Exploiting Medium Access Diversity in Rate Adaptive WLANs

1
Exploiting Medium Access Diversityin Rate
Adaptive WLANs

• Mobicom 2004, Philadelphia Aug 2004
• Zhengrong Ji, Yi Yang, Junlan Zhou, Mineo Takai,
  Rajive Bagrodia
• Computer Science Department
• University of California Los Angeles
• Los Angeles, CA 90095

2
Problem Statement

• Application demands higher capacity
• Increasing number of users
• Campus, Mesh network, hotspots like airport
  terminals
• Increasing application traffic
• Ubiquitous computing, multimedia, file-sharing,
  p2p applications
• Existing MAC solution Single-link rate
  adaptation
• Utilize multi-rate support at PHY (802.11b/g/a,
  MIMO, ABL-OFDM)
• Auto Rate Fallback (ARF) Lucent WLAN II card
• Receiver Based Auto Rate (RBAR) Holland et al,
  MobiCom01
• Opportunistic Auto Rate (OAR) Sadeghi et al,
  MobiCom02
• Can overall throughput be further improved in a
  wireless LAN with multiple users?

3
Exploiting Multi-user Diversity
What is Multiuser Diversity? In network with
multiple users, each user has an independent
fading channel
Channel Conditions
SNR
AP
TIME
USERS
4
Related Work

• Exploited multiuser diversity in cellular
  networks
• Downlink scheduling improvement
• CDMA2000 1xEV-DO High Data Rate (HDR)
• W-CDMA High Speed Downlink Packet Access (HSDPA)
• Closed-loop feedback of channel conditions via
  uplinks
• Channel-aware slotted ALOHA Qin et al
  INFOCOM03
• Assume symmetric channel conditions
• Each user knows their own channel gain as well as
  channel gain distribution of other users
• Uplink transmission probability of each user
  based on current channel gain

5
Exploit Multiuser Diversity in WLAN

• What are the major differences in WLAN?
• CSMA/CA no explicit channel feedback
• Rate control no power control to help user with
  weak channel
• Challenges to the implementation of multiuser
  diversity
• Provisioning of channel feedback
• Control overhead of channel feedback must be
  minimized
• Fairness among active users
• Maximally exploit high-link-rate channel
  conditions
• We propose an 802.11-based MAC solution
• Medium Access Diversity (MAD)

6
Conceptual Design of MAD
Assume channel condition of each user is known
Scheduling
Channel Probing
Data Transmission
SubScheduling
7
Channel Probing

• Facilitate channel probing by Group RTS
• Group RTS (GRTS)
• Query multiple users for CTS in RA list
• CTS
• Feedback of feasible Data Rate Relative Gain
  (current/avgSNR)

8
Data Transmission Schemes

• MAD using Packet Concatenation (PAC)

sender
GRTS
user 1
ACK 02
user 2
user k
GRTS
9
N-to-1 User Selection Criterion

• Objective
• Improve network throughput while maintain
  statistical temporal fairness
• Assumption
• AWGN channel with Rayleigh fading, allowed data
  rate follows Shannons law
• Channel conditions of all users are known to
  sender
• Maximum Relative Gain Formulation
• Relative Gain function
• User with maximum relative gain wins current data
  transmission
• Observation
• Statistical fairness is guaranteed (due to i.i.d.
  fast fading)
• Throughput is improved if sender chooses to send
  data to a user whose channel condition is near
  its peak

10
k-set Round-Robin Scheduler

• Information Gathering
• User i maintains instant and average SNR
• Average SNR obtained through exponential
  averaging (?0.2)
• More SNR samples obtained via overhearing tx
  from sender
• Proposed data rate ri determined from reception
  of latest GRTS
• Gi and ri passed to sender in CTS
• A naïve algorithm (k-set-round-robin)
• Choose at most k users from active queue for
  probing
• User with highest gain wins will be dequeued
• Repeat above steps, enqueue all users when queue
  is empty

11
Revenue Based Scheduler

• Revenue credit savings a user can use to pay
  for data transmission.
• Let Xi denote revenue of user i
• Xi 0 when output queue to user i is empty
• Candidate selection rule
• N-to-k selection
• 1st kth highest-revenue user wins
• K-to-1 selection
• Reward Ri ??(1Gi) (All rewards are forfeited
  after candidate selection)
• User with highest (XiRi) wins
• Revenue update after data transmission
• Let Ui denote time spent in last data
  transmission to user i.
• Xi 0 if output queue is empty else
• Xi max(Xi-Ui), 0
• Xj?i Xjmax(Ui-Xi), 0

12
Choosing Appropriate Value of k

• Practical constraints ? smaller number is
  desirable
• Channel coherence time
• Control overhead for query
• Assumptions
• A Tx node at center of a disk with radius D
• backlogged queue to every user
• A random set of k users (randomly located in the
  disk) are probed in every MAD transmission
• M physical rates with corresponding SNR threshold
• Free space path loss Rayleigh fading
• All packets are received correctly
• Expected improvement over
  baseline (k1)

13
Analyzing MAD Performance

• Derivation of expected network throughput C with
  MAD
• transmitters (6 in the following cases)
• Data rate distribution follows previous
  assumptions
• Contention modeling directly leveraged from
  Bianchis work
• Performance analysis of the IEEE 802.11
  distributed coordination function IEEE JSAC 2000
• Numerical results (Ts time of succ. service
  EH payload)

14
Simulation Study

• Setup
• Simulation conducted with QualNet
• Follow 802.11a specification radio spec from
  SENAO Inc.
• Free space pathloss and Rayleigh fading
• Topology
• Single sender (star topology)
• Multiple senders (random topology)
• All flows are backlogged with packets (of size
  1KB)
• Performance Metrics
• Aggregated network throughput
• Temporal share fairness

15
Capacity Improvement by MAD

• Varying traffic density
• Star Topology
• Distance between sender
• and user is fixed at 300m
• K-RR vs Rev based schedl
• ARF saturates at 3Mbps
• (Adaptive Rate Fallback)
• OAR at 7Mbps
• Variation of MADs increase
• capacity over OAR by up to
• 100

MAD
OAR
ARF
16
Impact of Transmission Distance on MAD

• Varying transmission
• distance D
• Star topology
• 3 users with equal
• distance to the sender
• Throughput gain over OAR
• increases from 30 to
• 120
• MADs improvement is the
• highest when it matters the
• most (users are far away)

17
Impact of Mobile Velocity on MAD

• Varying mobile speed
• Star Topology (9 users)
• Transmission distance
• fixed at 100m
• Coherence time is shorter
• with higher velocity
• As mobile velocity gets
• higher, performance of all
• schemes decline
• MADs improvement is
• obvious in the simulated
• velocity range

18
Fairness in a Random Topology

• MADPAC-Re with Random Topology Settings
• 25 nodes uniformly distributed in 200x200m2
  terrain.
• 16 Tx node, each has 5 traffic flows to randomly
  chosen 5 users.

19
Conclusion

• Proposed MAD to exploit Multiuser Diversity in
  WLAN
• Proposed efficient data transmission scheme PAC
• Analysis showed MAD throughput gain over OAR (avg
  50)
• Simulation results showed gain over existing rate
  adaptation scheme in range of (30120) for
  heavily loaded WLAN, while temporal fair share is
  maintained
• Possible application to multi-hop wireless
  networks with heavily loaded intermediate routers
• Unique issues in MAC, routing and end-to-end
  performance need to be addressed