WINLAB

Software-Defined Radio Access

WINLAB Fall 2015 Research Review

Ivan Seskar, Associate DirectorWINLAB

Rutgers, The State University of New JerseyContact: seskar (at) winlab (dot) rutgers (dot) edu

5G Wireless: Technical Challenges

Faster Cellular

Radios Access

~1-10 Gbps

~1000x capacity

Low-Latency/

Low-Power

Access Network

For Real-Time IoT

New

Spectrum

&

Dynamic

Spectrum

Access

Next-Gen

Mobile

Network

Wideband PHY

Cloud RAN arch

Massive MIMO

mmWave (60 Ghz)

Multi-Radio access

HetNet (+WiFi, etc.)

…

Custom PHY for IoT

New MAC protocols

RAN redesign

Light-weight control

Control/data

separation

Network protocol

redesign

….

60 Ghz & other new

bands

New unlicensed/shared

spectrum

Dynamic spectrum

access

Spectrum sharing

techniques

Non-contiguous

spectrum

Network/DB coordination

methods

….

Mobile network redesign

Convergence with Internet

Clean-slate Mobile

Internet

Software Defined

Networks

Open wireless network

APIs

Cloud services &

computing

Edge cloud/fog computing

Virtualization, NFV

….

WINLAB

Related Collaboration Projects

FLEX (FIRE LTE testbeds for open EXperimentation)

WiSHFUL(Wireless Software and Hardware platforms for Flexible and Unified radio and network controL)

GENI(Global Environment for

Network Innovation)

CREW(Cognitive Radio

Experimentation World)

METIS-II (Mobile and wireless communications Enablers for the Twenty-twenty Information Society)

JUNO(Virtual Mobile Cloud Network for Realizing Scalable, Real-Time Cyber Physical Systems)

WINLAB

• Wireless Software and Hardware platforms for Flexible and Unified radio and network controL

• Call: H2020-ICT-2014-1• Topic: ICT-11-2014 (FIRE+)• Type of action: RIA• Budget: 5.171 M€• Duration: 36 M• Partners

WiSHF L

Courtesy: WiSHFUL Consortium

WINLAB

– many radio devices, each of them with specific HW and SW platform

– many implementations of protocol stacks

– non-unified, limited or no control of radio and network

Current Software Architecture

APP

NET

MAC

PHY

TRANSPORT

SDRCustom

Courtesy: WiSHFUL Consortium

WINLAB

WiSHFUL software architecture

Local Control

Program

Local Monitoring & Configuration

Engine

APP

NET

MAC

PHY

TRANSPORT

Global Control Program

Global Monitoring & Configuration Engine

UPIG

UPIN

UPIR

Local Control

Program

Local Monitoring & Configuration

Engine

APP

NET

MAC

PHY

TRANSPORT

UPIN

UPIR

Remote UPI usage

UPIHC UPIHC

Radio platforms

…SDR Custom

Device specific (HW & SW platform), implemented by WiSHFUL

Device independent (within device class), implemented by experimenter

Courtesy: WiSHFUL Consortium

WINLAB

FLEX: FIRE LTE testbeds for open experimentation

Open and highly configurable LTE platforms

Interaction of the user with the real 4G/5G world.

Commercial and Open Source equipment

Courtesy: FLEX Consortium

WINLAB

Open Air Interface (OAI)• OpenAirInterface.org today and Ecosystem

• Open-source for 5G

• Software Alliance– Membership

– License

– Strategic member areas

Hardware Platforms Software Platforms Radio in a PC

Courtesy: Navid Nikaein, Eurecom/Open Air Interface

WINLAB

OAI Software platform

– Commercial UE OAI eNB + Commercial EPC *– Commercial UE OAI eNB + OAI EPC *– Commercial UE Commercial eNB + OAI EPC *– OAI UE Commercial eNB + OAI EPC *– OAI UE Commercial eNB + Commercial EPC *– OAI UE OAI eNB + Commercial EPC– OAI UE OAI eNB + OAI EPC

Courtesy: Navid Nikaein, Eurecom/Open Air Interface

WINLAB

OAI Soft LTE eNB and UE• Challenge : efficient base band unit• OpenAirInterface uses general-purpose x86

processors (GPP) for base-band processing– front-end, channel decoding, phy procedures, L2

protocols

• Key elements– Real-time extensions to Linux OS

• x86-64 multicore arch

– Real-time data acquisition to PC– SIMD optimized integer DSP

• 64-bit MMX 128-bit SSE2/3/4 256-bit AVX2• iFFT/FFT, Channel Estimation, Turbo Decoding

– SMP Parallelism • Master-worker model ©www.openairinterface.org

Courtesy: Navid Nikaein, Eurecom/Open Air Interface

WINLAB

OAI Strategic Vectors

5G Modem

Software-defined 5G system

Heterogeneous 5G Network

Large-Scale Emulation

Test and measurements

RF Platform

Massive MIMO

Full-duplex Radionew waveform

SDN/NFV

Cloud-native RAN Juju/OpenStack

Ethernet Fronthaul

MEC API

Ultra-dense network

Coexistence and Aggregation

Unlicensed bands

Relaying

Carrier aggregation

Realistic experimentation

PHY abstraction

Channel models

Interoperability / Compliance System Integration

System Integration

Design Validation Channel Sounding

Performance

Low cost BS

Soft RRH

Courtesy: Navid Nikaein, Eurecom/Open Air Interface

WINLAB

GENI Wireless Deployment

Wayne State

Clemson

U Michigan

Columbia

UMassU Wisconsin

Madison

U ColoradoBoulder

UCLA

Stanford

Rutgers

Temple

Drexel

NYU

• 26 WiMAX and LTE BS on 14 campuses• SDN (Click and OVS based) datapath/backbone• 10 mini-ORBIT deployments some with SDRs

Kettering

Utah

WINLAB

Ideal GENI Wireless Unit

Local processing

SDR front-end

RF “Firewall”Modest power amplifier

GENI Rack

WINLAB

Need for Inter-network Cooperation

Radio device A1

Type: Transmit/Receive (client)

Location: (xA1, yA1)

Power, BW, frequency, duty cycle

Radio device B1

Type: Transmit/Receive (client)

Location: (xB1, yB1)

Power, BW, frequency, duty cycle

Aggregate radio map

Range of operation: (xB, yB, rB)

Technology type: Wi-Fi

Device list: B1 params, B2

params, …

Per device parameters Per device parameters

Policy/capabilities

Controller type: C2

Information sharing enabled

Merge RRM enabled

Aggregate radio map

Range of operation: (xB, yB, rB)

Technology type: Wi-Fi

Device list: A1 params, A2

params, …

Policy/capabilities

Controller type: C2

Information sharing enabled

Merge RRM enabled

Network A Network B

Algorithm& policy negotiation

Spectrum infoexchange

• Interaction between managed wireless networks over the back-end wired link for making more efficient use of the spectrum

WINLAB

SDN Approach to Wireless Control Plane

Network OS with wireless abstractions

MobilityMgmt.

WirelessAccessControl

Wired + Wireless Network

QoSControl

ChannelAssign-ment

Tx PowerControl

Inter-Network

Coordination

Data Plane Apps Control Plane Apps

Through extension of OpenFlow Match/Action Fields

Through the ControlSwitch

framework

• Introducing flexibility in the wireless control plane by leveraging software defined networking techniques

• Inter-network cooperation translates to inter-controller interactions and setting of flow-rules

WINLAB

A: Distributed Control Plane

Extension of traditional Enterprise Controller:• Multiple copies of wireless controllers (WC) with mechanisms to cooperate,

scattered throughout (SDN) based control plane

• Reduced distance between device and a controller – reduced flow setup times (reduced control latency)

BS/AP Control Network (data plane)

WirelessController

WirelessController

Shared State

Control

WINLAB

B: Heterogeneous Distributed Control Plane

• Builds the control plane as a network of different controllers– Each controller is a part of a control stack

– Controllers communicate by message passing

– Multiple controllers process each event● SDN control plane can pick and choose services for each event,

avoiding conflict

• Wireless Control Stack (WCS) realized as a complex interaction between controllers rather than (single) monolithic application

WINLAB

C: Heterogeneous Hierarchical Control Plane

● Groups of functional WCS controllers arranged in tiers– Higher tier events = more global and less

frequent

– Supports natural hierarchical inter-controller relations

● Controllers connected with (SDN) inter-controller links– Joins pieces of wireless control stacks

together through SDN

– Additional benefit: dynamic provisioning and routing for events to be moved between controllers and/or tiers (if they can’t be handled at a tier)

C1

B2B2

A3A2A2Tier 1

Tier 2

Tier 3

BS/AP Control Network (data plane)

WINLAB

Heterogeneous Hierarchical Control Plane

• Example: Pair of enterprises with heterogeneous decomposed controllers

BS/AP Control Network (data plane)

BS/AP Control Network (data plane)

AI1

AI1

H2H1

Tier 1

Tier 2

H3H2H1

Authentication/Interference

Handoff

WINLAB

Distributed SD(W)Ns: What’s Missing?

Scalable and resilient, but not readily available– proprietary, experimental, or early-stage

Focuses on uniformity, expects controllers to be homogeneous

No standard mechanism to handle interaction of network stacks (SDX?)– Relies on specialized mechanisms

● network hypervisors - virtualize network

● network compilers - resolve flow conflicts

● semi-intelligent switches - delegate some work to switches

● SDN-to-IP network peering applications

WINLAB

Implications

A complex control plane structure requires:

– Orchestration of event handling across multiple controllers→ An inter-controller process chain

– Path selection across control plane→ A control plane routing mechanism

– Allow controllers to learn of available services/topology→ An information propagation scheme

WINLAB

Generic Wireless Ontology (GWO) http://www.orbit-lab.org:8080/tsc/resources/WirelessOntology ǂ

Wireless Domain Connector ontology

(WDSO)http://www.tasorone.com/tsc/resources/OpenAirInterface

Spectrum Sensing Capability ontology

(SSCO)http://www.tasorone.com/tsc/resources/SpectrumSensing

Capability

Command Line ontology (CLOnt)http://www.tasorone.com/tsc/resources/CLOnt

Meta Service Architecture ontology

(MSAO)http://www.tasorone.com/tsc/resources/msa

Resource Description Framework

ontology (RDF)

See also:

http://groups.geni.net/geni/wiki/GENIFireCollaborationWorkshopSeptember2015for more detailed treatment of a foundation ontology for wireless networking

CoordsSS Ontologies Framework (FLEX)

WINLAB

OBRIT Extension: Proposal

WINLAB

OBRIT Extension: Current• 40 USRP X310s

– Available FPGA resources:

– RF 2 x UBX-160 (10 MHz - 6 GHz RF, 160 MHz BB BW)

– 2 x 10G Ethernet for fronthaul/interconnect

– Four corner movable mini-racks (4 x 20 x 20 -> 1 x 80 x 80)

• > 500+ GPP Cores/CloudLab Rack (?)

• Number of GPU platforms

• 32x40G SDN aggregation switch

Resource Type Number

DSP48 Blocks 58K

Block Rams (18 kB) 14K

Logic Cells 7.2M

Slices (LUTs) 1.5M