation makes the simplifying assumption
nodes participating in a multi-hop wireless connection, with the of a scenario in which mobility does not greatly impact power
objective of minimizing the end-to-end wireless connection bit
error rate (BER). We compare the efficacy of our scheme with allocation decisions, the conclusions we present are neverthe-
two static schemes: one that distributes power uniformly, and less significant in the broader context of power management
one that distributes it proportionally to the square of inter-hop in wireless and ad-hoc networks.
distances. In our experiments we observed that the dynamic The remainder of the paper is organized as follows. We
allocation scheme achieved superior performance, reducing BER begin in Section II with an exposition of prior related research
by using its ability to distribute the power budget. We quantified
the sensitivity of this performance improvement to various envi- work. Then, in Sections III and IV we define the problem and
ronmental parameters, including power budget size, geographic the presumed network model. In Section V, we describe the
distance, and the number of hops. protocol by which power is redistributed dynamically, to attain
Index Terms— wireless ad-hoc networks, multi-hop path, bit minimum BER. In Section VI we describe the experimental
error rate, power budget, optimal power distribution. setup, and then analyze the results of the simulation study in
Section VII, by comparing the proposed protocol against other
traditional power distribution schemes.
I. I NTRODUCTION
New distributed computing/communication applications
drive the energy requirements of wireless ad-hoc systems II. R ELATED W ORK
ever upwards, while simultaneously, the batteries which power Approaches for efficient power management have been
wireless devices present a hard constraints on the operation of investigated at various protocol layers by several researchers,
mobile computing systems. Recent developments in devices (e.g. see [14], [13], [4]) 1. At the Physical layer: Using
with tunable transmission power enable us to manage the directional antennae, applying knowledge of spatial neighbor-
tension of power supply and power demand using dynamic hood as a hint in setting transmission power; 2. At the Data-
power redistribution schemes. In this paper, we present results link layer: Avoiding unnecessary retransmissions, avoiding
of recent investigations along this avenue. Our objective is to collisions in channel access whenever possible, allocating
optimize the bit error rate (BER) of connections—and hence contiguous slots for transmission and reception whenever pos-
the packet-level error rate (PER) experienced at the network sible; 3. At the Network layer: Considering route-relay load,
layer. Since many applications require a minimal Quality of considering battery life in route selection, reducing frequency
Service (QoS) to guarantee acceptable responsiveness, such an of control messages, optimizing size of control headers, route
improvement can greatly benefit network function. reconfiguration; 4. At the Transport layer: Avoiding repeated
Historically, reconciling the gap between power consump- retransmissions, handling packet loss in a localized manner,
tion and supply involved [14] solving the following issues: (i) using power-efficient error control schemes.
improving the power efficiency in the system; and (ii) prevent- One broad category of solutions consists of energy aware
ing the system deconstruction due to unfair power usage. In routing protocols (e.g. see [13], [6], [8]). In wired networks,
our earlier work [2], [3], we proposed addressing these issues the emphasis has traditionally been on maximizing end-to-end
through the principle of optimal allocation of budgeted power; throughput and minimizing delay. To maximize the lifetime
we introduced a model in which every connection request is of mobile hosts, however, routing algorithms must select the
assigned a fixed power budget to support its instantiation. best path from the viewpoint of power constraints and route
In this paper, we present a scheme which dynamizes these stability. Routes requiring lower levels of power transmission
approaches by enabling the redistribution of a power budget are generally preferred, but this can adversely affect end-to-
among the constituent nodes in a multi-hop connection, with end throughput. Transmission with higher power increases the
the objective of minimizing the wireless connection BER. probability of successful transmission, although high power
Standard models of wireless ad-hoc networks typically con- strategies also result in more cross-node interference, destroy
sider infrastructure-less networks in which every node assumes existing transmission bands, and thus cause the network to
the role of both a host and router, and every node is mobile. have blocked connections. In [5] and [1], Banerjee and Misra
showed that energy-aware routing algorithms that are solely
† Western Michigan University, MI.
∗ John Jay College of Criminal Justice, City University of New York, NY based on the energy spent in a single transmission are not
10019. able to find minimum energy paths for end-to-end reliablepacket transmissions, in both End-to-End and Hop-by-Hop received by node j is given by
retransmission settings. Pt (i)
Our own prior work [3] was a natural extension of Misra Prcv (j) = , (1)
c × dα
[1] and Banerjee [5], reframed by normalizing experimental ij
scenarios using a fixed power budgets for each connection. where dij is the distance between nodes i and j. α and c are
In [2], we presented an experimental evaluation of those both constant, and usually 2 ≤ α ≤ 4 (See [5]). In order to
techniques, showing how data replication along multiple paths correctly decode the signal at the receiver side, it is required
can be used to lower packet error rate of application layer that
connections in wireless ad-hoc networks under power budget
P (j) > β0 × N0 , (2)
constraints.
This work begins at the point where energy aware routing where β0 is the required signal to noise ratio (SNR) and N0
ends. Here we propose a new dynamic scheme that con- is the strength of the ambient noise. We denote the minimum
tinuously redistributes the power budget among all nodes signal power at which node i is able to decode the received
across a multi-hop wireless connection with the objective of signal as Pmin .
minimizing the wireless connection BER. Each link (i, j) has a computable bit error rate BER(i, j),
which represents the probability of the occurrence of an
error during the data transfer over that link. The relationship
III. P ROBLEM D EFINITION
between the bit error rate BER over a wireless channel and
Consider a single connection request between a source node the received power level Prcv is a function of the modulation
s and a destination node t, and assume that a transmission scheme. It can be expressed in general as follows [5].
power budget P has been specified for this connection. The s !
question to be answered is how should P be distributed among Prcv Cte
BER ∝ Q , (3)
intermediate nodes of the connection if the objective is to f Pnoise
minimize the end-to-end connection bit error rate? We shall
where Pnoise is the noise spectral density, f is the raw channel
assume, as assumed in other similar investigations (e.g. [11]),
bit error rate, and Q(x) is defined as follows.
that each node has the ability to send with dynamically tunable
2 x −t2
Z
transmission power, and that node mobility is insignificant Q(x) = 1 − e dt. (4)
when compared to routing convergence times. The proposed π 0
dynamic power distribution protocol is implemented on top of Since we are only interested in studying the general depen-
a routing protocol that is responsible for providing a multi-hop dence of the bit error rate on the received signal power, we
path between s and t, within total power budget constraints— will consider the non coherent binary orthogonal Frequency
designing such an energy-aware routing protocol is beyond the Shift Keying (FSK) modulation scheme. Other modulation
scope of this paper. schemes can be analyzed in similar way, however closed-
Our design idea is founded on the simple observation that form analysis may not be always possible. For this specific
in a multi-hop path the distance between two consecutive modulation scheme, the instantaneous channel bit error rate
intermediate nodes varies on a hop-by-hop basis. For nodes BER is given by [9], [10], [7] to be:
which are a short distance from each other, less power can be
1 − 2PPrcv
allocated while still attaining good channel bit error rate. When BER = e noise (5)
2
two consecutive nodes are far from each other, a weak trans-
mission power would result in a high wireless channel bit error A path consisting
Qrof a sequence of links L1 , . . . , Lr has a
rate. We present a dynamic power redistribution scheme based BER equal to 1 − `=1 1 − BER(L` )
on geographical distance, which allows nodes to negotiate the
amount of power they use (while remaining within connection V. DYNAMIC S CHEME
budget constraints) thereby optimizing overall connection bit The proposed protocol operates on all (overlapping) consec-
error rate. utive triplets of nodes within the connection (s, t). Within each
triplet, we denote the nodes to as the upstream node, the central
node, and the downstream node. This naming convention is
IV. N ETWORK M ODEL
illustrated in Figure 1.
We consider a wireless ad-hoc network consisting of N A node enters the protocol by simultaneously sending
nodes equipped with omni-directional antennas that can dy- an Update message to its upstream and downstream neigh-
namically adjust their transmission power. We model this bors. The Update message describes its present transmission
network as a linear geometric graph G = (V, E), where V strength. A node receiving an update uses its contents and the
is the set of nodes and E is the set of edges. Each node is actual received signal strength to deduce an estimate of the
assigned a unique ID i in {1, . . . , |V |}, and node i can send distance to the sender of the Update. Thus each node (viewed
data with a dynamically tunable transmission power in the in its central role) maintains estimates of distance to upstream
range [0, Pmax (i)]. and downstream nodes. When the central node receives an
Wireless propagation suffers severe attenuation [5] and [12]. update message informing it of the transmission power and
If node i transmits with power P (i), the power of the signal (implicitly) distance to a neighbor, it determines the optimalUpstream Central Downstream
Upstream Node Central Node Downstream Node
node node node
Update from neighbors
Get Initial
Get Initial Signal Get Initial Signal
Signal Power
Power strength Power strength
strength
s Update message Update message
t
(1) Estimate the
distance to the
upstream and
downstream nodes
Fig. 1. Multi-hop path description (2) Compute best
Pow er distribution negotiation
power allocation
between central and
upstream nodes
(3) if a significant
redistribution of power between itself and the upstream node. change is required
then:
This local optimization is computed on the basis of the analytic
BER model presented in the previous section. In effect the Decrease transmit
signal power
central node acts greedily to minimize the BER of the two
Power Transfer Message
hop sub-path from its upstream node to the downstream node.
Increase
If the local optimization shows that a significant redistribution transmit
signal power
of power is required, and this redistribution will not cause Ack message
Update to neighbors
Update Message
Update
the received signal strength to drop below Pmin at any node, message Update local
information
then the central node is able push power downstream (Figure about neighbor
2) or push power upstream (Figure 3). It accomplishes this
by Power Request and Power Transfer messages, respectively.
Receipt of a Power Request always causes a node to reduce its Fig. 2. Event sequence diagram: pushing power upstream
transmission power and reply with a Power Transfer Message.
Receipt of a Power Transfer Message always causes a node to
increase its transmission power and reply with a Ack Message. two end points. During the experiment, all network parameters
Receipt of Ack and Update Messages always result in further involved in the system are kept in the following ranges:
propagation of an Update Message. The power reallocation • Path Length: We consider path lengths ranging from short
process is negotiated concurrently between all (overlapping) (5 intermediate nodes) to long (25 intermediate nodes).
triplets of nodes via a distributed protocol. The protocol is • Power budget: We consider connection power budgets
said to have converged if the total power exchange drop below ranging from small (1 Watt) to large (10 Watts).
a user specified threshold. In the rest of this paper, we will • Distance: We consider scenarios in which the two end-
refer to the converged distribution attained by this distributed points range from nearby (100m) to distant (400m).
protocol as the Dynamic scheme. We compare the performance • α: A scaling constant is kept fixed at 2, as appropriate to
of the dynamic protocol against two static schemes. our connection scales.
• SN R: The Signal to Noise Ratio of the wireless channel
is kept fixed at 1mW, as appropriate to a typical SN R
A. Uniform Scheme
value for wireless channel.
Given a connection between nodes s and t with length
The graphs in the next section depict the average values
k + 1 hops and a total power budget P . The uniform power
collected from 104 trial runs of each experiment scenario.
distribution scheme consists of allocating to each of the k
We demonstrate how protocol optimally distributes this budget
nodes (excluding the destination node) a uniform fraction of
among the nodes of the multi-hop path under consideration.
the total power Punif = Pk .
VII. R ESULTS AND A NALYSIS
B. Sqr Scheme
To begin, we study the impact of the variance in inter-node
Under this power distribution scheme, the power is allocated
distances on the improvement (in connection BER) achieved
based on the square of the distance to the next hop along
by the Dynamic scheme when compared to the Uniform
the path towards the destination node. Specifically, given a
scheme. Intuitively, one might expect that in a high variance
connection between nodes s and t with length N − 1 hops
scenario the dynamic power distribution would outperform
and a total power budget P, each node j will be allocated a
PN −1 uniform allocation of power, because the negotiation process
power Psqr such that Psqr = P d2j / i=1 d2i , where dj is the
would converge to a significantly different power distribution.
distance from node j to node j + 1 along the path.
However, Figure 4 shows that the effects are more subtle and
cannot be captured by a single parameter of variance. For
VI. E XPERIMENTAL S ETUP instance, for a variance value of 37m, the improvement varies
In our simulations, we consider networks where the inter- from 4% to 24%. Similarly, when the variance is small (say
mediate nodes are randomly distributed along a line between 7m), the improvement varies from 3% to 17%. We concludeUpstream Node Central Node Downstream Node Power Distribution Scheme Efficacy
50
Dynamic/Sqr
Dynamic/Uniform
Update from neighbors
Get Initial
Signal Power
Get Initial Signal Get Initial Signal 40 Sqr/Uniform
Power strength Power strength
Percentage Improvement
strength
30
Update message
Update message
20
10
(1) Estimate the
distance to the
upstream and 0
downstream nodes
(2) Compute the best
power allocation
-10
between the upstream
Power distribution negotiation
and downstream nodes
(3) If a significant
-20
change is required 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
then:
Power (mW)
Power Request Message
Fig. 5. Percentage improvement vs. total connection power budget
Decrease
transmit
signal power
Power Transfer Message
conducted at that power setting. Similarly, the curve indicates
Increase transmit
power that with a 5W power budget, Sqr performed almost 20%
U pdate to neighbors
Update Message
Update
Message
Ack worse than Uniform.
Update local
information Figure 5 illustrates the impact of the power budget on the
about neighbor
performance of each power allocation scheme. The distance
between endpoints was fixed at 120m, and the number of
Fig. 3. Event sequence diagram: drawing power downstream
intermediate nodes was fixed at 9—thus the average internode
spacing was approximately 12m, in the range of present
25
path length = 5
54M b/s wireless technology. Considering the slopes of these
curves we conclude that the improvement of the Dynamic
scheme relative to the Uniform and the Sqr schemes increases
20
as the total connection power budget increases. For example,
comparing Dynamic to Sqr, we see that at 1W power budget
15 Dynamic outperforms Sqr by 8% in terms of BER, while by
improvement
9W the improvement rises to 40%. We note, however, that
10
the relative performance of Uniform and Sqr schemes is not
monotone: when the power budget is small, the Sqr scheme
outperforms the Uniform approach, but as the power budget
5
increases to 10W , the conclusion is reversed. Comparing the
heights of the curves, we conclude that the proposed dynamic
0
0 5 10 15 20 25 30 35 40 45 50
scheme outperforms both of the other power allocation tech-
variance niques in both fair and good wireless channel conditions.
Figure 6 illustrates the impact of varying the distance
Fig. 4. Percentage improvement of Dynamic over Uniform vs. variance
between the connection end points, while keeping constant
both the number of intermediate hops and the total power
that the performance of the dynamic scheme is not well- budget. The connection power budget was fixed at 2200mW ,
modeled by a coarse measure such as variance. and the number of intermediate nodes was fixed at 10—
We compare the performance of our Dynamic scheme with thus the average node transmission power was approximately
both the Uniform and the Sqr schemes. For each of these 220mW , in the range of present 54M b/s wireless technology.
schemes, we study the impact of considering different path Considering the slopes of these curves we conclude that the
lengths, connection power budgets, and end point distance. improvement of the Dynamic scheme relative to the Uniform
The legends of each curve indicate the average relative per- and the Sqr schemes decreases as the total distance increases.
formance of two schemes. For example in Figure 5, the curve For example, comparing Dynamic to Sqr, we see that at 100m
titled Dynamic/Sqr shows the average value of the quantity distance Dynamic outperforms Sqr by 16% in terms of BER,
but at 200m the improvement drops to 3%. We note, however,
BER(Sqr) − BER(Dynamic)
. that the relative performance of Uniform and Sqr schemes
BER(Sqr) is not monotone: The lower curve of Figure 6 reaches a
The fact that this curve passes through the point local minimum at distance 110m. At connection distances
(8000mW, 40%) indicates that when the power budget below this critical value, the improvement of Uniform over
was 8W , the BER achieved by Dynamic was (on average) Sqr decreases as the distance increase, but this behavior gets
40% lower than what was achieved by Sqr, over the 104 trials reversed for distances bigger than 110m. By comparing thePower Distribution Scheme Efficacy VIII. C ONCLUSION AND F UTURE W ORK
60 In all the experiments, the dynamic allocation scheme
Dynamic/Sqr
50 Dynamic/Uniform achieved superior performance relative to the uniform and
Sqr/Uniform
Percentage Improvement
40 distance-squared proportional schemes. This improvement re-
sulted from the dynamic scheme ability to reduce the BER by
30
dynamically allocating the power budget among the interme-
20 diate nodes.
10 In all the experiments conducted, the proposed scheme was
0 seen to converge in fewer than 10 iterations per node. The
convergence rates and communication overhead was tunable
-10
by adjusting the definition of “significant change” in the
-20 protocol. Because we were not considering mobility, this
50 100 150 200 250 300
Distance (meters) cost was taken as the one-time initialization cost for the
connection. In future, we intend to extend our consideration to
Fig. 6. Percentage improvement vs. total connection distance the fully mobile setting. Because our power allocation protocol
is decentralized and dynamic, it can react to node mobility by
Power Distribution Scheme Efficacy
redistributing power in a manner which optimizes the BER. To
20 evaluate the efficacy of the protocol in the mobile setting, we
Dynamic/Sqr are presently conducting experiments to quantify the tradeoffs
15 Dynamic/Uniform
Sqr/Uniform between convergence thresholds, control-traffic overhead, and
Percentage Improvement
10 resultant improvement in BER.
5
0 R EFERENCES
-5 [1] S. Banerjee and A. Misra. Energy Efficient Reliable Communication for
Multi-hop Wireless Networks. Journal of Wireless Networks (WINET),
-10 2004.
[2] G. B. Brahim and B. Khan. Budgeting Power: Packet Duplication and
-15
Bit Error Rate Reduction in Wireless Ad-hoc Networks. International
-20 Wireless Communications and Mobile Computing Conference, IWCMC,
5 10 15 20 25 Vancouver, Canada, 2006.
Path Length (hops) [3] G. B. Brahim, B. Khan, A. Al-Fuqaha, and M. Guizani. Using Energy
Efficient Overlay to Reduce Packet Error Rates in Wireless Ad-Hoc
Fig. 7. Percentage improvement vs. connection length Networks. International Conference on Communications, ICC, 2006.
[4] R. Cravets and P. Krishnan. Power Management Techniques for Mobile
Communication. NOBICOM 98 Dallas Texas USA, 1998.
[5] Q. Dong and S. Banerjee. Minimum Energy Reliable Paths Using
Unreliable Wireless Links. MobiHoc’05, Urbana-Champaign, Illinois,
heights of the curves, we conclude that the proposed dy- May 25-27, 2005.
namic scheme outperforms both of the other power allocation [6] C. E. Jones, K. M. Sivalingam, P. Agrawal, and J. C. Chen. A Survey
techniques in both small and large distances scenarios. As of Energy Efficient Network Protocols for Wireless Networks. Wireless
Networks 7, 343 358, 2001.
the distances become larger, the difference between power [7] G. Laurer. Packet Radio routing, Chapter 11, pages 351-396, Prentice
allocation schemes becomes immaterial. Hall 1995.
[8] Q. Li, J. Aslam, and D. Rus. Online Power-aware Routing in Wireless
Figure 7 illustrates the impact of varying the path length Ad-hoc Networks. Proceedings of ACM Mobicom’2001, pp97-107,
(in terms of the number of intermediate nodes) between the 2001.
source and destination nodes while keeping constant both the [9] S. Loyka and F. Gagnon. Performance Analysis of the V-BLAST
Algorithm: An Analytical Approach. IEEE Transactions onWireles
distance between the connection end points and the total power Communications, Vol.3 No.4, 2004.
budget. The connection power budget was fixed at 2200mW , [10] J. G. Proakis. Digital Communications, McGraw Hill, 2001.
and the number of distance was fixed at 120m–drawing upon [11] A. Srinivas and E. Modiano. Minimum Energy Disjoint Path Routing
in Wireless Ad-hoc Networks. MobiCom’03, San Diego, California,
the two experiment scenarios described earlier. Considering September 14-19, 2003.
the slopes of these curves we conclude that the improvement [12] J. Tang and G. Xue. Node-Disjoint Path Routing in Wireless Networks:
of the the Dynamic scheme relative to the Sqr scheme lightly Tradeoff between Path Lifetime and Total Energy. IEEE Communica-
tions Society, 2004.
decreases as the number of the intermediate hops increases. [13] C.-K. Toh. Maximum Battery Life Routing to Support Ubiquitous Mo-
However, in case of Dynamic versus Sqr and Sqr versus bile Computing in Wireless Ad Hoc Networks. IEEE Communications
Uniform schemes, the improvement increases as the length Magazine, June 2001.
[14] Y. Zhang and L. Cheng. Cross-Layer Optimization for Sensor Networks.
of the path increases. For example, when considering a 10 New York Metro Area Networking Workshop, New York, September 12,
hop path, Dynamic achieved an improvement of 7% over the 2003.
Uniform scheme, while for a 20 hop path, the improvement
was 10%. Comparing the heights of the curves, we conclude
that the proposed dynamic scheme outperforms both of the
other power allocation techniques for both short and long paths
scenarios.You can also read
