1. Understanding Impact of PHY/MAC Attributes on Performance
PI: Jennifer C. Hou
Postdoctoral research fellows: Tae-Seok Kim and Kyung-Joon Park
Graduate students: Jihyuk Choi, Yan Gao, Ray Lam, and Yong Yang
Department of Computer Science
University of Illinois at Urbana Champaign

2. Problems and Challenges
Interests in wireless ad-hoc networks have opened up new research venues for protocol design, implementation, and development.
However, several performance-related and deployment problems have been identified.
Unpredictable channel behaviors
Inability to locate stable and high-throughput paths due to the shortest path algorithm
Throughput degradation because of intra-flow and inter-flow interference
Lack of incentives (and a pricing mechanism) to forward transit packets
Security holes and lack of privacy protection mechanisms

3. The Notion of a Link No Longer Exists
A link is usually characterized by its bandwidth, latency, packet loss ratio and patterns.
In wireless environments, most links have intermediate loss rates, and the delivery ratio has little correlation with the SNR or the distance.
This is rooted in the fact that the wireless medium is a shared medium, and the sharing range is determined by
Each station’s transmit power and carrier sense threshold
Intra-flow and inter-flow interference (determined in turn by the node distribution and the traffic distribution)
Multi-path fading, small-scale fading, shadowing
Temperature and humidity variation
Existence of objects or obstacles

4. Current Research Work
We aim to take a bottom-up approach and tackle issues that work toward a better characterization of wireless links and its implication for higher-layer protocol design and optimization.
(1) Characterize the behavior of wireless links in such a way that they become amenable to rigorous analysis and reasoning.
(2) Identify control knobs in MAC/PHY layers with which the network capacity can be optimized.
(3) Design and implement modular, virtual device driver that encompasses research components in a city-wide wireless community network, CUWiN (

5. Representing Wireless Links in a Coordinate System

6. Unpredictable Channel Behavior
SNR versus distance
Delivery ratio versus distance

7. How to Better Characterize Wireless Links
Most links have intermittent loss rates.
The distance are not a good characterization of wireless links.
What would be a better characterization?
Whether or not one can still represent the network topology in a coordinate system?

8. A New Coordinate System Based on Signal Strengths
Each mesh node measures signal strength to the other nodes and reports to gateway nodes
A gateway node constructs the coordinate system with the use of measured signal strengths.
wireless node
A wireless node
measures RSSs to
other nodes

9. RSS Measurement
Through exchange of hello packets, a GN n gathers RSS measurement
between itself and a node m that can directly communicate with it.
between a neighbor node of m’s and m.
Node n constructs S=[sij], where sij is (-RSS) measurement made in dBm, 1<= i, j <= p, and p is the number of node n’s one-hop neighbors

10. Virtual Coordinate System
The jth column of S represents node j’s coordinates in a p-dimension.
These coordinates are correlated with each other  it is difficult to identify components that play an important role.
PCA comes to rescue.
PCA transforms a data set that consists of a large number of correlated variables to a new set of uncorrelated principal components.

11. Principal Component Analysis (PCA) of RSS Measurements
Principal components [Jolliffe86]
Using singular value decomposition
 The columns of U are the principal components
Reduction of dimensionality
The principal components become the orthogonal basis of the new space.
Projection using first n principal components: Un
where di = [di, , dp] is the jth column of S.

12. Determining Coordinates for Two-Hop Neighbors
A neighbor node of the GN measures RSSs from its neighbor nodes.
For a node k that is two hops from GN, we obtain the coordinate of k, xk, by minimizing the following objective function:

13. Tuning PHY/MAC Control Knobs for Capacity Optimization

14. Control Knobs in PHY/MAC Layers
To mitigate interference, increase spatial reuse, and maximize the network capacity, there are several control knobs:
Transmit power  power control
Carrier sense threshold  CS threshold tuning
Spatial and temporal domain in which a node transmits  scheduling
Channel diversity  use of non-overlapping channels
we seek a fundamental understanding of how, and to what extent, controlling these attributes impacts the capacity performance.

15. Spatial Reuse Through Controlling CS Threshold
The “contending area” can also be adapted through tuning
the carrier-sensing threshold A B C D
distance
Signal Strength
CS Threshold E F

16. How CS Threshold Controls Contending Area
Larger CS threshold leads to smaller contending area
Less nodes compete the channel in time  less collisions
Larger CS threshold also leads to higher interference
Transmission rate depends on Signal-to-Interference-Noise Ratio A B C D
distance
Signal Strength
CS Threshold E F

17. Tuning Transmit Power or Carrier Sense Threshold
One can increase the level of spatial reuse by either reducing the transmit power or increasing the carrier sense threshold 
The SINR decreases as a result of smaller received signal and the increased interference level.
Can the tradeoff between the increased level of spatial reuse and the decreased data rate each node can sustain be quantified?
Is there a relation between the transmit power and the carrier sense threshold?
Does increasing the transmit power have the same effect of increasing the carrier sense threshold?

18. Network Capacity as a Function of Transmit Power and Carrier Sense Threshold
Increasing Function D
Constants
Spatial Reuse
Link Capacity
As mentioned at the very beginning, the network capacity depends on the channel capacity of individual link and the level of spatial reuse.
Based on this fact, network capacity Γ_n can be formally expressed as this.
C0 and C1 are constants.
This term is the formal expression for carrier sense range D, and f is an increasing function.
In this expression, this part accounts for the level of spatial reuse, and the rests are for the link capacity.
The spatial reuse is derived by dividing the area of the whole network by the contending area of a Tx-Rx pair
The Link capacity part is based on the Shannon capacity formula.
Note that D is inversely proportional to the level of spatial reuse, but proportional to the link capacity .
This tradeoff makes optimal D exist achieving the maximum network capacity.
Note that D depends only on the ratio P_tx/T_cs, this implies that optimal D can be determined by adjusting one, while fixing the other parameter.
We plot the network capacity over transmit power and carrier sense threshold as shown in figure (a).
Capacity is maximized when the combination of the power and threshold fall in the dark brown strip as shown in figure (b).

19. Effect of Carrier Sense
When dcs > d + din, the per-node throughput S decreases with the increase of dcs
The optimal dcs* exists in [0, d + din] s r d
dcs
din

20. Power and Rate Control (PRC) Algorithm
We have devised a power and rate control algorithm, PRC
A localized algorithm that enables each transmitter to adapt to the interference level that it perceives and determines its transmit power.
The transmit power is so determined that the transmitter can sustain the highest possible data rate, while keeping the adverse interference effect on the other neighboring concurrent transmissions minimal.
The rationale of PRC is as follows.

21. Simulation Results
Simulation Results.
As the network gets more populated, the performance gap between the algorithms becomes more significant.
PRC obtained the best performance over all topologies.
The main reason for this is that PRC uses the smallest transmit power since it automatically detects the case
where high data rate can be sustained without the use of a large transmit power.
Using the smaller transmit power leads to the smaller carrier sense range and in consequence higher spatial reuse.
Also in figure (c) it is observed that CS threshold not properly tuned for the DSB causes the larger carrier sense range and lower spatial reuse.
The insight attainable from this results is that a unnecessarily high transmit power and CS Threshold not properly tuned can actually reduces the attainable level of spatial reuse.
In spite of the fact that GPC uses global information, it does not achieve good performance in the lower density network as shown in figure (a).
This is because lower density network makes less interference level, and this causes GPC to use the higher transmit power.
This also leads to lower spatial reuse.
However, the network gets populated, the level of spatial reuse becomes increased due to the reduced transmit power level.
Unnecessarily high transmit power and CS Threshold not properly tuned can actually reduce the attainable level of spatial reuse !!!

22. Real-Life Implementation and Experimentation

23. Urbana Champaign Wireless Community Networks
We are currently working with Champaign-Urbana Wireless Community Network to deploy a open, city-wide wireless community network in Urbana Champaign.
Currently 40 wireless nodes are operational in downtown Urbana, and we expect to extend to 100 nodes providing full coverage of Champaign and Urbana.
Both a research testbed and a production network.
*Supported by NSF Computing Research Infrastructure program.

24. Hardware The CUWiN rooftop router
Contains a Soekris Engineering net4526 single-board computer in a weatherproof enclosure and an b/g radio with Atheros chipset.
Operates in b-standard IBSS mode, and uses b rates, 1, 2, 5.5, and 11 Mb/s.
Is equipped with a CUWiN software solution to defeat IBSS network partitioning and powered by the power-over-Ethernet injector.
Can configure itself as either an Internet gateway or a client, depending on whether it detects a DHCP server on the Ethernet interface.
Approximately $370 per node ( $500 including installation)

25. Configuration database
CUWireless Software Architecture
Web UI
dhcpselect
Configuration database
dhcpd
/etc
zebra
routevizd
dhclient
hslsd
mDNS
responder
nsbridge
Routing
socket
Kernel
FIB
Ethernet NIC
Wireless NIC

26. Measurement Zebra Kernel Modified Madwifi Interference Frame transport
New Quagga
Clients
CUWireless Software
Environment
Channel behavior
modeling
Channel utilization
optimization
Zebra
FIB
Frame scheduling
Kernel
Modified Madwifi
Frame transport
Interference
Detection & mitigation
Measurement
Power assignment
& Parameter turning
Hardware abstraction layer
Wireless NIC
(Atheros Chipset)

27. Rationale for Transparent Device Driver
In order to optimize network performance, PHY/MAC attributes should be exported to higher layer protocols in order to
enable cross layer design and optimization,
promote spatial reuse through tuning of PHY/MAC parameters,
allow implementation of new MAC functions other than those provided by IEEE
Because new MAC functions may be in conflict with existing ones that have been implemented in the firmware of most IEEE interface cards, there should be mechanisms for disabling selected MAC functionalities in the firmware and/or setting various parameters that are originally controlled by the firmware.

28. Network Application TCP IP uniform extension datalink interface
manager
extension interface 1
extension interface 3
Network Application
extension interface 2
device driver
kernel mode proxy
user
module
Application Layer
User space
Kernel space
Transport Layer
Network Layer
Legends
Datalink Layer
standard layer interface links
extension interface links

29. Internals of TDD An extension interface A cross-layer module
registers (unregisters) itself with the extension manager via RegisterExtInterface() (UnregisterExtInterface()).
registers (unregisters) its set/get handlers with the extension manager via RegisterSetHandler() (UnregisterSetHandler()).
A cross-layer module
finds whether or not an extension interface exists via FindExtension().
sets/gets the value of certain PHY/MAC parameters exported by an extension interface via GetExtParam() and SetExtParam().
The uniform extension manager maintains
(i) the definition record of all the supported events in an event definition tree; and
(ii) the list of subscribers of each event.
An extension interface generates and delivers an event to the uniform extension manager by calling TriggerEvent().
A cross-layer control module (un-)subscribes to an event with a callback function by calling AddEventHandler() (RemoveEventHandler()).

30. user module 1 user module 2 dispatcher user module 3 device driver
uniform extension manager
user module 1
event definition tree
subscription record
user module 2
event queue
dispatcher
user module 3
synchronous events
asynchronousevents
access tunable parameters and extension interface
event trigger
tunable parameters
device driver

31. Desirable Features
Controlled transparency: The TDD provides a transparent and generic interface for higher-layer protocol modules to access, through well-defined APIs, a rich set of PHY/MAC attributes and functionalities in the device driver.
(i) the transmit power level,
(ii) the carrier sense threshold,
(iii) the data rate,
(iv) the receive signal strength index (RSSI), and
(v) the channel used to transmit a frame/upon which a frame is received, and
(vi) the time instant at which a frame is scheduled for transmission/receive.
Through an event subscription mechanism, higher-layer protocol modules can also receive timely update of channel status, without directly inserting callback functions in various places of the device driver.

32. Desirable Features
Flexibility: The event subscription mechanism is simple and elegant
allows multiple higher-layer protocol modules to (i) subscribe, and be alerted of, PHY/MAC events of interest; and (ii) register with the event subscription mechanism their callback functions, allowing adequate actions to be taken upon event occurrence.
allows the time granularity at which PHY/MAC properties are controlled to be on a per-packet or per-connection basis, or permanently (i.e., until the property is reset).

33. Our Plan for Next Year
Network Application
user
module
We will investigate other PHY/MAC attributes such as interfere-mitigated scheduling in the temporal/spatial domain and channel assignment/scheduling.
We will refine the transparent device driver and make open-source software release, so as to allow (in-)validation of theoretical results of cross-layer design and optimization.
Partially under the support of NSF/CRI, we will open up CUWiN as a city-wide research testbed.
kernel mode proxy
TCP
user
module IP
user
module
datalink interface
uniform extension
manager
extension interface 1
extension interface 3
extension interface 2
device driver

34. Questions and Comments?