A Simple and Effective Cross Layer Networking System for Mobile Ad Hoc Networks Wing Ho Yuen, Heung-no Lee and Timothy Andersen Outline Introduction to cross layer design Proposed channel model Rate adaptation scheme Routing metrics utilizing MAC info Simulation setup and results Why Cross-layer? Nature of wireless ad hoc networks Limited capacity, constrained energy, mobility … Application requirements Real time, mission critical … Current layered design paradigm is inflexible and sub-optimal for wireless networks Cross-layer design requires info exchanged across layers, thus allows protocols to adapt in a global manner, eventually achieves optimal network performance Cross Layer Example Channel Model Three signal strength attenuation factors are considered, namely, pass loss, shadowing and multipath fading For channel-adaptive protocols, A good time-varying channel model is needed for simulation A correlated shadowing channel model is proposed Correlated Shadowing Model Correlated Shadowing Model Shadowing attenuation Aj of a node j does not change until node j moves out of a disc of radius d from previously reference position Suppose node j moves out of the disc, new attenuation is A(n 1) i A(n) 1 i2W Suppose both node i and j move out of the disc A(n 1) ij A(n) 1 ij2 W ij exp( 2 / D0 ) Where i exp( / D0 ) W is a zero mean Gaussian random variable Rate Adaptation Scheme Modification on IEEE 802.11 protocol RTS,CTS and ACK packets sent at nominal rate SNR is estimated at when node receives RTS Transmission rate is mapped from the estimated SNR, and appended to the CTS The sender transmits data at the adapted rate Rate Adaptation Scheme An M-QAM scheme is used in which the constellation size changed with SNR Constellation size is decided by M ( ) K * * where K is a constant determined based on the power constraints, is SNR. Threshold rule is used, if M j M ( ) M j 1 , assign M j to NAV modification Routing Metrics Bandwidth awareness 1 Rij , Rij represents the rate of link between node i and j Interferences awareness Dij , Where Dij is the interval from the when the RTS packet is sent to when the data packet is received Congestion awareness Qi , where Qij is the queuing delay in the buffer of transmit node Interference awareness is implemented Implementation in DSR DSR route maintenance unmodified since only low mobility scenarios are considered Received SNR information are appended in RREQ, since RTS and CTS are not used when broadcasting RREQ, no rate adaptation is used in RREQ packets RREP packets are unicast packets to the source node using rate adaptation based on the SNR information along the route Source node compute the MAC delay of every RREP packets and choose the route with min delay Simulation Setup Ns-2 with wireless extensions by the Monarch Project, CMU Channel Model: Correlated shadowing, implemented in C++ MAC layer: 802.11b at 914MHz, 2MHz bandwidth Network layer: modified (dynamic source routing) DSR algorithm Transport layer: UDP agent Application layer: CBR application Simulation Environment 50 nodes Transmission range 250m Scenario size 1500X300m Channel model: Path loss model: 2 ray ground reflection model shadowing variance s=12 (severe shadowing) Correlated fading (slow fading at low mobility) Mobility model: Random waypoint model 2 values of node mobility s=0m/s and 1m/s corresponding to stationary and pedestrian scenarios Simulation Environment Each scenario has 20 flows (source destination pairs) Packet rate varies from 10 to 60 packet/s Each traffic flow starts at staggered time between 0s and 100s Performance metrics: throughput, delay, packet delivery ratio Three schemes are investigated: Plain DSR RA: rate adaptation IARA: interference aware rate adaptation Results: Stationary Scenario Results: Pedestrian Scenario Conclusions Spectrally efficient rate adaptation scheme leads to drastic improvement in throughput Simple routing metric incorporated to the DSR protocol, leading to modest decrease in packet delay