Lund University /

Presentation 2012

ETSN01

Advanced

Telecommunication

Network Constructs,

Paradigms

and shifts

Lund University /

Presentation 2012

Two ways of building networks

Licensed vs. Unlicensed Spectrum

Licensed Spectrum

• Solution for QoS sensitive

applications

• Exclusive right to spectrum

use

• Network engineering possible

– Predictability

– manageability

• Complex and costly systems

• For big players

Unlicensed Spectrum

• Inherent best effort systems

• “Some” QoS support possible

• No right to use spectrum

- Competition

- Collaboration

• Simple and cheap systems

• For small / medium players

Lund University /

Presentation 2012

Unlicensed Spectrum

• We will look at cellular networks later let’s start with unlicensed

spectrum constructs.

• Today, almost all spectrum licensed and controlled by authority (PTS

in Sweden, FCC in the US etc.)

• Some bands are available for unlicensed use around the world, most

common bands, ISM bands

– 2.4, 5.8, 24.1 GHz common

• Philosophy: equipment must operate in a manner so as to allow co-

existence

• A set of rules regulates equipment operation e.g. output power, use

of spread spectrum

Lund University /

Presentation 2012

The fall of the link paradigm

• Link paradigm migrated from fixed networks

• often used to model and describe wireless networks as well (wireless

links)

• Can fail miserably

• Consider the most basic case:

• two APs, one STA each

• single omnidirectional antennae

• constant output power

• no path loss

• no fading, shadowing etc.

Lund University /

Presentation 2012

This?

Lund University /

Presentation 2012

Or This?

Lund University /

Presentation 2012

Topological Links vs. spatial graphs

In fixed networks:

• “Links” are independent entities that may be connected through

routers/switches

• The properties of a single link does not affect other links properties

– apart from the traffic distribution shaping etc.

In broadcast based wireless networks:

• Transceivers cover an area

• other transceivers within area affected, dependencies

• overlap in coverage - dependency

Lund University /

Presentation 2012

Effect of interference between networks

• We have already studied the behaviour of 802.11 at the link layer

• However, ALL OTHER LAYERS ARE ALSO AFFECTED

• Lets consider TCP ARQ as an example:

• Assume N overlapping STA

• Each STA selfish, using DCF

• Traffic arrival in bunches

Activit

y

Tim

e

Lund University /

Presentation 2012

How does TCP react?

• High contention - transmissions delayed

• Enough delay - slow start

• TCP goes into exponential backoff

• “link” becomes good again, TCP still in backoff mode

• Remember this expression?

• TCP rate limited by RTT

Lund University /

Presentation 2012

Why is TCP so heavily affected?

• TCP congestion control is aggressive

– Backoff exponential

• Queue length increase also exponential

– Assume two contending wireless

stations

– Poisson arrivals of traffic

– linear increase in traffic means:

• i.e. exponential delay increase

compounded by TCP exp. backoff

• Old known problem with TCP congestion

control.

Lund University /

Presentation 2012

Antenna sectorisation

• Common strategy in wireless to use directional antennas

– Avoid overlap and contention

– Almost exclusively for fixed installations

– mobility tracking inherently difficult in small cells with highly

varying path characteristics

• Possible to create point to point links that really behave like “links”

Lund University /

Presentation 2012

Ad hoc networks

• Sometimes infrastructure not needed, available, desired

– Ad hoc = spontaneous network formation, as needed

– self configuration ability

• addressing

• routing / forwarding

– Can also be used to extend infrastructure in mixed mode

Lund University /

Presentation 2012

Ad hoc Networks

Problems / challenges

• Lack of infrastructure also brings

– fully distributed solutions (no central coordination)

• Mobility / instability of topology and connectivity

• Energy constraints

Lund University /

Presentation 2012

Use cases

• Disaster recovery

– No infrastructure working

• Developing areas

– No infrastructure available

• PANs (Bluetooth)

• WSN

• VANET

• ...

Lund University /

Presentation 2012

Wireless Sensor Networks (WSN)

• Small battery operated sensors with wireless capability

• deployed in:

– nature (bio monitoring)

– farms (agricultural monitoring)

– industry (process monitoring)

– Warehouses/ships/trucks (asset monitoring)

– etc.

• ad-hoc capabilities spontaneous routing formation

• biggest issue: energy conservation

Lund University /

Presentation 2012

WSN strategies

• Routing based on phy. layer constraints

– connectivity

– coverage

– residual energy

– energy usage (multihop or single high power Tx)

• Scalability and complexity important issues

– Common strategy, clustering

Lund University /

Presentation 2012

Mesh networks

• Multi hop wireless backbone network with wireless access

• Cheap alternative (equipment cheap)

• Very similar to MANET but backbone has low mobility

– routing paths more stable

• Can be combined with gateways to the Internet

Lund University /

Presentation 2012

ISSUES with Mesh

• In urban areas, deployment at or near

street level (there is where most people

are)

• Wireless channel heavily dependent on

vehicular movement

– doppler shift

– shadowing

• links can be unreliable

RT

T

Lund University /

Presentation 2012

Multi hopping

• mobile devices - limited coverage

• how get data through a network?

• multi hop routing / forwarding

Lund University /

Presentation 2012

Routing

• Mobility causes paths to be unstable

• Using “traditional” shortest path algorithms inefficient

– Dijkstra, Bellman-Ford etc.

• Two main approaches

• Proactive

– Continuously update about topology and changes

– good: low latency, bad: costly updates

• Reactive

– Only determine routing path when sending data

– good: no constant overhead, bad: delay due to route discovery

Lund University /

Presentation 2012

Example, Ad-Hoc On Demand Distance

Vector Routing (AODV)

• Sender broadcasts a Route request packet

– flooded throughout the network

• Nodes maintain route cache, use destination sequence number for

each route entry

• state installed in nodes per destination

• only reacts when routes fail

– try local recovery (nodes monitor neighbours, issue error message

on detected problem)

– sender repeats route discovery

• Cost directly related to rate of mobility, only works for smaller

networks with limited mobility

• currently popular in deployed systems

Lund University /

Presentation 2012

Performance of multihop networks

• Imagine a row of multi hop links

• Each node unit disk graph coverage, 802.11

• Let’s investigate the capacity of this network

Lund University /

Presentation 2012

Performance of multi hop networks

• Forwarding nodes n = 1-2-3-4-5

• 1 transmits to 2 => 3 silent

• 2 transmits to 3 => 4 silent, 1 cannot receive

• 3 transmits to 4 => 1 and 5 silent

• If infinite n, upper bound on capacity = 1/5 (one Tx, 4 silent hops)

1 2 3 4 5 6

Lund University /

Presentation 2012

Performance of interfering networks

• See Gupta & Kumar Trans. Info Theory for theoretical bounds

• Place n nodes arbitrarily on a unit disk area. Each node capacity is W

bps. The upper bound on capacity for the network is:

• Valid for surface and sphere

• Optimally placed nodes yield:

Lund University /

Presentation 2012

Work around

• Multi channel schemes

• Assign different channels to different links

– Significant capacity increase possible

– Added cost (multiple interfaces) and complexity

• Typically use some graph colouring algorithm or similar

• However, channel assignment not an easy task, see for example Manitpornsut and Landfeldt MSWIM 2009

• The work showed that minimising DOI increases the Hidden node

problem, severely impacts on voice performance

Lund University /

Presentation 2012

More complex routing protocols

• Traditional routing protocols (RIP, OSPF etc.) use no. hops as metric

• Consider other relevant metrics

– interference

– energy usage

– etc.

• Very complex task to construct such protocols that employ multi

constraint optimisation

– conflicting goals

– varying utility functions

• In WSN research, energy optimisation king

• In MESH research capacity king

Lund University /

Presentation 2012

Geographic Routing

• Routing based on nodes' physical positions, rather than network

addresses or routing tables

• Messages are then routed towards a destination location

Lund University /

Presentation 2012

Geographic Routing - Motivation

• Route discovery is costly in terms of time and energy

• Routing tables quickly become out of date in highly mobile networks

• Addressing a message to a position may be more useful than to a

network address

• Sending an accident or traffic jam notification to upstream vehicles

in a VANET

• Sending a request to collect sensor data in a particular region in a

WSN

• Tracking something through a WSN (e.g. an animal)

Lund University /

Presentation 2012

Localisation

• In order to perform geographic routing, we need to know the

locations of nodes

• Can do this using in-built location sensors such as GPS, IMU

• Or we can localise relative to other nodes by measuring e.g. TDOA

of beacons and then using trilateration

Lund University /

Presentation 2012

Greedy Forwarding

• Forward to the node that gets us the closest to the destination

• Guaranteed to have no loops

• Can use different metrics, e.g. distance to target, distance along the

source-destination line, smallest angle to destination (compass

routing)

Lund University /

Presentation 2012

Greedy Forwarding

• Greedy forwarding fails when we hit a local minimum

• Current node has no neighbours closer to the destination than it is

source

destination

Lund University /

Presentation 2012

Greedy Forwarding

• Greedy forwarding fails when we hit a local minimum

• Current node has no neighbours closer to the destination than it is

• We need to find a way to route around the obstacle.

source

destination

Lund University /

Presentation 2012

Face Routing

• Route message along the face of a polygon in a set direction

• When continuing around the face of the polygon would cause you to

cross the line between the source and destination, change to the

next face

Lund University /

Presentation 2012

Face Routing

• Face routing requires that the graph is planar, i.e. no crossing links

• We can planarise the graph by removing crossing edges (e.g.

Delauney triangulation), but this can lead to increased hop count

• Less efficient, both in terms of delay and energy usage

• Face routing is also sensitive to localisation errors since it relies on

geometric information to choose forwarding nodes.

Lund University /

Presentation 2012

Why not both?

• Use greedy forwarding until we reach a dead-end

• Then change to face routing to get around the obstacle

• Once we are closer to the destination than the dead-end node, we

can resume greedy forwarding.

• Whilst using greedy forwarding, we can employ all links – don't need

a planar graph.

Lund University /

Presentation 2012

Contention-Based Greedy Forwarding

• When using greedy forwarding, we want to select the forwarding

node to be the one closest to the destination

• One way is to keep track of neighbours' locations but this increases

overhead and breaks down when we have high mobility.

• Instead, the source node sends RTS and potential forwarding nodes

respond with CTS at a time determined by their distance to the

destination

• Nodes that are closer to the destination will have a shorter timer

• Using this method can increase delay (must wait for the timers to

expire).

• We still need to know neighbours' locations to do face routing when

recovering from a dead end.

• Can request locations from neighbours when required

Lund University /

Presentation 2012

Contention-Based Greedy Forwarding

Lund University /

Presentation 2012

Realistic Network Models

• The unit disk is not a good model of radio

transmission

• Transmission radius is not uniform

• Three regions: good signal, degrading

signal, no signal

• In the intermediate region, we have a link,

but it is unreliable

• We can use the lognormal

shadowing or log-distance

path loss model to give a

reception probability for a

given distance.

Lund University /

Presentation 2012

Realistic Network Models

• Using greedy forwarding, we will select a forwarding node that is far

from previous node

• Likely to land in the region with poor signal

• This can lead to increased retransmissions

• We need to modify our greedy metric to take into account the cost of

retransmissions, and balance this against increased hop count as

results if we choose a closer node.

Lund University /

Presentation 2012

Realistic Network Models

• Nodes can record the number of successfully received packets S and

the total number of transmitted packets T on a give link.

• Packet reception rate (PRR):

• The PRR can serve as a reception probability for future

transmissions.

• We can then combine this with the distance gained to select a

forwarding node

• e.g. use PRR x distance as forwarding node selection metric,

instead of only distance

Wireless Network Constructs

Wireless networks

Infrastructure networks

Licensed spectrum (e.g. cellular)

Unlicensed spectrum (e.g.

Wifi)

Mesh

Ad-hoc networks

WiMAX (802.11s)

Centralised

802.11 a/b/g/n/ac in infrastructure

mode

Wireless sensor

networks

Mobile ad-hoc

networks

Vehicular ad-hoc

networks

Delay tolerant

networks GSM

UMTS

LTE

Personal Area

Networks

802.11 a/b/g/n/ac in ad-hoc

mode

Bluetooth

ZigBee (802.15.4)

WAVE (802.11p)

Short range

NFC RFID

Lund University /

Presentation 2012

The future of wireless networks

Lund University /

Presentation 2012

Hetnets

• Cellular network scalability problem

– capacity limited (bps/Hz)

• need more and smaller cells

– sites too expensive

• what to do?

• Hottest trend in cellular network architecture

– Heterogeneous networks (Hetnets)

– incorporate different access technologies into the mix

– Wireless data offloading (use existing 802.11)

Lund University /

Presentation 2012

Hetnets

Cellular

WiFi

Femto

Lund University /

Presentation 2012

The Internet of Things (IoT)

• Physical objects embedded with sesnors, actuators

processing power, and network connectivity

• ”Smart objects” collect and exchange data with each

other

• Machine-to-machine communication (M2M)

• Uses a range of different, interacting technologies

• WSNs, MANETs, VANETs, WiFi, Bluetooth,

ZigBee, RFID, …

• Applications in healthcare, energy management,

transportation, building automation, industrial

processes, …everything!

[Image: http://www.seediscover.com]

Lund University /

Presentation 2012

Self Organising Networks (SON)

• Increasing density of nodes

– management very hard

– cell and access network planning inherently difficult

– Customers not capable of configuring equipment

– some technologies operate in unlicensed spectrum

• Network needs to be self-organising

• Access configuration can be dynamically configured

– Channel assignment

– Power management

– Traffic separation

– etc.