intelligent Converged network consolIdating Radio and ... KPIs, ... LTE Long Term Evolution ... SSB Single Side Band . Contract No: 644526 iCIRRUS 1 Jan 2015 – 31 Dec 2017

iCIRRUS Contract No. 644526 1 Jan 2015 – 31 Dec 2017

This project has received funding from the European Union’s Horizon 2020

research and innovation programme under grant agreement No 644526

intelligent Converged network consolIdating Radio and optical access aRound

USer equipment

DELIVERABLE: D3.4

Updated Low-Cost, Energy-Efficient Fronthaul Architecture

Contract number: 644526

Project acronym: iCIRRUS

Project title: Intelligent converged network consolidating radio and optical access

around user equipment

Project duration: 1 January 2015 – 31 December 2017

Coordinator: Nathan Gomes, University of Kent, Canterbury, UK

Deliverable Number: D3.4

Type: Report

Dissemination level Public

Date submitted: 07.07.2017

Editors: Kai Habel (HHI)

Authors / contributors

(contributing partners)

HHI: Kai Habel, Christoph Kottke, Malte Hinrichs, Luz Fernandez del

Rosal

ADVA: Daniel Münch, Nicklas Eiselt

Kent: Philippos Assimakopoulos, Nathan Gomes

VIAVI: Howard Thomas

UEssex: Mike Parker, Felix Ngobigha, Geza Koczian, Terry Quinlan,

Stuart Walker

Orange: Philippe Chanclou

Internal reviewers Mike Parker, Philippe Chanclou

Contract No: 644526 iCIRRUS 1 Jan 2015 – 31 Dec 2017

Page 3 of 78

This project has received funding from the European

Union’s Horizon 2020 research and innovation

programme under grant agreement No 644526

Document history

0.0 Document creation 14/12/2016

0.1 First input HHI 08/05/2017

0.2 First Input UEssex, VIAVI 10/05/2017

0.3 Update VIAVI 15/05/2017

0.4 Merge of input from Kent,

Viavi and ADVA

29/05/2017

0.5 Update HHI 30/05/2017

0.6 Input from ADVA 31/05/2017

0.7 Input from UEssex integrated 01/06/2017

0.8 Input from Orange 01/06/2017

0.9 Update from UEssex

integrated

06/06/2017

0.95 Final touch for

Introduction/Conclusion

15/06/2017

0.96 Input from UKent 16/06/2017

0.97 Internal Reviews (Orange,

UEssex)

29/06/2017

0.99 Resolving review comments 05/07/2017

Final Final clarifications applied

(ADVA)

07/07/2017


Page 4 of 78




Abstract

This deliverable “Updated Low-Cost, Energy-Efficient Fronthaul Architecture” gives, as the title

suggests, an update of the 5G fronthaul architecture for iCIRRUS. The fronthaul requirements and

key performance indicators (KPIs), such as data rates or parameters for synchronisation, timing

accuracy, latency, and bit error rate have been again evaluated and updated in this document. The

key building blocks for the iCIRRUS solutions have been investigated and experimentally verified. The

main results are given.


Page 5 of 78




Executive Summary

This deliverable “Updated Low-Cost, Energy-Efficient Fronthaul Architecture” gives an update of the

fronthaul architecture as compared to the previous iCIRRUS deliverable D3.2, in terms of the 5G

requirements, KPIs, and technical building blocks. The iCIRRUS fronthaul comprises data streams

with different performance requirements, including synchronization, legacy fronthaul, evolved

fronthaul user data, fronthaul control data, and potentially backhaul traffic thanks to the structural

convergence enabled by Ethernet. The iCIRRUS architecture combines Ethernet as the transport

protocol with modifications to the functional split in order to reduce data rates in the fronthaul

while making statistical multiplexing gains possible, and thus allows a more efficient use of the

network resources. Extensive investigations for the various technical building blocks of the proposed

solutions are given.

Our analysis indicates that most KPIs are feasible within the iCIRRUS architecture and that the

challenges will reside in meeting timing and synchronization requirements. More advanced 5G

demonstration scenarios are planned for the final test in the project to confirm that the proposed

architecture meets the demanding fronthaul requirements that are expected from 5G networking.


Page 6 of 78




Index of terms

ADC Analogue to Digital Converter

API Application Programming Interface

APTS Assisted Partial Timing Support

ATM Asynchronous Transfer Mode

AWG Arbitrary Waveform Generator

BBU Baseband Unit

BC Boundary Clock

BER Bit Error Rate

BES Best Effort Section

BESS Best Effort Sub-Section

BF Basic Frame

BL Bit Loading

BSC Base Station Controller

btb Back to Back

BTS Base Station

BW Bandwidth

CD Chromatic Dispersion

CDR Clock Data Recovery

CFO Carrier Frequency Offset

CPRI Common Public Radio Interface

C&M Control and Management

CRC Cyclic Redundancy Check

C&M Control and Management

C-RAN Cloud Radio Access Network

CSI Channel State Information

CP Cyclic Prefix

CCDF Complementary cumulative distribution function

CO Central Office

CoMP Coordinated MultiPoint

CPRI Common Public Radio Interface

CQF Cyclic Queuing and Forwarding

D2D Device-to-Device

D2I Device-to-Infrastructure

DAC Digital to Analogue Converter

DCI Downlink Control Information

DD Direct Detection

DFB Distributed Feedback Laser

DL Downlink

DLSCH Downlink Shared Channel

DMRS Demodulation Reference Signal

DMT Discrete Multi-tone

DQPSK Differential Quadrature Phase-Shift Keying

DSO Digital Storage Oscilloscope

DSP Digital Signal Processing

DU Digital Unit

EDFA Erbium Doped Fibre Amplifier

EML Electro-Absorption Modulated Laser


Page 7 of 78




eNodeB (eNB) Evolved Node B

EPC Evolved Packet Core

EVM Error Vector Magnitude

FCS Frame Check Sequence

FDV Frame-delay variation

FEC Forward Error Correction

FFE Feed Forward Equalizer

FFT Fast Fourier Transform

FIFO First In, First-Out

FIL Fronthaul Interface Library

FIR Finite Impulse Response

FLR Frame Loss Rate

FPGA Field Programmable Gate Array

FQTSS Forwarding and Queuing Enhancements for Time-Sensitive Streams

FRP Frame Result Packet

FSPL Free Space Path Loss

FTTx Fibre To The x

FUSION

FWHM Full-Width Half-Maximum

3GPP 3rd Generation Partnership Project

GNSS Global Navigation Satellite System

GP Guard Period

GPS Global Positioning System

G-PON Gigabit Passive Optical Network

GST Guaranteed Service Transport

HARQ Hybrid Automatic Repeat Request

HD Hard Decision

HetNet Heterogeneous Network

HP High Priority

ID Identifier

IDFT Inverse Discrete Fourier Transform

IFFT Inverse Fast Fourier Transform

IM Intensity Modulation

IMDD Intensity Modulation and Direct Detection

IF Intermediate Frequency

IFFT Inverse Fast Fourier Transform

IMT International Mobile Telecommunications

IPU Intelligent Processing Unit

ITU-T International Telecommunication Union-Telecommunication Standardization Sector

iRRH Intelligent Remote Radio Head

IQ In-phase / Quadrature

IP Internet Protocol

JT Joint Transmission

KPI Key Performance Indicator

LC Lucent Connector

LTE Long Term Evolution

LAN Local Area Network

LO Local Oscillator

LP Low Priority

MAC Media Access Control


Page 8 of 78




MAN Metro Area Network

MB Mobile Cloud

MC Multi Carrier

MCS Modulation and Coding index

MEF Metro Ethernet Forum

MIMO Multiple-Input Multiple-Output

MME Mobility Management Entity

MRP Metric Result Packet

MTU Maximum Transfer Unit

MZM Mach-Zehnder Modulator

NFV Network Function Virtualization

NGFI Next-Generation Fronthaul Interface

NGMN Next Generation Mobile Network

NRZ Non-Return to Zero

OAI Open Air Interface

OAM Operations, Administration and Management

OBSAI Open Base Station Architecture Initiative

ODN Optical Distribution Network

OFDM Orthogonal Frequency Division Multiplexing

OLT Optical Line Termination

OMC Operations and Maintenance Center

ONU Optical Network Unit

OOK On-Off Keying

OSS Operation Support System

OTA Over-The-Air

OTG OAI Traffic Generator

OTT Over The Top

PAM Pulse Amplitude Modulation

PAM-4 Four-level Pulse Amplitude Modulation

PAPR Peak-to-Average Power Ratio

PDCP Packet Data Convergence Protocol

PDU Protocol Data Unit

PDSCH Physical Downlink Shared Channel

PDV Packet Delay Variation

PHICH Physical Hybrid-ARQ Indicator Channel

PHY Physical (Layer)

PMA Physical Medium Attachment Sublayer

PMD Physical Medium Dependent Sublayer

PMI Precoding Matrix Indicator

PL Power Loading

PLL Phase-Locked Loop

PON Passive Optical Network

POX Python-based 1 network operation system

PRACH Physical Random Access Channel

PRC Primary Clock

PRE Packet Routing Engine

1 http://searchsdn.techtarget.com/definition/POX


Page 9 of 78




PS Protected Section

PSD Power Spectrum Density

PSFP Per-Stream Filtering and Policing

PSS Protected Sub-Section

PtMP Point-to-Multi-Point

PtP Point-To-Point

PTP Precision Time Protocol

PUCCH Physical Uplink Control Channel

PUSCH Physical Downlink Shared Channel

QAM Quadrature Amplitude Modulation

QoE Quality of Experience

QoS Quality of Service

RAN Radio Access Network

RAR Random Access Response

RAT Radio Access Technology

RAU Remote Aggregator Unit

RE Radio Equipment

REC Remote Equipment Controller

RF Radio Frequency

RLC Radio Link Control

RNC Radio Network Controller

RoE Radio over Ethernet

ROP Received Optical Power

RRC Root Raised Cosine

RRH Remote Radio Head

RRS Radio Remote System

RRU Remote Radio Unit

RSTD Reference Signal Time Difference

RTT Round Trip Time

RU Remote Unit

Rx Receiver

SC Single Carrier

SD Soft Decision

SDH Synchronous Digital Hierarchy

SDN Software Defined Network

SDQPSK Single DPQSK

SFO Sampling Frequency Offset

SFP Small Form-factor Pluggable

SI System Information

SISO Single Input, single Output

SLA Service Level Agreement

SM Statistically Multiplexed

SN Sequence Number

SNR Signal-to-Noise Ratio

SON Self Optimising Network

SONET Synchronous Optical Networking

SP Strict Priority

SRAM Static Random-Access Memory

SRS Sounding Reference Signal

SSB Single Side Band


Page 10 of 78




SSMF Standard Single-Mode Fibre

SyncE Synchronous Ethernet

TAS Time-Aware Shaping

TB Transport Block

TDD Time Division Duplex

TDM Time Division Multiplexing

TDMA Time Division Multiplexing Access

TG Traffic Generator

TIA Trans-Impedance Amplifier

TS Training Symbol

TSN Time-Sensitive Networking

TTI Transmission Time Interval

TW Transmission Window

Tx Transmitter

UDP User Datagram Protocol

UE User Equipment

UL Uplink

USRP Universal Software Radio Peripheral

UTC Coordinated Universal Time

vBBU Virtual Baseband Unit

VLAN Virtual Local Area Network

VM Virtual Machine

VNA Vector Network Analyser

VOA Variable Optical Attenuator

WDM Wavelength Division Multiplexing

WFQ Weighted Fair Queuing

WRR Weight Round Robin

XGS-PON 10-Gigabit-capable Symmetric Passive Optical Network


Page 11 of 78




Contents 1 Introduction .................................................................................................................................. 13

2 Architecture overview ................................................................................................................... 13

2.1 Access Network Solutions for Ethernet Backhaul ................................................................. 14

2.2 Fixed Access Network segments for RAN fronthaul ............................................................. 15

2.3 New Fixed Networks segments for virtual RAN .................................................................... 17

2.4 Fronthaul deployment options ............................................................................................. 18

3 Refined fronthaul requirements and KPIs .................................................................................... 20

3.1 Data rates .............................................................................................................................. 20

3.1.1 Data rate for low layer RAN functional split ................................................................. 20

3.1.2 Data rate for High layer RAN functional split ................................................................ 21

3.2 Fronthaul Security ................................................................................................................. 23

3.2.1 Introduction to Security in 5G Access Networking ....................................................... 23

3.2.2 Security Considerations for 5G Architectures ............................................................... 24

3.3 Fronthaul KPIs ....................................................................................................................... 26

4 Updated architectural building blocks .......................................................................................... 26

4.1 Ethernet mapping and encapsulation ................................................................................... 26

4.2 Timing and synchronization (priority/scheduling) ................................................................ 33

4.3 Fronthaul intelligent processing unit (IPU) ........................................................................... 33

4.4 Time sensitive Ethernet switching /aggregation .................................................................. 36

4.4.1 Deterministic Ethernet transport with low and fixed latency ...................................... 36

4.4.2 Time-aware-shaping (TAS) reference scenario ............................................................. 43

4.5 High-speed and low-cost transmission links ......................................................................... 48

4.5.1 100 Gbit/s per wavelength ........................................................................................... 48

4.5.2 Beyond 100 Gbit/s per wavelength .............................................................................. 53

4.5.3 Wireless transmission over millimetre wave ................................................................ 57

4.6 Test and performance monitoring ........................................................................................ 62

5 Further architectural considerations ............................................................................................ 65

5.1 Evolved digital fronthaul vs. analogue fronthaul .................................................................. 65

5.1.1 Reference system .......................................................................................................... 65

5.1.2 Analogue Fronthaul ....................................................................................................... 65

5.1.3 Evolved Digital Fronthaul .............................................................................................. 66

5.1.4 Optical Link .................................................................................................................... 66


Page 12 of 78




5.1.5 Results ........................................................................................................................... 67

5.1.6 Comparison ................................................................................................................... 69

5.1.7 Conclusions ................................................................................................................... 69

5.2 SON use cases ....................................................................................................................... 69

5.2.1 Representation of the radio environment .................................................................... 70

5.2.2 Representation of the fronthaul environment ............................................................. 70

6 Conclusions ................................................................................................................................... 72

References ............................................................................................................................................ 73

List of figures ......................................................................................................................................... 76

List of tables .......................................................................................................................................... 78


Page 13 of 78




1 Introduction This deliverable D3.4 updates the architecture for the evolved fronthaul that has been

developed within the iCIRRUS project. It is mainly based on the technical solutions addressed in WP3

and enhanced by results of the other work packages (WPs). An architecture overview of 5G fronthaul

follows in section 2, also covering the most recent discussions of 5G deployment options. An update

relative to the previous deliverable D3.2 [51] for the requirements and KPIs for the mobile fronthaul

are covered in section 3. The technologies to realize the iCIRRUS fronthaul solutions are evaluated in

section 4. These are namely: Ethernet mapping and encapsulation, Timing and synchronization,

Fronthaul intelligent processing unit (IPU), Time sensitive Ethernet switching and aggregation, High-

speed and low-cost transmission links, and Test and performance monitoring. The module or

subsystem tests are also covered there, in order to prepare the integrated tests planned for WP5.

Finally, in Section 5, further architectural aspects are considered.

2 Architecture overview A high capacity transport infrastructure is vital for efficient mobile network operation. While

capacity can be relatively easily increased to accommodate 2G, 3G and 4G Mobile generations, the

needed throughputs for the coming 5G networks will be a technical challenge. The promises of 5G

are expected to enable a “fibre-like” user experience, in its capacity to support the anticipated

requirements such as high throughputs, and low latencies etc. In that context, optical networks are

therefore expected to be the predominant technology for carrying the traffic to the antenna sites

with adequate bandwidth and quality of service (QoS). In the fixed access network segment,

Ethernet is the dominant protocol and interface technology for the Digital Unit (DU) backhaul of 2G,

3G, and 4G due to its capability for supporting QoS policies, aggregation, embedded synchronization

traffic, and security features. Optical access networks based on point-to-point and point-to-

multipoint topologies supports the current Ethernet-based mobile backhaul. In this overview, we

propose to present in a few words the different technical options available to the mobile Ethernet

backhaul for the access network (last mile) segment, in synergy (or not, as the case may be) with

Fibre-To-The-Home (FTTH) deployment dedicated to residential customers.

Figure 1: Optical access solutions for backhaul based on 1) PtP, 2) T(W)DM PON, 3) PtP WDM PON


Page 14 of 78




In contrast to the backhaul, the Radio Access Network (RAN) introduces new kinds of network

segments where optical fibre is becoming essential: the midhaul and fronthaul segments. We have

already described in the previous iCIRRUS deliverables D2.1 and D2.2 the network requirements,

whilst here we now briefly put the backhaul evolution into its perspective, and how these new

network notions with their related architectures along with the fixed optical access solutions can

work together to support the fronthaul. Finally, we focus our interest on the latest evolution trends

for RAN, particularly with respect to virtualization features. The evolved fronthaul based on Ethernet

is also discussed in the context of the RAN functional split. Finally, potential technical solutions for

future optical access networks are also presented to support this RAN evolution.

2.1 Access Network Solutions for Ethernet Backhaul

Mobile Backhaul is the transport network between the antenna cell site and the Base Station

Controller (BSC) or Radio Network Controller (RNC), respectively for 2G and 3G RAN. With 4G, there

are no radio network controllers, since controller functions have been incorporated into the evolved

Node B (n.b., eNodeB becoming synonymous with the Digital Unit) for the main part and also within

the serving gateway. Thus, the 4G RAN backhaul is the transport network segment between the

antenna cell sites up to the Evolved Packet Core (EPC). Here, we will focus our interest on the last

mile of this transport backhaul, such that we do not assess the technical discussions about the

aggregation backhaul network based on ring or mesh topologies with typically Internet

Protocol/MultiProtocol Label Switching (IP/MPLS) transport. Figure 1 shows three different technical

solutions based on:

1) A dedicated fibre supporting a point-to-point (PtP) connectivity between the access node,

equipped with an Optical Line Terminal (OLT), and the antenna site, equipped with an Optical

Network Unit (ONU). This PtP topology is the most straightforward solution for areas with plenty

of optical fibre resources.

2) For areas with available (but limited) optical fibre resource, a technical solution is required to

enable it to be shared. The most common and widespread deployed solution is that of Gigabit

capable Passive Optical Network (G-PON). Following the legacy G-PON, it is also widely

recognised that XGS-PON (PON working at 10Gbit/s downstream and 2.5 or 10Gbit/s upstream) is

the next deployable solution for an enhanced capacity fixed broadband, but now also for the RAN

backhaul. All these solutions are based on a wavelength channel pair to achieve the up- and

downstream and which are able to coexist on the same Optical Distribution Network (ODN). Time

Division Multiplexing/Mulitple Access (TDM/TDMA) is used for sharing the trunk part of the ODN

equipped with an optical power splitter at the branching node, and a single optoelectronic

interface at the access node. Since 2015, multi-wavelength PON solutions have been

standardized, either mixing the time and wavelength dimensions (TWDM-PON), or only

considering multi-wavelength (PtP WDM-PON) approaches.

3) This last technical solution based exclusively on Wavelength Division Multiplexing (WDM) could

support either a wavelength-routed topology, where the branching node is composed of a

wavelength multiplexer (WM) device such as an Arrayed-Waveguide Grating (AWG) or a

combination of thin-film filters (TFFs), or a wavelength-selected topology where the branching

node is composed of an optical power splitter or a bandpass wavelength filter. These PtP WDM-

PON solutions also require multi-transceiver interfaces at the access node.

It is mandatory that these three access solutions are based on a single fibre. In other words, the up-

and downstreams operate in bidirectional diplex mode in the fibre, with each ONU being associated

with a wavelength channel pair. In the case that wavelength is used in the access network (i.e. PtP


Page 15 of 78




WDM-PON, or TWDM-PON), a colourless optical module at the ONU capable of working at any

wavelength is essential. All these wavelength channels need to be controlled by the OLT to fix and

control the wavelength allocation and avoid any rogue wavelength behaviour in the concerned ODN.

For any access solution, the ONU must also wait for the OLT’s permission to start emitting optical

power, in order to be compatible with the mandatory “silent start” function. The power

consumption policy must also be optimized under control of the OLT, whereby the OLT is designated

the unique source of management, wavelength and time control and synchronization of the ONUs.

2.2 Fixed Access Network segments for RAN fronthaul The term midhaul has been defined by [3] as the carrier Ethernet network between antenna sites

(especially when one site is a small cell site). The MEF reference scenario in Figure 2 shows that the

midhaul is considered as a backhaul extension between a small cell DU and its master macro-cell DU.

Two other scenarios are also considered: i) the midhaul between two DU pools (illustrated in Figure

2); and ii) the midhaul between two DU pools through a network controller (not illustrated in Figure

2). All midhaul scenarios are Ethernet-based, and use the same fixed access connectivity as the

backhaul.

Figure 2: Mobile Backhaul, Midhaul and Fronthaul from MEF [3].

Figure 3: First three steps of mobile equipment evolution.

The term fronthaul [4] is used to designate the dedicated connectivity between the Digital Unit


Page 16 of 78




(DU) and the Radio Unit (RU). In the sections 3 and 4, we will only focus our interest on this fronthaul

segment. Figure 3 shows the different RAN arrangement scenarios based on short and long reach

fronthaul scenarios. The radio signal processing functions are commonly supported by equipment

localized at the base of the antenna. Concerning the RF amplifier inside the RU, we note that its

performance (power consumption and cost) depends on the RF attenuation of the coaxial cable

reaching the antenna (cf. Figure 3a). It is desirable that an outdoor form factor for this RU remains

close to the antenna. As far as the links between DU and RUs are concerned, the main requirements

are to allow for the lowest RF signal degradation possible and to reach several tens of metres of

propagation. The combination of DAC & ADC (Digital to Analog Converter, and vice versa) and digital

transmission over fibre with regular pluggable optoelectronic transceivers has allowed the meeting

of these requirements (cf. Figure 3b). This approach to splitting the RAN equipment is known as low

layer RAN functional split fronthaul, and uses CPRI, OBSAI or ORI interfaces [8]. Each fronthaul link

between DU and RU is based on a constant and symmetrical high bit rate serial digital interface

created from the digitization of the baseband, time domain radio signals. Typically, three times 2.5

Gbit/s (i.e. 3x2.5-Gb/s) are needed to transport a 20 MHz 2x2 Multiple Input Multiple Output

(MIMO) radio signal for three sectors whose maximum mobile peak bit rate is limited to about 150

Mbit/s. The reason behind such poor spectral efficiency can be straightforwardly explained by the

quantization and coding operations needed to convert the radio signals into the on-off keying (OOK)

sequences used in the optical link. Also, the clock of this OOK signal serves as a reference for mobile

RF generation inside the RU. Since commercial optical transceivers are available and can reach

several tens of kilometres at the required bitrates, the reach extension of the fronthaul becomes

possible (cf. Figure 3c). Optical fibres are used to reach the antenna sites within the limit of the

maximum round trip time allocated to the fronthaul (typ. 20 km for one way, given that this value

depends on the RAN implementation). Nevertheless, the feasibility of transporting these fronthaul

links over Ethernet network equipment remains a challenge due the bit rate quantity and the timing

requirements. The IEEE launched a standard specification action to draft a standard for Radio-over-

Ethernet to achieve encapsulations and mappings of Inphase/Quadrature (I/Q) radio samples

coming from variety of fronthaul interfaces. The IEEE document P1914.3, which addresses this issue,

is still under construction [5].

Figure 4 shows the most common fixed access solutions for the last mile transport of the native

low layer fronthaul without Ethernet encapsulation. The first one uses a dedicated optical fibre per

RU. We can see from Figure 4a that several fibres are required to reach all the RUs (one per sector,

per radio carrier, per radio technology). In order to achieve fibre sharing, PtP WDM-PON could be

used (cf. Figure 4b). The passive wavelength-based access solution is compatible with the required

line rate and low latency (no framing) by using colourised transceivers at the DUs and RUs.


Page 17 of 78




2.3 New Fixed Networks segments for virtual RAN

Nowadays, a new trend is paving the way for 5G RAN architectures based on different functional

splits. Rather than concentrating all the network intelligence at the DU, such as is the case with CPRI,

OBSAI and ORI, these splits consist of transposing some of the radio protocol layers to the antenna

site. Different possible functional split have been proposed and analysed, based on three main

drivers: i) the feasibility to implement some of the DU functions by software [6], which allows one to

dynamically change and optimize the functional split between a (now) virtual DU and RU, hosting the

real-time dependent hardware; ii) to provide an alternative to the bandwidth hungry and time

sensitivity of the current existing fronthaul solutions; iii) to reuse the widespread Ethernet-based

access ecosystem.

Putting these drivers together, a solution has appeared that is based on high protocol layer

functional splits. This choice allows a virtual DU (v-DU) to be placed at an edge node and to be

connected via Ethernet (access and aggregation network segments) to e-RUs (“e” for

evolved/Ethernet RU). Figure 5 shows such RAN evolution in combination with the already

mentioned fixed optical access solutions [7]. Similar to the previously shown fixed access solutions

for backhaul, we find once more that solutions are based on the following: (n.b. we remind that we

are not considering the aggregation network in this introduction)

1. A dedicated optical fibre between the access node and the antenna site, with the particular

aspect that the ONU must now collect (and prioritize) multiple Ethernet traffic from/to several e-

RUs.

2. A shared optical fibre using a T(W)DM-PON solution with a dedicated or shared ONU to collect

several e-RUs. If high layer Ethernet fronthaul traffic prioritization is required, this functionality

could be done by the OLT with a specific dynamic bandwidth allocation mechanism.

Figure 4 Optical access solutions for low layer fronthaul based on a) PtP fibre, b) PtP WDM PON


Page 18 of 78




3. A shared optical fibre using a PtP WDM-PON solution, also with dedicated or shared OLT to

collect several ONUs. The use of several wavelength channel pairs could allow the minimization

of Ethernet traffic congestion in the transport equipment, by implementing this function in

dedicated Ethernet equipment in either the access node or at aggregation network segments.

This transport solution also allows (on the same fibre) the support of distinct traffic with

different policies, such as backhaul and low- or high-layer fronthaul for different RAN carriers or

different (i.e. legacy) generations of RAN equipment. This would also provide a seamless

migration path.

2.4 Fronthaul deployment options In the previous section 2.2 and 2.3, the fronthaul deployment options have been already described

for the transport of a low and high layer RAN split. In this section we now propose to comment on

some of the key points for each of the transport RAN segments:

Backhaul: the preferred deployment scenario is based on PtP Ethernet. The option of using TDM-

PON (like G-PON) is that a synergy with FTTx is technically feasible, although not largely used due to

the following reasons:

- Availability of specific ONUs;

- Complex inventory and maintenance due to different skills and group involved from fixed and

mobile operation;

- RAN traffic is not guaranteed (congestion in combination with fixed traffic).

If G-PON were the adopted technology for the 2G, 3G and 4G backhauling, then that makes the XGS-

PON more appropriate for the 4G+ and 5G phase 1 (carriers at 700MHz and 3.5GHz) evolution. The

25G-PON and TWDM-PON would then be the most appropriate PON technology for 5G phase 2

(mmW carriers).

Figure 5: Optical access solutions based for high layer fronthaul based on: 1) PtP, b) T(W)DM-PON, 3) PtP WDM-PON.


Page 19 of 78




Low layer RAN split

The preferred deployment scenario is dark fibre with the option to use WDM-PON for fibre sharing.

This network segment is short range due to the RAN latency requirements.

High layer RAN split

This interface is “similar” to the backhaul, in term of the transport protocol (Ethernet), throughput

and latency. Hence, an existing backhaul solution which is natively based on PtP Ethernet will also be

the preferred option in order to maintain the existing operation and the reduced cost aspects.

Nevertheless, due to the required 5G throughput (and latency), the combination of multiple 5G RAN

equipment (several carriers are coming for 5G phase 1 and phase 2) and the need to reuse the

existing backhaul fibre, the option of point-to-point for several parallel Ethernet links using WDM, is

also now on the table for 5G. If TDM-PON for backhaul is adopted, the XGS for 5G phase 1, and 25G-

PON and TWDM for 5G phase 2 will also be the potential technology for high layer RAN split

transport in the access segment.


Page 20 of 78




3 Refined fronthaul requirements and KPIs

3.1 Data rates

3.1.1 Data rate for low layer RAN functional split

The required bandwidth capacity with different communication standards scenarios for a single base

station is shown in the table 5-1 below for the low layer RAN functional split.

Table 1: Overview of required CPRI line rate as a function of several RAN configurations

Radio access technologies (3

sectors)

CPRI line rate for

downlink only

configuration of the fronthaul

interface

LTE (20MHz & 40MHz with 2x2 MIMO) 3×2.5 = 7.5 Gbit/s &

3×5 = 15 Gbit/s

RUs without cascading：3 parallel links

working at 2.5 or 5 Gbit/s

RUs cascaded：one link working at 10 or

25 Gbit/s

W-CDMA (20MHz with SISO) 3×1.25 Gbit/s =3.75 Gbit/s


working at 1.25 Gbit/s

RUs cascaded：one link working at 5

Gbit/s

GSM (10MHz with SISO) 3 x 614.4Mbps


working at 1.25 or 2.5 Gbit/s due to the

fact that it is comon to fill the pattern with

idle bits in order to use the most common

1.25 or 2.5Gbit/s optical transceiver

RUs cascaded：one link working at

2.5Gbit/s

two RF carriers of

LTE(20MHz)+WCDMA+GSM

2 x 3 x 2.5 Gbit/s + 3 x 1.25

Gbit/s + 3 x 614.4 Mbit/s = 20.5

Gbit/s

RUs without cascading: 12 parallel links

working at 2.5 Gbit/s (uniform transceiver

line rate by using idle in the CPRI pattern)

RUs cascaded：4 parallel links working

at 10 (for LTE), 5 (for WCDMA) and 2.5

(for GSM) Gbit/s

This table shows that parallel links are required for the low level RAN functional split. This is due to

the fact that without cascading RUs, a bidirectional link needs to be provided for each sector and the

RF carriers. However, with cascaded RUs, the line rate of a link increases. Consequently, many

studies are now being undertaken to reduce the line rate on the fronthaul connection by introducing

compression and/or a different functional split between DU and RU.

Essentially two families of approaches are being considered in the literature:

o Compression techniques, which reduce the line rate at the expense of some loss of

information and additional delay. ETSI proposes that the IQ compression solution should

achieve a compression ratio of at least 50%, with an EVM degradation and a one-way


Page 21 of 78




supplementary latency caused by the compression/decompression algorithms of under

3% and 100µs (preferably 20µs), respectively,

o New functional split options, which modify the hardware split between central and

remote units at the expense of having new layouts for the remote units and changing

the interoperability paradigm.

Both approaches enable significant savings in terms of the capacity of the RAN interfaces. A

compression technique is described in [9] with a complete description of the functional split options.

The next section also discusses the requirements of these high levels RAN split options.

3.1.2 Data rate for High layer RAN functional split

We report in this section the potential network requirements of several high layer RAN splits as

shown in Figure 6.

Figure 6: Several high layer split fronthaul interfaces for the downlink and uplink.

Since these new functional splits are not defined yet by a standardisation document, the proposed

parameters of this white paper could be modified in the future. We propose also in this section to

position these high layer function splits [10] (named F1) with respect to the existing and well-known

RAN interfaces which are the backhaul (S1) and the currently adopted low layer split based on CPRI

or OBSAI.

Table 2 is based on a small cell forum analysis [11] and shows the latency and data rates of these

functional split interfaces, considering a 20 MHz channel bandwidth and 2x2 MIMO for downstream

and 1x2 MIMO for the upstream (more parameters are defined in Appendix C: Bandwidth

calculations of [11]). This data calculation for a split at the high layers, e.g. between PDCP and RLC or

RLC and MAC layers, highlights some interesting facts, namely:

o The traffic is dependent on the traffic load of the end users (i.e., it is not constant traffic

as is the case for low layer splits, e.g. CPRI/OBSAI and split PHY);

o Extra traffic must be considered in addition to the traffic associated to the traffic load of

the end users. This extra traffic between the modified DU and RU is used to achieve

control, scheduling, and security & synchronization features. An extra +10% of traffic

(with a maximum of 100 Mbit/s) must be used, in the first approximation, to dimension

for such features. An exact value still needs to be specified, and this is a function of radio

standardization efforts and also vendor implementation. Some of the extra traffic could

also be constant (i.e. independent of the traffic load of end users).

These high layer functional split interfaces are compatible to packet flow and thus Ethernet will be

the common protocol used. We note that the IEEE working group P1914 “Next Generation

Fronthaul Interface” is expected structure the support of Ethernet for some of these high layer

functional splits.


Page 22 of 78




Table 2: Data rate for different RAN functional split (20MHz, and 2x2 MIMO for downstream and 1x2

MIMO for upstream)

RAN functional split (cf.

Figure 6)

One-way latency

(maximum value)

Downlink

data rate

(Mbit/s)

Uplink

data rate

(Mbit/s)

Comments

backhaul 30 ms 150 48.5 Security and synchronization features

are required

service (RRC-PDCP) 30 ms 150.1 48.6 Traffic dependent on customer

demand

Plus 10% s extra traffic versus

backhaul for control, scheduling,

security, and synchronization (not

included here)

PDCP-RLC 30 ms 150.3 48.7

MAC 6 ms 151.5 49.4

MAC-PHY 2 ms 152,5 49.9

Split PHY (between

resource mapping and

FFT)

250 µs 1075.2 921.6 Constant traffic (not traffic load

dependent)

Synchronization natively include

CPRI/OBSAI (low layer

split)

250 µs 2457.6 2457.6

Additionally, from a mobile operator perspective, a split at the higher layers, e.g. between PDCP and

RLC or RLC and MAC layers, might be of interest. Such an Ethernet backhaul-like traffic forwarding

solution might offer a potentially easier and faster deployment. However, it is not capable of

supporting advanced radio interference techniques such as inter-site CoMP. It is also necessary to

pay attention to the traffic dimensioning of the optical access segment. This traffic dimensioning

refers to all the aggregation equipment, and could be different compared to what is used in the

current backhaul where a statistical multiplexing factor (typically between 3 and 5 for a macro cell

site with 3 sectors) is used. The statistical multiplexing factor can be applied at every point in the

access network that aggregates and combines Ethernet interfaces from multiple sources (e.g. also

from different backhaul mobile generations, or from several 5G RUs each equipped with an Ethernet

interface). What needs to be noted is that for traditional wireless backhaul, the statistical

multiplexing factor is applied at the L3 interface of the RAN equipment.

Nowadays, for the Ethernet transport of the high layer split, and due to the fact that L3 RAN is not

already deployed, the dimensioning and aggregation rules should also now be defined with attention

to the end-to-end quality of service (QoS).


Page 23 of 78




3.2 Fronthaul Security

3.2.1 Introduction to Security in 5G Access Networking

Future 5G networking features both wireless and fixed links, each of which has its own security

issues. The broadcast nature of the wireless domain and mobility of the users make them

susceptible to a wide variety of security attacks such as passive (Traffic analysis and Eavesdropping)

and active (Denial of service (DoS) attacks, Resource consumption, Masquerade attacks, Replay

attacks, Information disclosure and Message modification) attacks [12]. Traditional solutions to

mitigate the security challenges are usually handled at the upper layer using various types of private

and public secret keys via computation-based mechanisms (i.e. cryptography). The reliability in the

exchange of information between a source node (commonly denoted as “Alice") and an intended

destination node (commonly referred to as ”Bob"), and security in terms of confidentiality and

message integrity with respect to an adversary (commonly referred as “Eve") using the

computational security approaches have been reported, e.g. in [13][14] to be susceptible to attacks,

and in a dynamic mobile environment it is computationally complex and intensive [15][16]. Security

is seen from the viewpoint of layered network design as an add-on feature, in particular separating

the physical layer from the upper layers as a reliable bit pipe, and providing security at this layer

should be viewed as fortifying the existing computational security techniques.

Physical layer security (PLS) against passive and active attacks is classified into five major categories:

theoretical secure capacity, channel, coding, power and signal detection techniques. The

fundamental issues of theoretical secure capacity have drawn much attention and most of the works

in this area focuses on secrecy capacity; that is, the maximum rate achievable between the

legitimate transmitter-receiver pair subject to the constraints on information attainable by the

unauthorized receiver. In summary, information-theoretic security is an average-information

measure and it also requires the knowledge of the channel state information that may not be

necessarily accurate in practice. The channel approach is classified into three methods to provide

physical layer security based on exploitation of the channel characteristics, such as radio frequency

(RF) fingerprinting, Algebraic Channel Decomposition Multiplexing (ACDM) pre-coding, and

randomization of MIMO transmission coefficients. The main objective of the code approach is to

improve resilience against jamming and eavesdropping. The code technique is subdivided into the

use of error correction coding and spread spectrum coding. Information protection can also be

facilitated using power techniques. The usual schemes here involve the employment of directional

antennas and injection of artificial noise. A directional antenna facilitates receiving of data from the

direction not covered by the attacking signal. Finally, in the case of introducing artificial noise, secret

communication between legitimate nodes can also be achieved. In [18], a method has been

proposed in which discriminatory channel estimation is performed by injecting artificial noise into

the remaining space of the legitimate receiver's channel in order to degrade the estimation

performance of the eavesdropper. An improvement to this approach is discussed in [19], whereby

the channel feedback information from the legitimate receiver at the beginning of each

communication stage is exploited, and this is called a multi-stage training-based technique.


Page 24 of 78




3.2.2 Security Considerations for 5G Architectures

Security for a 5G hierarchical architecture presents a technical challenge as compared to the more

centralized 4G network architecture. A distributed e2e security approach depends on the algorithms

and methods implemented at the end-points of a connection, i.e. the user devices (UEs); as well as

those algorithms and technologies located between user devices and the services offered by the

network. To provide e2e security, user data needs to be encrypted at the mobile device, e.g. each

previous mobile network generation has used a different encryption method: the A5/1-A5/4

methods were used in GSM; 128 bit encryption and the KASUMI algorithm is used in 3G; while

SNOW/AES is used in 4G/LTE. A generic architectures view of 4G (centralised) and a decentralised

(hierarchical) 5G network is shown in the Figure 7 below.

Figure 7: Comparative architecture views of 4G (centralised) and 5G (decentralised) networks.

While UEs are authenticated in the EPC, they communicate via unsecured transport networks. In

addition, because of the flat 4G IP architecture, communication between eNodeBs and core network

is not authenticated. The Access Stratum is terminated in the eNodeB and thereby protects control

and data transport over the air interface. The user plane and control plane traffic between eNodeB

and EPC is, however, not protected in the Non-Access Stratum, with only the encryption keys and

the control traffic between eNodeB and Mobility Management Entity (MME) being protected.

Communication between decentralized eNodeBs and centralized EPC is realized, in general, via the

unprotected transport network, which consists of the fixed access-, metro- and core networks that

are partly or fully shared among multiple operators and services. 4G/LTE user traffic can therefore

be tapped most easily within unsecured microwave links, which are frequently part of the fixed

access network. 4G LTE networks represent a centralized core network architecture, with the radio

access network being entirely distributed. All base stations (eNBs) are connected via an IP-based


Page 25 of 78




network to the central node/cloud, which performs the functions of the IP packet core (EPC) and the

MME, which includes the routing.

5G networking is causing EPC functionalities to be shifted closer to the base stations; so that the EPC

is becoming less complex and more distributed in nature. However, the many small µEPCs therefore

need to be supervised by a central EPC/MME control network function. From the security point of

view, distributing the EPC functionality is an improvement because distributed security is harder to

attack. The µEPC can be much closer to the base stations, so that the path lengths are thereby

shorter and latency also significantly reduced.

Security improvements have also been suggested for 5G C-RAN architectures, such as iCIRRUS, which

features functional splitting of the eNodeB into a radio unit (RU) and a more centralized baseband

unit (BBU) or DU. In this way, numerous RUs are attached via a new fronthaul interface to one DU,

which allows significant improvements for interference management. Towards the core network,

the DU behaves like a giant base station, having a large number of distributed antennas. Hence, in

5G the DU also has the S1 F1 and Xx interfaces, analogous to the eNodeB in 4G LTE.

While the backhaul behind the eNodeB can be unsecured in 4G LTE (e.g. if a microwave link is used),

the compound link from the UE over the air to the RU and then via a microwave or fixed network

connection to the centralized DU is inherently secure. Only after the DU, is a secure uplink to the

µEPC needed again. The cloud-based iCIRRUS architecture enables instantiation of virtualized core

and RAN network functions among the clouds. In a shared network infrastructure, this implies that

these network functions have to be encapsulated by a secure transport protocol, such as IPSec.

Operators using the shared physical substrate of an infrastructure provider hence have to build

secure islands inside each cloud, which are isolated from the other operators using the same

physical substrate, and where their own virtual network functions (VNFs) can be operated. The only

function that needs no isolation is that of routing, which is natively safe when using IPSec. But all

other VNFs will need security encapsulation.

The cloud infrastructure is a remaining security weakness in the cloud-based iCIRRUS infrastructure,

because isolation between the tenants is virtual; but VNFs of different tenants can be physically

processed in the same machine. At the low processing level, tenants are therefore not physically

isolated. One way out is to only use certified cloud hardware in which interactions between the

tenants can be considered to be impossible. As a result, a combination of over-the-top (OTT) and

network-assisted end-to-end security enables the shared use of the 5G network infrastructure, while

guaranteeing low latency, with secure end-to-end communication.

Targeting an all-IP infrastructure also forces network designers to be aware of secure layer 3 and

layer 2 communications. From a security point of view, authentication and data integrity needs to be

provided by the network infrastructure, to provide data security beginning from the lowest level

possible.


Page 26 of 78




3.3 Fronthaul KPIs The following Table 1 gives an overview of the fronthaul KPIs.

Table 1: Compilation of Fronthaul KPIs

4 Updated architectural building blocks

4.1 Ethernet mapping and encapsulation LTE functional subdivisions or “splits” have been considered as a means of meeting fronthaul data

rate requirements for next generation mobile networks, and as such have attracted the interest of

standards bodies including both 3GPP and IEEE groups. A number of potential split points have been

identified with factors such as data rate, latency, ease of migration/deployment and ability to

accommodate advanced joint signal processing techniques, playing an important role in the choice.

A number of possible split options are shown in Figure 8(a) using the NGMN numbering scheme. In

general, split points further away from the antenna and towards the mobile core, offer the highest

reductions in data rates while starting from the radio side and moving towards the core, a number of

interesting interface points can be identified. The different LTE channels are demarcated at the

resource mapper (RM) (Split II), resulting in an aggregate data rate that depends on the cell load,

leading to statistical multiplexing gains. At the antenna-processing block (layer and port mapper) the

transition from per-antenna flows to per-user flows occurs (Split I), resulting in large reductions in

data rates as these stop depending (proportionally) on the number of antennas. In general,

frequency domain splits lead to data rates through reductions in sample widths (bits per sample)


Page 27 of 78




and sampling redundancy (time domain oversampling). These data rate reductions need to be

considered in unison with all the other performance factors of a given split point (latency, support

for CoMP, pooling and virtualisation gains), with splits closer to the radio side more able to

accommodate advanced joint processing features and larger virtualization/pooling gains.

The MAC/PHY split offers a good overall balance at the expense of the strict latency constraints that

the Ethernet fronthaul will need to meet. Ethernet features such as prioritized scheduling, may offer

means for guaranteeing timely delivery of packets to/from the RU.

The implemented MAC/PHY split is shown in Figure 8(a). The split interface resides between the

MAC layer processing, and the error correction block. The resulting processing module subdivision is

shown in Figure 8(b). The LTE eNodeB protocol stack, up to and including the MAC layer, runs within

the DU and generates MAC layer protocol data units (PDUs) (or MAC transport blocks (TBs)). The

PDUs are encapsulated into Ethernet packets, sent over the Ethernet network and received by a

remote aggregator unit (RAU) which de-packetizes the PDUs and performs all the physical layer

processing (forward error correction (FEC), quadrature amplitude modulation (QAM), antenna

processing, mapping of resources to resource blocks and inverse-fast Fourier transformation (IFFT)).

The resulting IQ radio samples are sent to the remote radio head (RRH) for radio frequency (RF)

processing. The RAU and RRH together then form a remote unit (RU).

The networking entity subdivision is shown in Figure 8(c). The EPC runs in a separate processing

node that is connected through GbE (gigabit Ethernet) to the DU, which in turn is connected to the

RAU through GbE.

The testbed is flexible and can run with different options of emulated, simulated or real hardware

implementations. For example, it can include the EPC, hardware-based RF (e.g. Universal Software

Radio Peripheral , USRP) and commercial 4G phones. Alternatively, it can employ emulated UEs and

simulated air interfaces with and without S1 interface.

For the results presented here, the EPC, DU, RU and UE entities are software emulations, while the

RF processing and air interface are simulated. There is no S1 and F1 interface, and instead internet

protocol (IP) data is fed directly at the PDCP layer at the DU.


Page 28 of 78




Figure 8 (a) Different LTE functional subdivisions (function splits) options, (b) The implemented split processing module

subdivision and (c) the implemented split networking entity subdivision.

A different view of the fronthaul network, concentrating more on the networking part and the

subdivisions of the traffic flows is shown in Figure 9. The LTE functionality in the DU, RU and user

equipment (UE) runs in a software emulation environment based on the open source

OpenAirInterface (OAI) software libraries (see the OpenAirInterface software alliance) and

specifically on the ‘OAI5G’ source code. A Fronthaul Interface Library (FIL) is used to encapsulate the

data exchanges between the functional split entities and to provide a useable abstraction (mapping

functions) to the new functionality. The DU performs all the eNodeB processing up to and including

the MAC layer. It then generates a number of flows and packetizes them into the MTU section of an

Ethernet frame. The resulting packet-types include, downlink shared channel (DLSCH), downlink

control information (DCI), system information (SI) and random access response (RAR). The flows are

then transported over an Ethernet network and are received by the RU, which performs all the LTE

physical layer (PHY) processing. For simplicity, in the uplink (from RU to DU) a single packet type is

used to aggregate all uplink transmissions. The network comprises standard Ethernet switches

forming trunk links where different data flows can contend. A background traffic generator is used in

the figure to indicate how contention with background traffic can be tested in a testbed

environment. The PKT_DCI packet is processed at the RU before retrieval of any of the other packets

is attempted, as this packet, in addition to carrying the DCI data, also acts as a MAC/PHY primitive

carrier. The primitives are used by the RU to extract information on the type and number of

allocations to expect for the current subframe.

The encapsulation format used is common to all packets and is shown in Table 2. The system has the

flexibility to identify flows at varying “resolutions” by combining VLAN IDs and packet-types. An

example configuration entails “bundling” all packet types destined for the same RAU, within the

same VLAN. The standard Ethernet headers are indicated in Table 2 using italicised fonts. The other

headers form part of the Ethernet payload, which means that the network socket at either end of


Page 29 of 78




the fronthaul is aware of the header boundaries. A number of these header values are used in the

buffer management algorithm at the two end-points of the fronthaul network. The Ethernet payload

section contains packet-type specific data (fields) in addition to the MAC PDU. An example is shown

in Table 3 for the PKT_DLSCH packet.

Figure 9 (a) The evolved fronthaul and (b) high-level view of the buffering stages and measurement interface points


Page 30 of 78




Table 2: The 28 (32)-Octet common packet header for all packets sent/received through the fronthaul interface.

Field Size/Octets Description

Dst MAC 6 The destination H/W address, source H/W

address, and EtherType - as per IEEE 802.3.

EtherType is fixed to hex ’08 00’ alluding to

IPv4 datagram.

Src MAC 6

VLAN ID (Optional) 4

EtherType 2

SFN (TX) 2 The LTE SFN and subframe the data in the

packet is part of (for Tx processing). LTE Radio Subframe

(Tx)

1

SFN (RX) 2 The LTE SFN and subframe the data in the

packet is part of (for Rx processing). LTE Radio Subframe

(Rx)

1

Packet-type 2 An unsigned 16-bit enumeration of the packet

types.

Packet Length 2 The size of the packet in Octets, as an unsigned

16-bit integer.

Payload N Packet payload including packet-type specific

data (see Table 3 for example, for PKT_DLSCH)

CRC 4 Cyclic redundancy check

Table 3: Ethernet frame payload fields for PKT_DLSCH

PKT_DLSCH (Ethernet Frame Payload Section)

Field Size/Octets Description

UE index 1 Index of the UE the data in the packet is

intended for

RNTI 2 UE Cell radio network temporary

identifier

Length 2 Length of the payload

Payload N-5


Page 31 of 78




Figure 10 shows a comparison of the data rates produced by the application layer (here using the

OAI traffic generator) and those over the fronthaul interface, for different numbers of UEs. The

downlink application data rate per UE is approximately 1.2 Mb/s while the uplink data rate is entirely

due to control data, and LTE and FIL encapsulation overheads. The ‘total overhead’ trace shows the

increase, as a percentage, between the application and fronthaul data rates, and is approximately

43% for the different number of UEs. It is due to the encapsulation overheads added by the LTE

protocol stack and the FIL.

Figure 11 shows the results of three different tests for a single UE, with each test representing a

different data rate from the OAI traffic generator (OTG). Each processing stage adds some overhead,

resulting in a higher data rate. The first stage includes the MAC PDU encapsulation with a resulting

data rate increase of 34% and is a result of the addition of all LTE headers (PDCP, RLC and MAC).

Following this stage, the PDU is processed by the FIL with a resulting overhead increase of 3%. Both

of these increases are constant for all three tests, as the DLSCH size is fixed to approximately 1000-

octets. The last stage includes all packet types transmitted over the fronthaul, and in this case, the

percentage increase varies form one test to the next. While the amount of SI data is independent of

application data rate, the amount of DCI data is not resulting in a different percentage increase in

each test.

Figure 12 shows the subframe processing latency of the FIL for different numbers of UEs. The latency

is measured between the first packet in each subframe (a DCI packet-type) at interface point 3 and

the last packet for that same subframe at interface point 3 (see Figure 9 (b)). The latency increases

approximately linearly with number of UEs.


Page 32 of 78




Figure 10 Fronthaul and application (OTG traffic generator) data rate measurement results for different numbers of UEs.

The traffic generator is producing traffic only for the downlink direction.

Figure 11 Data rates and percentage increases at different points in the processing chain, for three different tests of

ascending application layer data rates.


Page 33 of 78




Figure 12 Fronthaul processing latency per LTE subframe for different numbers of UEs

4.2 Timing and synchronization (priority/scheduling) The combination of functional splitting and Ethernet means that new traffic mechanisms will

become available due to statistical multiplexing gains in aggregation nodes (Ethernet switches).

However, this means that fronthaul links will need to be provisioned for the timely delivery of

fronthaul traffic to the end stations. To this extent, IEEE 802.1 CM [20] is defining/adapting time-

sensitive networking (TSN) profiles for fronthauling. Currently focusing on CPRI flows, similar profiles

will be required for functional split traffic and any in-line timing protocol (e.g. PTP).

Previous work has investigated the effects of using priority based scheduling in the fronthaul for IQ-

based traffic using weighted round-robin (WRR) and strict priority (SP) [21] algorithms. The latter is

attracting more attention recently as it is (tentatively) part of the IEEE 802.1CM profile A [20].

Additional techniques include frame pre-emption (IEEE 802.1Qbu) [22] and time-aware shaping

(IEEE802.1Qbv) [23]. For the latter, simulation results with CPRI [24] and traffic flows emulating

functional split and PTP traffic [25] have been presented.

4.3 Fronthaul intelligent processing unit (IPU) The purpose of the fronthaul intelligent processing unit (IPU) is to Collect information from the

fronthaul and over-the-air (OTA) sections, run the algorithms that extract the necessary metrics and

produce statistics for the different KPIs. These statistics are in turn used to adapt the operation of

the fronthaul and OTA sections by employing software-defined networking (SDN) techniques and

handovers for load balancing. The IPU also reports the statistics in real time and presents them in a

user-friendly manner.


Page 34 of 78




Figure 13 The fronthaul intelligent processing unit (IPU) and the different feeds from the fronthaul and OTA sections

The PacketPortal Python API is used to collect metadata from the filtered result packets (FRPs) and

metric result packets (MRPs). These include timestamps, probe IDs, sequence numbers and packet

counts. The Dynamic KPI extraction block is then used to calculate the different KPI statistics which

include inter-frame delay, latency, frame-delay variation (FDV) and throughput.

The SDN controller VM runs the POX2 controller. The controller sets the initial flow table

configuration in the SDN switch and thus operates in a pro-active manner. That is, the switch does

not have to send packets from newly received flows to the controller so that its flow tables are set,

thus avoiding this initial set-up delay. The controller is fed average KPI values (e.g. for latency) by the

KPI extraction VM and once a KPI value exceeds a certain threshold in triggers traffic steering to take

place by steering a flow (or flows) to another trunk link. The controller also has the ability to obtain

KPIs from the switch (e.g. number of packets, throughput) and steer based on these. An example of

traffic steering in an evolved fronthaul is shown in Figure 14. Note the latency variation due to

contention with background traffic and the corresponding significant reduction in both latency and

latency variation once the steering occurs.

2 http://searchsdn.techtarget.com/definition/POX


Page 35 of 78




Figure 14 The result of SDN-enabled traffic steering of the background traffic in an evolved fronthaul on the latency and

latency variation of the split traffic. Three packet type traces are shown, pkt_DCI (downlink control information), pkt_DLSCH

(downlink-shared channel) and pkt_SI (system information).

The OTA remote API is implemented through a web socket connection between the DU and the IPU.

The IPU continually interrogates the DU regarding a number of KPIs. These include HARQ

retransmissions and cell load among others. The IPU can then send commands to initiate handover

and/or dynamically change the power level of each cell (recursive cell stretching). An example of

dynamic KPI monitoring and how the KPIs are presented in graphical format is shown in Figure 15.

For this example, the monitored traffic is from a centralised split (IQ transport) with an LTE

bandwidth of 5 MHz, while the fronthaul includes one GbE switch.


Page 36 of 78




Figure 15 Example of live KPI performance monitoring.

4.4 Time sensitive Ethernet switching /aggregation As previously stated, in an Ethernet-based fronthaul for 5G networks, latency and especially latency

variation or packet delay variation or frame delay variation are major issues to fulfil the stringent

timing and synchronization requirements. Time-sensitive networking means are being discussed in

standardization like IEEE 802.1 (see iCirrus Deliverable D3.2 Preliminary Fronthaul Architecture [26]

and Section 4.4.2 below). A a novel approach to provide low deterministic latency for time sensitive

traffic has been identified and investigated as romising technology (see [27] and Section 4.4.1

below).

4.4.1 Deterministic Ethernet transport with low and fixed latency

4.4.1.1 Concept

FUSIONA combination of packet and circuit switching is used to multiplex high priority (HP) traffic

streams (the circuit switched part) with low priority-statistically multiplexed (SM) streams (the

packet switched part) and to transport them over an Ethernet network [27]. The HP traffic is also

called GST (Guaranteed Service Transport). The main idea is to take advantage of the inter-packet

gaps between HP frames to transport low priority (LP) frames of SM streams while leaving the HP

streams essentially unaltered (see Figure 16). As the approach uses the best properties of circuit and

packet switching, it is called “FUSION”.


Page 37 of 78




Figure 16: FUSION: Exploiting the inter-packet gaps between HP frames to transmit LP frames

This is accomplished by adding a deterministic delay to an outgoing HP stream, which is equal to the

maximum transmission time of a SM frame. A gap detector obtains the inter-packet gaps and an SM

scheduler chooses an SM frame that fits within each inter-packet gap. The receiver extracts the HP

streams with the inter-packet gaps preserved. As a result, latency variation is significantly reduced

(max about of 160 ns, but depends on the number of aggregated streams). Furthermore, this

approach achieves a significantly improved utilisation compared to a fully provisioned circuit

switched network and does not require any additional (out-of-band) form of synchronisation, like is

the case in a time-triggered communication or in a TDMA (time division multiple access)-based form

of communication.

4.4.1.2 Theoretical latency and latency variation considerations

Figure 17 depicts the test setup in order to investigate the FUSION approach.

Figure 17: FUSION: Test setup investigating latency and latency variation

The test setup consists of two aggregator or switch instances, with three 10G Ethernet ports and one

100G Ethernet port. Each aggregator instance uses a FUSION 100G FPGA-based IP provided by the

Norwegian company Transpacket A/S. The presented setup considers of one GST link and two SM

links.

Figure 18 provides an overview of the theoretical estimation of the latency and latency variation of

the FUSION IP.


Page 38 of 78




Figure 18: FUSION: Theoretical estimation of the latency and latency variation of the FUSION IP

The GST aggregation consists of a variable part depending on the MTU (maximum transfer unit) and

two fixed processing delays (Agg GST fixed processing delay and Agg fixed processing delay). The GST

de-aggregation encompasses only a fixed processing delay (Deagg fixed processing). The SM

aggregation consists of a variable part (depending on the packet size and number of SM streams)

and one fixed processing delay (Agg SM fixed processing delay). The SM de-aggregation is the same

as for GST. The values of the fixed delays provided by Transpacket are dependent on the FUSION IP-

Core Version. The considered FUSION IP-Core Version is V1.5.


Page 39 of 78




The aggregation case using (one FUSION IP instance) is depicted in Figure 19.

Figure 19: FUSION: Theoretical latency and latency variation for aggregation, neglecting PHY and MAC delays and cable

delays

The table part of this figure describes minimum latency, maximum latency and packet delay

variation. The middle column shows the values, the calculation equation and an example calculation

for GST with parameters for the present test setup (1 GST link, 2 SM links, GST MTU with 16000

Byte). The right column describes the values, the calculation equation and an example calculation for

SM for the used test setup (1 GST link, 2 SM links, SM packet size with 9622 Byte (9600 Byte payload

and 22 Byte header overhead).

Figure 20 shows the aggregation and de-aggregation case using two FUSION IPs that we also have in

the considered test setup (see also Figure 17). The table part of this figure describes values,

calculation equation and example calculation for minimum latency, maximum latency and packet

delay variation for the considered test setup for GST and SM. The middle column depicts GST,

whereas the left column describes SM. Similar to the single aggregation consideration before, PHY

(physical) / MAC (media access control) layer delays are neglected as well as cable delays.


Page 40 of 78




Figure 20: FUSION: Theoretical latency and latency variation for aggregation and deaggregation neglecting PHY and MAC

delays and cable delays

Figure 21 includes the additional PHY/MAC delay for GST, and Figure 22 presents it for SM. Both

figures provide total end-to-end values for minimum latency, maximum latency and packet delay

variation for the considered test setup without considering cabling. The values for 10G MAC/PHY

and 100G MAC/PHY are measured values on a Xilinx VCU110 platform (Virtex UltraScale XCVU190-

2FLGC2104E). For GST, the minimum latency results in 14175 ns, the maximum latency in 14250 ns

and the latency variation in 95.6 ns. In the case of SM, the minimum latency results in 8769 ns, the

maximum latency in 11177 ns and the latency variation in 2408 ns.


Page 41 of 78




Figure 21: FUSION: Theoretical end-to-end latency and latency variation including PHY/MAC delays and neglecting cable

delays for GST links

Figure 22: FUSION: Theoretical end-to-end latency and latency variation including PHY/MAC delays and neglecting cable

delays for SM links


Page 42 of 78




4.4.1.3 Latency and Latency variation measurements

To check the compliance of the theoretical considerations with real systems, a lab setup investigates

the described test setup with 1 GST link and 2 SM links (see also Figure 17). The network tester in

this setup is a Xena Networks Xenabay with M2SFT+T (10G LAN) line cards. A Xilinx VCU110 (Virtex

UltraScale XCVU190-2FLGC2104E) serves as hardware platform. The used FUSION IP has the version

number V1.5. The GST MTU of 16000 Bytes is chosen since this is the maximum value of the FUSION

IP and leads to worst-case values. The setup selects 9622 Bytes for the packet size for the SM links,

as this is the maximum size the network tester that was used can process; and which would also

cause worst-case values for latency and latency variation. Further, for investigating high load

conditions, a constant bitrate traffic pattern with 7864.32 Mbit/s (CPRI option 7 equivalent; 9830.4

Mbit/s without 8Bit/10Bit line coding -> 7864.32 Mbit/s) is used for the GST link as well as for the SM

links. The fibre connections of the lab test are shorter than 2m.

Figure 23 shows the measured results for latency and latency variation using an MTU of 16000 Byte

for GST and a packets size of 9622 Byte for SM links. Since the two results for SM1 and SM2 look

almost identical, the figure shows only the evaluation for SM1. The upper part of the figure showing

the GST results indicates a minimum latency of 14180 ns, and a maximum latency of 14316 ns

resulting in a maximum latency variation of 136 ns. The lower chart describing the SM results with a

minimum latency of 8769 ns and a maximum latency of 11177 ns, and introduces a latency variation

of 2408 ns. In contrast to the GST chart, the latency histogram for SM clearly shows the range of the

variance in the latency indicated by the red arrow.

Figure 23: FUSION: Measured results for latency and latency variation using an MTU of 16000 Byte for GST and a packet

size of 9622 Byte for SM


Page 43 of 78




4.4.1.4 Discussion

The comparison of the results of the theoretical estimation with the measured results leads to the

conclusion that the values for minimum latency and maximum latency reside in the same order of

magnitude. The fibre connections of the lab test are smaller than 2 m (per connection). Therefore,

the fibre delay is negligible (<10 ns) for the comparison for the absolute values, and has no effect on

the relative values for the latency variation. In the case of the latency variation for SM, this is almost

the same value. The latency and latency variation results are within the expected limits.

4.4.2 Time-aware-shaping (TAS) reference scenario

An example reference scenario for TAS use-cases is shown in Figure 24(a). The TAS is applied

towards the edge of the mobile network where the fronthaul networks are formed. The fronthaul is

made up by a pool of digital units (DUs) which are connected, through Ethernet, to remote units

(RUs) or remote radio head (RRHs). These distributed RAN entities can perform different split

functionalities, with some using a centralised processing (IQ radio transportation, DU to RRH pair)

while others use a split at the LTE MAC/PHY interface (DU to RU pair). Some of the RAN processing is

carried out in nodes that are closer to the core (for example implementing higher layer splits at the

PDCP/RLC interface). The DUs then perform the rest of the LTE processing up to (and including) the

LTE MAC layer. MAC/PHY split data flows are then transported to the RUs (which perform the PHY

layer processing) over the fronthaul links. At the same time, timing flows (e.g. PTP) are provided over

the fronthaul, with PTP boundary clocks (PTP BCs).

Local and global scheduling over the TAS application area is provided though scheduling entities

(these can be SDN-type controllers). The global scheduler communicates configuration parameters,

between the switch and the end-stations, regarding window section configurations.

Figure 24 (a) Reference architecture for the time-aware shaper use-cases presented in this work and (b) Scheduling design

concept.


Page 44 of 78




The IEEE 802.1Qbv standard offers a solution to tackling contention-induced FDV in a bridged

network. Traffic flows are assigned to different window sections (or sub-sections within). High

priority (HP) traffic is assigned to a protected window section (PS) or subsection (PSS) while low

priority (LP) traffic is assigned to a best effort section (BES) or subsection (BESS) within. Port gating is

applied such that a traffic flow is only allowed to pass through the switch in its allocated section. For

this scheme to work, an overlaid time synchronisation network is assumed present. The division of

the total transmission window (TW), encompassing all traffic sources, into the different sections is

shown in Figure 25. To prevent the best effort traffic from overrunning into the PS, a guard period

(GP) is used, where no transmission is allowed.

An example of high-priority traffic in the network is PTP traffic while control and management

(C&M) traffic will usually be treated at a lower priority setting. It is possible within the BES to assign

priority levels to the different lower priority streams and employ an “intra-section” scheduler such

as SP (strict priority), weighted-round-robin (WRR) or weighted fair queuing (WFQ).

Figure 25 Generic time window, window section and subsection plan based on IEEE 802.1Qbv

The baseline for the results presented here is the SP algorithm. With SP, the different queues

transmit in the order of priority setting. Thus, an LP queue has to wait for all the higher priority

queues to finish their transmissions before it is allowed to transmit. The network implementation in

OPNET is shown in Figure 26. It consists of two traffic generators (TGs); one of them representing

the PTP grandmaster (TG1) while the other (TG2) the best effort traffic generator. TG1 sends data

over VLAN ID 10 while TG2 sends data over VLAN ID 20 in a port-based configuration (i.e. the end

stations do not tag the frames).


Page 45 of 78




Figure 26 Network Scenario implemented in Opnet. All network interfaces are 1 GbE

In the first scenario, background traffic is generated as a burst of fifty frames, with an inter-frame

gap of 20 μs and a frame size of 1000 octets. This traffic source may represent either CPRI-type

traffic or C&M traffic. The PS duration is set to 50 µs.

Figure 27 shows the peak and average FDV results for SP and for TAS with different GPs. The results

show that the worst case TAS performance (zero GP) is equivalent to that of SP. This makes sense as

in both cases ongoing transmissions cannot be resolved. The step-like behaviour for the TAS results

is an effect of the resizing of the BES in order to accommodate the GP (i.e. the TW remains

constant). As the GP is increased, there is no change in FDV until the GP “eliminates” the frame from

the burst that is closer (in time) to the GP boundary. This can be seen by observing that the step

changes for the peak FDV occur at GP values, that when added to the corresponding FDVs, are

approximately equal to one background frame serialization.

As the GP is increased, both the average and max FDV with TAS reduce steadily until they reach zero

at a GP of 6 μs, which corresponds to a serialization of a large part (75%) of a background traffic

frame.


Page 46 of 78




Figure 27 Average and peak FDV for the PTP traffic with SP and TAS with different GPs. The background traffic source is

constant frame-rate and constant frame size

The second scenario is similar to the first, but with a varying frame size for the background traffic.

The traffic source is meant to represent functional split traffic, e.g. for fifty user allocations per LTE

subframe (i.e. 50 frames every 1 ms), in a MAC/PHY split (3GPP option 6). Note also that a constant

(or close to constant) number of allocations could arise as a result of employing statistical

multiplexing gains over a trunk link. Two different settings are used: The first follows a normal

distribution with a mean value of 1000 octets and variance of 200 octets (Figure 28). The second is

similar, albeit with an increased variance of 500 octets (Figure 29).

The results show that the peak and average FDV is increased (compared to the first scenario) for

both SP and TAS with zero GP, and approaches the serialization delay of a full background traffic

frame. Furthermore, the peak FDV for the results of Figure 29 reaches zero at a GP that is equivalent

to the serialization delay equivalent to that of a frame with a size equal to the average value of 1000

octets. This is indicative of the dependence of the scheduler performance, with regards to FDV, on

the transmission pattern characteristics of the traffic sources.

Figure 30 is a zoom-in of Figure 29 in the x-axis range from 0 to 1 μs. The small inset shows the

resulting time-stamping error with PTP for the peak FDV values, assuming that this peak FDV is

encountered in one direction of traffic (either downlink or uplink) while there is zero FDV in the

opposite direction.

This result shows the main limitation of SP which, although it can reduce significantly the average

FDV, the peak FDV remains constant and can potentially result in large PTP time-stamping errors

(depending on the size of the background traffic frame). TAS on the other hand looks promising in its

ability to reduce FDV (and thus time-stamping errors) as the GP is increased, or eliminate FDV

entirely when the GP is sufficient to eliminate contention. The drawback in this case is the increased

end-to-end latency, especially if the number of aggregation nodes becomes large.


Page 47 of 78





constant frame-rate with a varying frame size following a normal distribution with mean of 1000 octets and variance of 200

octets


constant frame-rate with a varying frame size following a normal distribution with mean of 1000 octets and variance of 500

octets


Page 48 of 78




Figure 30 Zoom-in in the region of GPs from 0 to 1 μs for the results of Figure 29. The inset shows the worst-case PTP time-

stamping error that would result from the peak FDV values

4.5 High-speed and low-cost transmission links

4.5.1 100 Gbit/s per wavelength

Low-cost requirements are leading to increasing considerations being given to using intensity

modulation and direct detection (IMDD) together with grey optics for 100G trunk lines. To make the

system as simple as possible, optical amplification and dispersion compensation are also not an

option. Consequently, the 1300nm transmission window is required to prevent severe limitations

from chromatic dispersion. For the modulation format, non-return to zero (NRZ) would be the

simplest solution; however this requires expensive high-bandwidth optics and electronics. Hence,

advanced modulation formats combined with DSP (digital signal processing) and FEC (forward error

correction)-encoding are considered here, with potential candidates being PAM-4 (Four-level pulse

amplitude modulation) and DMT (discrete multi-tone).

Figure 31 shows the employed experimental setup. Offline DSP is applied at the transmitter as well

as at the receiver side and requires the use of a high-resolution DAC (digital analogue converter) and

ADC (analogue digital converter). Both operate at a sampling speed of 84 GS/s, have a nominal bit

resolution of 8 bit, and show a 3-dB bandwidth of around 15 GHz and 18 GHz, respectively. At the

transmitter, the differential outputs of the DAC are first amplified by a linear, differential input and

single-ended output modulator driver (MAOM-003115) driving directly the succeeding electro-

absorption modulated laser (EML). The integrated high frequency coils of the driver allow control of

the bias of the EML and the driver. Furthermore, the gain of this driver is adjustable up to a

maximum of 9 dB, and delivers a maximum output swing of 2V. The 3-dB bandwidth of the driver is

around 25 GHz. The EML operates at a fixed temperature of 45°C, and the current of the distributed

feedback laser (DFB) section is set to a maximum of 100mA, which gives the highest linear range and

highest optical output power. At these operating conditions, the transmission wavelength of the


Page 49 of 78




Table 4: DMT System Parameters.

Modulation Formats BPSK to 64QAM

Frame Length (data

symbols) 128

Training Symbols (TS) 4

FFT Length 512

Usable carriers 255

Cyclic Prefix 1/64

Clipping Ratio 15dB

Equalizer 1-tap decision-directed

EML is around 1308nm and the 3-dB bandwidth is measured to be around 27 GHz, but with a

smooth roll-off. The optical link setup consists of conventional standard single-mode fibre (SSMF)

with an attenuation of around 0.32dB/km at 1300 nm, a variable optical attenuator (VOA) with an

integrated power monitor, and a PIN-photodetector (Picometrix PT-40E) integrated with a linear

trans-impedance amplifier (PIN/TIA) with a combined bandwidth of 35 GHz. Finally, the signal is sent

back to the ADC and stored for offline processing.

Figure 31: Experimental transmission setup for 100G.

Discrete Multi-Tone Transmission (DMT)

Discrete multi-tone transmission (DMT) as a special variant of orthogonal frequency division

multiplexing (OFDM) employs the properties of Hermitian symmetry and the IFFT (Inverse Fast

Fourier Transformation) to create a real-valued signal with the frequency spectrum divided into

orthogonal subcarriers. Each subcarrier can be modulated and the power of each subcarrier can be

allocated based on the water filling method. This process is known as bit and power loading (BL, PL)

and enables the effective compensation of channel impairments and component bandwidth

limitations without applying complex

signal processing, e.g. a simple 1-tap

equalizer at the receiver side is efficient.

To apply BL and PL, the transfer function

of the transmission system is first

estimated in terms of the signal-to-noise

ratio (SNR) at the receiver with 16-QAM

(quadrature amplitude modulation)

constellations with equal power on each

subcarrier. Afterwards, Chow's margin-

adaptive bit loading algorithm and

Cioffi's power loading are applied to efficiently distribute the bits and allocate the power. Figure

32b) shows the estimated SNR (signal noise ratio) of the transmission setup for the optical back-to-

back case, together with the corresponding bit and power allocation for a 112-Gb/s DMT signal.

From the estimated SNR, we can also estimate the available bandwidth of the transmission system:


Page 50 of 78




an SNR of 15 dB or more is available up to 25 GHz, while it drops below 0dBm for frequencies above

30 GHz. Figure 32a) illustrates the DSP blocks of the analysed DMT system and Table 3 summarizes

the most important DMT system parameters. To meet the previously mentioned memory

requirements of the DAC and the ADC, a DMT frame consists of 124 data symbols and four training

symbols (i.e. 128 DMT symbols in total), which are used for channel estimation and synchronization.

Figure 32: a) DSP blocks of the implemented DMT system and b) i) estimated SNR per subcarrier at the receiver optical back-

to-back and the applied ii) bit loading and iii) power loading

Figure 33 shows the achieved BER (bit error rate) vs. received optical power (ROP) into the PIN/TIA

for different data rates as well as for different transmission distances. Since DMT allows one to easily

switch between different data rates by loading a different number of bits onto each subcarrier, the

performance of 112 Gb/s, 89.4 Gb/s, 74.7 Gb/s, and 56 Gb/s is investigated and compared. Two

different FEC thresholds are added as a solid and a dashed line, representing the standardized KP4-

FEC (RS(544,514,10)) with a BER-limit of 2E-4 and the continuously-interleaved BCH FEC (CI-

BCH(1020,988)) with a BER-limit of 4.4E-3 [28][29]. Transmitting at a data rate of 112Gb/s, BERs

below the CI-BCH FEC are achieved only for the optical back-to-back (b2b) case, while BERs around

the FEC threshold are achieved in the case of a 10-km transmission distance. At a bias of 1.25V, the

output power of the deployed EML is around 1dBm, which results in a maximum achievable input

power of 5 dBm after 20 km transmission at this wavelength (0.32 dB/km*20 km).

Figure 33: Transmission results of DMT at different data rates and for different transmission distances.


Page 51 of 78




This input power is not sufficient to achieve BERs below the FEC-limits in case of 112 Gb/s. Indeed, a

very similar performance is demonstrated for the different transmission distances up to the

achievable input powers. The performance improves with decreasing bitrate.

PAM-4

Four-level pulse amplitude modulation (PAM-4) encodes 2-bits into one symbol, resulting in a four-

level signal and reducing the transmission bandwidth by a factor of two compared to on-off-keying.

Utilizing Nyquist pulse shaping with a small roll-off factor (β=0.1), the signal bandwidth can be

further reduced, resulting in an electrical bandwidth of around 30 GHz for a 112-Gb/s PAM-4

signal. Figure 34(a) shows the implemented offline DSP blocks for the Nyquist PAM-4 system. A 4-ary

deBruijn sequence of order eight (48=65536 symbols) is used and grey-mapped onto a PAM-4 signal.

Compared to DMT, PAM-4 offers the possibility of easily compensating the nonlinear transfer

function of the modulator by adjusting the levels towards equally spaced power levels after the

modulator. Afterwards, the signal is upsampled to 3 samples/symbol, undergoing raised cosine

shaping in the frequency domain with β=0.1, and downsampled by a factor of two, generating a

112 Gb/s Nyquist-PAM-4 signal with the 84GS/s DAC. Furthermore, digital pre-emphasis

compensates the bandwidth limitations of the DAC and driver, and the signal is quantized into

integer values between 0 and 255, in order to use the full 8-bit resolution of the DAC. Figure 34(c)

illustrates the obtained eye diagrams after the driver amplifier as well as after the EML, exhibiting

the typical over- and undershoots of a Nyquist PAM-4 signal. Furthermore, the power spectrum

density (PSD) of the transmit-signal before the DAC demonstrates the effect of pre-emphasis (grey

area = uncompensated signal (Figure 34(c)). At the receiver, the signal is resampled to 2-fold

oversampling, clock-recovery by means of the Gardner loop, and adaptive symbol-spaced feed-

forward equalization (FFE) is applied to recover the PAM-4 signal. The BER is calculated from the

detected and transmitted bits.

Figure 34: a) DSP blocks of the PAM-4 system, b) digital PSD of transmit signal and c) eye diagram after the EML.

The performance of the pre-equalizer (Tx-FFE) in combination with the applied FFE at the receiver

(Rx-FFE) is evaluated in terms of BER-performance as a first step. In principle, the question of how

many Tx-FFE and Rx-FFE coefficients are necessary for such a transmission scenario in order to

achieve BERs below the desired FEC threshold is answered. Figure 35(b) and (c) depict the BER vs.

ROP results for optical b2b, using 5 and 61 Tx-FFE coefficients, respectively, in combination with a

different number of applied Rx-FFE coefficients. The number after the term "FFE" notates the

number of used coefficients. Again, the previously discussed FEC thresholds are shown as black lines.


Page 52 of 78




Applying 5 Tx-FFE coefficients, up to 31 Rx-FFE coefficients, is necessary to achieve BERs below the

CI-BCH FEC-limit; while 11 Rx-equalizer coefficients are required when 61 Tx-FFE coefficients are

used. The interaction between the number of Tx-FFE and Rx-FFE coefficients is further illustrated as

a contour plot in Figure 35(d), where the achieved BERs for different Tx-FFE/Rx-FFE combinations at

a fixed ROP of 0dBm are shown. Basically, no significant BER improvement is seen with more than 11

Tx-FFE coefficients, while at the receiver at least 21 Rx-FFE coefficients are required. To achieve BERs

below 1E-3 however, more than 40 coefficients for both Tx-FFE and Rx-FFE are necessary.

Figure 35: Optical back-to-back transmission results of 112Gb/s PAM-4 employing different numbers of pre- and post-FFE

coefficients: a) shows the optical eye diagrams obtained directly after the EML using a pre-equalizer of 5 coefficients and 61

coefficients, b) and c) illustrate the BER vs. ROP results using different numbers of post-FFE coefficients and d) depicts the

BER performance for different Tx-FFE/Rx-FFE combinations at an input power of 0dBm.

Based on the results of Figure 35 Tx-FFE and 21 Rx-FFE coefficients offer a good trade-off between

performance and complexity and are used for further evaluation. With these settings, the

performance for optical b2b, 10 km, and 20 km is compared in Figure 36. For optical b2b and for 10-

km SSMF, the results stay well below the CI-BCH FEC threshold, however, the KP4-FEC threshold is

not reached. In addition, the limited output power of the EML prevents the possibility of

transmitting over 20 km. Up to the achievable input power a similar performance of the different

transmission distances is shown.

Figure 36: Transmission results of 112 Gb/s PR PAM-4: a) using different MLSE memory length after the FFE in case of

optical back-to-back transmission

Conclusion


Page 53 of 78




In summary, transmission over 10 km of SSMF was successfully demonstrated for both modulation

formats (DMT and PAM-4) if an HD-FEC of 4.4E-3 is assumed, allowing error free transmission. The

CI-BCH FEC is required for a transmission distance of 10km.

4.5.2 Beyond 100 Gbit/s per wavelength

Transmission rates appreciably beyond 100 Gbit/s per wavelength require the utilization of new

signal generation schemes, together with advanced modulation formats and digital signal

processing. This is especially the case if simple optical components are mandatory, i.e. intensity

modulation and direct detection schemes, to avoid higher costs [30][31]. In order to exploit the

available bandwidth offered by the latest optical components, high-speed digital-to-analogue

converters (DAC and ADC) with analogue bandwidths beyond 30 GHz are required. So far, low-cost

CMOS technology has only achieved bandwidths of around 25 GHz [32]. To overcome the bandwidth

bottleneck, two techniques have been proposed. The first one, spectral up-conversion [33], enables

140 GBd optical BPSK transmission, using electrical up-conversion of multiple sub-bands. In [34] a

combined discrete multi-tone (DMT) and orthogonal frequency division multiplexing (OFDM)

approach with independent sub-bands has achieved 178 Gbit/s. The second technique utilizes a

high-speed analogue multiplexer together with two DACs to generate a wideband signal.

Transmission rates of 214 Gbit/s using pulse amplitude modulation (PAM) or 300 Gbit/s using DMT

have also been shown [35][36].

In the work presented here, the feasibility of the first technique: the utilization of electrical up-

conversion to generate wideband signals, is investigated. The performance of spectral up-conversion

is demonstrated by utilizing a commercially available electrical IQ-mixer, together with single-carrier

(SC) modulation, like PAM and QAM and multi-carrier (MC) modulation like DMT and OFDM. In

addition, soft decision forward error correction (SD-FEC) is used as it enables error free transmission

at higher BERs, at the price of additional overhead [37]. The BER limit is then at 2.7x10-2, as opposed

to the standard hard-decision FEC (HD-FEC) where it is at 3.8x10-3. Altogether, we demonstrate a

record high transmission rate of 200 Gbit/s and 224 Gbit/s, for SC and MC modulation of two sub-

bands for spectral up-conversion in an optical back-to-back experiment.

A. Concept and Setup

Figure 37: System concept for spectral up-conversion utilizing IQ mixers and several independent sub-bands. a) Electrical

spectrum after DAC, b) after IQ-mixing, c) after signal combining, d) after optical modulation.


Page 54 of 78




The general system concept for spectral up-conversion is shown in Figure 37. Each DAC generates a

baseband signal, i.e. the in-phase (I) or quadrature (Q) of a complex signal. The signals are up-

converted onto the local oscillator (LO) frequency by means of an electrical IQ-mixer. After

combining multiple sub-signals, the compound wideband signal drives an optical intensity

modulator. After propagating along the fibre, a direct detection receiver performed optical-to-

electrical (O-E) conversion. The signal is split passively, filtered and the individual sub-bands are

down-converted. Finally, the resulting baseband signals are digitized with an array of ADCs.

Figure 38: Experimental setup for spectral up-conversion based IM/DD transmission links.

Figure 38 shows the experimental setup of the proposed multi-band architecture. Three channels of

an arbitrary waveform generator (AWG), working at 80 GS/s, were used to generate two

independent signal bands: a baseband signal at 0-21 GHz and an up-converted signal at 21-39 GHz.

The up-converted signal was formed with the help of an electrical IQ-mixer, driven by two baseband

signals and a local oscillator (LO) at 30 GHz. The baseband signal and the up-converted signal were

set to a fixed power ratio and combined using a diplexer.

The amplified electrical signal drives a Mach-Zehnder modulator (MZM). The MZM was biased at the

quadrature point to operate as an intensity modulator with an average output power of 4 dBm. The

laser was a distributed feedback laser (DFB) at 1550.5 nm. After optical modulation, the signal was

transmitted over different standard single mode fibre (SSMF) lengths. An optical filter was applied

after the MZM to enable a single sideband (SSB) transmission at fibre lengths beyond 2 km. This was

necessary to reduce the influence of the chromatic dispersion (CD). The receiver consisted of a

wideband photodiode (PD 45 GHz) with an additional optical amplifier (EDFA). After O-E conversion,

the signal was amplified and split by a passive coupler. One signal was low-pass filtered and

recorded by a digital storage oscilloscope (DSO), whereas the second signal was down-converted by

an IQ-mixer, filtered and recorded likewise. The DSO had a sampling rate of 80 GS/s at 30 GHz

bandwidth. All digital signal processing (DSP) was done offline.

For the transmission experiments two different scenarios were considered. In the first scenario the

baseband signal was PAM modulated at a fixed rate of 40 GBd and the up-converted signal was QAM

modulated at a fixed rate of 16 GBd. The DSP for PAM and QAM consisted of pre-equalization to

compensate for the DAC frequency roll-off, a root-raised cosine filter at the Tx and Rx (roll-off 0.1),

and an additional linear finite impulse response (FIR) based equalizer at the receiver, to compensate

for the remaining system frequency response.

The same bandwidth assignment for baseband and up-converted signals has also been used for the

second scenario. Here, a DMT signal was used in the baseband and an up-converted QFDM signal in

the upper band. Both consisted of 256 subcarriers. The modulation order of each subcarrier was


Page 55 of 78




adapted to the channel properties by SNR estimation of training symbols. At the receiver, channel

estimation and correction were performed for every subcarrier.

For the upper signal band with QAM or OFDM modulation, an additional IQ-imbalance correction

was performed, in two steps prior to equalization. First, a 2x2 multiple-input multiple-output

(MIMO) time domain correction of the signal (I, Q) was applied, eliminating the phase difference of

the Tx and Rx Los, and compensating amplitude variations of the signals relative to each other.

Second, a 2x2 MIMO correction of each frequency with respect to the mirror frequency, compared

to the LO, was applied. This was necessary to compensate for frequency-dependent IQ-imbalance of

the mixers, which results in a crosstalk between mirroring frequencies.

B. Results

First, we evaluated the performance of the single carrier approach, i.e. scenario #1 with PAM and

QAM. Figure 39 (a+b) shows the measured BERs for 40 GBd of 4/8-PAM in the lower band and

16 GBd of 8/16/32-QAM in the upper band for different lengths of optical fibre. The following can be

observed for the lower band at 0-21 GHz: 4-PAM is possible for back-to-back (btb), 2 and 10 km

reaches using a HD-FEC; 8-PAM permits only btb transmission using a SD-FEC. For the upper band at

21-39 GHz the observations are: 8-QAM is possible at all fibre lengths using HD-FEC; 16-QAM at btb

and 2 km using HD-FEC; and 32-QAM allows only btb transmission using SD-FEC.

Figure 39 (a+b) Measured BERs for PAM/QAM modulation at different fibre lengths (scenario #1). (c+d) Measured data

rates for different target BERs using DMT/OFDM (scenario #2).

The total transmission rates for scenario #1 below the HD-FEC limit are thereby 144, 144, 128 and

88 Gbit/s for btb, 2, 10 and 20 km of SSMF, respectively. For the SD-FEC limit and btb, 2, 10 and

20 km of SSMF, data rates of 200, 160, 144 and 144 Gbit/s were measured, respectively.

For the MC based approach, i.e. scenario #2, the performance of the DMT and the OFDM signal was

determined for different target BERs and fibre lengths as well. Figure 39 (c) shows the results for the

lower band, i.e. the DMT signal and Figure 39 (d) for the upper band, i.e. the OFDM signal. At the

HD-FEC BER limit a total transmission rate, i.e. the sum of the DMT and OFDM data rates, of 179,

171, 141 and 134 Gbit/s for 0, 2, 10 and 20 km of SSMF respectively, can be determined. For the SD-

FEC case 224, 219, 175 and 167 Gbit/s can be achieved. The relatively strong degradation of the BERs

in both scenarios at fibre lengths of 10 km and 20 km can be primarily attributed to the finite slope

of the optical SSB filter.


Page 56 of 78




Figure 40: (a+b). Measured BERs for PAM and DMT at 80 and 120 Gbit/s for the lower band at different fibre lengths. (c+d)

Measured BERs for QAM and OFDM at 64 and 80 Gbit/s for the upper band at different fibre lengths.

In order to compare the SC and MC approach, we determined which BERs can be achieved for

DMT/OFDM, at the data-rates of the PAM/QAM transmission. This was done by interpolating the

curves of Figure 39 (c+d). Figure 40 (a) shows the results for 80 Gbit/s, which relates to 40 GBd 4-

PAM in the lower band. It can be seen that the determined BERs of the DMT signal are slightly

worse. A similar behaviour can be observed for 8-PAM in Figure 40 (b). These findings are in

accordance with the theoretical considerations, since PAM offers a better performance in terms of

SNR in a system with a relative flat channel, due to the much better peak-to-average-power ratio.

For the upper band, consisting of the QAM or the OFDM signal, a different behaviour can be

observed. Figure 40 (c) shows the BERs for a data rate of 64 Gbit/s, which relates to the 16 GBd of

16-QAM. Here, the performance of the OFDM modulated signal is clearly better. For 16 GBd of 32-

QAM, a similar examination can be made as shown in Figure 40 (d).

The superior performance of the multi-carrier based approach in the upper band can be attributed

to several effects. First, the total system frequency response with a 3-dB bandwidth at around

32 GHz, suits the OFDM signal much better due to the easy adaption in terms of subcarrier bit-

loading. For QAM, the drop in the frequency response results in a noise enhancement due to the

necessary equalizer at the Rx. Secondly, the characteristics of the electrical IQ-mixer at the Tx and

the Rx causes further degradations. Primarily, these are penalties due to the frequency-dependent

IQ-imbalance and port isolation. Both penalties can also be much better addressed with the

subcarrier-based structure of OFDM, which results in lower penalties as compared to QAM.

C. Summary

The proposed electrical up-conversion based technique enables the generation of signals with

bandwidths beyond the state-of-the-art systems. Further, due to the use of mature components,

generally lower costs as compared to a purely optical solution can be achieved. The investigation of

SC and MC modulation formats for the presented concepts showed a better performance for the SC

approach in the lower band (baseband) and a better performance of the MC approach in the upper

band (up-converted band). This can be primarily attributed to the different channel characteristics in

both bands. The achieved transmission rates at the SD-FEC limit are 200 Gbit/s and 224 Gbit/s for

the SC and MC approach, respectively.


Page 57 of 78




4.5.3 Wireless transmission over millimetre wave

The iCIRRUS project foresees an ever-increasing use of wireless (including both mmW and optical

wireless communications) to support both fixed and mobile applications [38][39] leading to the use

of interlinked small cell architectures [40][41][42], as a means to achieving high bandwidth and low-

cost transmission. The bi-directional nature of this traffic will require that these small cell network

elements will need to be capable of supporting backhaul as well as fronthaul applications in order to

satisfy many of the likely applications. As a result, the demands on antenna designs have become

extreme. To some extent, short to medium range meshed topologies can negate the necessity for

very high gain antenna systems of the type seen in point-to-point links, but any device must provide

service up to a nominal distance of around 100 metres to be useful. Also the requirement for

omnidirectional signal capture and a very wide bandwidth remains, if the antenna is to support high

data rate, multi-user applications. Solutions to this issue mostly include the use of multi-element

antenna arrays that rely on the use of a great deal of signal processing to implement complex beam

steering, beam forming and MIMO techniques. The device described here embraces a different

approach in that it is inherently capable of operating over a 360 degree capture range whilst

maintaining a gain of 13dBi at 57 GHz, which is then maintained at over 12 dBi at 64 GHz, the upper

limit of the IEEE802.11ad frequency allocation. Uniquely, this has been accomplished by treating the

circular radiator as a horn antenna and concentrating the resulting radiation horizontally to form a

flattened disk-like radiation pattern with a full-width half-maximum (FWHM), far field, frequency

dependent divergence angle of between 6.5 and 8.2 degrees. Unlike many multi-element designs

this radiation pattern is inherently stable across the IEEE802.11ad frequency band, and so

unaffected by phase-related beam anomalies (squint). This work represents further development

and resulting practical embodiment of the device recently described in [43]. The unique capability of

this device is its suitability to be deployed in a mesh network communicating with a number of user

scenarios simultaneously. Coupled with its high gain properties that enable a potential 100 metre

operating range, this device represents a useful bridge between the backhaul, mesh cell, mobile and

fixed users of mmW wireless systems as exemplified by the iCIRRUS architecture.

Figure 41: Antenna model graphic showing conical top and bottom sections with integral matching rings, antenna feeding

arrangements and modified support pillars, with surface current plot showing dissipation effect of the matching rings.

D. Antenna Model Structure / Design Philosophy

In order to accomplish the omnidirectional and wide bandwidth requirements, features related to a

bi-conical design have been adopted. A high gain performance was achieved by treating the device

as a horn structure, so using a monopole feed rather than the dipole-like feeding arrangement


Page 58 of 78




normally associated with conventional bi-conical designs [44][45][46]. Here it can be seen that two

opposing metallic conical structures provide the basic propagating mechanism. An integral part of

these are the matching rings shown on the upper and lower outer aspects of the conical sections

[47][48][49]. These rings also provide convenient anchoring points for the Perspex (Ɛr 3.4) support

pillars used to separate the two cones, with these being spaced away from the main body of the

structure as well as being reduced in thickness to 3mm to minimize the predicted “shadowing”

effect of the pillars on the radiation pattern. Having dimensions of 5 mm x 10 mm, the matching

rings provide a predicted increase in gain of between 1.5 dB and 2.2 dB across the 57-GHz to 64-GHz

range [43]. As can be seen in Figure 41 this was achieved by the effective dissipation of disruptive

surface currents traveling backwards across the outer face of the structure.

The horn feed is formed from a section of Huber and Suhner SR 86 semi rigid coaxial cable with a

section of the central conductor exposed to form a central monopole radiator. It was determined

that the optimal length for the monopole launch was 1.15 mm. This was based on a dimension that

gave a maximum gain / bandwidth product. This feeding arrangement was positioned at the centre

of the lower conical structure around which is a flat section that, in conjunction with the reciprocal

arrangement in the top cone formed a symmetrical quarter wavelength waveguide launch. With a

central operating frequency of around 60 GHz this central flat section was assigned a diameter of 2.5

mm so as to allow for a quarter wavelength dimension in all directions. The height of the waveguide

section was initially taken from that of WR15 waveguide as 1.88 mm. These dimensions allow for an

excellent broadband performance, good impedance match and desirable propagation

characteristics. Finally, dimensions of the conical disk section were determined as having a radius of

75 mm (15 wavelengths at 60 GHz) with an empirically determined critical mouth dimension of 21.94

mm, so giving a 15.5 degree flare angle after the waveguide section.

Figure 42: Antenna assembly, showing all component parts and the support pillars and probe clamping arrangement.

As can be seen in Figure 42, the body of the new device was fabricated in aluminium, with the

modified Perspex support pillars and feed probe clamp also clearly visible. A commercially available

MMPX connectorized semi-rigid cable was used for the feed probe, with a specified -16dB return

loss. The feed and clamp arrangement was positioned in a closely fitting centrally drilled hole in one


Page 59 of 78




of the aluminium conical sections so as to ensure contact with the outer casing of the semi-rigid co-

ax. Using an Anritsu 37397D Vector Network Analyser (VNA) this was then carefully positioned so as

to give the best S11 measurements. The newly developed clamping arrangement then locked the co-

ax this in position.

E. Antenna Performance Measurements

With the VNA calibrated across a 30-GHz to 67-GHz range, measurements were conducted with

Figure 43 showing the resulting S11 return loss plot with markers at the same frequencies as for the

simulation [6]. Overall this indicated good agreement with the modelled result across the frequency

range. With return losses ranging from -16.8 dB at 64 GHz falling to -28.9 dB at 60 GHz good coupling

across the frequency range was expected. Corresponding input impedance measurements were

taken as shown in the Smith chart in Figure 44. This shows a well-matched device with similar

properties to those previously predicted. The impedances at the four spot frequencies are closely

grouped around the ideal 50 Ω point; the only deviation being at 61 GHz that gave a measurement of

60.5 Ω.

Figure 43: Measured S11 return loss over the IEEE 802.11ad frequency range with markers at four spot frequencies.

Figure 44: Corresponding Smith chart showing measured antenna feed impedance values at four spot frequencies


Page 60 of 78




The real life gain and polarization characteristics of the antenna have also been assessed. Using the

Anritsu 37397D VNA and two Steatite QSH-SL-50-75-V-20 linearly polarized 20-dB gain horn

antennas set up at a distance of 1 metre, a reference channel was established. Although in a non-

anechoic environment this showed an essentially flat channel up to 65 GHz with a low-end -3 dB

point at 42.75 GHz with a steep roll off thereafter. One of the reference antennas was then removed

and substituted with the antenna to be evaluated. As can be seen in the co-polar trace in Figure 44

the antenna under test mimicked the frequency response of the reference horn antenna. Evident is

the low-end roll-off being the same as for the reference antenna used as the other control

(reference) antenna. It can also be seen that the gain difference of around 5 dB to that of the horn is

commensurate with that expected. The amplitude of the received signals was then measured at the

four frequencies shown in Table 5 and the antenna gains calculated taking into account the free

space path loss (FSPL), horn antenna gain and connector losses due to the MMPX socket. It can be

seen that apart from a gain measurement that is often around 1 dB higher than that predicted, there

is again good agreement with the predicted gains at all four spot frequencies; this being maintained

around the device periphery. Maximum gain variations of 0.43 dB at 57 GHz, 0.83 dB at 60 GHz, 0.69

dB at 61 GHz and 0.58 dB at 64 GHz were recorded. With this established in conjunction with the

other validation evidence gathered, the predicted radiation patterns appear to be credible.

Table 5: Comparison of Peripheral Measured Antenna Gain

As discussed earlier, this device was mostly aimed towards high capacity wireless mesh network

applications. In such situations a polarization diversity function may also be desirable for functions

that require frequency reuse. With this in mind, measurements were conducted to ascertain the

polarization isolation characteristics of the antenna. Again, using the same measurement setup the

linearly polarized horn antenna was rotated through 90 degrees and a plot was taken of the received

signal levels. This can be seen in the bottom trace in Figure 45. This determined that polarization

isolation ranges from 25.15 dB at 64 GHz rising to 29.14 dB at 57 GHz, which is adequate for most

applications. Also apparent from Figure 45 is the lower cut-off frequency limit measured, which is

that of the test horn antenna, while the upper limit is that of the VNA.

Frequency Reference

0 degrees

90 degrees

180 degrees

270 degrees

57 GHz 13.24 dBi 13.18 dBi 12.81 dBi 13.12 dBi





Page 61 of 78




Figure 45: Comparison of co-polar and cross-polar transmission characteristics between 30 GHz and 65 GHz, also illustrating

the wide bandwidth capability of the omnidirectional antenna.

C. Proof of Concept

To ascertain the novel omnidirectional aspects of this design a four-channel transmission system was

implemented using a variety of antenna types and frequencies simultaneously. This was designed to

represent the operation of the device in a mesh network setting where both fixed and mobile users

could be accommodated. These were positioned to cover 360-degree operation around the centrally

positioned omnidirectional antenna, which was connected to a spectrum analyser. The received

signals were captured as shown in Figure 46. Channels 1 & 2 represents a wide bandwidth user

requirement such as that of backhaul usage and were served by horn antennas. Channel 3

represents a mobile user and used a small coaxial slot array antenna, and channel 4 used another

identical omnidirectional antenna and so represents the mesh deployment.

Figure 46: Showing four channel capability over an operating angle of 360 degrees, demonstrating backhaul, mobile, and

mesh deployment usages.

D. Summary

The proposed linearly polarized omnidirectional antenna design described here operates without

the need for beam steering and possesses a gain in excess of 12 dBi gain over a frequency range

between 57 GHz and 64 GHz. A set of antenna measurements was additionally carried out to


Page 62 of 78




demonstrate that the radiation patterns are also reliable. The device discussed here has been shown

to possess novel high gain, and omnidirectional properties that will enable its deployment in a mesh

femtocell environment.

4.6 Test and performance monitoring This deliverable develops aspects of the “SLA and SON Concept for iCIRRUS” set out in deliverable

D3.3 towards practical experimentation in WP5, taking into account the constraints and

opportunities likely to be encountered in a putative real-world deployment.

The goal of SON in iCIRRUS, whether the SON objective is configuration, live improvement or

healing, is the creation of solutions that jointly account for the constraints of the fronthaul network

and the radio access network (RAN).

Changes or adaptation to the constraints in both fronthaul and RAN domains that delimit the

configuration space for optimisation are characterised by different timescales. Examples of changes

that occur on a slow timescale of the order of months to years include: deployment of new optical

fibre, deployment of new base stations, and acquisition of new radio spectrum. In these cases the

slow speed of change is due to the need for manual intervention and potentially even regulation.

Whereas, other changes may occur on a timescale of seconds to hours: for example, modification to

QoS parameters on a link, changes to transport path routing, or change to the maximum permitted

transmit power of a base station.

Consequently, in some scenarios, the test and performance monitoring related to a constraint may

be represented by semi-static measurements, appropriate for storing in a database of

configurations, whereas another constraint may be represented by “real-time” measurements.

The following Table summarises the timescales over which exemplary principal elements of the

fronthaul network may change, and the type of measurements that may be performed on a one-off

basis to characterise the property of that element, and may be stored in a database; and those

measurements that that are required to be conducted in an on-going manner to characterise the

real-time performance or failure of that element.


Page 63 of 78




Table 6: Characteristics of measurements in fronthaul domain

Measurement

object

Configuration

space

Changes Timescale Slow / one-off

measurements

On-going

measurements

Topology of

physical access

network

ducts, fibres,

splitters, MMW

links, copper

add fibres / add

PON technology/

change optical

split / change OLT-

OLU devices

Add MMW links…

Add copper links …

Months-

years

• Validate

inventory

• Discover

topology

• Check

connectivity

• (Audit 3rd

party

provider)

• Detect failure

(alarm)

Logical access

network

Link and

redundancy

routing

Change routing Seconds-

months

• Check

connectivity

• Discover

routing

• Detect failure

(alarm)

Logical access

network

Resource

allocation / QoS

Change resource

allocation

Milli-

seconds (eg

NGPON2)

-months (eg

3rd party)

• Service

activation

• Real-time

performance

measurement

The following Table performs the same task as that above, but for the exemplary principle elements

of the RAN network.


Page 64 of 78




Table 7: Characteristics of measurements in RAN domain

Measurement

object

Configuration

space

Changes Timescale Slow / one-off

measurements

On-going

measurements

Set of RAN

elements

Base stations &

antenna

systems,

ducts, fibres,

splitters, MMW

links, copper

Add base stations,

Add antenna

systems (support

MIMO, DAS)

Months-

years

• Validate

inventory

• Discover

geolocation

• Check

connectivity

• (Audit 3rd

party

provider)

• Detect failure

(alarm)

Cell level time

configuration

Time/ frequency

alignment

Frequency

synchronisation,

time of day

alignment

Months-

years

• Check synch

operation

• Detect out of

limit operation

(alarm)

Cell level

mechanical

configuration

Antenna height,

antenna

direction,

Change antenna

height, tilt,

antenna azimuth,

Days-

months

• Audit settings

• Check

connectivity (if

appropriate)

• Detect effect

of changes on

RF

performance

(RF level,

quality,

dropped call

etc)

Cell level

parameter

configuration

Base station

activation,

Transmit power,

inter-cell

synchronisation

Activate/

deactivate base

stations / small

cells, change max

transmit power,

change control

channel power

Minutes-

days

• Audit settings • Detect effect

of changes on

RF

• Detect failure

(alarm)

Inter-cell

parameter

configuration

Permitted CoMP

set size

Number of cells

allowed in CoMP

set

Minutes-

days

• Inter-cell

transport/

fronthaul

bandwidth

• Detect CoMP

performance

benefit

• Monitor CoMP

bandwidth

requirement

• Detect out of

bandwidth

(alarm)

The joint configuration space resulting from the allowed changes to the fronthaul and to the RAN on

a variety of different timescales gives rise to a large number of SON scenarios that may be exploited.

These are reviewed and summarised in the Section 5.2 example use-cases, that may be investigated

in WP5.


Page 65 of 78




5 Further architectural considerations

5.1 Evolved digital fronthaul vs. analogue fronthaul The evolved digital fronthaul concept investigated in D3.1 [50] and D3.2 [51] has been evaluated in

simulations [52] and experiments [53] against an analogue approach that promises similar

opportunity for reduction of the optical bandwidth on the fronthaul. In this section, the

experimental results are presented.

5.1.1 Reference system

The reference system for the experiments is a custom real-time 5 Gbit/s millimeter-wave (mm-wave)

transceiver as described in [54]. This system concept provides the background for a 10 Gbit/s

system; however not in real-time. The data rate of 10 Gbit/s is chosen with respect to 10 Gigabit

Ethernet, which is widely used in access networks and also considered for realization of the evolved

digital fronthaul [50].

The extended mm-wave system transports four channels at a data rate of 2.5 Gbit/s each. A π/4-

shift DQPSK (π/4-DQPSK) modulation is used with a symbol rate of 1.25 GBd per channel. To

minimize bandwidth usage, a channel spacing of 1.5 GHz together with pulse shaping and bandpass

filtering is applied. The digital signal processing (DSP) at the transmitter (Tx) and receiver (Rx) is

realized in Matlab. The following steps are taken in the Tx separately for each channel: first the user

data is encoded for forward error correction (FEC), then modulated using the π/4-SDQPSK

modulation scheme, and filtered using a root-raised cosine (RRC) filter with a roll-off factor of 0.25

for pulse shaping. In this step, the sampling rate of the baseband signals is increased to 2.5 GS/s.

At the Rx, a frequency domain equalizer is applied to each baseband channel along with another RRC

filter. In this step, the oversampling is removed, so that the sampling rate is 1.25 GS/s again.

Afterwards, previously to demodulation and FEC, the error vector magnitude (EVM) of the π/4-

SDQPSK symbols is estimated.

In this reference system both fronthaul concepts are integrated between the π/4-SDQPSK

transmitter and receiver, and are experimentally evaluated.

5.1.2 Analogue Fronthaul

The analogue fronthaul is modelled by up-converting the four analogue baseband signals of the

reference system to separate intermediate frequencies (IFs) and transmitting them on a common

optical carrier as shown in Figure 47. After baseband processing and oversampling to 20 GS/s, each

Figure 47: Signal processing for analogue fronthaul


Page 66 of 78




baseband signal is up-converted to its respective IF, and then added onto a common signal vector. A

random time offset is also added to every baseband signal in this step in order to avoid constructive

interference between the training sequences used for synchronization. The IFs used are 1.25 GHz,

2.75 GHz, 4.25 GHz and 5.75 GHz for the four channels. The resulting signal vector is exported and

digital-analogue converted with an arbitrary waveform generator (AWG). After transmission through

the optical fronthaul link (please refer to section 2.3), the signal is recorded with a real-time

oscilloscope at 80 GS/s. The Rx DSP consisted of band pass filtering, to separate the four channels,

down-conversion to baseband and RRC filtering. Afterwards, the signals are further processed as

specified for the reference system.

5.1.3 Evolved Digital Fronthaul

The signal processing for the digital fronthaul is shown in Figure 49. Compared to the analogue

approach, here the fronthaul Tx is located after the FEC encoding. The FEC-encoded data from all

channels is serialized as one data vector and OOK NRZ modulated at 10 Gbit/s. Afterwards the signal

is up-sampled to 20 GS/s and a training sequence for the Rx equalizer is inserted. After digital-

analogue conversion with the AWG, the signal was transmitted over the optical link (please refer to

section 2.3). At the Rx side, a simple finite impulse response (FIR) equalizer is applied, before down-

sampling to the symbol rate. After hard decision, the FEC encoded user data is recovered. To

evaluate the link performance, it is sufficient to determine the BER after the decider (dashed line),

since any accumulated errors are evenly distributed on the four wireless channels in the subsequent

signal processing.

5.1.4 Optical Link

The fronthaul signals are transported over an intensity modulation / direct detection (IM/DD) link.

Figure 48 shows a block diagram of the setup: the beam from a DFB laser with a wavelength of

1550.52 nm is first passed through a polarization filter and then modulated using a Mach-Zehnder

modulator (MZM). The MZM is driven using the output signal from the AWG, which is previously

Figure 49: Signal processing for evolved digital fronthaul

Figure 48: Experimental setup (optical link)


Page 67 of 78




low-pass filtered and passed through an amplifier. The MZM is biased at the quadrature point to act

as a simple intensity modulator. The launch power of the signal was around 2 dBm. After the MZM

the signal is transmitted over various lengths of SSMF. Afterwards an Erbium doped fibre amplifier

(EDFA) with a constant output power was used to achieve a sufficient signal level at the Rx. To vary

the optical power, an additional variable optical attenuator (VOA) was applied in front of the Rx. The

Rx consisted of a PIN photodiode with an integrated trans-impedance amplifier (TIA). After electrical

amplification, the signal was recorded using a real time oscilloscope at 80 GS/s.

The fibre used in the experiments had a dispersion coefficient of ~17 ps/(nm∙km) and an attenuation

of ~0.2 dB/km. Three patches of 25.5 km length each were used to achieve total fibre lengths of

25.5, 51.0, and 76.5 km. For a fronthaul scenario in the access network, 25.5 km are considered as

sufficient. The higher distances of 51.0 and 76.5 km address, e.g. large scale DU pooling and

fronthaul transmission in the metro network.

5.1.5 Results

In the following, the experimental results for both fronthaul solutions are presented.

Figure 50 shows the EVM over the received power for optical back-to-back (btb), 25.5, 51.0, and

76.5 km of SSMF. The solid lines depict the average EVMs over all four channels, the dashed lines

represent the maximum EVMs observed at any channel.

It can be seen that the penalty of 25.5 km fibre compared to btb is relatively low for higher received

power, and virtually inexistent at lower power. However, even at 0 dBm received power, an EVM

floor of around -22 dB for btb and -20 dB for 25.5 km occurs, due to general system limitations.

While this is not problematic for the radio link considered here to operate at low SNR, it is difficult

for the analogue fronthaul when the radio link is operated at high spectral efficiency. After 51 km

and 76 km of fibre, a clear impact of the CD can be seen. At 0 dBm received power the average EVM

is -16 dB after 51 km, and -12 dB after 76 km. More notably, the maximum EVM increases drastically

with distance: it is at -14 dB for 51 km, and at -7 dB for 76 km. As a reference, the EVM limit stated

for QPSK by the LTE standard is 17.5%, or around -15 dB [55], which is only fulfilled on all channels

for btb and 25.5 km, at a received power of at least -8 dBm.

EVM increasing with fibre length can be explained by the effects of chromatic dispersion, which

accumulates over fibre distance and causes signal fading at certain frequency ranges, starting at high

frequencies and moving to lower frequencies with increasing distance [56].


Page 68 of 78




Figure 50: Estimated EVM for analogue fronthaul over Rx power and for different fibre lengths

Figure 51: Estimated BER for analogue fronthaul over Rx

power and for different fibre lengths

Figure 52: Estimated BER for digital fronthaul over Rx

power and for different fibre lengths

This also explains the discrepancy between average and maximum EVM in the analogue system

towards longer distances: since dispersion induced fading is frequency selective, it only affects some

of the channels at a time.

Figure 52 shows the BERs associated with the fronthaul transmission, estimated from the EVM

values of each channel and then averaged over all channels. For 25.5 km of fibre, only minor

penalties compared to btb exist, even at very low BERs. In detail this is 1 dB (-7 vs -6 dBm) at a BER

of 10-12. With 51 km fibre, the estimated BER is at least ~10-7 for all power levels; and with 76 km

always above 10-3. Below a BER of 10-4 all errors can be corrected by the applied FEC algorithm

(Reed-Solomon 255/239). This condition is fulfilled for a received power of at least -10 dBm for btb

and 25.5 km, of -8 dBm at 51 km, and not at all at 76 km.

Figure 51 shows the BER of the received NRZ signal on the digital fronthaul over received power for

the different fibre lengths. For btb and 25.5 km, the performance is similar and unproblematic for

sufficient received power: at -10 dBm, the BER is below 10-12 for both distances. At 51 km fibre

length, the BER ranges between 10-6 and 10-5 for powers greater than -9 dBm. An estimated BER of


Page 69 of 78




below 10-4 is achieved for a received power greater than -14 dBm for btb and 25.5 km, and -12 dBm

for 51 km. Again, transmission over 76 km produces errors for all received power levels.

5.1.6 Comparison

Both fronthaul solutions perform reasonably well over distances of up to 25.5 km for sufficiently

high received power levels. Regarding the fronthaul alone, an acceptable error rate can be achieved

for a higher received power at 51 km, too; however, this distance appears to be the limit of the

system in both cases.

The necessary Rx power below the BER threshold of 10-4 is clearly lower for the digital fronthaul as

compared to the analogue fronthaul. In detail, this is for digital vs. analogue: -15 vs. -11.5 dBm for

btb; -14.5 vs. -11 dBm for 25 km, and -12.5 vs. -8.5 dBm at 51 km. This can be explained by the more

robust modulation format of OOK compared to π/4-SDQPSK. Especially, the far better peak-to-

average power ratio (PAPR) of OOK compared to analogue signals results in a significant gain.

For the distance of 51 km, the BER at high Rx power levels is better for the analogue fronthaul. This

can be explained by the smaller bandwidth of the analogue fronthaul signal as compared to the

digital signal, which results in smaller penalties due to the CD. At 76.5 km, both fronthaul concepts

experience severe penalties due to the CD, so only minor BER difference can be observed at high

power levels.

5.1.7 Conclusions

The findings outlined above show that despite the larger signal bandwidth, the NRZ-based digital

solution is more resilient, especially at the low received powers that might occur in a low-cost

system without optical amplifiers on the Rx side. The analogue fronthaul significantly affects the

transported baseband signal, especially for low received power levels. Both solutions, however,

show little dependency on fibre length regarding dispersion effects, at least up to 25.5 km distance.

In metro networks with covered distances of up to 100 km, the tested solutions do not hold up. Both

reach their limitations at about 50 km distance.

5.2 SON use cases As described in [57], SON includes families of solutions that address self-configuration, self-

optimisation and self-healing. The deliverable presented a summary of SON use cases and provided

a table that listed some exemplary use-case experiments to be used as a tool to help design use-case

experiments for the testbed phase of the iCIRRUS project in WP5. In this Section, we examine more

of the detail of the SON experiments that may be conducted the testbeds, considering the greater

knowledge that is available on the physical constraints of the testbeds, and also sources of data that

are expected to be available which may be analysed to construct a “real-world” background scenario

for the SON experiments.

There are certain prerequisites for experiments in the SON use-cases, including for example, that the

system exhibits stable operation in a static non-impaired environment. Further, that impairments

can be added into the radio and fronthaul domains in a controlled manner, and that the testbed

environment itself can be “placed” in the context of a more realistic network deployment. Then,

with a knowledge of the fronthaul topology and the possible RAN modes, the system determines


Page 70 of 78




configuration options. Whether a configuration is changed dynamically depends on which type of

SON solution is under investigation. Additionally, depending on the capability of the testbed, a

dynamic reconfiguration can be emulated piecewise as a set of static configurations.

A further input that is required to determine an optimal network configuration (and as a trigger to

reconfiguration) is a knowledge of the variation of the traffic demand across the network. The

extent to which traffic demand is time varying and unevenly distributed across the cells of the

networks affords opportunities

5.2.1 Representation of the radio environment

The mobile radio environment is characterised by sets of wanted and interfering signals that

experience time varying multipath propagation. In contrast, the testbeds in iCIRRUS have few

mobile devices and fewer cells.

The simplest way to approximate radio channel impairment in a controlled measurable manner is to

use a variable attenuator to statically attenuate a wanted signal. Interference may be represented

by the noise floor, or an interfering signal maybe injected. An alternative possibility is to set a lower

power level in the base station and mobile devices using configurable software parameters.

More accurate emulation may be achieved by use of a radio channel fader that creates multi-path

interference; the approach can be extended for MIMO and CoMP channel emulation. However, the

cost and complexity of a radio channel fader is outside of the iCIRRUS scope, even for the simple

case.

Additionally, radio channel measurement data from real networks can be analysed to approximate a

realistic radio “context” in which to operate the testbed. This would include, for example, a set of

signal power and quality values representative of operation across a cell. Placing the measurements

taken from the testbeds in a context representative of a real-world deployment allows the testbed

results to be extrapolated to estimate real-world performance. Adjustment to the performance to

represent degradation in fading channels (rather than static channels) is also required, which can be

approximated based on published data.

5.2.2 Representation of the fronthaul environment

The fronthaul network in a real network deployment is characterised by a set of possible routes over

physical infrastructure, which may include optical fibre, for example, as point-to-point links or using

optical splitters for multi-point transmission, copper, and millimetre-wave radio links. The

consequent fronthaul topology and the reconfiguration and redundancy opportunities define the set

of paths that may be configured and reconfigured. Each link over a path through the network may

be characterised by a set of KPIs as discussed in earlier deliverables, including: packet delay, delay

variation, packet loss, and from time to time by link outage. The traffic flows on the links may be

associated with Class of Service parameters that dynamic network elements such as switches

attempt to assure. In contrast, the testbeds in iCIRRUS are characterised by simple point-to-point

links.


Page 71 of 78




The simplest way to approximate impairments in fronthaul network impairment in a controlled

manner is to use an impairment generator that can impose variable latency, packet ordering, packet

loss and error injection.

Additionally, transport performance data from real networks can be analysed to approximate a

realistic fronthaul network context in which to operate the testbed. This would include, for example,

typical topologies, passive optical network (PON) split depth, and short and long term link

performance KPIs. Placing the measurements taken from the testbeds in a context representative of

a real-world fronthaul deployment allows the testbed results to be extrapolated to estimate real-

world performance.


Page 72 of 78




6 Conclusions An update of the iCIRRUS fronthaul architecture in terms of requirements, KPIs and building blocks

has been given in this deliverable, in order to meet the requirements of 5G (and beyond) mobile

networks in a cost effective and energy-efficient manner. The intelligent iCIRRUS fronthaul

architecture combines Ethernet as a transport protocol with the modification to the functional split

to reduce data rates in the fronthaul, while making statistical multiplexing gains possible, and thus

allows a more efficient use of the network resources. The intelligence in the proposed fronthaul will

assist in the provision of further services such as network-assisted D2D communications and mobile

cloud networking at the network edge. The various technical building blocks of the proposed

fronthaul solutions have been described and characterized in this deliverable. The investigations of

the time sensitive Ethernet technologies have demonstrated that the fronthaul latency requirements

can be met. Experimental evaluations of different modulation formats for data rates of 100G and

beyond have shown, that transmission lengths of 10 km can be achieved, which is within the

required range for a fronthaul application.

The architecture of the fronthaul network was reviewed in this deliverable. The work of WP5,

namely the different test scenarios, will also be based on the findings of WP3 and WP4 in the final

phase of iCIRRUS.


Page 73 of 78




References

[1] J-I Kani, J. Terada, K-I Suzuki, A. Otaka, “Solutions for future mobile fronthaul and access-network

convergence”, Journal of Lightwave Technology (2016) [early access article]

[2] P. Chanclou, L. Anet Neto, K. Grzybowski, Z. Tayq, F. Saliou, N. Genay, “Mobile Fronthaul Architecture

and Technologies: a RAN Equipment Assessment”, Th4B.1, OFC 2017, 19-23 March, 2017, USA

[3] Metro Ethernet Forum: MEF 22.1.1, Mobile Backhaul Phase 2, Amendment 1, July 2014

[4] A. Pizzinat, P. Chanclou, F. Saliou, T. Diallo, “Things you should know about fronthaul”, Journal of

Lightwave Technology, 1st March 2015, vol. 33, Issue: 5, 1077-1083 (2015).

[5] N. J. Gomes, P. Chanclou, P. Turnbull, A. Magee, V. Jungnickel “Fronthaul evolution: From CPRI to

Ethernet”, Optical Fibre Technology, Volume 26, Part A, December 2015, Pages 50–58

[6] M. Garyantes, “Virtual Radio Access Network opportunities and challenges”, 36th IEEE Sarnoff

Symposium, pp. 24-28 (2015)

[7] Z. Tayq, L. Anet Neto, B. Le Guyader, A. De Lannoy, M. Chouaref, C. Aupetit-Berthelemot, M.

Nelamangala Anjanappa, Si Nguyen, K. Chowdhury, P. Chanclou “Real Time Demonstration of the

Transport of Ethernet Fronthaul based on vRAN in Optical Access Networks”,Th3A.2, OFC 2017, 19-23

March, 2017, USA

[8] Common Public Radio Interface (CPRI): CPRI Specification V7.0 (2015-10-09),

- Open Base Station Architecture Initiative (OBSAI) : System specification v2.0

- Open Radio equipment Interface (ORI) : ETSI GS ORI 001 V4.1.1 (2014-10)

[9] Uwe Dötsch, Mark Doll, Hans-Peter Mayer, Frank Schaich, Jonathan Segel and Philippe Sehier,

“Quantitative Analysis of Split Base Station Processing and Determination of Advantageous Architectures

for LTE”, Bell Labs Technical Journal, Special Issue: General Papers, Volume 18, Issue 1, pages 105–128,

June 2013

[10] 3GPPP TR 38.801, Technical Specification Group Radio Access Network; Study on New Radio Access

Technology; Radio Access Architecture and Interfaces; (Release 14)

[11] Small Cell Forum, “Small Cell Virtualization Functional Splits and Use Cases,” January 2016

[12] Y.-S. Shiu, S. Y. Chang, H.-C. Wu, S. C.-H. Huang, and H.-H. Chen, “Physical layer security in wireless

networks: a tutorial," IEEE Wireless Communications, vol. 18, no. 2, pp. 6674, 2011.

[13] A. Biryukov, A. Shamir, and D. Wagner, “Real time cryptanalysis of a5/1 on a pc," in International

Workshop on Fast Software Encryption, pp. 1-18, Springer, 2000.

[14] D. Wagner, B. Schneier, and J. Kelsey, “Cryptanalysis of the cellular message encryption algorithm," in

Annual International Cryptology Conference, pp. 526-537, Springer, 1997.

[15] C. Adams and S. Lloyd, Understanding PKI: concepts, standards, and deployment considerations.

Addison-Wesley Professional, 2003.

[16] T. Austin, PKI: A Wiley Tech Brief. John Wiley & Sons, Inc., 2000.

[17] A. D. Wyner, “The wire-tap channel," The bell system technical journal, vol. 54, no. 8, pp. 1355-1387,

1975.

[18] T.-H. Chang, Y.-W. P. Hong, and C.-Y. Chi, “Training signal design for discriminatory channel

estimation," in Global Telecommunications Conference, 2009. GLOBECOM 2009. IEEE, pp. 1-6, IEEE,

2009.

[19] I. Csiszar and J. Korner, “Broadcast channels with confidential messages," IEEE transactions on

information theory, vol. 24, no. 3, pp. 339-348, 1978.

[20] “Time-Sensitive Networking for Fronthaul,” IEEE Standard P802.1CM [Online]. Available:

http://www.ieee802.org/1/pages/802.1cm.html

[21] M. K. Al-Hares, P. Assimakopoulos, S. Hill, and N. J. Gomes “The Effect of Different Queuing Regime

On a Switched Ethernet Fronthaul,” in IEEE Int. Conf. on Transp .Optic.Net. (ICTON), Trento, Italy,

2016, pp. 1-4

[22] “Frame Preemption," IEEE standard 802.1Qbu [Online]. Available:

http://www.ieee802.org/1/pages/802.1bu.html.

[23] "Enhancements for Scheduled Traffic," IEEE standard 802.1Qbv [Online]. Available:

http://www.ieee802.org/1/pages/802.1bv.html

[24] T. Wan and P. Ashwood-Smith, “A Performance Study of CPRI over Ethernet with IEEE 802.1Qbu and

802.1Qbv Enhancements,” in Global Commun. Conf. (GLOBECOM), San Diego, CA, 2015, pp. 1-6


Page 74 of 78




[25] M. K. Al-Hares, P. Assimakopoulos, D. Muench and N. J. Gomes, “Scheduling in an Ethernet Fronthaul

Network,” in European Conf. on Networks and Commun. (EUCNC), Oulu, Finland, 2017. Accepted for

publication

[26] L. Fernandez del Rosal, V. Jungnickel, D. Muench, H. Griesser, P. Assimakopoulos, N. Gomes, Y. Kai, H.

Thomas, M. Parker, C. Magurawalage, K. Wang, P. Chanclou and V. D, “D3.2 iCirrus - Preliminary

Fronthaul Architecture Proposal,” 2016.

[27] R. Veisllari, S. Bjornstad, J. P. Braute, K. Bozorgebrahimi and C. Raffaelli, "Field-Trial Demonstration of

Cost Efficient Sub-wavelength Service Through Integrated Packet Circuit Hybrid Network," Journal of

Optical Communications and Networking, 2015.

[28] “IEEE P802.3bs/D1.2 Draft Standard for Ethernet Amendment: Media Access Control Parameters,

Physical Layers and Management Parameters for 400 Gb/s Operation,” pp. 1–269, 2016. [Online].

Available: http://www.ieee802.org/3/bs/

[29] M. Scholten, T. Coe, and J. Dillard, “Continuously-interleaved BCH (CI-BCH) FEC delivers best in class

NECG for 40G and 100G metro applications,” in 2010 Conference on Optical Fiber Communication

(OFC/NFOEC), collocated National Fiber Optic Engineers Conference, 2010, paper NTuB3.

[30] J. Wei et al., “400 Gigabit Ethernet Using Advanced Modulation formats: Performance, Complexity, and

Power Dissipation, “Commun. Mag. 53(2), 182 (2015)

[31] W. Kobayashi et al., “25 Gbaud/s 4-PAM (50 Gbit/s) modulation and 10 km SMF transmission with 1.3

µm InGaAlAs-based DML,” Electron. Lett., 2014, 50, (4), pp. 299-300

[32] H. Huang et al., “An 8-bit 100 GS/s Distributed DAC in 28 nm CMOS for Optical Communications,”

Trans. Microw. Theory. 63(4), 1211 (2015)

[33] X. Chen et al., “All-electronic 100-GHz Bandwidth Digital-to-Analog Converter Generating PAM Signals

up to 190-GBaud,’ Proc. OFC, Th5C.5 (2016)

[34] C. Kottke et al., “178 Gbit/s Short-Range Optical Transmission Based on OFDM, Electrical Up-

Conversion and Signal Combining,” Proc. ECOC, W.4.P1.SC5 (2016)

[35] S. Kanazawa et al., “Transmission of 214-Gbit/s 4-PAM signal using an ultrabroadband lumped-electrode

EADFB laser module,” Proc. OFC, Th5B.3 (2016)

[36] H. Yamazaki et al., “300-Gbps Discrete Multi-tone Transmission Using Digital-Preprocessed Analog-

Multiplexed DAC with Halved Clock Frequency and Suppressed Image,” Th3B.4, ECOC 2016

[37] D. Chang et al., “LDPC convolutional codes using layered decoding algorithm for high speed coherent

optical transmission,” OW1H.4, OFC, 2016

[38] M. Sauer, A. Kobyakov, and J. George, "Radio over fibre for picocellular network architectures," IEEE

Journal on Lightwave Technology, vol. 25, no. 11, 2007.

[39] "5G: Personal Mobile Internet beyond What Cellular Did to Telephony", Gerhard Fettweis, Siavash

Alamouti, IEEE Communication Magazine, February 2014.

[40] T. Marzetta, "Noncooperative cellular wireless with unlimited numbers of base station antennas," IEEE

Transactions on Wireless Communications, vol. 9, no. 1, pp 3590-3600, Nov 2010.

[41] M.C. Parker et al. CHARISMA: "Converged Heterogeneous Advanced 5G Cloud-RAN Architecture for

Intelligent and Secure Media Access," EuCNC2016.

[42] Upamanyu Madhow. "Broadband Millimeter Wave Networks: Architectures and Applications." 2008 2nd

International Symposium on Advanced Networks and Telecommunication Systems. Pages: 1 - 3

[43] Terence Quinlan and Stuart Walker. "A Monopole Fed Omnidirectional 13dBi Gain Bi-Conical Horn

Antenna for IEEE802.11ad Applications," Loughborough Antennas and Propagation Conference, Nov

2016.

[44] Constantine Balanis. Antenna Theory, Third Edition. Wiley. Pages 498 - 501.

[45] O.B. Jacobs, J.W. Odendaal and J. Joubert. "Analysis and Design of a wideband omnidirectioal Antenna."

Microwave and Optical Technology Letters, June 2011. wileyonlinelibrary.com. DOI 10.1002/mop.25990

[46] Sheldon S. Sandler and Ronold W. P. King. "Compact Conical Antennas for Wide-Band Coverage." IEEE

Transactions on Antennas and Propagation, Vol. 42, No. 3, March 1994. Pages 436 - 439.

[47] W. D. Burnside and C. W. Chuang, "An aperture-matched horn design," IEEE Trans. Antennas Propag.,

Vol. AP-30, No. 4, Jul. 1982 Pages. 790-796.

[48] Constantine Balanis. Antenna Theory, Third Edition. Wiley. Pages 792 -794.

[49] Matthew A. Morgan and Tod A. Boyd. "A 10-100-GHz Double-Ridged Horn Antenna and Coax

Launcher," IEEE Transactions on Antennas and Propagation, Vol. 63, No. 8, Aug. 2015, Pages 3317 - 34.

[50] P. Turnbull, H. Thomas, D. Venmani, P. Chanclou, V. Jungnickel, L. Fernandez del Rosal, M. Parker, P.

Assimakopouluos, K. Kenan Al-Hares and N. Gomes, “D3.1 iCIRRUS - Verification of Ethernet as

transport protocol for fronthaul / midhaul,” 2015.


Page 75 of 78




[51] L. Fernandez del Rosal, V. Jungnickel, D. Muench, H. Griesser, P. Assimakopoulos, N. Gomes, Y. Kai, H.

Thomas, M. Parker, C. Magurawalage, K. Wang, P. Chanclou and V. D, “D3.2 iCirrus - Preliminary

Fronthaul Architecture Proposal,” 2016.

[52] M. Hinrichs, L. Fernandez del Rosal, C. Kottke, and V. Jungnickel, “Analog vs. Next-Generation Digital

Fronthaul: How to Minimize Optical Bandwidth Utilization,” 21st International Conference on Optical

Network Design and Modeling, 2017.

[53] M. Hinrichs, L. Fernandez del Rosal, C. Kottke, V. Jungnickel, and R. Freund, “Experimental

Investigation of New Fronthaul Concepts for 5G,” 11. ITG-Fachkonferenz Breitbandversorgung in

Deutschland, 2017.

[54] Fernández del Rosal, L., Habel, K., Weide, S., Wilke Berenguer, P., Jungnickel, V., Farkas, P., Freund, R.:

Multi-Gigabit Real-Time Signal Processing For Future Converged Networks. ITG-Fachkonferenz Breit-

bandversorgung in Deutschland 2016

[55] ETSI Technical Specification 136 104 v12.6.0 (2015-05)

[56] Gliese, U., Nørskov, S. and Nielsen, T.N.: Chromatic Dispersion in Fiber-Optic Microwave and

Millime¬ter-Wave Links. IEEE Transactions on Microwave Theory and Techniques 44.10 (1996), pp.

1716–1724

[57] Howard Thomas, et al. “D3.3 SLA and SON Concept for iCIRRUS”, 2016


Page 76 of 78




List of figures Figure 1: Optical access solutions for backhaul based on 1) PtP, 2) T(W)DM PON, 3) PtP WDM PON ................ 13

Figure 2: Mobile Backhaul, Midhaul and Fronthaul from MEF [3]. ...................................................................... 15

Figure 3: First three steps of mobile equipment evolution. ................................................................................. 15

Figure 4 Optical access solutions for low layer fronthaul based on a) PtP fibre, b) PtP WDM PON..................... 17

Figure 5: Optical access solutions based for high layer fronthaul based on: 1) PtP, b) T(W)DM-PON, 3) PtP

WDM-PON. .................................................................................................................................................. 18

Figure 6: Several high layer split fronthaul interfaces for the downlink and uplink. ............................................ 21

Figure 7: Comparative architecture views of 4G (centralised) and 5G (decentralised) networks. ....................... 24

Figure 8 (a) Different LTE functional subdivisions (function splits) options, (b) The implemented split processing

module subdivision and (c) the implemented split networking entity subdivision. .................................... 28

Figure 9 (a) The evolved fronthaul and (b) high-level view of the buffering stages and measurement interface

points ........................................................................................................................................................... 29

Figure 10 Fronthaul and application (OTG traffic generator) data rate measurement results for different

numbers of UEs. The traffic generator is producing traffic only for the downlink direction. ...................... 32

Figure 11 Data rates and percentage increases at different points in the processing chain, for three different

tests of ascending application layer data rates. .......................................................................................... 32

Figure 12 Fronthaul processing latency per LTE subframe for different numbers of UEs .................................... 33

Figure 13 The fronthaul intelligent processing unit (IPU) and the different feeds from the fronthaul and OTA

sections ........................................................................................................................................................ 34

Figure 14 The result of SDN-enabled traffic steering of the background traffic in an evolved fronthaul on the

latency and latency variation of the split traffic. Three packet type traces are shown, pkt_DCI (downlink

control information), pkt_DLSCH (downlink-shared channel) and pkt_SI (system information). ............... 35

Figure 15 Example of live KPI performance monitoring. ...................................................................................... 36

Figure 16: FUSION: Exploiting the inter-packet gaps between HP frames to transmit LP frames ....................... 37

Figure 17: FUSION: Test setup investigating latency and latency variation ......................................................... 37

Figure 18: FUSION: Theoretical estimation of the latency and latency variation of the FUSION IP ..................... 38

Figure 19: FUSION: Theoretical latency and latency variation for aggregation, neglecting PHY and MAC delays

and cable delays ........................................................................................................................................... 39

Figure 20: FUSION: Theoretical latency and latency variation for aggregation and deaggregation neglecting PHY

and MAC delays and cable delays ................................................................................................................ 40

Figure 21: FUSION: Theoretical end-to-end latency and latency variation including PHY/MAC delays and

neglecting cable delays for GST links ........................................................................................................... 41

Figure 22: FUSION: Theoretical end-to-end latency and latency variation including PHY/MAC delays and

neglecting cable delays for SM links ............................................................................................................ 41

Figure 23: FUSION: Measured results for latency and latency variation using an MTU of 16000 Byte for GST and

a packet size of 9622 Byte for SM ................................................................................................................ 42

Figure 24 (a) Reference architecture for the time-aware shaper use-cases presented in this work and (b)

Scheduling design concept........................................................................................................................... 43

Figure 25 Generic time window, window section and subsection plan based on IEEE 802.1Qbv ....................... 44

Figure 26 Network Scenario implemented in Opnet. All network interfaces are 1 GbE ..................................... 45

Figure 27 Average and peak FDV for the PTP traffic with SP and TAS with different GPs. The background traffic

source is constant frame-rate and constant frame size .............................................................................. 46


source is constant frame-rate with a varying frame size following a normal distribution with mean of 1000

octets and variance of 200 octets ................................................................................................................ 47


Page 77 of 78





source is constant frame-rate with a varying frame size following a normal distribution with mean of 1000

octets and variance of 500 octets ................................................................................................................ 47

Figure 30 Zoom-in in the region of GPs from 0 to 1 μs for the results of Figure 29. The inset shows the worst-

case PTP time-stamping error that would result from the peak FDV values ............................................... 48

Figure 31: Experimental transmission setup for 100G. ........................................................................................ 49

Figure 32: a) DSP blocks of the implemented DMT system and b) i) estimated SNR per subcarrier at the receiver

optical back-to-back and the applied ii) bit loading and iii) power loading ................................................. 50

Figure 33: Transmission results of DMT at different data rates and for different transmission distances. ......... 50

Figure 34: a) DSP blocks of the PAM-4 system, b) digital PSD of transmit signal and c) eye diagram after the

EML. ............................................................................................................................................................. 51

Figure 35: Optical back-to-back transmission results of 112Gb/s PAM-4 employing different numbers of pre-

and post-FFE coefficients: a) shows the optical eye diagrams obtained directly after the EML using a pre-

equalizer of 5 coefficients and 61 coefficients, b) and c) illustrate the BER vs. ROP results using different

numbers of post-FFE coefficients and d) depicts the BER performance for different Tx-FFE/Rx-FFE

combinations at an input power of 0dBm. .................................................................................................. 52

Figure 36: Transmission results of 112 Gb/s PR PAM-4: a) using different MLSE memory length after the FFE in

case of optical back-to-back transmission ................................................................................................... 52

Figure 37: System concept for spectral up-conversion utilizing IQ mixers and several independent sub-bands. a)

Electrical spectrum after DAC, b) after IQ-mixing, c) after signal combining, d) after optical modulation. 53

Figure 38: Experimental setup for spectral up-conversion based IM/DD transmission links. .............................. 54

Figure 39 (a+b) Measured BERs for PAM/QAM modulation at different fibre lengths (scenario #1). (c+d)

Measured data rates for different target BERs using DMT/OFDM (scenario #2). ....................................... 55

Figure 40: (a+b). Measured BERs for PAM and DMT at 80 and 120 Gbit/s for the lower band at different fibre

lengths. (c+d) Measured BERs for QAM and OFDM at 64 and 80 Gbit/s for the upper band at different

fibre lengths. ................................................................................................................................................ 56

Figure 41: Antenna model graphic showing conical top and bottom sections with integral matching rings,

antenna feeding arrangements and modified support pillars, with surface current plot showing

dissipation effect of the matching rings. ..................................................................................................... 57

Figure 42: Antenna assembly, showing all component parts and the support pillars and probe clamping

arrangement. ............................................................................................................................................... 58

Figure 43: Measured S11 return loss over the IEEE 802.11ad frequency range with markers at four spot

frequencies. ................................................................................................................................................. 59

Figure 44: Corresponding Smith chart showing measured antenna feed impedance values at four spot

frequencies .................................................................................................................................................. 59

Figure 45: Comparison of co-polar and cross-polar transmission characteristics between 30 GHz and 65 GHz,

also illustrating the wide bandwidth capability of the omnidirectional antenna. ....................................... 61

Figure 46: Showing four channel capability over an operating angle of 360 degrees, demonstrating backhaul,

mobile, and mesh deployment usages. ....................................................................................................... 61

Figure 47: Signal processing for analogue fronthaul ............................................................................................ 65

Figure 48: Experimental setup (optical link) ......................................................................................................... 66

Figure 49: Signal processing for evolved digital fronthaul.................................................................................... 66

Figure 50: Estimated EVM for analogue fronthaul over Rx power and for different fibre lengths ...................... 68

Figure 51: Estimated BER for digital fronthaul over Rx power and for different fibre lengths ............................ 68

Figure 52: Estimated BER for analogue fronthaul over Rx power and for different fibre lengths ....................... 68


Page 78 of 78




List of tables Table 1: Compilation of Fronthaul KPIs ................................................................................................................ 26

Table 2: The 28 (32)-Octet common packet header for all packets sent/received through the fronthaul

interface. ...................................................................................................................................................... 30

Table 3: Ethernet frame payload fields for PKT_DLSCH ....................................................................................... 30

Table 4: DMT System Parameters. ....................................................................................................................... 49

Table 5: Comparison of Peripheral Measured Antenna Gain ............................................................................... 60

Table 6: Characteristics of measurements in fronthaul domain .......................................................................... 63

Table 7: Characteristics of measurements in RAN domain .................................................................................. 64

Documents

intelligent Converged network consolIdating Radio and ... KPIs, ... LTE Long Term Evolution ... SSB Single Side Band . Contract No: 644526 iCIRRUS 1 Jan 2015 – 31 Dec 2017