Cognitive Wireless Network 15-744: Computer Networking L-19 Cognitive Wireless Networks Optimize wireless networks based context information Assigned reading White spaces Online Estimation of Interference (3 sections) DIRC: Optimizing Directional Antennas (2 sections) Optional reading Centaur: Hybrid optimization 2 Overview Background on cognitive wireless Building conflict graphs DIRC: directional antennas White space networks Cognitive Wireless Networks Performance of wireless network depends on deployment environment and use Try for any complex system, but especially for wireless given interactions with physical world Node density, mobility, physical infrastructure, traffic load, wireless technologies,.. Often not practical to hand tune the system Ok on campus: fairly predictable, strong control Home, hotspots, vehicular, industrial,.. 3 4 1
Generic Cognitive System Abstractions Modeling Analysis Sensing Measurement Scale/Efficiency Heterogeneity System Intent based Management Optimization Management Stability and manageability Actuation Control How about Cognitive Wireless? Sensing and measurement Limited by capabilities of wireless NICs Low resolution, lack of calibration Cost of exchanging measurements Modeling and analysis Abstraction of link, network Minimizing cost in building, maintaining model Basis for CS optimization, global management Cognitive Wireless continued Optimization and management Focus is often network capacity plus fairness, but can consider other factors Often uses heuristics (always exponential) Actuation and control Execute the plan devise above Minimize overhead in distributing instructions, coordination in execution Low resolution control, lack of calibration Cognitive Wireless Networking Very common for tuning of low level PHY and MAC parameters Optimization if local one or two nodes Example: rate adaptation More global optimizations, e.g. dealing directly with interference Use a conflict graph as the abstraction Dynamic spectrum access Adapt to presence of primary users 7 8 2
Rate Adaptation Goal Existing Approaches 54 Mbps 48 Mbps 36 Mbps 24 Mbps 18 Mbps 11 Mbps 5.5 Mbps 2 Mbps 1 Mbps Trial-and-error Probing Use recent packet loss history to pick rate Short history oversensitive to loss Long history slow to react to changes How to differentiate poor SINR from collision? SINR: Signal to Interference + Noise Ratio Need to know SINR at the receiver Use RTS/CTS to learn channel quality Overhead RTS/CTS rarely used in practice 9 RTS T Data CTS R What are the models used? 1 CHARM Channel-aware rate selection algorithm Transmitter passively determines SINR at receiver by leveraging channel reciprocity Determines SINR without the overhead of active probing (RTS/CTS) Select best transmission rate using rate table Table is updated (slowly) based on history Needed to accommodate diversity in hardware Jointly considers problem of transmit antenna selection SINR: Noise and Interference SINR = RSS Noise + Interference Noise Thermal background radiation Device inherent Dominated by low noise amplifier noise figure ~Constant Interference Mitigated by CSMA/CA Reported as noise by NIC 11 12 3
CHARM: Channel-aware Rate Selection Some Examples 11 Mbps 5.5 Mbps 2 Mbps 1 Mbps R SINR T Per-node History Time Leverage reciprocity to obtain path loss Compute path loss for each host On transmit: Predict path loss based on history Select rate & antenna Update rate thresholds More global optimizations requires more complex models Routing and opportunistic forwarding based on signal propagation conditions Maximizing spatial reused based on transmit power adaptation, directional antennas, carrier sense tuning, Packet scheduling as a replacement for carrier sense 14 15 Overview Background on cognitive wireless Building conflict graphs DIRC: directional antennas White space networks What is a Conflict Graph? A graph that summarizes the interference between wireless links Nodes are wireless links and edge indicate that the links interfere Many variants depending on the nature of the network and optimization Definition of interference, e.g. MAC vs PHY Directed versus undirected edges Annotations of the edges 16 17 4
Example Conflict Graph Constructing CGs Passive techniques Monitors to collect traces Interference inferencing algorithms E.g., Jigsaw, WiT Active techniques Send broadcast traffic in parallel Use Broadcast Interference Ratio (BIR) metric Measurements of bandwidth Typically off line 18 19 Tradeoffs Micro-Probing Architecture Active: better accuracy, captures weak interference Passive: low overhead, no downtime, no client mods Micro-probing: combine the best of both 2 21 5
Basics of Micro-Probing Testing for CS interference AP i initiates broadcast transmissions at some set of times AP j initiates transmission at same times with offset Use MAC service time (MST) estimates Testing for collision interference Initiate transmission between AP i and client AP j sends broadcast frame at same instant Example: Transmit Power Control Adjust transmit power for each transmission to maximize spatial reuse Maximize simultaneous transmission What model should we use? Circle model : each transmitter has a transmission and interference range Widely used in (bad) simulators SINR model: packet reception depends on SINR at the receiver 22 23 Transmit Power Control Take 1 Using SINR Take 2 Idea: reduce transmit range to minimum needed to reach destination so interference is minimized Right? Signal SINR = Noise + Interference 24 Decreasing transmit power reduces interference for other nodes but makes transmission more vulnerable to interference! Should increase transmit power to maximize chances of reception Right? But you cannot have it both ways? 6
Automatic Power Control AP 1 n 1 L 21 AP 2 L 11 L 22 L 12 n 2.2 Iterative,.72 Min Power,.62 Equal Power,.46.5 1 1.5 Link Throughput (Mbps) Any transmission creates interference on all other links Must consider pair-wise interference between all links Concurrent transmission is possible if SINR 1 +SINR 2 2*SINR threshold How use this observation for a distributed power control algorithm? CDF of Link Throughput 1.8.6.4 Improving Spatial Reuse Create pairwise conflict graph for links in the wireless network Edges defined using the SINR model Uses path loss very efficient Remove links by adjusting transmit power Based on equation of previous slide 27 Distributed Automatic Power Control Distribute algorithm by having each node operate on a local conflict graph Includes transmitters up to two hops away Direct observations + gossiping of signal strength Final step is to adjust the CCA threshold to allow simultaneous transmissions (messy) Implementation is very tricky Calibration of registers, limited control, deafness, mobility, etc. Overview Background on cognitive wireless Building conflict graphs DIRC: directional antennas Presentation Richard White space networks Potential Interferer Peer Me Peer Potential Interferer 28 29 7
Overview What are White Spaces? Background on cognitive wireless TV Wireless Mic ISM (Wi-Fi) Building conflict graphs DIRC: directional antennas Presentation Richard 54-9 17-216 47 7 24 25 518 53 MHz 5 TV Channels 6 White spaces Each channel is 6 MHz wide dbm 7 MHz White space networks Based on slides by authors 45 FCC Regulations* Sense TV stations and Mics 1 Recent ruling 47 allows MHz use Frequenc of databases 7 MHz Portable devices on channels y 21-51 White Spaces are Unoccupied TV Channels TV Stations in America 46 The Promise of White Spaces TV Wireless Mic ISM (Wi-Fi) Higher Frequency Broadcast TV Wi-Fi (ISM) 54- MHz 9 47 174-216 7 More Spectrum Up to 3x of 82.11g 24 25 518 53 7 MHz Longer Range at least 3-4x of Wi-Fi 47 48 8
Fraction of Spectrum Segments White Spaces Spectrum Availability.8.7.6.5.4.3.2.1 Urban Suburban Rural 6 >6 # Contiguous Channels Differences from ISM(Wi-Fi) Fragmentation Variable channel widths White Spaces Spectrum Availability TV Tower Differences from ISM(Wi-Fi) Fragmentation Variable channel widths Spatial Variation Cannot assume same channel free everywhere Each TV Channel is 6 MHz wide Spectrum is Use Fragmented multiple channels for more bandwidth Location impacts spectrum availability Spectrum exhibits spatial variation 49 5 White Spaces Spectrum Availability Channel Assignment in Wi-Fi Differences from ISM(Wi-Fi) Fragmentation Variable channel widths Spatial Variation Cannot assume same channel free everywhere Incumbents appear/disappear over time Temporal Variation Same Channel will not always be free Any connection can be disrupted any time Must reconfigure after disconnection 1 6 11 1 6 11 Fixed Width Channels Optimize which channel to use 51 52 9
Spectrum Assignment in WhiteFi Accounting for Spatial Variation Fragmentation Include Assign Spatial Variation Spectrum Assignment Problem Goal Maximize Throughput Spectrum at clients Center Channel & Width Optimize for both, center channel and width BS must use channel iff free at client = 53 54 Intuition Discovering a Base Station Intuition Use widest possible channel But Limited by most busy channel BS Carrier Sense Across All Channels Discovery Time = (B x W) All channels must be free ρ BS (2 and 3 are free) = ρ BS (2 is free) x ρ BS (3 is free) Tradeoff between wider channel widths and opportunity to transmit on each channel 55 Can How BS and we does optimize Clients the must new this use client discovery same discover channels time? channels used by the BS? Fragmentation Try different center channel and widths 56 1
SIFT, by example 1 5 MHz ADC SIFT SIFT Does not decode packets Pattern match in time domain Amplitude Data SIFS ACK Time 57 11