Scaling the Throughput of Wireless Mesh Networks via Physical Carrier Sensing and Two-Radio Multi-Channel Architecture

1
Scaling the Throughput of Wireless Mesh Networks
via Physical Carrier Sensing and Two-Radio
Multi-Channel Architecture

• Jing Zhu, Sumit Roy, Xingang Guo, and W.
  Steven Conner
• Department of Electrical Engineering
• U of Washington, Seattle, WA
• Communications Technology Lab
• Intel Corporation, Hillsboro, OR

2
Outline of Presentation

• Mesh Networks Introduction, Architecture

• Enhancing Aggregate (Network) Throughput
• 1. Enhance spatial reuse via optimal
  physical carrier
• sensing
• 2. Multiple Orthogonal Channels (frequency
  reuse)
• Channel Allocation with clustering

• Multi-Radio, Multi-channel Architecture ?
  Towards a soft-radio architecture for
  high-performance MESH

3
Mesh Networks Salient Features

• Scalability for coverage
• Single hop ? Multi-hop (mesh)
• Heterogeneous Nodes, Hierarchy
• Mobile Clients, APs, SoftAPs (router)
• Multiple PHY technologies
• WiFi, WiMAX, UWB,
• Challenge for MAC in Mesh
• - Current MAC Protocols (e.g. 802.11) are not
  optimized for Mesh
• low efficiency, poor fairness,
• Key Solution Approach Spatial Reuse Channel
  Reuse

4
Example1 AP-MT MeshEnterprise

1. As clients (laptops) increase, more APs are
   needed in the same area.
2. Available orthogonal channels is very limited
   (3 or 8 in 11b/a) ? increased multiple acccess
   interference.

5
Example 2 Wireless AP-AP Mesh
6
Link Capacity
How to scale a MESH?
Our Focus

X
X
Network Throughput
Frequency (Channel) Reuse
Spatial Reuse
7
Outline

• CSMA/CA the core of 802.11 MAC
• Spatial Reuse and Physical Carrier Sensing
• Implementation of PCS in OPNET Simulation of
  Spatial Reuse
• Enhance Physical Carrier Sensing Scheme
• Optimal PCS threshold through tuning PCS
  adaptation
• Channel Reuse Two-Radio Multi-Channel Clustering
  Architecture
• Next-gen Adaptive MAC Framework for Mesh

8
CSMA/CA basic 802.11 MAC

• Carrier Sensing Multiple Access / Collision
  Avoidance
• Physical Carrier Sensing (PCS) for Interference
  Avoidance
• Binary Exponential Back-off (BEB) for Collision
  Avoidance
• (Optional) RTS/CTS Handshaking
• Advantages
• Asynchronous, Distributed, Simple
• Disadvantages
• Low Spatial Reuse (due to Non-optimized PCS)
• No QoS Support (due to pure contention-based
  access)

9
Spatial Reuse

• Multiple communications using the same
  channel/freq happen simultaneously at different
  locations w/o interfering each other
• Received SINR Model
• Physical Carrier Sensing
• A station samples the energy in the medium and
  initiates transmission only if the reading is
  below a threshold ? threshold optimization

10
Hidden node Problem Revisited
Hidden Node A node that cannot hear the current
transmission but will cause the failure of the
transmission if it transmits.
Any node outside of transmission range of Tx and
Rx could be a hidden node, which cannot be
prevented by using RTS/CTS!
11
Hidden Nodes in a MESH

• Multiple (group) of hidden nodes in a mesh
• Accumulation of interferences
• Impossible to identify due to the unknown number
  of contributors.
• Instead of preventing all hidden nodes, the goal
  of the interference avoidance/mitigation is
  pro-actively avoiding the worst-case
  interference
• Sensed energy during PCS is a good indicator of
  interference level on the coming transmission.
• The lower the sensing threshold, the higher the
  received SNIR on average

12
Effect of PCS threshold on Network Throughput

• Has a great impact on the performance
• PHY improvement does NOT necessarily mean
  proportional improvement at MAC
• Optimal PCS threshold varies with data rates and
  topology
• How to set the optimal carrier sensing threshold
  dynamically?

13
Comparison with analytical estimates

• Analytical estimate of end2end tput
• Observations
• Near optimal results can be achieved by simply
  tuning the carrier sensing threshold without
  using RTS/CTS

(simulation is for 90-node Chain)
1 Xingang Guo, Sumit Roy, W. Steven Conner,
"Spatial Reuse in Wireless Ad-hoc Networks," IEEE
VTC 2003, Orlando, FL, October, 2003.
14
Optimal PCS Threshold

• Assumptions
• Co-location of receiver and transmitter
• Homogenous links (same reception power)
• Ignore background noise
• Saturation traffic load
• Result
• Optimal PCS Threshold 1/S0, where S0 is the
  SINR threshold for sustaining the maximum link
  throughput
• S0 11dB, 14dB, 18dB, and 21dB for 802.11b
  1Mbps, 2Mbps, 5.5Mbps, and 11Mbps, respectively.

15

• 10x10 Grid with Local Only Traffic and Homogenous
  Links

16
Comparison of 1/S0 with the Simulation Optimal
PCS threshold
1/S0(dB)
Simulations match the theoretical estimates !
17
Enterprise Network AP-MT Mesh
3 Channels 16 / 30 / 72/ 110 APs per
channel 11Mbps, So 21dB 154 m x 154 m
Office Path Loss Exponent 3
18
Scale the Capacity of Enterprise AP Network
73
60
40
28

1. Network capacity is proportional to of APs
2. The optimal PCS achieves best per-AP capacity

19
Summary Spatial-Reuse for a single-channel MESH

• Spatial-Reuse the key to improve the aggregate
  throughput of a single-channel mesh
• links sufficiently separated can transmit
  simultaneously without interfering each other
• Limitations
• Not effective for a small scale network, i.e. the
  required minimum separation distance could be
  high.
• For example, gt7 hops in a regular chain network
  with 802.11b 1Mbps and path loss exponent 2.
• Further Scaling the Throughput with Multiple
  Channels!

20
Scaling the Throughput with Multiple Channels

• Takes advantage of multiple channels (even
  multiple bands)
• 8 orthogonal channels in 802.11 a
• 3 orthogonal channels in 802.11 b
• UWB, 802.11, and 802.16
• Channel Bonding (wider channel BW) is another
  alternative
• Increases peak link rate but does not translate
  to proportional MAC throughput increase
• Lack of backward compatibility proprietary
  solution
• Multi-channel Approaches Our Choice
• No change on channel BW
• Use all available channels through the network
• Key issues channel allocation

21
Feasible Multi-Channel Architectures

• One-Radio Multi-Channel Approaches
• Efficient, but will require new MAC (hence not
  backwards compatible)
• Still cannot do full-duplex transmission
  (e.g.difficult to conduct channel sensing
  consistently due to channel switching)
• Control overhead per-packet channel swtiching
• Multi Radio One Channel per NIC(Network
  Interface Card)
• Simple to implement
• Each NIC channel is fixed (i.e. comes hard-coded
  from manufacturer)
• no negotiation required for channel selection
• Fully compatible with legacy
• But costly, will not scale (number of NICs
  number of channels)
• Our Approach Two Radio Multi-Channel
• Scale, i.e. number of NICs fixed at 2
• Backwards compatible
• Assumptions ad-hoc scenario, irregular but not
  random topology, homogenous traffic ? No need to
  frequently update the channel allocation!

Jiandong LI, Zygmunt J. Haas, and Min Sheng
Capacity Evaluation of Multi-Channel Multi-Hop
Ad Hoc Networks '' IEEE International Conference
on Personal Wireless Communications, ICPWC 2002.
A. Adya, P. Bahl, J. Padhye, A. Wolman, and
L. Zhu, A Multi-Radio Unification Protocol for
IEEE 802.11 Wireless Networks, Microsoft
Research, Technical Report MSR-TR-2003-44, July,
2003.
22
Two-Radio Based Network Cluster

• Channel Allocation with Clustering
• Each cluster is identified a common channel
  i.e. all inter-cluster communications using the
  default (primary) radio
• Intra-cluster communications on different
  channels using the secondary radio
• Interference Mitigation
• Interference among co-channel clusters is
  minimized through an efficient channel selection
  algorithm MIX (min. interference channel
  select).
• Interference within the cluster is prevented by
  Physical Carrier Sensing.
• Legacy compatible legacy APs connect to mesh via
  default radio.

23
Framework

• Semi-distributed clustering channel assignment
  distributed MAC mechanisms (802.11 DCF)
• Semi-distributed channel on secondary radio is
  assigned by the local cluster-head within the
  cluster
• Distributed CSMA/CA MAC protocols
• Default vs. Secondary Radio
• Both radios are for data transmission
• The secondary radio has no administrative
  functionality, such as association,
  authentication, etc.
• The common channel on the default radio is
  determined a-priori.
• Layer 3 (IP) routing between the nodes

24
Distributed Highest Connection Clustering (HCC)
Algorithm

• A node is elected as a clusterhead if it is the
  most highly connected (has the highest number of
  neighbor nodes) node of all its uncovered"
  neighbor nodes (in case of a tie, lowest ID (e.g.
  MAC address) prevails).
• A node which has not elected its clusterhead is
  an uncovered node, otherwise it is a covered
  node.
• A node which has already elected another node as
  its clusterhead gives up its role as a
  clusterhead.

M. Gerla and J.T.-C. Tsai, "Multicluster,
mobile, multimedia radio network", ACM/Baltzer
Journal of Wireless Networks. vol. 1, (no. 3),
1995, p. 255-265.
25
Clustering Procedure

• Step 1 All nodes have their neighbor list ready
  (every node should know its neighbors, how many)
• Step 2 All nodes broadcast their own neighboring
  information, i.e., the number of neighbors, to
  its neighborhood.
• Step 3 A node that has got such information from
  all its neighbors can decide its status
  (clusterhead or slave)

26
MIX Minimum Interference Channel Selection

• On-Air energy estimation per channel
• t0 estimation starting time
• T estimation period
• Ei(t) on-air energy at time t on channel i
• k Selected Channel

27
Forwarding Table (MAC Extension)
Neighbor MAC/PHY
192.168.0.2 Default
192.168.0.4 Secondary
Neighbor MAC/PHY
192.168.0.2 Secondary
192.168.0.3
192.168.0.1
192.168.0.4
Neighbor MAC/PHY
192.168.0.3 Secondary
Cluster 1
Cluster 2
192.168.0.2
Neighbor MAC/PHY
192.168.0.1 Secondary
192.168.0.3 Default

• An IP packet will be forwarded to default or
  Secondary MAC/PHY according to the forwarding
  table in the MAC Extension layer.

28
Example 10 x 10 Grid
Cluster-Slave
Cluster-Head

• Transmission range d
• d neighboring distance

29
Simulation Topology

• Random, Local, and Saturate Traffic
• 10 x 10 Grid
• 802.11 b 1Mbps
• 3 orthogonal channels
• Path Loss Exponent 3
• Packet Size 1024 Bytes
• Dash Circle Cluster
• Dark node Cluster-Head

30
Tracing One-Hop Aggregate Throughput

• The new multi-channel and two radio architecture
  achieves 3X performance, compared to a
  traditional single-channel and single-radio mesh.

31
Throughput Distribution

• Location-dependent fairness problem Links Ai
  experience worse interference environment than
  links Bi and Ci, leading to the oscillation of
  the throughput distribution.
• Future Work How Physical Carrier Sensing could
  mitigate the location dependent fairness problem?

32
200m x 200m 100 nodes Random Topology
33
Performance Comparison in Random Topology

• a) Tracing Aggregate Throughput
  b) Throughput Distribution
• Performance gain of aggregate throughput is
  almost 3x (10Mbps vs. 3.5Mbps)