Artificial Intelligence for 5G Wireless Networks�-Why, When and How?

Dr. Thyaga Nandagopal

Deputy Division Director

Division of Computing and Communication Foundations (CCF)

National Science Foundation

​

IEEE 5G World Forum Keynote September 10, 2020

The new buzz - AI

• Artificial Intelligence, and more specifically, Machine Learning
• Is this hype? Is this a game-changer?

​

• Yes and Yes!

​

​

AI in 5G - Why, When, How?

2

09/10/2020

Why is AI needed for 5G?

• 5G networks are getting bigger, complex and multi-domain
• Spectrum management alone is a time-consuming, data-intensive (and perhaps, mundane) task.

AI in 5G - Why, When, How?

3

09/10/2020

NITRD Workshop on Artificial Intelligence & Wireless Spectrum:

Opportunities and Challenges, August 2019

When is AI helpful?

• Tasks that are challenging/boring for humans to deal with
• Human actions might be either sub-optimal or time-consuming/expensive (or both)
• e.g., Chess/Go, Large-scale Image Recognition
• 5G Wireless Networks are inherently complex
• Tight Integration of Core, Edge and RF
• Virtualization leads to flexibility (and more tunable parameters)
• Spectrum, Site Management, Network Tenancy, NFV, Traffic Planning, App Deployment, Content management
• Complex issues that need cross-optimization

​

AI in 5G - Why, When, How?

4

09/10/2020

Potential Impediments for AI in 5G

• Inability to explain prevent failures in AI/ML models.
• Difficulty detecting spurious correlations in hidden data.
• Limiting understanding of human-AI system interactions.
• Stability
• Interpretability
• Incorporation of real-world conditions in training models.
• Leveraging human feedback into the multi-domain training/learning process.

​

AI in 5G - Why, When, How?

5

09/10/2020

How to get AI into 5G

• Practice, Practice, Practice
• We need “AI Gyms” in 5G Networks
• DARPA Spectrum Collaboration Competition (SC2) Challenge is an example
• Colosseum RF Emulator to repeatedly test various AI strategies under emulated conditions.
• More dimensions, more datasets, more models
• Testbed resources are key!

AI in 5G - Why, When, How?

6

09/10/2020

Principal gaps today

• Large, well-labeled, data sets
• Representative of real-world environments
• And associated corner-cases/complexities
• Reproducibility
• Leading to verifiability

​

AI in 5G - Why, When, How?

7

09/10/2020

NSF Platforms for Advanced Wireless Research

Three platforms funded:

• POWDER/RENEW: University of Utah – Available for experimenter use since Nov 2019
• COSMOS: Columbia University – Available for experimenter use since May 2020
• AERPAW: NCSU – Available by March 2021
• Fourth platform on rural broadband technologies to be announced Spring 2021.

Plus:

• Colosseum
• Belongs to the PAWR family, at Northeastern University – available from March 2020
• NSFCloud and FABRIC testbeds
• Provides Cloud Computing and internet-scale experimentation capabilities

AI in 5G - Why, When, How?

8

09/10/2020

AI relevant areas enabled by PAWR

AI in 5G - Why, When, How?

9

09/10/2020

mmWave R&D and systems testing at the millimeter-wave bands that are about 28GHz, 60GHz with a target of 100 Gbps in data rates for small-cell networks that cover a few city blocks.

​

Network Slicing to focus on the providing differential isolated Micro services to multiple users from RAN to Network slicing .

​

NFV MANO provide support for ETSI and other MANO implementations to orchestrate end-to-end VM, container, VNF deployment in a cloud native environment including radio resources that operate on the wireless edge.

​

Microservices Architecture assembling, controlling, and composing services. PAWR provides a service control plane that is layered on top of a diverse collection of back-end service implementations, including VM-hosted VNFs, container-based micro-services, and SDN-based control programs that embed functionality in white-box switches

Massive MIMO 2.5-2.7GHz and 3.5-3.7GHz 128 antenna element fully programmable radio to allow PHY/MAC/network FDD, full duplex research to design, build and demonstrate high bandwidth connectivity to multiple users simultaneously.

​

​

RAN CU-DU Split to advance capabilities of baseband-RRH and other functional splits being debated n different communities e.g. eCPRI, OTN backhaul, O-RAN.

​

​

Applications/Services – Smart and Connected Community networks that demonstrate potential applications/services including Cyber-Physical Systems, Cyber-Security, Internet of Things, Robotics, Smart and Connected Health, and Big Data.

Summary

• AI can enable rapid advances for network policy, design, operations and management
• AI needs extensive training for emerging 5G networks
• Need to start now
• NSF PAWR platforms are a unique resource
• Can enable AI solutions for modern networks
• Open to all
• Wireless networks as well as wired
• Open, unbiased, reproducible data collection and datasets
• Unleash them in the controlled ‘wild’ – enabled by PAWR

AI in 5G - Why, When, How?

10

09/10/2020

NSF PAWR TESTBED HIGHLIGHTS

Backup Slides

AI in 5G - Why, When, How?

11

09/10/2020

POWDER/RENEW

AI in 5G - Why, When, How?

12

09/10/2020

POWDER: Platform for Open Wireless Data-driven Experimental Research

RENEW: A Reconfigurable Eco-system for Next-generation End-to-end Wireless

• RENEW Massive MIMO base station
• End-to-End Programmable
• Diverse Spectrum Access 50 MHz-3.8GHz
• Hybrid Edge computer composed of FPGA and GPU/CPU-based processing,
• Hub Board aggregates/distributes streams of radio samples

• Next Generation Wireless Architecture
• Dynamic Spectrum Sharing
• Distinct environments: a dense urban downtown and a hilly campus environment.

Control Framework with Hardware + Software Building Blocks

IRIS software-defined radio modules

Architectural view of RENEW base station

Deployment Area: UofU Campus +Downtown SLC + Connected Corridor

POWDER/RENEW: Open for experimenters

AI in 5G - Why, When, How?

13

09/10/2020

8 Rooftop Base station and Fixed End Point sites

• Available since Nov. 2019

​

Software Profiles Available:

- Open Air Interface

- Worked with ONF to provide basic XRAN functionality in OAI

- Open Network Automation Platform (ONAP) [LF]

- Converged Multi-Access and Core (COMAC)/Open Mobile Evolved Core (OMEC) [ONF]

- Akraino Edge Stack, Radio Edge Control (REC)

- RAN Intelligent Controller (RIC)

- O-RAN [O-RAN Alliance]

COSMOS: Cloud-Enhanced, Open, Software-Defined Mobile Testbed for City-Scale Deployment

AI in 5G - Why, When, How?

14

09/10/2020

Deployment Area: West Manhattan/Harlem

• A multi-layered computing system with an RF thin client; flexible signal processing; network function virtualization (NFV) between a local SDR (with FPGA assist) and a remote cloud radio access network (CRAN) with massive CPU/GPU and FPGA assist
• Deployed in New York City, one of the country’s most populated urban centers
• Wideband radio signal processing (with bandwidths of ~500 MHz or more)
• Support for mmWave communication (28 and 60 GHz)
• Optical switching technology (~1µs) provides passive WDM switch fabrics and
• radio over fiber interfaces for ultra-low latency connections

​

28GHz phased-array ICs and phased-array antenna modules (PAAM)

COSMOS Radio Site Design All-Optical Network Design

COSMOS: Available today

Base Configuration

• 2 Large and 3 Medium Nodes
• 16 port Space Switch
• ROADMs: 1 fiber pair each, 2 total
• Direct CRF connections: 6 fiber pairs
• Ethernet switch: 2 fiber pairs

AI in 5G - Why, When, How?

15

09/10/2020

AERPAW: Aerial Experimentation and �Research Platform for Advanced Wireless

AI in 5G - Why, When, How?

16

09/10/2020

Goals

• Accelerate the integration of unmanned aerial systems (UAS) into the national air-space
• Enable new advanced wireless features for UAS platforms, including flying base stations for hot spot wireless connectivity

​

Focus areas

• Advanced wireless communication technologies that enable beyond-VLOS and autonomous UAS operations and three-dimensional mobility for UAS
• New use cases for advanced wireless technologies for UAS

​

Tactics

• Create a one-of-a-kind aerial wireless experimentation platform and a proving ground and technological enabler for emerging innovations, including package delivery platforms and urban air mobility
• Accelerate development, verification, and testing of transformative advances and breakthroughs in telecommunications, transportation, infrastructure monitoring, agriculture, and public safety

AERPAW at a glance

​

• Led by North Carolina State University (NCSU) with three other universities
• Start date 9/01/2019
• Approximately 20 fixed nodes at 3 main sites in the RDU Triangle area
• 20+ unmanned autonomous vehicles (drones) with advanced wireless tech through the coverage area

​

AERPAW: Deployment Zone

AI in 5G - Why, When, How?

17

09/10/2020

Colosseum

AI in 5G - Why, When, How?

18

09/10/2020

Colosseum is the world’s largest wireless network emulator with granularity at the RF signal level

• 256 x 256 100 MHz RF channel emulation
• 128 Programmable Radio Nodes
• Computing resources (CPU, GPU, FPGA)
• Access control and scheduling infrastructure
• Supports remote shared access
• Colosseum is a General Purpose Cooperative Radio Development and Testing Environment
• https://www.darpa.mil/program/spectrum-collaboration-challenge

Envisioned experiment lifecycle �for future wireless research

AI in 5G - Why, When, How?

19

09/10/2020

Find out more

• PAWR Project Office: http://advancedwireless.org
• POWDER/RENEW:
• http://powderwireless.net
• http://renew.rice.edu
• COSMOS: http://cosmos-lab.org
• AERPAW: http://aerpaw.org
• Colosseum: https://www.northeastern.edu/colosseum

​

AI in 5G - Why, When, How?

20

09/10/2020