Design Issues for Wireless Networks Across Diverse and Fragmented Spectrum Collaborators: Bell Labs India: Supratim Deb, Kanthi Nagaraj Bell Labs USA: Piyush Gupta All Rights Reserved.
Download
Report
Transcript Design Issues for Wireless Networks Across Diverse and Fragmented Spectrum Collaborators: Bell Labs India: Supratim Deb, Kanthi Nagaraj Bell Labs USA: Piyush Gupta All Rights Reserved.
Design Issues for Wireless
Networks Across Diverse and
Fragmented Spectrum
Collaborators:
Bell Labs India: Supratim Deb, Kanthi Nagaraj
Bell Labs USA: Piyush Gupta
All Rights Reserved © Alcatel-Lucent 2006, #####
Mobile Data Explosion Will Result in Diverse & Fragmented Spectrum
Examples:
AT&T Spectrum in New York – 700MHz band, 800MHz, 1.7GHz and 2.1GHz
Unlicensed Spectrum in US – DTV Whitespaces (500-700MHz), 2.4GHz and 5.1GHz
2 | Connecting the Next Billion 2010 |
All Rights Reserved © Alcatel-Lucent 2009
Research Goal
Design a network stack that operates across
1) Fragmented and spatially varying spectrum with diverse propagation
2) Devices with different tunable center freq. and b/w range
Routing + Flow
Control
MAC + Spectrum
New Design
Perspective
Selection
PHY + Sensing
3 | Connecting the Next Billion 2010 |
Lot of research
All Rights Reserved © Alcatel-Lucent 2009
No fixed interference graph
Frequency dependent propagation
Complicates spectrum allocation and MAC design
No fixed communication graph
Next-hop (hence, routing) depends on frequency
Leakage power at a distance depends on
frequency
Adjacent channel interference (hence, guard
band) varies with frequency
Cross layer optimization is highly complex
slope = -20
log fc
Power Spectral Density
Devices can tune frequency and bandwidth
Non-channelized system complex MAC
Received Power at
fixed distance, power (dB)
Designing MAC and above: Unique Challenges
All traditional network stack design issues require a
fresh look.
4 | Connecting the Next Billion 2010 |
All Rights Reserved © Alcatel-Lucent 2009
Interference Management
Standard Approach:
Measurement based interference map
But, if the spectrum is diverse …
Two devices may interfere only in certain frequency bands
Design Principle:
Generate different interference maps for different bands
Ideally a single/small number of control channels
Can use measurements over a single control channel
Pr(f1) - Pr(f2) = 20 log(f2/f1)
Interference maps in a higher freq. band can be
deduced from interference map in a lower band (and
knowledge of ambient interference)
5 | Connecting the Next Billion 2010 |
All Rights Reserved © Alcatel-Lucent 2009
Agile and Efficient MAC for
Unlicensed Access in TV
Whitespaces and ISM Bands
All Rights Reserved © Alcatel-Lucent 2006, #####
Gist of FCC Mandate
Usable Free Spectrum: Unused TV channels between 500-698 MHz for
unlicensed portable access
Varies from city to city
Limit Interference to DTV receiver:
Tx power: 16 dBm in adjacent band, 20 dBm elsewhere
Out of Band Emission: Has to fall by 55 dB in the adjacent 6 MHz band
Spectrum Mask
7 | Connecting the Next Billion 2010 |
All Rights Reserved © Alcatel-Lucent 2009
Single Link Capacity With ACI Constraints
Fundamental Question: What’s the optimal transmit power anyway?
High power-> lower bandwidth
Guard band required to prevent interference to the TV channel
16 dBm
Freq •
DTV Channel
Freq •
8 | Connecting the Next Billion 2010 |
Optimal capacity is independent on center frequency
For FCC’s spectrum mask, the power cap is not too
sub-optimal
Design Implications:
•
DTV Channel
•
Observations:
Choose fixed power (marginal loss in capacity)
DTV Channel
20 dBm
DTV Channel
Low power -> reduced system capacity
Ensure a minimum bandwidth (leakage depends on PSD)
All Rights Reserved © Alcatel-Lucent 2009
MAC Design Considerations
Fragmented spectrum + limitations on
maximum tunable bandwidth of a radio
implies
Multi radio AP, single radio client (for
cost considerations)
Need to account for frequency
dependent propagation
Resilient to disruptions
E.g. wireless microphones
Evolution over WiFi
Easy adoption path, quick time to
market, can interoperate with ISM
bands
IEEE 802.11af standard already in
progress
9 | Connecting the Next Billion 2010 |
All Rights Reserved © Alcatel-Lucent 2009
Joint Spectrum Selection and Client Allocation
4
3
2
1
5
6
(1,2,3,4,5,6,7)
(1,2,3,4,5,6,7)
(1,2,3,4,5,6,7)
(1,2,3,4,5,6,7)
Joint Spectrum Selection and Client Assignment is a non-trivial problem
10 | Connecting the Next Billion 2010 |
All Rights Reserved © Alcatel-Lucent 2009
7
Capturing the Utility of a Band
Important Observation:
For most technologies, data rate/Hz
Client-1
depends on SNR as
Data Rate / Hz = a × (SNR in dB) - b
Client-2
ASE = Data rate averaged over all clients / Hz
= a × (SNR averaged over all clients in dB) – b
SNR(f1) - SNR(f2) = 20 log(f2/f1)
ASE in f1 can be generated for ASE in f2
Design Implication:
Two step ASE generation:
Each AP generates ASE in control channel band.
Use frequency dependence and ambient interference measurements to
compute ASE in all bands.
11 | Connecting the Next Billion 2010 |
All Rights Reserved © Alcatel-Lucent 2009
Proportional Fairness is Key for Whitespaces
More likely that
far away client
will bring down
performance of
system
6Mbps
600MHz
6Mbps
2.4GHz
12 | Connecting the Next Billion 2010 |
All Rights Reserved © Alcatel-Lucent 2009
Very simple
design with
minimal
information
overhead on
top of 802.11
ensures
proportional
fairness
Simulation Results
Comparison scheme
White space selection algorithm is
optimal
Number of clients assigned per band is
proportional to number of clients
60-90% improvement in total throughput
Proportionally fair
Frequency agnostic schemes do not work
well!
13 | Connecting the Next Billion 2010 |
All Rights Reserved © Alcatel-Lucent 2009
Making it Work in Practice
Designed frequency translator with < 2
microsecond switching delay
Dynamic range of 100- 900 MHz
Integrates with a WiFi card on Sokeris
box
Extensive indoor trials done
Waiting for experimental license for
outdoor trials
14 | Connecting the Next Billion 2010 |
All Rights Reserved © Alcatel-Lucent 2009