MCMSDA: A MultiChannel MultiSector Directional Antenna Wireless LAN

1
MCMSDA A Multi-Channel Multi-Sector Directional
Antenna Wireless LAN

• Yong Huang, Weibo Gong, Dev Gupta?
• University of Massachusetts, Amherst
• ? Newlans Inc.

IEEE WoWMoM 06
2
Outline

• Motivation
• Idea and MAC design
• Scheduling problem and solution
• Boosting scheduling algorithm
• Conclusion

3
Motivation

• Current Wireless LANs
• Pros convenient, low cost
• Cons Low bandwidth and poor security
• Mostly using one channel
• Opportunity
• Multiple non-overlapping channels
• Multi-Sector Directional Antennas
• Why dont we utilize them simultaneously?
• MCMSDA WLAN

4
A simple MSDA model
5
MCMSDA WLAN Architecture

• Each sector of the user node will never see
  two APs with same color

6
IDEA of MAC
Scheduling
7
MAC Design -Time Division Multiple Access (TDMA)

• Channel time is divided into frames
• Frame includes
• Beacon broadcasting scheduling results
• CP contention period to transmit control
  messages
• CFP contention free period for data transmission

• Scheduling algorithm is executed every frame to
  provide channel assignment for next frame

8
Scheduling problem

• How to detour the traffic when some APs are
  congested?
• Goal Minimize total transmission time
• Assumptions
• Channel switching time is ignored (but any
  request can ONLY be carried by one AP during one
  scheduling period)
• User nodes can estimate effective throughput to
  every APs
• Decision variables X(i,j) whether node i uses AP
  j or not (0-1 variable)
• An integer programming problem

9
Problem formulation

• Notations
• m number of APs
• n number of stations
• qi,j effective throughput between station i and
  AP j
• bi requested data length of station i
• xi,j 0-1 decision variable
• ti,j transmission time used by station i via AP
  j
• Tj available CFP time of AP j

• Goal Minimize the total transmission time
• Subject to
• AP constraints
• Station constraints

10
Input and output

• Input
• Available channel time of every AP Tj
• User loads bi
• Throughput between every ltUser, APgt qi,j
• Output
• Decision variable Xi,j

11
Scheduling solution

• Why not using well known solutions? Scheduling
  time is very limited
• Lagrangian Relaxation method
• Good for decomposable problem
• Lagrangian multipliers (?1, ?2,, ?m) can be
  viewed as virtual price and can be re-used to
  reduce scheduling time
• These virtual prices provide more information for
  network management

12
Scheduling algorithm

• Lagrangian function

13
Simulation

• 4 areas
• 11 Mbps to closest AP and 5M to nearby APs, 1Mbps
  to farthest AP
• 70 load in area 1, 10 load in every other area
• Users have same channel request length
• Superframe length 10ms

14
Simulation (contd)

• Parameters
• Network Size number of node
• Channel request length
• Two set of simulations
• Set 1Fixed network size (80 users), various
  channel request length (1kb-20kb)
• Set 2 Fixed channel request length (6kb),
  various number of nodes (100-250 users)
• Metrics
• Aggregated Throughput
• Scheduling time
• Compare to Best WAP strategy
• Best WAP user node always choose the AP with
  highest Throughput
• Duality Gap (related error between scheduled
  objective value and the optimal value) lt 5 even
  when of users is 200

15
Aggregate throughput
Set 1
Set 2

• Higher aggregate throughput than Best WAP strategy

16
Scheduling Time
Set 1
Set 2

• Complexity of solving all sub-problems o(mn)
• How to accelerate the scheduling process?

17
Incremental-adjusting method

• Key idea use l of previous round as initial
  value assuming the user statistic behavior is not
  varying much

18
Backup-resource method

• Key idea
• Use same l for several superframes
• Reserve part of channel time to accommodate the
  channel resource violation
• a percentage of reserved resource
• l is not the accurate price and some requests
  will not be granted
• How the a affect the scheduling time and request
  granted ratio?

19
Backup-resource method (contd)
Scheduling time
Request granted ratio
20
Conclusion

• Wireless LAN performance can be improved by
  combining Multi-channel and Multi-sector
  directional antennas
• Lagrangian relaxation based pricing algorithm can
  quickly reach a good sub-optimal solution
• Accelerate the scheduling process is possible
  with minor performance degradation

21
Thank you!!

• http//tennis.ecs.umass.edu/yhuang
• yhuang_at_ecs.umass.edu