End-to-End Design and Evaluation of mmWave Cellular Networks - PowerPoint PPT Presentation

End-to-End Design and Evaluation of mmWave Cellular Networks Michele Polese Department of Information Engineering End-to-end mmWaves University of Padova, Italy polesemi@dei.unipd.it Supervisor: Prof. Michele Zorzi

Outline • Introduction • A case for end-to-end, full-stack evaluations • Architectures for 5G mmWaves End-to-end mmWaves • End-to-end protocols for mmWaves • Data-driven 5G network optimization • Conclusions and research directions 2

3GPP NR: novelties Frame – 10 ms Physical Resource Block 5G Core Network options Subframe – 1 ms N subcarriers NSA deployment SA deployment 4G EPC 5G Core Bandwidth 𝐶 − max 400 MHz per carrier Examples of slot numbers with AMF SMF PCF different subcarrier spacing PGW/SGW AMF SMF PCF Slot – 0.25 ms AMF SMF PCF Subcarrier MME UPF spacing 60 UPF kHz UPF Slot – 0.125 ms HSS Subcarrier Network slicing and NFV spacing 120 kHz End-to-end mmWaves Symbol – 8.9 μs Flexible frame structure NR LTE Multi RAT access mmWave directional communications NR 3

3GPP NR: timeline Goal: deployment by 2020 5G phase 1 5G phase 2 March 2020 Dec. 2017 June 2018 Non Stand-alone Stand-alone End-to-end mmWaves specifications specifications Release 15 Release 16 4

3GPP NR: mmWaves in cellular networks 3GPP NR Release 16 will support frequencies up to 52.6 GHz Z. Pi and F. Khan, "An introduction to millimeter-wave mobile End-to-end mmWaves broadband systems," in IEEE Communications Magazine , vol. 49, no. 6, pp. 101-107, June 2011. 5G 5G increases the decreases the datarate [Gbps] latency [ms] 5

3GPP NR: challenges for mmWaves Ground station End-to-end mmWaves UAV with mmWave radio blockage 6

End-to-end design and performance: why? • Sometimes, link-level is enough • Real networks, however, have several components in-between the user , the link and the content he/she needs APP Internet End-to-end mmWaves TCP/IP TCP/IP Core network APP RRC PDCP RLC MAC PHY Focus of my research end-to-end, system-level design & evaluation of 5G mmWave networks 7

The Architecture System Level Design of 5G mmWave Networks The Protocols End-to-End and Cross-Layer Analysis of 5G mmWave Networks End-to-end mmWaves The Intelligence Data-Driven 5G Networks Optimization 8

9 The tool: ns-3 mmWave module Packet Application and Application and network stack network stack 3GPP cellular stack 3GPP cellular stack Channel model End-to-end mmWaves Error model Propagation Interference SINR Fading Beamforming https://github.com/nyuwireless-unipd/ns3-mmwave https://github.com/signetlabdei/quic https://github.com/signetlabdei/ns3-mmwave-iab https://github.com/signetlabdei/mmwave-psc-scenarios

The Architecture System Level Design of 5G mmWave Networks The Protocols End-to-End and Cross- Layer Analysis of 5G mmWave Networks The Intelligence Data-Driven 5G Networks Optimization System Level Design of 5G mmWave Networks Multi connectivity, beam management and Integrated Access and The Architecture Backhaul

System-level challenges at mmWaves Issues: high propagation loss and blockage Ultra-dense deployments Large antenna arrays increase the link budget, but the power is focused on narrow beams 1 High number Provide backhaul to all the base stations 2 of handovers The Architecture Need to track the narrow beams when moving 3 11

System level solutions at mmWaves Beam management Seamless tracking 3 Integrated Access and 2 Backhaul The Architecture Low cost, high density mmWave deployments Multi-connectivity Low-latency, highly 1 reliable handovers 12

1 Mobility management • Goal: design a system resilient to fluctuations and outages • Contribution: Multi-connectivity architecture to combine sub-6 GHz and mmWave benefits The Architecture 13

Results: latency with TCP traffic Proposed solution The Architecture • No handover (always keep the same BS) • Single connectivity (traditional HO architecture) • Multi connectivity (fast handovers – no service interruption) 14

2 Integrated Access and Backhaul 3GPP Work Item for Release 16 • Goal: provide wireless backhaul to ultra-dense mmWave networks • Contributions: IAB module for ns-3 mmWave Analysis of IAB end-to-end performance Distributed path selection policies The Architecture Core Network Internet IAB-donor UE SDAP IAB-node IAB-node SDAP PDCP PDCP GTP-U GTP-U Adaptation Adaptation Adaptation Adaptation RLC RLC RLC RLC RLC RLC UDP UDP Uu F1* F1* MAC MAC MAC MAC MAC MAC IP IP PHY PHY PHY PHY CU-UP DU MT

End-to-end Performance for IAB Impact of synchronous vs. bursty traffic The Architecture Webpage loading time with Throughput with full buffer source browsing model 20

3 Beam management in 3GPP NR • Goal: perform directional initial access and tracking • Contributions: Study of 3GPP NR beam management schemes Analysis of their performance with design insights The Architecture 28 GHz omnidirectional range 28 GHz directional range 15

Beam Management in NR The 3GPP has specified a set of procedures for the control of multiple beams at mmWave frequencies which are categorized under the term BEAM MANAGEMENT 1. Beam sweeping Initial Access in a standalone deployment gNB UE SS Burst Beam sweep and 2. Beam measurement measurement UE decides which Beam is the best beam determination SS Blocks to get RACH The Architecture 3. Beam determination resources UE receives RACH resource allocation Beam reporting RACH preamble 4. Beam reporting 16

Accuracy-reactiveness tradeoff in NR Accuracy: what is the probability Reactiveness: how much time of receiving an SS block? does it take to perform IA? Number of antennas at gNB and UE gNB density The Architecture Number of SS blocks per burst 17

Beam management for UAVs Proposed location-based beam management for UAVs Experimental evaluation The Architecture

The Architecture System Level Design of 5G mmWave Networks The Protocols End-to-End and Cross- Layer Analysis of 5G mmWave Networks The Intelligence Data-Driven 5G Networks Optimization End-to-End and Cross-Layer Analysis of 5G mmWave Networks TCP issues in mmWave networks The Protocols

TCP issues in mmWave networks End-to-end data plane Host-to- host “abstract” view • Congestion and flow control APP • Retransmissions TCP IP How do these components interact? SDAP PDCP RLC Link view MAC Retransmissions • PHY Reordering • The Protocols Buffering • mmWave channel: volatile, highly variable capacity, large bandwidth 22

23 NLOS LOS After transition from LOS NLOS Large buffer 1 Bufferbloat High latency LOS time congestion RLC buffer window occupancy Small buffer NLOS 2 Buffer overflow The Protocols a) Low throughput b) LOS LOS time time a) DUPACK retx (CW/2) RLC buffer congestion b) RTO retx (CW=1) occupancy window 3 Slow ramp-up when back in LOS

Possible solutions • Goal: improve TCP end-to-end performance on mmWaves • Contributions: Edge deployments : a shorter control loop, to react faster CC algorithms : faster window ramp-up mechanisms Exploit multiple paths: mobility management or MP-TCP milliProxy: cross-layer approach to better control the TCP sending rate milliProxy instance milliProxy instance Flow window management module flow window Flow management module Buffer flow The Protocols Buffer ACK management module server ACK management module UE end-to-end flows UE server 24

25 milliProxy – a TCP proxy for mmWaves Reduce buffering latency + increase goodput ▪ Transparent to the end-to-end flow ▪ Installed in the gNB – or at the edge ▪ Cross-layer approach Per-UE data rate ▪ RLC buffer occupancy ▪ milliProxy instance milliProxy instance RTT estimation ▪ Flow window management module flow window management ▪ Modular Flow module Buffer flow Buffer Plug-in different ▪ ACK management module The Protocols server ACK management module flow control algorithms (inspired to [1]) UE end-to-end flows UE server [1] M. Casoni et al., “ Implementation and validation of TCP options and congestion control algorithms for ns- 3,” in Proc. WNS3, 2015

milliProxy – flow control ▪ Interaction with the TCP sender Advertised window (receiver’s ▪ TCP sending rate is min(CW,ARW) feedback sent on ACK packets) Congestion window (computed ▪ milliProxy modifies the ARW in the by the sender) ACKs, according to the flow control policy used Bandwidth-Delay ▪ Product (BDP) based ARW = BW*RTT More conservative ▪ ARW = The Protocols min([RTT*PHY rate ]-B, 0) 26

Results: scenario with many LOS/NLOS transitions Throughput Latency The Protocols Throughput gain w milliProxy Latency reduction w milliProxy 27

The Architecture System Level Design of 5G mmWave Networks The Protocols End-to-End and Cross- Layer Analysis of 5G mmWave Networks The Intelligence Data-Driven 5G Networks Optimization Data-Driven 5G Networks Optimization Machine Learning at the Edge The Intelligence