team6

advertisement
Next-Generation Wireless Networks
Topic: Vehicular Networks
Team 6
R99945051 洪晧瑜
R99922041 陳彥璋
R99922083 梁逸安
CARS: Context-Aware Rate Selection for Vehicular Networks
P. Shankar , T. Nadeem , J. Rosca , L. Iftode , Proc. IEEE ICNP , Oct. 2008 , pp. 1 – 12.
Introduction
Rate Adaption
IEEE 802.11 allows multiple transmission rate at the physical layer (PHY). Figure 1 shows
the relationship between link quality and transmission rate.
Higher data rates allow high quality links to transmit more data, but have a higher loss
probability on low quality links. On the other hand, a low data rate is more resilient to low
quality links, but fails to achieve a high throughput in a high quality link.
Rate adaption is the problem of selecting the best transmission rate based on the
real-time link quality, so as to obtain maximum throughput at all times.
Link Quality
The link quality is an important factor
in rate adaption. However, the link quality
changes rapidly due to fading, shadowing,
and distance between two stations.
In vehicular networks, when the two
vehicles are stationary and close to each
other, the link quality still oscillated by
small-scale fading. In the case of a moving
car, the large-scale path loss is much more
significant than small-scale fading.
Fig.1 Relationship between link quality and transmission rate.
Key Challenges
Rate adaption in vehicular networks faced the following three key challenges.
1. Rapid variation of the link quality.
In addition to the fading effect, the mobility at vehicular speed mainly variates the link quality.
So the transmission rate must adapt fast in order to be effective.
2. Few information in the estimation window during infrequent and burstytransmission.
The usage of the estimation window is to record the recent transmission results of
packet and then estimate the link quality. When a client periodically transmit packets in short
bursts and then stay quiet, there are no recent samples available in the estimation window,
causing anomalous behavior in rate adaption algorithms.
3. Distinguish losses due to environment from those due to hidden-station induced collision.
A frame loss induced by environmental factors should trigger rate adaption to recover
from the loss. However, a hidden-station induced loss should not trigger rate adaption, as
reducing the transmission rate cannot solve the contention problem.
Context-Aware Rate Selection
Overview
The core idea of CARS is to make use of context
information from the application layer, in addition to
the frame transmission statistics received from the
lower layer. Figure 2 shows the overall system
architecture of the CARS scheme.
The context information used in CARS broadly
consists of information about the environment that is
available to the node. Vehicles can obtain the position
and speed from a GPS device and periodically
broadcast to the neighbors.
Fig.2 The CARS system architecture
Rate Adaption Algorithm
Algorithm 1 describes the CARS
algorithm.The key ideaof the CARS
algorithm is to estimate the link quality
usingboth context information as well as
past history. The CARSrate selection
algorithm estimates the packet error by
meansof a weighted decision function
involving two functions, ECand EH(line 4 of
algorithm). The function ECuses thecontext
information, transmission rate and packet
length asinput parameters, and outputs the
estimated packet error rate.The function
EHuses an exponentially weightedmoving
average (EWMA) of past frame transmission
statisticsfor each bitrate.The algorithm
calculates estimated throughput for each bitrateand selects the bitrate that it predicts will
provide the mostthroughput.
Context information is represented by the variable ctx.
Theweight α determines whether to give preference to the contextinformation or to the
EWMA. αis assigned based on thevehicle speed.More precisely,
α = max(0,min(1, speed/S)). The authors select S = 30 (m/s) as the best value.
N is the maximum number of retransmissions, andavg_retriescomputes the average
number of retransmissions(line 5 of algorithm).
ρis the weight that signifies the penaltygiven to unsuccessful packet transmission. The
authors select ρ = 8 as the bestvalue.
Performance
Experimental Result
Figure 3 shows that CARS consistently
outperforms all the existing rate adaption
algorithms at all distances. However, there
is still a lot of scope for improvement
compared with the supremum throughput.
“Supremum throughput” means the
maximum goodput possible at each
distance range.Fig.3Measured good put averaged over distance for
CARS compared with different state-of-the-art rate
adaption algorithms.
Fig.4 Packet distributions with selected data
rates for Hotspot uploading scenarios.
Fig.5 Packet distributions with selected data
rates for Highway scenarios.
Vehicle-to-Infrastructure Scenario(Simulation)
In this scenario, all vehicles act as clients. The scenario consists of a road of length 5000
meters with multiple lanes. The base station is located at the middle of the road.
As shown in figure4, CARS scheme exploits the available higher data rates better than
the other schemes.
Vehicle-to-Vehicle Scenario(Simulation)
The authors describe the scenarios of Vehicle-to-Vehicle communication over a highway.
Highway scenariosconsist of a bi-directional highway of 10Km length, whereeach direction has
the same number of vehicles.
As shown in figure 5, CARS manages to use higher data rates for transmission more
optimally than both SampleRate and AARF.
Conclusion
In this paper, the authors showed why existing rate adaption algorithms underutilize the
wireless link capacity in vehicular environments, and introduced Context Aware Rate Selection
(CARS), a novel scheme that uses context information from the environment to perform fast
rate adaption in dynamic environments.
Delay-bounded Routing in Vehicular Ad-hoc Networks
Antonios Skordylis, Niki Trigoni , Oxford University Computing Laboratory
ACM International Symposium on Mobile Ad hoc Networking and Computing, 2008
Introduction
VANETs
Vehicles are equipped with wireless
transceivers that will enable them to
communicate with each other and form a
special class of wireless
VANET-based applications
The authors classify VANET-based
applications into two categories:
Figure 1 Model. Shaded circles indicate the Communication range
1) those that require broadcasting of information from one vehicle to many nearby vehicles
2) those that require the propagation of information hop-by-hop to a single destination
Model (Figure 1).
In the urban scenario the authors considerthe network consists of mobile nodes
(vehicles) and a few stationary access points (APs) that do not provide full city coverage
Objective
The authors’ objective is to devise carry-and-forward algorithms that leverage
knowledge of traffic statistics in an urban setting to enable timely delivery of messages from
vehicles to stationary access points, whilst minimizing wireless transmissions and optimizing
bandwidth utilization.
Vehicles :
 Obtain geographical position from a GPS receiver.
 Get digital map(G(V,E)) and historical traffic statistics , including the average speed u and
average vehicle density d.
 Can communicate with neighboring vehicles or APs within a range of 250 meters.
APs: Infrastructure nodes whose absolute location in known to all vehicles.
Message:
 tg :Represent message generation time.
 λ : Represent the time-to-live value(or the message delay threshold).
Algorithm
D-Greedy : Delay-bounded Greedy Forwarding
The D-Greedy algorithm defines a forwarding strategy that assumes no knowledge of
traffic information beyond node speed.
Each vehicle maintains a neighbor list by periodically broadcasting beacons. A beacon
message contains the unique vehicle identifier (id) and the length of the shortest path
between the vehicle’s current location and the location of the closest AP (distToAP).
Let distT oInt be the remaining length, until the next intersection, of the current street
segment e on which the vehicle is traveling.
D-Greedy computes the available delay budget :
Del = = λ × distToInt/distToAP
Then calculate the expected delay if the Data Muling strategy were to be used :
DelDM = distToInt/u.
If DelDM ≤ Del then the algorithm opts for the Data Muling strategy. Otherwise, the
Multihop Forwarding strategy is chosen.
D-MinCost : Delay-Bounded Minimum Cost Forwarding
The D-MinCost algorithm leverages the knowledge of global traffic statistics, i.e.
estimated values of average vehicle speed u and density d for all edges of the street graph G.
Graph extension
The authors convert the original
directed graph G(V,E) that represents
the street map to a new graph G’(V,E’),
which contains the same set of vertices,
and twice as many edges. For each
directed edge e Є G that connects two
vertices, they create a new sibling edge
e’ Є G’ connecting the same two vertices.
Figure 2 Replacing edge (d, a) in G by two sibling
edges, one per strategy.
Then the authors annotate each edge with two metrics:
1) cost (C), representing the number of message transmissions along the edge, and
2) delay (Del), denoting the time required to forward a message along the edge.
For Data Muling strategy :
 CDM = 1, DelDM = l/u, (l : the length of the edge and u : the average vehicle speed)
For Multihop Forwarding strategy : (if l > R and d ≥ l/R )
 CMH = l/R, DelMH = CMH × q, (R : communication range, d : the average vehicle density for
the edge, q : the time required for the node to check its neighbor list and identify the
best next hop)
Path selection
D-MinCost utilizes the Delay Scaling Algorithm (DSA) to compute delay-constrained least
cost paths from the vehicle’s location to all APs on the network. Then the AP that can be
reached with the least cost, the exact min-cost path to that access point, and the strategy that
should be followed at each edge can be obtained.
EVALUATION
Compared with Epidemic routing and MinDelay :
 D-Greedy, DMinCost and MinDelay exhibit very similar behavior, never falling behind
Epidemic’s optimal values by more than 10%.( Figure 3)
 The rate of increase of bandwidth utilization for both D-Greedy and D-Mincost is
significantly lower than that of MinDelay and Epidemic.( Figure 4)
 D-Greedy, DMinCost exhaust the available delay threshold, by delivering messages as
late as possible, so delivering messages later than MinDelay and Epidemic. ( Figure 5)
 The messages that need to travel long distances to an access point make aggressive use
of the Multihop Forwarding mode, while messages closer to an access point alternate
between the two modes. ( Figure 6)
Figure 3 Message Delivery Ratio
Figure 5 Average delivery delay
Figure 4: The total number of bytes sent
Figure 6: Strategy chosen
Conclusion
In this paper, the authors proposed two packet forwarding schemes for urban vehicular
networks, D-Greedy and D-MinCost, to route packets towards fixed infrastructure nodes. The
algorithms leverage local and global traffic statistics to choose strategies that minimize
communication cost and adhere to delay threshold. The schemes outperform Epidemic and
MinDelay in terms of communication cost and achieving a similar delivery ratio to Epidemic.
Download