Transcript slides
Challenges in Enabling and Exploiting Opportunistic Spectrum MANETs An Industry Perspective NSF “Beyond Cognitive Radio” Workshop June 13-14, 2011 Ram Ramanathan Principal Scientist BBN Technologies Cambridge, MA Cognitive Radio at BBN • DARPA neXt Generation Communications (XG) [20032005] – Spectrum Agility: Cross-layer MANET protocols for dynamic spectrum access – Policy Agility: Framework for machine-readable policies and reasoning • DARPA Wireless Network after Next (WNaN) [2008 - ] – Large-scale MANET using low-cost cognitive radios – Multiple wideband (0.9 - 6 GHz) tunable transceivers per node – Opportunistic Spectrum Access (OSA): dynamic sensing and allocation, evacuation, reconstitution, policy conformance – Scalable routing, Disruption tolerant network, content-based… – 100 node mobile network demonstration last year 2 OSA Model • Primaries and Secondaries – Secondaries can use spectrum, conforming to interference policy • Spectrum overlay access – Sense and transmit on unoccupied frequencies to each of transceivers – Dynamic, distributed, frequency assignment algorithm • Evacuate immediately to another frequency if primary is sensed Primary on f1 7 7 2 1 6 2 1 4 4 6 2 1 5 Primary on f4 3 5 3 5 3 Enabling CR MANETs 4 Challenges in Sensing • Sheer number of channels (up to 1000) to sense – Detection/classification takes time – Need to evacuate quickly (200 ms) – Prioritization, multiple detectors etc. • Sensing with CSMA/CA – Need secondary MANET nodes to be “quiet” when sensing – No dedicated slot like in TDMA – Detector more sensitive, so large “quiet” radius • Collaborative sensing – Team formation, collaboration overhead – Linking nodes by fiber (static networks) 5 Allocation and Evacuation • Allocation and Use – Some frequencies are inherently “clearer” than others – Cross-channel interference cannot be ignored • choice of frequency to use depends upon frequencies assigned to other transceivers in same node and neighboring nodes – Wide range of frequencies => wide range in range! • Evacuating a frequency – Jammers vs Primaries – Coordinating evacuation, ripple effects • Unified view of primaries, interference, noise – MAC protocol that simply moves to the clearest frequency and transmits 6 Tools to help enable CR • Multiple Hypothesis testing – Dynamic probabilities of multiple hypothesis of primaries on each channel – Can be used with cooperative sensing as well • Artificial Intelligence in CR – Why: Human engineers cannot predict and design for all contingencies – Distributed planning and optimization to organize and coordinate – Machine learning to observe prior experience and modify behavior – Machine learning to rapidly update models – Scenario: Network trained in a desert can learn how to perform well under water – BBN work showing promise 7 Exploiting CR-enabled MANETs 8 E2E Throughput and Latency • End to end throughput in excess of per transceiver data rate via multiple frequency-orthogonal paths S D D – Challenge: Cross-layer solution for assignment and multipath route generation • Full duplex possible – Challenge: Picking non-interfering channels, isolation • Cut-through routing possible – Can reduce E2E latency dramatically – Challenge: New radio architecture Proc Network Proc Proc MAC/ Link Physical Proc Proc 1011111011101 Queueing Wait for access Backoff Retransmission Proc 1011011011 Packet Forwarding Queuing Packet 10101101111111 CUT-THRU using Orthogonal channels 9 Make your own topology • Sensing, allocation, etc. can be controlled by the network layer • Throw in power control, beamforming etc… a general architecture for controlling the topology with a global viewpoint Can do or Can do or ( and if or Can’t do ) Constraints Network Layer Get the topology that best matches the need Create Topology Is a controllable parameter Negotiate Routes 10 Summary • Several challenges in enabling a large scale real-world CR MANET.. – Sensing: scalability, “quieting” in CSMA, collaborative sensing – Allocation: channel interference, no control channel, evacuation – Multiple hypothesis and AI • … but perhaps more challenging is to go beyond and make sure we fully exploit what we get – True multipath routing – Full duplex, cut through – Topology by design • Acknowledgments: Chip Elliott, Karen Haigh, Jason Redi and others at BBN 11