Millimeter Wave Wireless Networks: Potentials and Challenges Sundeep Rangan, NYU-Poly Joint work with Felipe Gomez-Cuba, Ted Rappaport, Elza Erkip March 11, 2015 Rutgers Colloquium 1
Outline Millimeter Wave: A New Frontier for Cellular Can it Work? Measurements in NYC NYU WIRELESS Relaying Revisited Other Projects and Future Work 2
MmWave: The New Frontier for Cellular Massive increase in bandwidth Near term opportunities in LMDS and E-Bands Up to 200x total over long-time Spatial degrees of freedom from large antenna arrays From Khan, Pi Millimeter Wave Mobile Broadband: Unleashing 3-300 GHz spectrum, 2011 Commercial 64 antenna element array 3
Millimeter Wave Before Cellular Jagadis Bose at Royal Institution, London 1898 Demonstration of 60 GHz transmission 60 GHz Wireless LAN 802.11ad Photo from WiLocity mmwave backhaul Image from http://mobilebackhaul.blog.com/ 4
Key Challenges for Mobile Cellular All transmissions are directional: Friis Law: PP rr PP tt = GG tt GG rr λλ 4ππππ 2 Path loss λλ 2 Can be overcome with beamforming: GG tt, GG rr λλ 2 But requires directional search, tracking to support mobility Shadowing Mortar, brick, concrete > 150 db Human body: Up to 35 db NLOS propagation relies on reflections and scattering 5
Millimeter Wave Cellular Vision Uday Mudoi, Electronic Design, 2012 http://www.miwaves.eu/ Small cells Directional transmissions Relaying / mesh topology 6
Outline Millimeter Wave: A New Frontier for Cellular Can it Work? Measurements in NYC NYU WIRELESS Relaying Revisited Other Projects and Future Work 7
NYC 28 and 73 GHz Measurements Focus on urban canyon environment Likely initial use case Mostly NLOS Worst-case setting Measurements mimic microcell type deployment: Rooftops 2-5 stories to street-level Distances up to 200m All images here from Rappaport s measurements: Azar et al, 28 GHz Propagation Measurements for Outdoor Cellular Communications Using Steerable Beam Antennas in New York City, ICC 2013 8
Isotropic Path Loss Models Standard linear path loss model PPPP dd = αα + ββ log dd + ξξ Measures total power Aggregate across all directions Separate LOS and NLOS models Akdeniz, Liu, Samimi, Sun, Rangan, Rappaport, Erkip JSAC 2014 9
Isotropic Path Loss Comparison 20 db loss Isotropic NLOS path loss measured in NYC ~ 20-25 db worse than 3GPP urban micro model for fc=2.5 GHz But beamforming will offset this loss. Bottom line: mmw has no effective increase in path loss 10
Hybrid LOS-NLOS-Outage Model mmw signals susceptible to severe shadowing. Not incorporated in standard 3GPP models New three state link model: LOS-NLOS-outage Form derived from random shape theory (Bai, Vaze, Heath 13) Outages significant only at d>150m Will help small cell by reducing interference 11
Angular Spread RX power at different angles Measured powers at different TX-RX angular pairs. Avg. of 2 clusters of paths detected More likely with time resolution Typical beamwidth in each cluster: ~10 deg in AoA RX ~ 7 deg in AoATX Typical 2-3 spatial DoFs 12
Simulations: SNR Distribution Simulation assumptions: 200m ISD 3-sector hex BS 20 / 30 dbm DL / UL power 8x8 antenna at BS 4x4 (28 GHz), 8x8 (73 GHz) at UE A new regime: High SNR on many links Much better than current macro-cellular Interference is non dominant 13
Comparison to Current LTE Initial results show significant gain over LTE Further gains with spatial mux, subband scheduling and wider bandwidths System antenna Duplex BW fc (GHz) Antenna Cell throughput (Mbps/cell) Cell edge rate (Mbps/user, 5%) DL UL DL UL mmw 1 GHz TDD 28 4x4 UE 8x8 enb 1514 1468 28.5 19.9 73 8x8 UE 8x8 enb 1435 1465 24.8 19.8 Current LTE 20+20 MHz FDD 2.5 (2x2 DL, 2x4 UL) 53.8 47.2 1.80 1.94 10 UEs per cell, ISD=200m, hex cell layout LTE capacity estimates from 36.814 ~ 25x gain ~ 10x gain 14
Outline Millimeter Wave: A New Frontier for Cellular Can it Work? Measurements in NYC NYU WIRELESS Relaying Revisited Other Projects and Future Work 15
NYU WIRELESS Exciting new center Ted Rappaport, founder Faculty across ECE, Courant and Med school Pioneering in mmwave for cellular First to demonstrate feasibility Measurements in NYC Research includes: Cellular design, capacity studies, prototyping Biological impact, medical applications Bringing technology to reality 16
Industrial Affiliates 14 industrial affiliates Vendors, carriers & test NYU WIRELESS is Leading Industry Leading contributions to industry Channel modeling groups FCC Notice of Inquiry Host: Brooklyn 5G Summit New entrants: SiBeam: Leader in mmwave RF Ics StraightPath: Nationwide holder of 38 GHz spectrum 17
Outline Millimeter Wave: A New Frontier for Cellular Can it Work? Measurements in NYC NYU WIRELESS Relaying Revisited Other Projects and Future Work 18
Multihop Relaying for mmwave Significant work in multi-hop transmissions for cellular Gains have been minimal Why? Current cellular systems are bandwidth-limited Millimeter wave may be different Overcome outage via macrodiversity Many degrees of freedom 19
Consider Two Possible Protocols ISH: Infrastructure single hop IMH: Infrastructure multi hop Direct transmissions to mobiles Similar to current cellular Multi-hop transmissions using UEs Rarely used today [Gomez-Cuba, Rangan, Erkip, Gonzalez-Castano, ISIT 2014] 20
Scaling Laws Analysis Randomly drop nn mobiles. Area, bandwidth, number of antennas scale with nn Estimate scaling on RR nn 1 lim log RR nn nn n Estimate rate to a constant factor. Assume EGoS Base stations have multiple antennas Mobiles have single antenna Param Bandwidth Num BS antenans Area Num BS Scaling WW = WW 0 nn ψψ l = l 0 nn γγ AA = AA 0 nn νν mm = mm 0 nn ββ 21
Protocols for Downlink Infrastructure Single Hop BSs transmit using to UEs directly using MU-MIMO Frequency reuse one. Treat out-of-cell interference as noise Infrastructure Multi Hop Cell is divided into subcells Traffic is routed from BS to mobiles via multi-hop MU-MIMO on first hop from BS. Partial frequency reuse and time-division for half-duplex constraint and interference Sub-cell size is optimized 22
Achievable Rates: IMH vs. ISH Gain with widebands and large num mobiles / BS. Rate per user Param Bandwidth Num BS antenans Area Scaling WW = WW 0 nn ψψ l = l 0 nn γγ AA = AA 0 nn νν Num BS mm = mm 0 nn ββ Fixed DoFs (current cellular) Increasing DoFs (e.g. mmw) Path loss CCCC αα Degs of freedom per BS (bandwidth * antenna) 23
Regimes Interference vs power limit regime: When DoFs per BS is large, rate scaling ceases to increase There is a limit to very large DoFs Multi-hop transmissions can better exploit high DoFs Shortens range reduces power limit But, requires large scaling on number of mobiles / base station 24
What is the Right Protocol for 5G? Area constant. Increase mobile density Technology Scaling Protocol Rate per user No action Densification BSs density fixed DoF per BS is fixed Increase BS per mobile DoF per BS is fixed ISH or IMH ISH or IMH Decreases with mobiles per cell Improves due to increased BS density Massive MIMO or mmwave 25 BS per mobile fixed Increase DoF per BS ISH IMH Increases until the SNR to furthest mobile hits threshold Increases until the SNR to first mobile hits threshold IMH is not needed for densification But, IMH is required for Massive MIMO or mmwave
Link Layer Capacity Assume point-to-point MIMO link has capacity scaling: Θ(WW min l rr, l tt ) RR = PP Θ l rr min(l rr,l tt ) Critical bandwidth NN II WW = Θ WW WW WW WW l rr PP WWNN II min(l rr, l tt ) Bandwidth limited Power limited Applies to coherent and non-coherent channel For non-coherent, number of channel parameters can grow linearly with bandwidth Assumes CSIT at transmitter Channel estimation and feedback overhead is a constant factor 26
Estimating the Network Capacity Assume path loss model PP rr = CCCC tt dd αα, αα > 2 Assume MU-MIMO can use full spatial DoFs Treat interference as noise Use worst case distances For IMH, use partial frequency reuse to separate neighbors in adjacent sub-cells (most a constant capacity loss) For IMH, optimize subcell size / number of hops Ensure every sub-cell has at least one mobile 27
Can We Do Better than IMH? IMH is capacity achieving. Use a cut-set argument Rate per user Fixed DoFs (current cellular) Increasing DoFs (e.g. mmw) Hierarchical cooperation not necessary Difference with infrastructure vs ad hoc IRH: Infra relay hop Use if IMH is not practical Multihop from fixed relays not mobiles 28
Making Multihop Work Many issues Interference-to-noise Network discovery Beamforming synchronization & tracking Directional isolation Dynamic duplexing 29 Qualcomm FlashLinQ frame structure
Significant Potential Gains Num RNs Capacity Cell edge DL UL DL UL 0 2090 1890 10.0 3.40 2 2370 2280 28.2 5.99 4 2440 2330 238.6 244.5 Significant gains possible 10 UEs per cell, 4x4 antenna array, single stream Esp. cell edge Cell capacity limited by single stream Dynamic duplexing Multi-hop optimal scheduling Garcia-Rois, Gomez-Cuba, Akdeniz, Gonzalez-Castano, Burguillo-Rial, Rangan, Lorenzo, IEEE TWC, in revision 30
Outline Millimeter Wave: A New Frontier for Cellular Can it Work? Measurements in NYC NYU WIRELESS Relaying Revisited Other Projects and Future Work 31
Channel Measurements Outage Current measurements: Outage from a single base station Next steps: Joint probability multiple base stations Changes over time, distance Combine with ray tracing? Needed to assess: Macro-diversity, handover 32
Channel Measurements Deployment Models Current measurements: Rooftop antenna Traditional microcell Wide coverage, but NLOS Next set of measurements: Street furniture Lower range, but greater LOS paths More likely deployment model 33
Low-Power Beamforming Architectures A/D present major power bottleneck Wide bandwidths, large number of antennas Current solutions use analog BF via phase shifters Hybrid BF Many research questions Low-rate fully digital architectures? Synchronization, channel estimation? Implications for multiple access, cell search? 34
Directional Cell Search and Adaptation Synchronization may be key limiting factor Rapid changes from blockage Obstructions by buildings, hands, Other challenges: Demands for very low latency Multi-flow connectivity Must discover many elements 35
Spatial Channel Estimation Estimating directions of arrival is essential for: Tracking, synchronization, Analog BF, can look in only direction at a time Spatial covariance via non-negative matrix completion min pp zz ii ww QQ>0 ii QQww ii + λλλλλλ QQ, QQ = EE HHHH ii Beamforming with a 8x8 array Multipath NLOS channel NYC measurements [Amir-Eliasi, Rangan, Rappaport 2015] 36
Cell Search with Low-Rate Digital Architectures Significant gains for low-rate digital architectures Enable searching in multiple directions Also beneficial for control messaging / multiple access Detection in initial cell search Digital BF vs. analog BF 12 db gain Barati, Hosseini, Rangan, SPAWC 2014 37
Other MAC Layer Work How to design LTE for mmw? Key concerns Modifications for directionality Intermittency in links Areas of focus for next year: Cell search, synchronization Dynamic duplexing Multi-flow, carrier aggregation Qualcomm FlashLinQ frame structure 38
Adaptive and MultiPath TCP Rapid changes in rate: Cells intermittently blocked. Current TCP is slow to adapt Gateway UE New research directions New stochastic TCP mechanisms Multi-path congestion control Tingting Lu, Subramanian, Panwar 39
Heterogeneous Networks Many aspects of heterogeneity 4G and 5G Indoor / outdoor Third party ISP vs Cellular operator Many questions: New spectrum license models? Load balancing? Pricing? 40
Prototyping This year: Demonstrated mmw LTE-like signal Next steps: Integrated NI platform with ns3 for current LTE Demonstrated at EuCNC, June 2014 Will modify for mmw 41
A Big Question What is the killer app for mmwave? What applications can drive huge amounts of data? Many applications for human interaction are limited. Video < 20 Mbps. Much lower on mobile devices. Will data be driven by machine to machine? Many users bursty vs. few users continuous? What will drive very low latency (e.g. ~1ms)? What will be the network delays? What is the partition of mobile vs. cloud? Where will data be located? What about power, form factor, cost? 42
Summary MmWave offers tremendous potential A once in a generation technological advance But, many new challenges New regime where degrees of freedom are plentiful Dominant challenge is synchronization & intermittency Capacity tied closely with front-end capabilities Cellular will be re-designed Many opportunities for research, commercialization & theory 43
Thanks Faculty: Ted Rappaport, Elza Erkip, Shiv Panwar, Pei Liu Michele Zorzi (U Padova) Postdoc: Marco Mezzavilla Students: Felipe Gomez Cuba (U Vigo) Mustafa Riza Akdeniz, Parisa Amir Eliasi, Russell Ford, Yuanpeng Liu, George McCartney, Oner Orhan, Matthew Samimi, Shu Sun 44
References Khan, Pi, Millimeter-wave Mobile Broadband (MMB): Unleashing 3-300GHz Spectrum, Feb 2011, http://www.ieee-wcnc.org/2011/tut/t1.pdf Rappaport et al. "Millimeter wave mobile communications for 5G cellular: It will work!." Access, IEEE 1 (2013): 335-349. Rangan, Rappaport, Erkip, Millimeter Wave Cellular Systems: Potentials and Challenges, Proc. IEEE, April 2014 Akdeniz, Liu, Rangan, Rappaport, Erkip, Millimeter Wave Channel Modeling and Cellular Capacity Evaluation, JSAC 2014 Gomez, Rangan, Erkip, Scaling Laws for Infrastructure Single and MultihopWireless Networks in Wideband Regimes, ISIT 2014, http://arxiv.org/abs/1404.7022 Barati, Hosseini, Rangan, et al, Directional Cell Search for Millimeter Wave Cellular Systems http://arxiv.org/abs/1404.5068 Eliasi, Rangan, and Rappaport. "Low-Rank Spatial Channel Estimation for Millimeter Wave Cellular Systems." http://arxiv.org/abs/1410.4831 45