Upload
mandy
View
64
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Mobility Increases The Capacity of Ad-hoc Wireless Networks By Grossglauser and Tse. Gautam Pohare Heli Mehta Computer Science University of Southern California. Outline. Characteristics & Technical challenges of mobile adhoc networks. Diversity and its analysis Gupta and Kumar Model - PowerPoint PPT Presentation
Citation preview
Mobility Increases Mobility Increases The Capacity of Ad-The Capacity of Ad-
hoc Wireless hoc Wireless NetworksNetworks
ByByGrossglauser and TseGrossglauser and Tse
Gautam PohareGautam PohareHeli Mehta Heli Mehta
Computer ScienceComputer ScienceUniversity of Southern CaliforniaUniversity of Southern California
Outline• Characteristics & Technical challenges
of mobile adhoc networks. • Diversity and its analysis• Gupta and Kumar Model• The authors new approach• Session and Transmission Model
Outline (Cont’d)• Results • Two-phase Scheduling • Sender/Receiver centric approach• Distributed Implementation• Related Work• Conclusion and Critiques
Wireless Internet accessWireless LANsWireless Video/Music Multimedia to the CarSensor Networks Smart Homes/AppliancesAutomated Vehicles/HighwaysAll this and more…
Wire less Networks
Mobile AdMobile Ad--Hoc NetworksHoc Networks
No backbone infrastructure
Routing can be multihop.
Network is dynamic (links change, nodes move around) Fully connected network but with different link SINRs
Adaptive techniques can be exploited
Important for ubiquitous computing
Technical Challenges• Hardware
– Small, lightweight,low power
• Communication Systems– Harsh-time varying propagation
environment
– Scarce radio spectrum
– Data rates
Mobile Wireless Network• Characteristic:
– time variation of the channel strength of the underlying communication links.
• Reasons:– Multi path fading– Path loss via distance attenuation– Shadowing by obstacles– Interference from other users.
Use of Diversity for Time Variation
• Point-to-point Diversity can be obtained by:– Interleaving of coded bits– Combining of multipaths– multiple antennas or multiple basestations
• Multiuser diversity: Basic Idea: performance ↑ thro many independent
signal paths between tx-rc. Multiple users are communicating to the base station
via time-varying fading channel. Motivation from Knopp and Humblet’s paper.
Diversity Analysis:• Obtained by:
- Allocating channel to the user who can best exploit it.
- Overall throughput maximized.
• But Problems in this …… It incurs additional delay. Additional delay [buffer pkts until channel becomes
strong] More diversity can be obtained on asynchronous networks Examples: Email, Event notification
More diversity can be obtained on asyn
n/w reasons: • N/w topology changes due to user
mobility• Theorems we will see which proves
this.• Focus is on applications that can
tolerate delays of several mins/hrs like email,….
Gupta and Kumar Model:• Fixed Ad-hoc network with Relaying
– Nodes are immobile– Each node can be source, destination or
relay
• Result:– As the number of nodes per unit area n
increases, the throughput per S-D pair decreases approximately 1/n.
Problems with fixed Adhoc N/w :• Excessive interference in long range direct
communication.• If relaying used - # of hops increase (root of
n)• Nodes carry lot of relay traffic. Hence, more delay, less throughput.• Scalability problems !!! - Traffic rate per S-
D actually goes down to Zero as n increases.
New Approach by the authors is :• Avg long-term throughput per S-D pair can be kept
constant even if n increases !• Totally contrasting to fixed model by G & K.
How is this achieved ???• Source node sends packets to every other nodes
within its radius which act as relays.• Relays, when come near to the destination node,
deliver packet.• Many relay nodes => more probability its close to
dest node.• Imp: Max 1 Relay => Max 2 Hops S-D . So η ↑.• Fault Tolerance !!!
Models• Session Model:
– Each node is source node for one session and destination node for another session.
– Source-Destination association does not change with time.
– Each node has an infinite buffer capacity.– Nodes themselves move.
Transmission Model(Time slots present)
• At time t, node I transmits data at rate R packet/second to node j if
Pi(t) - transmission power, ij(t) - channel gain, - signal to interference ratio, No - background noise power, L - processing gain
Transmission Model (Cont’d)
The channel gain is given by (α > 2 , Xi and Xj are node locations)
2 Models used for experiments:
• One with packets directly transmitted from S-D
• Another nodes can be relays.• Scheduler, depending on scheduling policy,
chooses senders and receivers based on power levels• Long term throughput improvement is
feasible.
Results:
• Thus within a factor of log n, the throughput per S-D pair goes to zero like R/ n in the case when the nodes are fixed.
Fixed Node with Relaying ( G & K Model):– The following results yield upper and lower
bounds on the throughput.• Theorem : There exits c and c’ such that
Results (Cont’d):Fixed Node with Relaying ( G & K
Model):Reason for throughput becoming zero for fixed
nodes:• As Model scales, relay nodes ↑ by √n
In mobile nodes with Relaying, max # of hops is 2.Without relaying, no way of high throughput as n
increases.
Results• Mobile nodes without relaying::• Theorem: Direct transmission between the source and
destination nodes, and no relaying !
ResultsMobile nodes without relaying::• Theorem (Contd) : If c is any constant satisfying
This result says that without relaying, the throughput per S-D pair goes to zero at least as fast as
Problems with mobile nodes w/o relaying
• Too much long range communication• Resulting interference limits #
concurrent transmissions.• Hence throughput low.
Results Mobile Nodes with Relaying
• Why you need the relaying?• the source and destination become nearest
neighbors for small fraction of time(1/n)• Concurrent O(n) transmission are possible• How???• Theorem: For the scheduling policy , the expected
number E[Nt] of feasible sender-receiver pairs is (n).
Mobile Nodes with Relaying gives :
• Transmission to nearest nodes. • O(n) Concurrent successful
transmissions per time slot. • Receiver power at the nearest
neighbor is of same order as total interference from O(n) interferers.
• At max 2 hops for packet to reach from source to destination
Two phase scheduling
• Scheduling of packet transmission from source to destination : 1st phase
• Scheduling of packet from relays to final destination: 2nd phase
• Interleaved phase• Scheduling policy depends only on
location of the node
Two-phase scheduling policy
Throughput analysis:• Throughput = probability of two
nodes to be selected as sender receiver pair
• Throughput per each S-D pair turns out to be O(1) : theorem 3.5
Sender-Centric Vs. Receiver Centric Approach
• Sender-Centric means it is senders that selects the closest receiver to send to.
• Receiver-Centric means it is receivers that selects the closest sender from which receive.
Analysis
• Problem with receiver centric approach:-Two receivers may select the same sender • Adv of receiver centric approach:• In receiver centric approach, signal from
selected sender is always the strongest.• SIR Ratio – the interference in receiver
centric approach is stochastically smaller
Throughput for sender centric approach
Numerical Results• For given α there is an optimal sender
density Θ that maximizes the throughput.
• If density is too low the channel reuse has not been exploited.
• If density is too large the interference power becomes too dominant
• Select Θ carefully
Distributed Implementation– Node decide whether it wants to be sender or
receiver– Even phase: Sender forward packets from
source to relays.– Odd phase: Sender forward packets from relays
to destinations. – Uncoordinated access– Future work: Study of local scheduling strategies and their
impact on end to end delay.
Related Work• Infostation focused on maximizing the capacity of a
point to point channel in fix power budget• Our focus: spatial reuse of channel to
achieve higher throughput.• Related interference model studies the probability of capturing a single
receiver• Gupta & Kumar model
Conclusion• Results show that direct communication
between sources and destinations alone cannot achieve higher throughput because they are too far apart most of the time
• Delay tolerant applications can take advantage of node mobility to significantly increase the throughput capacity of networks.
Critiques:• The scheduling policy given is totally
abstract. • There will be duplicate packets problem
when multiple relays send the same pkt to the destination. No explanation for this.
• No way for real-time data – audio,video etc.• No communication protocol specified which
is crucial for efficiency in wireless networks• All Theoretical concepts, no practical
implementation model !!