Towards Multi-Hop Free-Space-Optical (FSO) Mesh Networks and MANETs: Low-Cost Building Blocks <…or how to communicate w/ your laser pointer …> Shiv Kalyanaraman shivkuma@ecse.rpi.edu : “shiv rpi” Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 1 : “shiv rpi” Students and Collaborators Jayasri Akella (PhD) Murat Yuksel (post-doc, now at Univ. Nevada, Reno) Bow-Nan Cheng (PhD) David Partyka (MS) Chang Liu (MS) Prof. Partha Dutta (optoelectronic devices) Prof. Mona Hella (RF/photonic circuits) Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 2 : “shiv rpi” Scope of Talk Understanding and overcoming limitations of FSO Orthogonal Geographic Routing Auto Node Localization Rendezvous Routing Network Error correction to Improve multi-hop link performance Use of directionality concept in the network layer: routing and localization PHY schemes Data-link Error Correction Schemes 2-D Multiple Node Localization Line -Of3D-LOS Element Alignment Multiple Element Antennas FSO Antennas Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 3 : “shiv rpi” Free Space Optical (FSO) Communications Open spectrum: 2.4GHz, 5.8GHz, 60GHz, > 300 GHz Lots of open spectrum up in the optical regime! Data transfer through atmosphere OOK Modulated light pulses. Line of sight “optical wireless” technology. Visible to near infrared regions. Currently terrestrial point-to-point links bridging connectivity gaps between buildings in a metro area medical imaging disaster recovery DoD use of FSO: Satellite communications DARPA ORCL project: air-to-ground, air-to-air, air-to-satellite 802.11a/g, 802.16e, Cellular (2G/3G) Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 4 : “shiv rpi” FSO vs RF: Directional Antenna Sizes: 2.4 Ghz, 5.8 Ghz 2.4 Ghz 802.11b Pringles Can antennas Dual Band 802.11a/b/g Directional antennas 5.8 Ghz 802.11a Directional antennas Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 5 : “shiv rpi” FSO Trans-receivers: Much Smaller! 2-d Array of LEDs Transreceivers: LED +PD (packed on a 3d sphere) Higher frequency: smaller antennas Small size => Can pack in 2-d array and 3-d structures ! Increasing use of HBLEDs in solid state lighting: can leverage low cost devices. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 6 : “shiv rpi” Elementary FSO: sending multi-channel music Audio Mixing: Tabletop laboratory systems used for propagating music via multiple channels through free space Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 7 : “shiv rpi” Why Free Space Optical Communication? FSO potential: Multi-Gbps System capacity Spatial re-use/minimal interference Suitable form factors (power, size and cost) Quick and easy installation. If interference-limited, then attractive for the last mile access or home networking where LOS exists. If power-limited, then attractive for sensor networks: much lower-power vs RF Challenges: FSO Needs line-of-sight (LOS) alignment Poor performance in adverse weather conditions: reliability How to seamlessly integrate and leverage FSO in the context of multi-hop networks? From LightPointe Optical Wireless Inc. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 8 : “shiv rpi” Apps: Opportunistic Links & Networks Opportunistic links Air-to-air or air-satellite Flying over oceans… Opportunistic links to cell towers. Expensive sat-com links for most urgent data, and delay-tolerant links to offload delay-tolerant data: DARPA ORCL program is already looking at some of this Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 9 : “shiv rpi” FSO Advantages High-brightness LEDs (HBLEDs) and VCSELs are very low cost and highly reliable components 35-65 cents a piece, and $2-$5 per transceiver package + up to 10 years lifetime Amenable to high density integration (eg: VCSEL arrays) Very low power consumption 4-5 orders of magnitude improvement in energy/bit compared to RF, e.g. 100 microwatts for 10-100 Mbps. Huge spatial reuse => multiple parallel channels for huge bandwidth increases due to spectral efficiency Not interference limited, unlike RF More Secure: Highly directional + small size & weight => low probability of interception (LPI) Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 10 : “shiv rpi” FSO Issues/Disadvantages Limited range (no waveguide, unlike fiber optics) Need line-of-sight (LOS) Any obstruction or poor weather (fog, sandstorms, heavy rain/snow) can increase BER in a bursty manner Bigger issue: Need tight LOS alignment over long distances: Directional antenna on steroids! LOS alignment must be changed/maintained with mobility or sway! Received power Spatial profile: ~ Gaussian drop off ~1km Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 11 : “shiv rpi” Geometric Attenuation due to Beam Spread • Divergence of light beam is primary cause for geometric attenuation. • When an energy detector is used, only a fraction of transmitted power is received. θ SAT SAR Receiver Source Laser I (Y ) I 0e 2Y 2 R ( Z )2 LED I ( ) I 0 cosn ( ) Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 12 : “shiv rpi” Typical FSO Communication System Receiver (Photo Diode/ Transistor) Transmitter (Laser/VCSEL/LED) Digital Data ON-OFF Keyed Light Pulses Light beam is “directional” (-) Line-of-sight is always needed between the transceivers. (+) Spatial re-use, diversity, and neighbor position estimation. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 13 : “shiv rpi” Elementary FSO System: Block Diagram 4 1 2 3 5 1. 2. 3. 4. 5. 6. 6 LED Module Collimating Lens External Magneto-Optic Modulator Pulsed Light Focusing Lens Detector Unit Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 14 : “shiv rpi” Link Design Issues 4 1 2 LEDs 5 3 Attenuation 6 Photodetector Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 15 : “shiv rpi” • Output Optical Power P LEDs 1.24 • P — Output Optical Power • — wavelength • I — Input Electrical Current I Output Optical Power is dependent upon the choice of wavelength. Longer wavelengths are also more safer to humans, but roomtemperature devices don’t exist. • Output Optical Spectral Width Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 16 : “shiv rpi” Photodetector Responsivity Responsivity is dependent upon the choice of wavelength Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 17 : “shiv rpi” Atmospheric Windows Future devices 1.55um: today’s devices Optical Loss is dependent upon the choice of wavelength. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 18 : “shiv rpi” Error Probability over Single Hop Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 19 : “shiv rpi” Link Budget PRC = PTX –Llens– LGS – Latt • PRC — Output Optical Power in transmitter • PTX — Received Optical Power in receiver • Llens — Optical Loss Due to Lens Used in transmitter and receiver • LGS — Optical Loss Due to Geometrical Spreading in the propagation distance • Latt — Optical Loss Due to attenuation in atmosphere Bottom Line: Trying to Achieve Greater Distance and Reliability With a Single FSO Hop is Tough! Change the game: Use shorter hops, multi-hops, low-cost BBs, and engineer reliability by using diversity at higher layers Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 20 : “shiv rpi” 3d & 2d Designs: Alignment & Capacity LOS 3-d Spheres: LOS detection through the use of 3-d spherical FSO Antennas Node 1 Node 2 D D/N … Node 1 Repeater 2 Repeater 1 Node 2 Repeater N-1 2d Array: 1cm2 LED/PIN => 1000 pairs in 1ft x 1ft square structure MultiGbps capacity possible, with different color LEDs (simple static WDM). Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 21 : “shiv rpi” 3-d Spheres for Auto-Alignment LED Micro Mirror PhotoDetector Spherical Antenna Initial 3-d FSO prototypes with autoalignment circuitry Optical Transmitter/Receiver Unit Design of 3-d FSO antennas: Honeycomb (tesselated) arrays of transceivers Auto-alignment Process: Step 1: Search Phase (pilot pulses) Step 2: Data Transfer Phase Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 22 : “shiv rpi” 3d-Sphere Auto-Alignment Circuit (cont’d) E.g.: 4-circuit block diagram Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 23 : “shiv rpi” 3d Spheres: Mobility Tests Misaligned Aligned Prior work obtained mobility in FSO for indoor using diffuse optics technology: [Barry, J.R; AlGhamdi, A.G.] Limited power of a single source that is being diffused into all the directions. Suitable for small distances (typically 10s of meters), but not suitable for longer distances. Our approach can scale to longer, outdoor distances and consumes less power. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 24 : “shiv rpi” 3d Spheres: Mobility Contd Received Light Intensity from the moving train. Light Intensity (lux) 70 Aligned 60 Not aligned 50 40 30 Detector Threshold 20 10 128 121 112 105 97.5 88.5 79 72 65 51.5 40.5 33 23 17 11 0 0 Angular Position of the Train (degree) • Denser packing will allow fewer interruptions (and smaller buffering), but more handoffs… • Even w/ buffering: becomes a “disruption”-tolerant/lossy networking problem over multiple hops. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 25 : “shiv rpi” Toy Train Experiment Contd. tA 2 D 2 tA : Time duration of alignment θ: Divergence angle of LED. D: Circuit delay Ω: Train's angular speed φ: Angular separation between transceivers on sphere. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 26 : “shiv rpi” FSO Node Designs Important node design questions: How good the node can be in terms of coverage or range? How many transceivers can/should be placed on the nodes? Do the placement patterns of transceivers matter? Goal: maximize capacity Various factors: Visibility – weather conditions source power and receiver sensitivity angles of devices – small angles are costlier packaging density Goal: maximize coverage Tradeoff: interference vs. angles vs. Tradeoff: interference vs. angles vs. packaging density packaging density Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 27 : “shiv rpi” 2-D Arrays: Increased Capacity Consider transmission from transceiver T0 on array A (TA0) to transceiver T0 on array B (TB0). The cone not only covers intended receiver TB0 , but also TB1 , TB2 , TB4 , TB7 . Parameters: d: distance between arrays θ: divergence angle ρ: Package density r d tan Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 28 : “shiv rpi” Array Designs : Helical Vs Uniform Transceiver Placement Helical array design gives more capacity for a given range and transceiver parameters due to reduced inter-channel interference. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 29 : “shiv rpi” Inter-channel Interference & Capacity w/ OOK Interference occurs when a subset of these potential interferers transmit when TA0 is transmitting. Probability that such an event occurs gives error probability due to crosstalk. pe (1 r YSep YSep2 j 2 ( j 1) 2 0 p ) p0 j where p0 is probability(ZERO transmitted). BAC capacity: C max H ( p0 . pe ) p0 H ( pe ) p( x) 0 1-pe pe X 1 1 Rensselaer Polytechnic Institute 0 Y 1 Shivkumar Kalyanaraman 30 : “shiv rpi” Uniform Array layout: Uncoded, Per-Channel capacity drops quickly with Package density Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 31 : “shiv rpi” Helical Array layout: Channel capacity drops slowly with Package density Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 32 : “shiv rpi” OOC (Optical Orthogonal Codes) can further improve the capacity between arrays. Two OOCs with weight 4 and length 32. Each transceiver uses a unique code similar to CDMA wireless users in a cell. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 33 : “shiv rpi” FSO Arrays and Space-Time Diversity Link 1 Link 3 Link 2 Link 4 Per-Link: Code over Time and Across Multiple Spatial Channels Per-Hop Per-Path Across a network: Build a virtual link composed of several FSO hops, and possibly perform FEC coding and mapping across multiple routed-paths. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 34 : “shiv rpi” Multi-hop Channel Model For small errors Pe <10e-2 , the channel is approximated as: N 1 P 1 ie i 1 1 N P i 1 0 N P ie i 1 ie N 1 P ie i 1 0 Visibility is modeled as a two Gaussians for clear and adverse weather. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 35 : “shiv rpi” Bit Error Rate versus Number of Hops Assume fixed e2e range that is split up into hops (2.5km) most gains with a few hops (~500m/hop) Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 36 : “shiv rpi” BER distribution Multi-Hop Error Distribution: more concentrated Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 37 : “shiv rpi” Multi-Hop Offers Robustness to Weather Multi-hop significantly outperforms single hop Number of Hops Mean BER Clear Weather Mean BER Variance Variance Adverse Weather Clear Weather Adverse Weather 1 1.5e-3 0.27 0.02 0.1176 5 9e-27 0.005 8e-50 0.0045 Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 38 : “shiv rpi” Using Multi-directional Communications @ Layer 3 Tessellated FSO Transceivers Multi-directional Antennas Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 39 : “shiv rpi” FSO-Meshes: Localization RF triangulation: needs THREE neighbors Granular tessellation allows accurate detection of angle of arrival. FSO localization: needs ONE neighbor FSO-based localization system with granular tessellation of transceivers Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 40 : “shiv rpi” FSO Localization Problem (x6, y6) (x7, y7) (x11, y11) (x10, y10) (x9, y9) (x4, y4) (x5, y5) FLA (x8,, y8) (0,0) (x3, y3) (x2, y2) After localization Shivkumar Kalyanaraman Before localization Rensselaer Polytechnic Institute 41 : “shiv rpi” FSO-Meshes: Orthogonal Rendezvous Routing Rendezvous point The source and destination sends probe packets at North-South and East-West directions based on their local sense of direction. Orthogonal/Directional Routing using FSO nodes Essentially choosing random orthogonal directions in the plane for dissemination and discovery. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 42 : “shiv rpi” ORRP vs Geo-Routing Classification of Research Issues in Position-based Schemes L3: Geographic Routing using Node IDs (eg. GPSR, TBF etc.) L2: ID to Location Mapping (eg. HDT, GLS etc.) ORRP N/A L1: Node Localization Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 43 : “shiv rpi” Void Navigation & Deviation Correction Basic Example VOID Navigation/Sparse Networks Example min(+4t, 4t min(+4t, 4t = g + 4t = g + 4t m = +2 m = +3 min(+4t, 6t min(+4t, 6t = g + 4t = g + 4t m = +2 m = +3 Void S min(+4t, 0 =g+ m=0 R min(+4t, 0 =g+ m = +3 Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 44 : “shiv rpi” ORRP: Reachability Analysis P{unreachable} = P{intersections not in rectangle} 4 Possible Intersection Points Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 45 : “shiv rpi” Path Stretch Analysis Average Stretch for various topologies • • • • Square Topology – 1.255 Circular Topology – 1.15 25 X 4 Rectangular – 3.24 Expected Stretch – 1.125 Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 46 : “shiv rpi” State Complexity Analysis GPSR DSDV XYLS ORRP Node State O(1) O(n2) O(n3/2) O(n3/2) Reachability High High 100% High (99%) Name Resolution O(n log n) O(1) O(1) O(1) Invariants Geography None Global Comp. Local Comp. Notes: • ORRP scales with Order N3/2 • ORRP states are fairly evenly distributed – no single pt of failure Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 47 : “shiv rpi” Summary FSO has interesting/complementary properties w.r.t. RF wireless Single Hop Issues: LEDs, PDs, Transmittance Windows Building Blocks: 3-d Sphere: LOS Auto-alignment, Coverage 2-d Array: Capacity, Co-channel interference due to geometric spread Helical Designs and Orthogonal Coding mitigates interference Low-cost Multi-hop FSO Networks: Simple OEO Repeaters, Error correction at electronic hops Use of directional PHY property at higher layers: Localization Routing: orthogonal rendezvous routing Low stretch, high connectivity, O(N1.5) state complexity Future work on multi-path routing, Wifi backup, coded-multiple parallel channels, WDM for capacity etc Dual-mode systems for opportunistic V2V links (vehicular ad-hoc) Extensions of our PHY and L3 mechanisms for higher mobility. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 48 : “shiv rpi” Thanks ! Papers, PPTs, Audio talks: : “shiv rpi” Ps: downloadable VIDEOS of all my networking courses available freely at the above web site Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 49 : “shiv rpi” Reliability through Diversity at Higher Layers Channel Performance Diversity Modes Continuous: Time, Frequency, Space ... Discrete: Code, Antenna, Paths, Routes … Standard technique: code across diversity modes and use degrees of freedom efficiently Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 50 : “shiv rpi” Erasure Coding >= K of N received RS(N,K) Recover K data packets! FEC (N-K) Block Size (N) Lossy Network Data = K Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 51 : “shiv rpi” HARQ Window (eg: window = 2 original data pkts) Packets Fragments Per-packet Data fragments FEC fragments Random-linear coded (RLC) FEC Fragments, coded across subsets of data/fec fragments in window Interleaved RLC Fragments Opportunistic mapping FSO sub-channel mm-wave RF subchannel Lossy, variable bit-rate sub-channels Fragments suffer bursty loss: data, FEC and RLC fragments lost Pkt 1 recovered w/o RLC (5 fragments Rensselaer Polytechnic Institute received) Pkt 2 needs RLC (only 4 fragments received). But, 5 RLC fragments, with pkt 1 fragments can recover these 4 missing fragments Shivkumar Kalyanaraman 52 : “shiv rpi” Hybrid FSO/RF-Mesh and MANETS Vision RF Communications High reliability Free-Space-Optical Communications High bandwidth Low power Directional – secure, Not i/f limited •802.1x with omni-directional RF antennas •High-power, Interference limited •Low bandwidth – typically the bottleneck link on a path •Error-prone, Disruptions •Less secure – very vulnerable to interception Mobile Ad-Hoc Networking Mobile communication Auto-configuration Hybrid Free-SpaceOptical/RF Mobile Ad-Hoc Networks Spatial reuse and angular diversity in nodes Electronic auto-alignment (auto-configuration) Optical auto-configuration (switching, routing) Low-power and highly secure Interdisciplinary, cross-layer design Bringing optical communications and RF adhoc networking together… Legacy RF MANETS Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 53 : “shiv rpi” 3-d Sphere Node Design Parameters Maximum possible range R tanθ R tanθ R R Half lobe area θ θ Interference area Transceiver ρ Not covered area τ r φ Case 1: No overlap, C=L Case 2: Overlap, C=L-I Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 54 : “shiv rpi” Sphere: Analysis Figures Lollipop design! Max communication range (m) for optimal node designs given P = 32mWatts, = 170.1mRad. Installed to ceilings, may be as lamps.. Reasonable coverage possible: For P=32mWatts, coverage as high as: 0.7 km2 (adverse) 2.10km2 (normal) 3.24km2 (clear) ~500m practical with cheap LEDs On top of towers.. Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 55 : “shiv rpi”