Download as pdf or txt
Download as pdf or txt
You are on page 1of 69

ESA STUDY CONTRACT REPORT

No ESA Study Contract Report will be accepted unless this sheet is inserted at the beginning of
each volume of the Report.

ESA Contract No: SUBJECT: CONTRACTOR:

4000123383/18/NL/CLP Final Report of Activity University of Luxembourg


“Demonstrator of Precoding
Techniques for Flexible
Broadband Systems”

* ESA CR( )No: No. of Volumes: 1 CONTRACTOR’S


REFERENCE:
n/a This is Volume No: 1
FlexPreDem

ABSTRACT:

A consortium consisting of University of Luxembourg (prime) and SES as supporting company


designed and developed a software simulator platform –referred to in the following as FlexPreDem
SW Tool- that demonstrates the benefits of precoding techniques over broadband HTS systems with
some kind of time-flexibility built into them. The FlexPreDem experimental campaign has evidenced
how much cluster-hopping with precoding within clusters can contribute to the minimization of the
unmet and the unused throughput under specific traffic patterns.
The activity was extended with a CCN to design and test new algorithms for flexible cluster hopping,
allowing clusters of different shapes and sizes to improve the gains achieved with the pre-defined
clustering used in the FlexPreDem SW Tool.

EUROPEAN SPACE AGENCY CONTRACT REPORT

The work described in this report was done under ESA Contract. Responsibility for the contents
resides in the author or organisation that prepared it.

Names of authors:
Dr. Eva Lagunas (eva.lagunas@uni.lu)
Prof. Symeon Chatzinotas (symeon.chatzinotas@uni.lu)
Dr. Nicola Maturo (nicola.maturo@uni.lu)
Dr. Hayder Al-Hraishawi (hayder.al-hraishawi@uni.lu)
Dr.Vu Ha Nguyen (vu-nguyen.ha@uni.lu)
Dr. Vaibhav Kumar Gupta (vaibhav.gupta@uni.lu)
Dr. Stefano Andrenacci (stefano.andrenacci@ses.com)
Dr. Joel Grotz (joel.grotz@uni.lu)

** NAME OF ESA STUDY MANAGER: ** ESA BUDGET HEADING:

Dr. Daniel P. Arapoglou ARTES Core Competitiveness Element -


Component A: Advanced Technology
Dr. Alberto Mengali

DIV:

DIRECTORATE:

* Sections to be completed by ESA


** Information to be provided by ESA Study Manager
Demonstrator of Precoding Techniques for
Flexible Broadband Systems

Final Report

AO/1-8936/NL/17/CLP

ESA ARTES Core Competitiveness Element - Component A: Advanced Technology

ESA Contract No: 4000123383/18/NL/CLP

Contractor:

University of Luxembourg
Interdisciplinary Center for Security, Reliability and Trust
29, Ave. J.F. Kennedy
1855 Luxembourg
Luxembourg

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means
Page 2 of 69
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Demonstrator of Precoding Techniques for Flexible Broadband Systems
Title:
Final Report
Document
Draft
Status:
Document
Code:
Issue
01
Number:
Revision
02
Number:

Issue Date:

Dr. Eva Lagunas (eva.lagunas@uni.lu)


Prof. Symeon Chatzinotas (symeon.chatzinotas@uni.lu)
University of Dr. Nicola Maturo (nicola.maturo@uni.lu)
Author(s):
Luxembourg Dr. Hayder Al-Hraishawi (hayder.al-hraishawi@uni.lu)
Dr.Vu Ha Nguyen (vu-nguyen.ha@uni.lu)
Dr. Vaibhav Kumar Gupta (vaibhav.gupta@uni.lu)
Dr. Stefano Andrenacci (stefano.andrenacci@ses.com)
SES
Dr. Joel Grotz (joel.grotz@uni.lu)

Document Change Record

Modified Pages Numbers,


Changes
Issue Rev. Issue Date Change Explanations and
Approved by
Status

01 00 03.Feb.2021

01 01 13.Feb.2021

01 02 17.May.2022 Adding CCN details

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means
Page 3 of 69
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Table of Contents
1 Abstract.................................................................................................................................... 6
2 Executive Summary ................................................................................................................ 7
2.1 Background ...................................................................................................................... 7
2.2 Objectives of the Activity ................................................................................................ 7
2.3 Study Logic ...................................................................................................................... 7
2.4 Results .............................................................................................................................. 8
3 Introduction, System ArCHitecture and Scenario Definition ............................................ 10
3.1 Introduction ................................................................................................................... 10
3.2 System Architecture ....................................................................................................... 11
3.2.1 Space Segment ........................................................................................................... 12
3.2.2 Ground Segment ........................................................................................................ 12
3.3 FlexPreDem Scenario .................................................................................................... 14
4 FlexPreDem Design .............................................................................................................. 16
4.1 FlexPreDem High-Level Design ................................................................................... 16
4.1.1 Cluster Formation and illumination ratio ................................................................... 18
4.1.2 Illumination Pattern Design ..................................................................................... 21
4.1.3 User Scheduling Within a Cluster ............................................................................ 22
4.2 Proposed Scheduling: Weighted Semi-Orthogonal scheduling ................................. 23
5 FlexPreDem - Software Tool Demonstrator........................................................................ 25
5.1.1 FlexPreDem Design Block ............................................................................................ 25
5.1.2 FlexPreDem Performance Evaluation Block ........................................................... 26
5.1.3 Benchmark Schemes ................................................................................................. 27
5.1.4 FlexPreDem Detailed Design .................................................................................... 27
6 Experimental Campaign ....................................................................................................... 31
6.1 Summary of Results ...................................................................................................... 31
6.1.1 CH versus BH ................................................................................................................ 31
6.1.1.1 Case 1: Demand fitting the Cluster Design ....................................................... 31
6.1.1.2 Case 2: Demand not fitting the Cluster Design................................................ 35
6.1.2 Impact of Inter-Cluster Interference........................................................................40
6.1.3 Demand-Based User Scheduling Within a Cluster ................................................. 42
7 Technology Roadmap ...........................................................................................................48
7.1.1 Time-Flexibility .............................................................................................................48
7.1.2 Precoding ................................................................................................................... 49
7.1.3 Joint Precoding and Time-Flexibility ...................................................................... 50

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means
Page 4 of 69
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
7.2 Next Steps ...................................................................................................................... 51
8 CCN: Flexible Cluster Hopping Techniques ....................................................................... 52
8.1 Flexible CH Techniques ................................................................................................ 54
8.1.1 Extension of BH-SnT solution ...................................................................................... 54
8.1.2 Penalty-Based Minimization Techniques ................................................................ 56
8.1.3 Queue-Based Techniques .......................................................................................... 57
8.2 Numerical Evaluation (Summary) ............................................................................... 58
8.2.1 Flexible CH versus Non-Flexible CH ....................................................................... 58
8.2.2 Computational Complexity ....................................................................................... 64
8.2.3 Comparison with 4 Color scheme............................................................................. 64
8.2.4 Evaluation of Transmit Power per Time-Slot ..........................................................68
8.2.5 Conclusions ................................................................................................................68
9 References ............................................................................................................................. 69

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means
Page 5 of 69
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
1 ABSTRACT

This document presents the final report of the ESA FlexPreDem project. The main goal of this activity
was to design and develop a software-based tool able to test the benefits of precoding techniques as
part of a flexible satellite communications system.

In particular, this document starts with a brief overview of the general mission, potential payload and
system architecture, and the definition of the key techniques involved in FlexPreDem. Once the most
promising architecture is selected, the FlexPreDem software-based demonstrator is designed to test
the behaviour of precoding and cluster-hopping in such scenario. The document also provides an
overview of the final FlexPreDem GUI and concludes with a summary of the findings extracted from
the testing phase of the project and with a technology roadmap to bring such technology to a reality
in future satellite communications systems.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means
Page 6 of 69
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
2 EXECUTIVE SUMMARY

2.1 BACKGROUND

Manufacturers and satellite operators are working hard to launch the next generation of flexible Very
High Throughput Satellite (VHTS) systems that able to offer Terabit per second in-orbit capacity when
and where needed.

In this context, we are currently going through a very interesting and exciting period where the main
goal is two-fold: (1) to drastically decrease the cost of satellite services by increasing the satellite
capacity while improving the system spectral efficiency, and (2) to deliver satellite services in a more
flexible manner by allocating the satellite resources based on the per-beam traffic demands.

FlexPreDem project focuses on two of the most promising disruptive techniques to tackle these
specific challenges: linear precoding and beam hopping. Beam hopping provides the means to flexibly
adapt the offered capacity to the time and geographic variations of the traffic demands, while
precoding exploits the multiplexing feature enabled by the use of multiple antenna feeds at the
transmitter side to boost the spectral efficiency. These two effective strategies can create unique
opportunities if they are properly combined.

2.2 OBJECTIVES OF THE ACTIVITY

FlexPreDem activity main objective is to select the appropriate scenario and demonstrate the benefits
at various levels of adding a layer of complexity (due to precoding) in a flexible broadband satellite
communications system. The benefits are evaluated not only from the capacity/throughput system
capabilities but also in terms of offered flexibility in accommodating variable traffic demands over the
coverage area.

More precisely, the introduction of precoding in beam-hopped systems is considered, particularly


when a set of adjacent beams is illuminated at the same time with the same frequency resources. We
define such set of adjacent beams as cluster and we refer to the concept of ``Cluster Hopping''. The
cluster-hopping performance is evaluated with a software-based tool identifying the specific demand
distributions where this kind of system make sense. Finally, a technology roadmap to bring the
combination of precoding and flexible payload technology to a reality in future satellite
communications systems is prepared.

2.3 STUDY LOGIC

Figure 1 presents the FlexPreDem summarized study logic, where the four main phases of the project
are highlighted with green blocks. In particular, FlexPreDem is organized into: (1) Mission, payload,
system and techniques definition, (2) FlexPreDem SW-Tool Definition and validation plan, (3)

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means
Page 7 of 69
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
FlexPreDem development, validation and integration; and (4) Experimental Campaign and technology
roadmap. In Figure 1, each main block is linked to the main outputs of the project.

Figure 1. FlexPreDem summarized study logic

2.4 RESULTS

The project evidence that the integration of precoding within the so-called cluster hopping concept is
beneficial when extended high demand areas cannot be covered with a single beam.

In terms of technical challenge, the most complex design part resides on two axes: (1) the appropriate
beam illumination pattern, and (2) the appropriate cluster shape and size. While these two aspects
can be optimally obtained via mathematical optimization, the complexity of the resulting problem
may not be acceptable unless certain assumptions of on the cluster type are made.

In FlexPreDem, we proposed a simplified design focusing on compact shaped, non-overlapping and


equal size cluster definition. The evaluation for a single GEO cluster-hopped multibeam satellite
system brought us to the following conclusions:
o Cluster size and shape should be as close as possible to the expected demand clusters. In
doing so, the demand-matching performance of the CH system significantly improves. An
example of the aforementioned cluster design is shown in Figure 2.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means
Page 8 of 69
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 2. Clustering Design to Match the Demand Clusters

o Regarding the illumination ratio, this has to be carefully selected based on (i) Payload
limitations in terms of maximum number of beams that can be simultaneously activated;
(ii) the power that is needed in a particular area to satisfy the demand. The second point
refers to the fact that the illumination ratio basically allows to devote more or less power
to the clusters. With a low illumination ratio, less clusters are activated in a particular time
instance, allowing to focus the payload power on fewer clusters. Therefore, if we face high
cluster demands, we may need to consider lower illumination ratios.
o While the beam illumination pattern design targets demand matching at beam level, in
general, to achieve a good user-level demand satisfaction, proper user scheduling needs
to be considered considering the specific users’ demand requests.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means
Page 9 of 69
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
3 INTRODUCTION, SYSTEM ARCHITECTURE AND SCENARIO
DEFINITION

This section presents the mission, payload, system and technique definition as included in [5].

3.1 INTRODUCTION

Conventional Beam Hopping (BH) satellite systems exploit the time-domain flexibility to provide all
available spectrum to a selected set of beams as long as they are not adjacent to each other. BH and
precoding are two trending technologies for the satellite community. While BH enables flexibility to
adapt the offered capacity to the heterogeneous demand, precoding acts as an enabler for aggressive
frequency reuse by mitigating the inter-beam interference . The main idea of FlexPreDem project was
to consider a high throughput satellite (HTS) system that employs BH in conjunction with precoding.

Clearly, beam hopping provides the means to flexibly adapt the offered capacity to the time and
geographic variations of the traffic demands, while precoding exploits the multiplexing feature
enabled by the use of multiple antenna feeds at the transmitter side to mitigate the inter-beam
interference that results from aggressive frequency reuse systems. These two effective strategies can
create unique opportunities if they are properly combined.

In particular, the concept of Cluster-Hopping (CH) was proposed in FlexPreDem, which seamlessly
combines the BH and precoding paradigms and exploits their individual competencies.

▪ Conventional BH: With BH, all the available satellite resources are employed to provide service to
a certain subset of beams, which is active for some portion of time, dwelling just long enough to
fill the demand in each beam. The conventional BH illumination pattern is illustrated in Figure 3,
where the active spot beams are designed to have a border area formed by inactive beams such
that degree of isolation exists between each sector (i.e. active beams do not interfere to each
other). In conventional BH, there are strict constraints in the possible illumination patters due to
interference avoidance

Figure 3. Typical illumination pattern of a BH system

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 10 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
▪ Proposed CH: The introduction of precoding in BH systems is required when a set of adjacent
beams is illuminated at the same time with the same frequency resources. We define such set of
adjacent beams as cluster and we refer to the concept of “Cluster Hopping”. An example of CH
illumination pattern is illustrated in Figure 4. The main advantage of the proposed cluster-based
hopping is the resource allocation flexibility within a cluster, meaning that when a group of
adjacent beams is simultaneously illuminated, one can implement a beam-free approach where
all the resources of the cluster can be shared across the cluster’s coverage area.

Figure 4. Proposed CH illumination pattern

3.2 SYSTEM ARCHITECTURE

The FlexPreDem system architecture considering the combination of beam hopping and precoding is
composed of the following:
▪ The space segment which consists of a single GEO satellite equipped with time-flexibility.
▪ The control segment which consists of all ground facilities for the control and monitoring of
the satellites. This includes:
o TT&C (tracking, telemetry and command)
o Management of the traffic and the associated resources on-board the satellite:
This is a key component of the system; it is the entity that determines the beam
hopping transmission plan. It needs at its input the demand requirement from each
terminal user and provides the plan to gateway, which allocates beams and time slots
for the whole transmission cycle. The satellite is informed about the illumination
pattern via TT&C.
▪ The ground segment consists of multiple broadband user terminals. FlexPreDem has focused
on a single-GW assumption. The GW collects CSI estimates which are used, together with the
traffic demand statistics, to generate the beam illumination pattern.

A simplified architecture is depicted in Figure 5, where the TT&C is collocated with the resource
management unit.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 11 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 5. Flexible broadband multi-beam satellite architecture

3.2.1 SPACE SEGMENT

The techniques developed in FlexPreDem have been designed assuming a conventional bent-pipe
Single-Feed per-Beam (SFPB) architectures with switching matrices to select the appropriate beam.
Both TWTAs or MPAs can be used as power amplification technology, being the second more flexible
in terms of power distribution across beams. While, in principle, the DTP is not part of the proposed
scheme the architecture could be updated with the inclusion of the DTP simply to provide routing
flexibility, especially for the case of multi-gateway system architectures. An example of payload
architecture assuming MPA and switching matrices is provided in Figure 6.

Figure 6: Proposed flexible payload architecture to support Cluster Hopping.

3.2.2 GROUND SEGMENT

Beam-hopping is a time-domain technique, for which time alignment and time slot allocation are very
important.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 12 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
From the UT perspective, it has to be synchronized with the transmitter in frequency, time, and in
coherent receivers in phase as well. It also has to be adapted to the channel impairment. A satellite
terminal typically performs this synchronization constantly, using the forward link that is transmitted
continuously. In a hopping environment, this is not the case, as the satellite beam illuminates the
terminal only during its dwell time. The rest of the time the terminal is in the dark without receiving
any signal.

The terminal has to acquire the satellite signal in two different circumstances:
1. Cold start, wherein the terminal needs to acquire the satellite signal without any prior
information. Synchronization procedures mainly include carrier frequency correction,
sampling timing correction, frame synchronization, equalization and fine phase correction.
2. Signal loss, wherein the signal is lost for a short period. In this case, most of the parameters
are available, and after reception is resumed full acquisition can be readily achieved.

It can be quite safely assumed that the hoppings receiving conditions are more of the signal loss type
rather than cold start. User terminals do not have to be accurately synchronized to the network. In
that case, a terminal would be required to search for the satellite signal and re-acquire it every dwell
time. However, network synchronization would enable the terminal management to be aware of
transmission times and to be power down or revert to sleep mode when possible, thus saving battery
power.

Most of the research works on synchronization for beam-hopped systems rely on the super-frame
format 4 introduced in DVB-S2X-Annex E, e.g. the works in [2][3]. From another point-of-view, some
works argue that the long and fix length of the super-frame structure limits the flexibility of beam
hopping [4]. To tackle this problem without using a specific format, the authors in [4] introduce a
dedicated dummy frame with a synchronization field. Discussion on the pros and cons of each air
interface can be found in [6].

Regarding the CSI acquisition process, outdate CSI increases with bursty tranmissions of beam-
hopping. In particular, at the moment that a user passes from the non-illuminated to the illuminated
phase, it has to rely initially on the CSI calculated in the previous illumination phase, which in general
will be outdated depending on the structure of the illumination pattern (e.g. assuming non-
overlapping clusters, CSI will always be outdated at the initial frame). Having said so, the outdated CSI
is not expected to have a strong impact. This is because a single super frame is enough to obtain a
good channel estimation and, therefore, the outdated CSI will only be needed for the initial (single)
super-frame.

The synchronization aspects and discussion have been included in this document for completeness,
but they are not included in the FlexPreDem software tool.

From the GW perspective, BH inherently requires synchronization, especially when multiple teleports
and many terminals are involved [5]. Synchronization of the teleports and gateways can be done
either via standard synchronization means in an open loop or using a satellite loopback. Furthermore,
this procedure can be supported by a satellite auxiliary transmitter which provides to the teleports
information about actual hop times and transmission times.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 13 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
3.3 FLEXPREDEM SCENARIO

We consider a multi-beam GEO satellite scenario, whose antenna pattern and link budget parameters
have been provided by ESA. The beam pattern, which is shown in Figure 7, is a Direct Radiating Array
(DRA) based pattern generated in software in the 20 GHz band, assuming 750 elements spaced 5
lambda. The pattern includes 100 beams. Satellite position is 13 degrees East.

Figure 7. FlexPreDem beam pattern, also called ESA100 beam pattern

It is assumed that the selected beams would be served with the full 19.7 – 20.2 GHz band (i.e. 500
MHz of user link bandwidth). A summary of the key parameters of the selected scenario is provided
in Table 1.

As a benchmark scenario, a conventional beam-hopped system (without precoding) is considered. A


1-color scheme with beam hopping capabilities, where illumination of adjacent beams is avoided.
Table 1. Key system parameters
Parameter Value
Orbit GEO, 13E
Coverage Europe
Max. Beam Pattern Gain 52 dBi
User Link BW 500 MHz
Roll-off 20%
Air Interface DVB-S2X
Carriers per Beam 1
Payload Losses 5 dB
Noise Power -118.4 dBW

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 14 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
UT Antenna Diameter 0.6 m
UT Antenna Gain 39.55 dBi

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 15 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
4 FLEXPREDEM DESIGN

Herein, we summarize the contents of [8].

The proposed cluster-hopping concept is depicted in Figure 8, where only a subset of clusters
composed of adjacent beams are illuminated at each dwell time. The set of illuminated beams may
change in each timeslot based on a time-space transmission pattern that is periodically repeated. By
modulating the period and the duration of illumination, different capacities can be offered in different
beams.

Figure 8. Example of beam illumination pattern for the proposed Cluster-Hopping concept

The proposed cluster hopping will be designed from 2 different perspectives:

1) Long-term flexibility: Aiming at the management and assignment of satellite resources


according to the predicted beam traffic demands. This mainly focusses on the design of the
appropriate illumination pattern so as to match the offered capacity with the average beam
demands.
2) Short-term flexibility: Once the illumination pattern is fixed, the optimization of the user
scheduling within the clusters can further improve the overall traffic matching at a user-level.
In this part, we show how the cluster-hopping approach provides another level of flexibility,
where the resources of a cluster can be used to serve users on the cluster coverage area in a
very flexible manner.

We believe that the short-term flexibility achieved with user-scheduling for user rate matching is an
innovative contribution which seems not been considered in previous works. Note that the main goal
is to match the offered capacity with the traffic demand, while the latency issue will only be considered
in the long-term flexibility by making sure that all beams are served within a beam hopping cycle.

4.1 FLEXPREDEM HIGH-LEVEL DESIGN

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 16 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
We assume that the BH cycle is denoted as 𝑇H , which is equally divided into 𝑁slot time-slots, and we
set duration of each time slot equal to 𝑇slot . So, we have 𝑇H = 𝑁slot 𝑇slot . Figure 9 illustrates the
relations among these hopping parameters.

Figure 9. Illustration of the hopping parameters.

The goal is to design the optimal cluster-based illumination pattern, i.e., to find the sets of clusters to be
activated simultaneously at different hopping events and the corresponding illumination period or dwell
time in order to offer an average capacity as close as possible to the requested demand.

By defining a “snapshot” as the particular arrangement of illuminated and non-illuminated beam-


clusters, the objective is to find which snapshots appear on the optimal illumination pattern and for
how many time-slots are they activated, as illustrated in Figure 10, where 𝜓𝑛 is an integer variable
indicating the duration of each selected snapshot.

Figure 10. Example of CH illumination pattern composed of 4 snapshots

The proposed FlexPreDem high-level system design scheme is illustrated in Figure 11. The first step is
to define the clusters (size and shape) across the satellite coverage area. To do so, we need a block
with input being the total number of beams in the coverage area and the illumination ratio (which is
given based on the payload capabilities).
Once the clusters are defined, we need to decide how many of them and which of them will be
illuminated at the same time. In other words, the definition of the snapshots is required. The more
snapshots, the higher the flexibility of the system to adapt to the heterogeneous traffic demands.
In parallel, we can run the user scheduling and precoding within each pre-defined cluster individually
in order to know the average cluster / beam / user offered capacity per each time-slot that this
particular cluster is illuminated. Obviously, this part requires knowledge of the user demands.
Next, the snapshot definition is combined with the results obtained by the user scheduling block such
that the average offered capacity at snapshot-level is obtained. Finally, the cluster hopping

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 17 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
illumination design is optimized by selecting the best snapshots and their corresponding time of
illumination.

Figure 11. FlexPreDem high-level system design scheme

4.1.1 CLUSTER FORMATION AND ILLUMINATION RATIO

The cluster definition is an integral part of the cluster-hopping approach and will determine its overall
performance. The size (number of beams in a cluster) of a cluster, the number of clusters in the
system, and the geographical distribution of beams within a cluster have a strong impact on the
flexibility as well as complexity of the illumination pattern design. Therefore, cluster formation needs
to be performed carefully taking into consideration the trade-off between resource allocation
flexibility and operational complexity. As an example, a larger number of non-overlapping cluster
(thus, cluster with small sizes, two beam per cluster is the extreme case) will yield to a bigger search
space for the illumination pattern design, since there will be multiple possible combinations of
simultaneous clusters. On the opposite, a small number of non-overlapping clusters (thus, clusters
with a lot of beams) will provide less flexibility in the operation resulting in poor rate-matching
performance.

Regarding the cluster definition, a cluster is defined by a particular shape and size. However, in order
to maximize the precoding gain, compact (round shaped) clusters are highly preferred from the
practical perspective.

Without restricting the cluster definition, the number of possible clusters is huge. As an example,
considering clusters of 6 beams/cluster and the ESA-71 beam pattern illustrated in Figure 12, it is
possible to form 17163 clusters (any shape allowed). If we consider all these possible clusters to be
activated in a particular time-slot, there will be 217163 possible snapshots. Note that in this considered
example, one beam can be part of multiple clusters.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 18 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
One way to restrict the number of possible snapshots is to consider the payload limitations in terms
of maximum number of beams that can be simultaneously illuminated. Assuming only 3 clusters can
be activated at each time-slots, we will have (17163
3
) possible snapshots, where the clusters can be
both adjacent and non-adjacent.

Furthermore, we can restrict the snapshot to only those with non-adjacent clusters, and those with
sufficiently compact shape such that the use of precoding makes more sense.

Figure 12. ESA71 beam pattern

Hence, for future satellite systems with a large number of beams, the optimal clustering design is a
very challenging task because of the clustering options mentioned earlier. Figure 13 provides the
rationale of the cluster design by summarizing the issues highlighted in this section.

Figure 13. Rationale followed for the cluster design

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 19 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
In FlexPreDem, we opt to have compact shaped, non-overlapping and equal size cluster. Assuming
the 67 beams out of 100 beams in Figure 7 that have actual demand, FlexPreDem considers the two
clustering options below (which have been illustrated in Figure 14):
▪ Clustering Option 1: 17 clusters (16 clusters of 4 beams each and 1 cluster with 3 beams)
▪ Clustering Option 2: 11 clusters (10 clusters of 6 beams and 1 cluster with 7 beams)

Figure 14. FlexPreDem clustering options

In FlexPreDem, we allow three different illumination ratios, which are 1/4, 1/6 and 1/8. The number
of clusters and number of beams that are activated for each of these illumination ratios is given in
Table 2.

Table 2. Number of clusters and number of clusters per snapshot for two different clustering options.
Number of
Illumination Ratio
Configuration beams per
cluster 1/4 1/6 1/8
Number of active clusters per
4 3 2
Clustering snapshot
4
Option 1 Average number of active beams
16 12 8
per snapshot
Number of active clusters per 3 2 1
Clusturing snapshot
6
Option 2 Average number of active beams 18 12 6
per snapshot

Once the number of active clusters per snapshot is fixed, we have to discuss how these are selected
from the total number of clusters. Assuming that two adjacent clusters cannot be activated at the
same time instance, Table 3 summarizes the number of possible snapshots for the two proposed
clustering options for each of the considered illumination ratios.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 20 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Table 3. Number of possible snapshots
Illumination Ratio
Configurations
1/4 1/6 1/8
Clustering Option 1 304 263 101
Clustering Option 2 36 35 11

Once the snapshots are defined, their offered capacity per beam can be determined by applying the
user scheduling and precoding within each cluster independently (note that the inter-cluster
interference is neglected in this step assuming proper geographical separation between clusters).

4.1.2 ILLUMINATION PATTERN DESIGN

Let 𝜓𝑖 be a nonnegative integer variable indicating the illumination duration for snap-shot “𝑖”, 𝑖 =
1,2, ⋯ , 𝑁s𝑠 , in terms of the number of time-slots within a hopping window of duration 𝑇𝐻 (see Figure
10). The offered capacity vector 𝒔 can be writen as,
𝑁
𝒔 = ∑𝑖=1
s𝑠
𝒍𝑖 𝜓𝑖 ,

where 𝒍𝑖 (𝑗 ) denotes the offered capacity per time-slot 𝑇s𝑙𝑜𝑡 at beam “j” when considering the
snapshot “i”.

Assuming 𝑑𝑗 being the demand at beam “j”, the illumination pattern design problem is formulated as
a demand satisfaction fairness problem, where the goal is to maximize the minimum ratio between
the offered capacity and the requested demand among the beams and can be expressed as follows,
𝑠𝑗
Max min
𝜓1 ,…,𝜓𝑁s𝑠 𝑖 𝑑𝑗
𝑁s𝑠

subject to 𝒔 = ∑ (𝑙𝑖 𝜓𝑖 )
𝑖=1

∑𝑁s𝑠
𝑖=1 𝜓𝑖 𝑇s𝑙𝑜𝑡 = 𝑇H

We can simplify the max − min optimization problem in terms of turning into a maximization problem
with the help of introducing a new slack variable 𝑡 along with a new constraint 𝑠𝑖 ≥ 𝑑𝑖 𝑡. The
optimization variable is an integer, thus resulting in an integer programming problem.

In FlexPreDem, we make use of the advanced GUROBI solver combined with CVX software 1 to solve
the aforementioned problem.

1
http://cvxr.com/cvx/doc/gurobi.html

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 21 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
4.1.3 USER SCHEDULING WITHIN A CLUSTER

Assuming that the beam-hopping pattern is fixed, the allocation of system resources in a short-time
basis remains a non-trivial task. Focusing on the time dimension, the latter is usually referred as user
scheduling. The need for spontaneous decision on the user scheduling combined with system
constraints imposed by user demand matching and the dynamic nature of the satellite link quality add
to the complexity of the scheduler design.

Practically, satellite systems are operating in vast multiuser environments and covering wide areas by
multibeam satellites that illuminate large group of users. Hence, scheduling these large numbers of
users, over subsequent in time transmissions, is an essential procedure to design a realistic system.

In this section, we focus on the user scheduling design within a particular cluster that is active at a
certain time. Therefore, no hopping is considered. More precisely, we address the user scheduling
within a set of adjacent beams that aggressively reuss the user link frequency resources at PHY layer,
and thus considers the implementation of interference mitigation techniques at the transmitter side
to exploit the multi-antenna multiplexing via precoding. In parallel, the system aims at traffic matching
at a user level, which puts some constrains on the scheduling at a link layer level.

The main objectives of the user scheduling proposed in this project target two main goals, one specific
to PHY layer and one specific to Link layer, which are indicated in Table 4.
Table 4. Scheduling objectives
Scheduling Objective
Multiple users to be jointly served by multiple beams should produce low
interference.
PHY Layer
(in case of multicast) Multiple users to be multiplexed in the same frame should
be able to support the same ModCod.
The provided user throughput should be proportionally fair to the requested rate
Link Layer
at a user-level.

In addition, there is an inherent advantage on the proposed cluster-based hopping approach, which
has rarely been considered in the literature. This is the resource allocation flexibility within a cluster.
Clearly, with the conventional non-clustered beam-hopping scheme there is no way that the resources
assigned to a particular beam can be used to support other beams. In our proposed cluster-based
model, we can distribute the cluster resources according to different traffic demands in the cluster
area. This corresponds to a “Hot-Spot Cluster Hopping” scenario, which considers the exploitation of
the low-demand-beam resources to provide more capacity to the high-demand beams.

To put this into perspective, the precoded cluster provides a rather different view compared to the
conventional frequency orthogonalization. More specifically, the conventional system has a clear
association of users to beams. Each beam handles a non-overlapping set of users and thus it can be
represented by a separate queue resulting in one queue per beam from a queueing theory point of
view (see Figure 15). On the other hand, the precoded cluster jointly serves all the users in the cluster.
This can be seen as a beam-free approach within the coverage area of the specific cluster. This has
been illustrated with a single queue per cluster in Figure 16. In that respect, depending on the

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 22 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
scheduling objectives, it is perfectly feasible that multiple users are scheduled in one beam while no
users in another.

Figure 15. Conventional beam-limited queue model

Figure 16. Queue model for cluster-based beam-hopping: beam-free approach

4.2 PROPOSED SCHEDULING: WEIGHTED SEMI-ORTHOGONAL


SCHEDULING

In this section, we propose a scheduling approach which jointly takes into account the channel
vectors of the users and their traffic demand. In this context, considering a cluster including M users,
we denote the traffic demand of the generic user m, in bps, as dm. In order to take into account the
traffic demand dm in the scheduling, we associate to the generic user m a coefficient 𝛼𝑚 ∈ [0,1], which
gives an indication on the priority that shall be given to the user m in the scheduling. Such coefficient
𝑑
is simply a normalized version of the demand dm, i.e., 𝛼𝑚 = max𝑚𝑑 , for each m = 1, …, M. Thus, 𝛼𝑚 =
𝑚
𝑚
1 is associated to the highest traffic demand and 𝛼𝑚 = 0 to the lowest.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 23 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
The proposed scheduling approach is a weighted version of the semi-orthogonal scheduling, which on
one hand aims to orthogonalize as much as possible the users’ channels, while on the other and
prioritizes the users requiring higher traffic (thus, having higher α). Given a cluster of K beams including
M users, K users are scheduled based on the following algorithm:
2
• The first scheduled user is the one maximizing the metric 𝛼𝑘 ||ℎ𝑘 || .
• After the first user has been scheduled, in order to schedule the remaining K-1 users the
𝐻|
|ℎ𝑗ℎ𝑚
following metric is sequentially calculated: 𝑤𝑚 = 𝛼𝑚 (1 − ∑𝑗∈Λ ) for each m = 1, …,
||ℎ𝑗||||ℎ𝑚||
M, with Λ denoting the set of indexes of the previously scheduled users.
• At each step the scheduled user is the one maximizing the metric 𝑤𝑚 , which jointly accounts
for the priority and for the orthogonality to the previously scheduled users.

Once the K users are scheduled within the cluster, they are served after applying the precoder
procedure previously described. After a user k is served, it is also important to update its priority 𝛼𝑘 ,
and specifically to reduce it accounting for the traffic delivered to the user.

▪ Priority update based on average offered rate


The priorities in a given instant are updated considering the average rate provided to the users until
that instant, rather than the instantaneous rate. This approach, which requires the storage of the
offered rates for a certain temporal window, is able to dynamically account for the average demand
satisfaction at each user. More specifically, the coefficient of a served user k at the scheduling instance
𝑑
t is updated as: 𝛼𝑘𝑛𝑒𝑤 = 𝐸 (𝑅𝑘 ) , with dk denoting the demand and 𝐸𝑡 (𝑅𝑘 ) the average rate until instant
𝑡 𝑘
t.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 24 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
5 FLEXPREDEM - SOFTWARE TOOL DEMONSTRATOR

While Figure 11 provides a general overview of the FlexPreDem tool, in Figure 17 we provide the
detailed FlexPreDem sequence diagram. In particular, the first step is to select a clustering option (as
we will see in section 4.1.1, to reduce complexity we rely on predefined clustering options tailored to
a specific beam pattern). Once the clustering definition is given, we execute the FlexPreDem design,
whose main outcome is the CH illumination pattern. To evaluate the proposed solution, we next
execute the FlexPreDem evaluation of the resulting CH illumination pattern and the results are
displayed though the GUI.

Figure 17. FlexPreDem sequence diagram

5.1.1 FLEXPREDEM DESIGN BLOCK

In this section, we briefly introduce the FlexPreDem design block included in Figure 17. With the
purpose of visually representing the main blocks and function within the FlexPreDem design, we
provide a diagaram in Figure 18, where we use the color “green” to indicate those blocks that are
executed on real time in FLexPreDem, and color “red” to indicate those blocks that are executed off-
line due to computational complexity issues. Figure 18 also illustrates the sequence of execution from
left to right. Note that each on-line block in Figure 18 corresponds to a single specific matlab function.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 25 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 18. FlexPreDem design scheme

5.1.2 FLEXPREDEM PERFORMANCE EVALUATION BLOCK

In this section, we briefly introduce the FlexPreDem evaluation included in Figure 17. With this aim,
we have decomposed the FlexPreDem evaluation block into sub-blocks in Figure 19. Essentially, the
FlexPreDem evaluation takes as input the resulting illumination pattern from the design phase and
executes it.
In particular, the evaluation will sequentially execute the following steps:
1) Extraction of the selected snapshots: From the illumination pattern, we can easily extract
which snapshot has been selected for activation at each time-slot.
2) For each selected snapshot, serve users according to the weighted semi-orthogonal
scheduling algorithm and MISO precoding (short term flexibility) and obtain the average
snapshot offered capacity (this time taking into account the inter-cluster interference).
3) Aggregate the snapshot offered capacity according to the illumination pattern (i.e. how many
times each snapshot is activated) and obtain the per-user performance in terms of rate
matching and achieved SINR.

The sequence of execution is also captured in Figure 19, which corresponds to the numbering of the
bullets above. Again, we used color “green” color “red” to indicate those blocks that are executed in
real time and non-real time in FlexPreDem, respectively.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 26 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 19. FlexPreDem Performance Evaluation Scheme

5.1.3 BENCHMARK SCHEMES

The FlexPreDem solution will be compared with the conventional BH. For the BH benchmarking, we
will have two alternatives:

▪ BH design proposed by the team: The design follows the same objective as the cluster-
hopping design but the main difference is that the predefined snapshots do not consider
adjacent active beams.
▪ BH design proposed by ESA: Unlike the previous one, the ESA’s BH design allow adjacent
beams to be illuminated together at a particular time instance. However, this is an heuristic
iterative algorithm which at each iteration, the beam with highest unmet is assigned a single
time-slot in a way that the SINR is Maximized. Therefore, the SINR maximization prevents to
have too many adjacent beams activated simultaneously.

5.1.4 FLEXPREDEM DETAILED DESIGN

Table 5 summarizes the main functionality of each of the key blocks included in FlexPreDem SW Tool.

Table 5. Description of FlexPreDem main blocks


Module Description
We opt to have compact shaped, non-overlapping and equal size cluster,
Definition of
which we believe provides a good trade-off in terms of complexity and
predefined clusters
flexibility.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 27 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
We allow three different illumination ratios, which are 1/4, 1/6 and 1/8,
Definition of which result in different snapshots. For reducing the number of possible
predefined snapshots snapshots, we impose some contraints (i.e. active clusters should be far
enough to assume negligible inter-cluster interference).
User scheduling This provides the short-term flexibility of FlexPreDem by allowing a beam-
within each free approach where the resources of a cluster can be user to serve any user
individual cluster withing the cluster.
Combination of This is a trival but required step to combine the average offered capacities
offered capacity at per cluster given by the previous block and obtain the average offered
snapshot level capacity per snapshot.
Illumination pattern This provides the long-term flexibility of FlexPreDem by determining which
design snapshots to be used and their corresponding dwelling time.
GUI The FlexPreDem will be presented in a user-friendly GUI

The FlexPreDem GUI is illustrated in Figure 20.

Figure 20. GUI overview

The input parameters are divided into fixed and adjustable parameters. Here is the list of fixed system
parameters:
• The parameters listed below are set by default.
• Satellite orbit: GEO, 13-degree East
• Beam Pattern: ESA-100
• Polarization: single
• Loss: 5 dB (OBO+Payload chain)

Where adjustable parameters are:


• Payload Power [Watt]
• Transponder Band Width [MHz]
• Operating frequency [GHz]
• BH Window [time slots]
• Time Slot Duration [msec]

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 28 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
• Mode of Operation
o Cluster Hopping
o Beam hopping by SnT
o Beam hopping by ESA
• Clustering Option
o Option 1 with 17 clusters of approx. 4 beams per cluster
o Option 2 with 11 clusters of approx. 6 beams per cluster
• Illumination Ratio
o 1/4 (one out of four active beams)
o 1/6 (one out of six active beams)
o 1/8 (one out of eight active beams)
• Traffic Time Instance
o 1am
o 8 am
o 12pm
o Artificially Generated Demand 1
o Artificially Generated Demand 2
• Scheduling
o Simplified (single user in the center aggregating all beam demand)
o Demand-based scheduling (multiple users with different demands)
• Capacity Selection
o Shannon Capacity
o DVB-based ACM capacity
• Model Selection
o Results including inter-cluster (or inter-beam) interference
o Results ignoring inter-cluster (or inter-beam) interference

Once the program completes the requested calculations, the main results are shown directly on the
GUI, including:
▪ Demand: Overall system demand (aggregation of the 67 beams). Measured in Gbps.
▪ Offered Capacity: This is the offered or so-called supplied capacity that is provided by the
satellite and includes both used and unused capacity. Measured in Gbps.
▪ Unmet Capacity: This output shows the amount of overall system unmet capacity. Measured
in Gbps.
▪ Unused Capacity: This output shows the amount of overall system unused capacity.
Measured in Gbps.

Visualization of the resulting illumination pattern is also possible, as well as different figures as listed
in Figure 21.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 29 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 21. Results Selection Drop-down Menu

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 30 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
6 EXPERIMENTAL CAMPAIGN

Below we present the main conclusions extracted from the testing of FlexPreDem project, which is
fully captured in [12].

Table 6 shows the summary of the tests carried out within FlexPreDem activity.

Table 6. Summary of FlexPreDem Tests


Test Description Goal
Observe the benefits of cluster hopping versus
CH versus BH with demands
1 conventional BH schemes considering the demands
from traffic emulator
provided from the SnT Traffic Emulator.
Same as 1 but with artificially -
2
generated demand
Observe the impact of of BH hopping window on CH
3 CH window length impact performance in terms of unmet and used capacity, and
compare the performance with conventional BH.
Inter-cluster interference at To analyse the impact of the inter-cluster interference at
4.1
snapshot level a snapshot level.
Inter-cluster interference at Same as 4.1 but at a system level
4.2
system level
Analyse the performance-complexity trade-off of the
5 Heuristic Approach heuristic algorithm versus the optimal solution for CH
illumination pattern design problem
To verify the operation and performance of the
proposed scheduling scheme (priority based) and
6.1 User Scheduling at cluster-level
compare its performance to the simplified scheduling
scheme at a cluster level
6.2 User Scheduling at system-level Same as 6.1 but at a system level
Shannon versus DVB-S2X Compare the proposed CH results in terms of DVB-S2x
7
capacity capacity and Shannon Capacity

6.1 SUMMARY OF RESULTS

6.1.1 CH VERSUS BH

6.1.1.1 Case 1: Demand fitting the Cluster Design

We initially test the proposed CH solution with a beam demand instance that fits well with the
proposed clustering design. Figure 22 shows an artificially generated demand which intentionally
contains clusters of high demand data (the so-called “Artificially Generated Demand Instance 1”).

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 31 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 22. Artificially generated demand 1 – Designed to fit a 4-beam cluster

We run the proposed FlexPreDem cluster-hopping design considering the clustering definition of 4-
beams per-cluster provided in Figure 14 and an illumination ratio of 1/6.

Figure 23. Clustering Option 1: 4 Beams per Cluster

A comparison of the beam demand and the beam supplied capacity obtained results are shown in
Figure 24, showing a very good beam demand matching. For comparison purposes, Figure 25 shows
the results obtained with a conventional beam hopping scheme designed by the team, where adjacent
beams cannot be illuminated simultaneously. It becomes evident that cluster hopping outperforms
the conventional beam hopping for this particular demand instance. Figure 26 provides the results for
The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 32 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
the ESA conventional BH design and it can be observed that this benchmark fails in satisfying the high-
demand beams.

Figure 24. Beam Demand versus Beam Supplied Capacity for proposed Cluster Hopping design (Clustering Option 1,
Artificial Demand 1, 1/6 illumination ratio)

Figure 25. Beam Demand versus Beam Supplied Capacity for Conventional BH design

Figure 26. Beam Demand versus Beam Supplied Capacity for ESA BH design

The system unmet and unused results of the CH technique for the artificially generated demand
instance 1 are reported in Table 7. The best satellite resource efficiency is achieved with illumination
ratio 1/6 and clustering option 1, with 1 Gbps of unmet capacity and 1 Gbps of unused capacity. When
The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 33 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
the illumination ratio decreases to 1/8, the unmet capacity increases as the number of simultaneously
active clusters decreases. Similarly, when the illumination ratio increases to 1/4, the unmet capacity
reduces as the number of simultaneously active clusters increases. This is a direct consequence of the
changes of the supplied capacity. In general, clustering option 1 outperforms clustering option 2, as
the demand was specifically designed to match the 4-beam clusters considered in clustering option 1.

Table 7. CH Results for artificially generated demand 1 (values in Gbps)


Illumination Ratio
Artificially Generated Demand 1
¼ 1/6 1/8
Clustering System Unmet Capacity 0 1 7
Option 1 (17 System Unused Capacity 6 1 0
clusters) Supplied Capacity 31.37 26.36 18.65
Clustering System Unmet Capacity 6 8 13
Option 2 (11 System Unused Capacity 15 9 3
clusters) Supplied Capacity 35.23 26.50 15.04

The average and minimum beam demand satisfaction results are summarized in Table 8 and Table 9.
Clustering option 1 also outperforms Clustering option 2 in terms of beam satisfaction, as expected.
Table 8. Average Beam Demand Satisfaction Percentage for artificially generated demand 1
Illumination Ratio
Artificially Generated Demand 1
¼ 1/6 1/8
Clustering Option 1 (17 clusters) 99.95% 97.27% 73.28%
Clustering Option 2 (11 clusters) 87.25% 78.72% 65.53%

Table 9. Minimum Beam Demand Satisfaction Percentage for artificially generated demand 1
Illumination Ratio
Artificially Generated Demand 1
¼ 1/6 1/8
Clustering Option 1 (17 clusters) 96.39% 75.25% 62.64%
Clustering Option 2 (11 clusters) 49.35% 39.11% 31.21%

The results obtained with the benchmarks for the same demand instance are reported in Table 10. In
this case, neither the BH solution proposed by the team nor the ESA BH solution are superior to the
proposed CH design. In fact, in all illumination ratios, the proposed CH design achieves lower unmet
capacity.

Table 10. Benchmarks Results in Gbps for artificially generated demand 1 (values in Gbps)
Illumination Ratio
Artificially Generated Demand 1
¼ 1/6 1/8
System Unmet Capacity 9
ESA BH System Unused Capacity 0
Supplied Capacity 17.34
System Unmet Capacity 8 7 9
SnT BH System Unused Capacity 9 3 0
Supplied Capacity 27.52 22.71 17.07

Finally, Table 11 and Table 12 show the average and minimum beam demand satisfaction percentage
for the benchmark schemes, respectively. Clearly, the benchmark solutions provide much lower beam
average satisfaction than the CH design (~10% less).

Table 11. Average Beam Demand Satisfaction Percentage for Benchmarks with Artificial Demand Instance 1
Illumination Ratio
Artificially Generated Demand 1
¼ 1/6 1/8
ESA BH(*) 85.88%
SnT BH 82.61% 85.47% 64.67%

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 34 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
(*) Illumination ratio does not apply

Table 12. Minimum Beam Demand Satisfaction Percentage for Benchmarks with Artificial Demand Instance 1
Illumination Ratio
Artificially Generated Demand 1
¼ 1/6 1/8
ESA BH(*) 47.88%
SnT BH 52.89% 59.22% 47.90%
(*) Illumination ratio does not apply

This experiment confirms the benefits of the CH design for specific clustered-demand configurations.

6.1.1.2 Case 2: Demand not fitting the Cluster Design


Herein, we consider the demand instance depicted in Figure 27, which corresponds to the traffic
demand at 1am extracted from the SnT Traffic Emulator2 [16]. Clearly, this demand instance does not
fit any of the clustering options considered in the FlexPreDem SW Tool. As an example, Figure 28
shows the same demand instance overlapped with the clustering option 1 with 4 beams/cluster.It can
be observed that clusters 12, 15, 16 include beams with very high demand and therefore, we expect
them to be illuminated more often than the others.

Figure 27. Demand Instance 1am

2
https://wwwfr.uni.lu/snt/research/sigcom/sw_simulators/satellite_traffic_emulator

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 35 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 28. Clustering Option 1 (17 Clusters) with demand instance 1am (cold=blue, yellow=warm, red=hot)

Assuming ¼ illumination ratio, the CH solution selects 7 snapshots for the final illumination pattern.
Details on the selected 7 snapshots are provided in Table 13 and Figure 29. In particular, Table 13
shows the clusters that are active on each snapshot (highlighting the high demand ones in red)
together with the assigned Tslots; while Figure 29 shows the geographical location of the active beams
corresponding to the selected snapshots. From Table 13, it is evident that high demand clusters are
simultaneously illuminated with low demand clusters, thus it is difficult to obtain a perfect demand
matching.

Table 13. Selected snapshots for CH solution (Clustering Option 1, Demand 1am, 1/4 illumination ratio, TH=256 time-
slots)
Snapshot ID Assigned
Clusters Active
(from 1 to 304) Tslots
65 [ 3 6 12 14 ] 75
104 [ 4 6 11 15 ] 64
287 [ 1 9 16 17 ] 47
95 [ 3 6 8 15 ] 24
116 [ 2 5 13 15 ] 21
293 [ 2 10 16 17 ] 17
89 [ 2 4 7 15 ] 8

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 36 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 29. Snapshots selected for CH solution (Clustering Option 1, Demand 1am, 1/4 illumination ratio)

Next, we evaluate the unmet and unused capacity at a Cluster level. A bar-plot of the cluster demand
and supplied capacity is shown in Figure 30, where the clusters have been ordered according to
decreasing demand. Clearly, there is not a good demand-supplied matching and this is because, as
anticipated, the high demand beams belong to clusters where also low demand beams are included.
Therefore, the conclusion of this analysis is that, ideally, the design of the clusters and the illumination
ratio should be targeted to the expected demand distribution. Otherwise, it is difficult to achieve a
sactisfactory traffic matching

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 37 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 30. Cluster demand and supplied capacity for CH solution (Clustering Option 1, Demand 1am, 1/4 illumination
ratio)

Table 14 shows the system level results obtained with CH solution assuming the demand instance at
1am.
Focusing on the clustering option 1, we can see that the unmet capacity does not suffer strong
variations when changing the illumination ratio (and consequently, the number of active beams on
each snapshot). However, it can be observed that the unused capacity is reduced as the illumination
ratio decreases, which seems to be a direct consequence of reducing the supplied capacity at each
snapshot (via reducing the number of beams that are simultaneously illuminated).
Focusing on the clustering option 2, we now illuminate bigger clusters but keeping similar number of
active beams compared to clustering option 1. This suffers the same problems as clustering option 1
but we note that the illumination ratio 1/8 considers only one cluster illuminated per snapshot, which
avoids the issue of illuminating hot spots together with cold spots. If we have a look at the results in
Table 14, we can confirm that the Clustering Option 2 with 1/8 illumination ratio provides the lowest
unused capacity value, while the unmet remains quite high (which is justified by the size of the cluster
~6 beams).

Table 14. CH Results in Gbps for demand instance 1am (values in Gbps)
Illumination Ratio
Demand Instance 1am
¼ 1/6 1/8
Clustering System Unmet Capacity 12 12 13
Option 1 (17 System Unused Capacity 20 15 7
clusters) Supplied Capacity 32.13 27.28 18.87
Clustering System Unmet Capacity 10 10 13
Option 2 (11 System Unused Capacity 23 14 4
clusters) Supplied Capacity 37.13 27.49 14.85

Table 15 shows the average beam demand satisfaction percentages for the CH solution. The
satisfaction percentage decreases as the illuminated ratio decreases. In other words, the less number
of active beams per snapshot, the lower the satisfaction percentage. Table 16 provides the minimum
of the beam demand satisfaction percentages. Note that the minimum beam satisfaction percentage
refers to the beam with the worst demand satisfaction percentage. It can be observed that the values
are between 18-26 %.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 38 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Table 15. Average Beam Demand Satisfaction Percentage for demand instance 1am
Illumination Ratio
Demand Instance 1am
¼ 1/6 1/8
Clustering Option 1 (17 clusters) 83.30% 73.43% 70.34%
Clustering Option 2 (11 clusters) 86.85% 84.68% 72.36%

Table 16. Minimum Beam Demand Satisfaction Percentage for demand instance 1am
Illumination Ratio
Demand Instance 1am
¼ 1/6 1/8
Clustering Option 1 (17 clusters) 22.07% 18.33% 26.45%
Clustering Option 2 (11 clusters) 25.25% 25.69% 21.61%

Next, we have the results obtained with the benchmark beam hopping schemes. Table 17 shows the
values of unmet and unused capacity for the benchmark schemes. As expected, the loose constraints
of ESA BH design provide very good demand matching with zero unused capacity and only 10 Gbps
of unmet capacity. This means that all supplied capacity is provided where needed but yet this is not
enough to satisfy the demand.
The SnT BH solution provides worse numbers that the CH solution, with higher unused and unmet
capacity in general. Again, the strict snapshot definition of the conventional BH not tailored to the
specific demand instance generates a very unbalanced situations.

Table 17. Benchmarks Results in Gbps for Demand Instance 1am (values in Gbps)
Illumination Ratio
Demand Instance 1am
¼ 1/6 1/8
System Unmet Capacity 10
ESA BH(*) System Unused Capacity 0
Supplied Capacity 14.02
System Unmet Capacity 14 15 15
SnT BH System Unused Capacity 17 14 9
Supplied Capacity 27.59 23.81 17.77
(*) Illumination ratio does not apply

Finally, Table 18 and Table 19 show the average and minimum beam demand satisfaction percentage
for the benchmark schemes, respectivey. Clearly, the SnT BH solution provides much lower beam
average satisfaction than the CH design (~10% less).

Table 18. Average Beam Demand Satisfaction Percentage for Benchmarks with Demand Instance 1am
Illumination Ratio
Demand Instance 1am
¼ 1/6 1/8
ESA BH(*) 92.03%
SnT BH 73.30% 70.74% 63.49%
(*) Illumination ratio does not apply

Table 19. Minimum Beam Demand Satisfaction Percentage for Benchmarks with Demand Instance 1am
Illumination Ratio
Demand Instance 1am
¼ 1/6 1/8
ESA BH(*) 12.23%
SnT BH 21.2% 22.64% 26.38%
(*) Illumination ratio does not apply

The results of this test revealed that a cluster-fixed CH, namely the strict clustering definition, results
in constraining the flexibility of the technique, which is degrading the performance. Cluster design
shall be more flexible and ideally dynamically adapted to the demand.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 39 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
6.1.2 IMPACT OF INTER-CLUSTER INTERFERENCE

We focus here on the artificially generated demand instance 1 (see Figure 22). All results will be based
on the Clustering Option 1, where we have 17 Clusters of 4 beams each. We run the CH solution for
different illumination ratios and the results are reported in Table 20 for illumination ratio ¼, and in
Table 21 for illumination ratio 1/6. As expected, the supplied capacity reduces when the inter-cluster
interference is considered. This could be a problem in cases where the demand is very high and we
cannot satisfy it with the considered spacing between clusters. However, in the cases reported n Table
20 and Table 21, the reduction of supplied capacity due to inter-cluster interference is not really a
problem, because the demand is low and in general we have enough resources to satisfy the demand.

Table 20. Results in Gbps w/ and w/o inter-cluster interference (Clustering Option 1, Illumination Ratio 1/4, Artificially
Generated Demand 1)
w/o inter-cluster interference w/ inter-cluster interference
Demand
Unmet Unused Supplied Unmet Unused Supplied
Cluster 1 0.3077 0 0.1533 0.4610 0 0.1224 0.4301
Cluster 2 3.6923 0 1.7245 5.4168 0 1.1208 4.8131
Cluster 3 1.2308 0 0.5748 1.8056 0 0.3383 1.5691
Cluster 4 0.3077 0 0.1533 0.4610 0 0.1033 0.4110
Cluster 5 0.3077 0 0.1533 0.4610 0 0.0416 0.3493
Cluster 6 1.2308 0 0.5748 1.8056 0 0.2790 1.5098
Cluster 7 3.6923 0 1.7245 5.4168 0 0.7084 4.4007
Cluster 8 1.2308 0 0.5748 1.8056 0 0.1762 1.4070
Cluster 9 0.3077 0 0.1533 0.4610 0 0.0653 0.3730
Cluster 10 3.6923 0 1.7245 5.4168 0 0.3561 4.0484
Cluster 11 0.3077 0 0.1533 0.4610 0 0.0958 0.4035
Cluster 12 1.2308 0 0.5748 1.8056 0 0.0702 1.3010
Cluster 13 3.6923 0 1.7245 5.4168 0 0.8484 4.5407
Cluster 14 1.2308 0 0.5748 1.8056 0 0.3028 1.5335
Cluster 15 0.3077 0 0.1533 0.4610 0 0.0635 0.3712
Cluster 16 0.3077 0 0.1456 0.4533 0 0.0778 0.3855
Cluster 17 2.7692 0 1.2934 4.0626 0 0.7564 3.5257
Total 25.8462 0 12.1312 37.9774 0 5.5264 31.3726

Table 21. Results in Gbps w/ and w/o inter-cluster interference (Clustering Option 1, Illumination Ratio 1/6, Artificially
Generated Demand 1)
w/o inter-cluster interference w/ inter-cluster interference
Demand
Unmet Unused Supplied Unmet Unused Supplied
Cluster 1 0.3077 0 0.0381 0.3458 0 0.0072 0.3148
Cluster 2 3.6923 0 0.3415 4.0338 0 0.2080 3.9003
Cluster 3 1.2308 0 0.1138 1.3446 0 0.0197 1.2504
Cluster 4 0.3077 0 0.0381 0.3458 0 0.0342 0.3419
Cluster 5 0.3077 0 0.0381 0.3458 0 0.0052 0.3129
Cluster 6 1.2308 0 0.1138 1.3446 0 0.0597 1.2905
Cluster 7 3.6923 0 0.3415 4.0338 0 0.0119 3.7042
Cluster 8 1.2308 0 0.1138 1.3446 0 0.1113 1.3420
Cluster 9 0.3077 0 0.0381 0.3458 0.0401 0 0.2676
Cluster 10 3.6923 0 0.3415 4.0338 0.3255 0 3.3669
Cluster 11 0.3077 0 0.0381 0.3458 0 0.0149 0.3226
Cluster 12 1.2308 0 0.1138 1.3446 0.1291 0 1.1016
The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 40 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Cluster 13 3.6923 0 0.3415 4.0338 0 0.1253 3.8176
Cluster 14 1.2308 0 0.1138 1.3446 0 0.0690 1.2997
Cluster 15 0.3077 0 0.0381 0.3458 0.0212 0 0.2865
Cluster 16 0.3077 0 0.2302 0.5378 0 0.2061 0.5138
Cluster 17 2.7692 0 0.2561 3.0254 0 0.1540 2.9233
Total 25.8462 0 2.6499 28.4960 0.5159 1.0264 26.3566

To gain insight, Figure 31 compares the difference between the final per-cluster supplied capacity with
and without inter-cluster interference for two illumination ratios, namely 1/4 and 1/6. As expected,
lower illumination ratio provides lower inter-cluster interference effect. In general, the average loss
due to inter-cluster interference on the overall cluster capacity is of the order of 0.39 Gbps with
illumination 1/4 and 0.13 Gbps with illumination 1/6.

Figure 31. Difference of Supplied capacity w/ and w/o inter-cluster interference (Clustering Option 1, Artificially
Generated Demand 1)

For the sake of completeness, Table 22 analyses the impact of inter-cluster interference at the final
system performance, assuming the artificially generated demand instance 1 and clustering option 1.
As seen before, neglecting the inter-cluster interference generates an increase on the supplied
capacity, which in this particular case reduces the system unmet but increases the system unused.

Table 22. Inter-Cluster interference Results in Gbps with CH for Artificially Generated Demand 1 and Clustering Option 1
(values in Gbps)
Illumination Ratio
Artificially Generated Demand 1
¼ 1/6
w/o inter- System Unmet Capacity 0 0
cluster System Unused Capacity 12 3
interference Supplied Capacity 37.98 28.50
System Unmet Capacity 0 1
w/ inter-cluster
System Unused Capacity 6 1
interference
Supplied Capacity 31.37 26.36

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 41 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
6.1.3 DEMAND-BASED USER SCHEDULING WITHIN A CLUSTER

In this section, we present numerical results to validate the proposed weighted semi-orthogonal
scheduling scheme. The considered benchmarks are semi-orthogonal scheduling, which follows the
same approach as the proposed scheduling but neglecting the demand coefficients 𝛼𝑘 , and demand-
based scheduling, which instead considers solely the priority coefficients 𝛼𝑘 while neglecting the
channel orthogonality.

For the numerical results, the considered cluster is depicted in Figure 32. A specific traffic demand has
been assigned to the different users, and the average traffic demand varies across the different beams.

To give an idea of the demand distribution, and of the temporal variation of the priority coefficients
based on the previously described algorithm, Figure 33 and Figure 34 show two snapshots of the
priorities 𝛼𝑘 for the different users, in the initial instant and in a successive instance respectively. It
can be seen how the priority coefficients are reduced in time based on the provided traffic. Further,
the 6 users scheduled by the algorithm are highlighted by black circles in the figures.

The scheduling performance is assessed both at beam and user level, by evaluating the average per-
beam user rate as well as the rate for each user, and by comparing them with the demand. We
consider a sum power constrained (SPC) system with a per-beam power of 20 dBW and a bandwidth
of 500 MHz.

The per-beam average user rate is shown in Figure 35 and compared with the average per-beam
demand. It can be noticed how the proposed scheme always meets the average demand for each
beam and tends to follow the demand trend. On the other hand, the benchmarks achieve an offered
rate either completely mismatched from the demand (this is the case of the semi-orthogonal scheme)
or not meeting the demand in every beam (this is the case of the demand-based scheduling). It is
worth highlighting how the demand-based scheduling attains a rate that follows quite well the average
demand, however it does not satisfy the demand in beam 4. Moreover, when the constraint of
selecting one user per beam is applied, the demand-based scheduling does not follow the demand
trend anymore.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 42 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 32: Considered beam cluster, with users' positions considered for scheduling.

Figure 33: Snapshot of initial priorities distribution.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 43 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 34: Snapshot of priorities distribution in a successive instant.

Figure 35: Per-beam Average User Rate vs. Average Demand

The per-user rate achieved by the proposed scheme, with respect to the demand, is shown in Figure
36. It can be seen that, also when considering the demanded rate for each single user, the proposed
approach follows the demand trend and satisfies the demand for most users. The per-user rate for
the semi-orthogonal benchmarks is shown in Figure 37, where it can be clearly seen how these
schemes do not follow the demand at all. The per-user rate for the demand-based benchmarks is
shown in Figure 38. It can be observed that the demand-based scheme, follows a bit the per-user

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 44 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
demand trend. However, more users experience unmet capacity with respect to the proposed
scheme, due to the fact that the channel vectors orthogonality are not accounted for in this
benchmark.

Further insights are given in Table 23 where the sum rate and the Jain’s fairness index of users’
satisfaction are provided for the considered scheduling schemes. The latter metric is a measure of
how the provided rate matches the demand at a user level. Specifically, defining the satisfaction um of
the generic user m as the ratio between the offered rate sm and the demanded one dm, the Jain’s
2
(∑𝑀
𝑚=1 𝑢𝑚 ) 1
fairness index is defined as 𝐽 = 𝑀 ∑𝑀 2 . Such index ranges between 𝑀 and 1. The obtained Jain’s
𝑚=1 𝑢𝑚
indexes clearly show how the proposed scheme is the one matching better the demand, with an index
of 0.8, while the achieved sum rate is lower than most of the benchmarks (since most benchmarks
provide unused capacity).

Figure 36: Per-user Rate vs. Demand, Weighted Semi-orthogonal Scheme

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 45 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 37: Per-user Rate vs. Demand, Semi-orthogonal Scheme (one user per beam)

Figure 38: Per-user Rate vs. Demand, Demand-based Scheme (one user per beam)

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 46 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Table 23: Sum rate and Jain's fairness index of users' satisfaction for the considered schedulig schemes, with the second
update approach.
Weighted Semi- Semi-orthogonal Demand- Demand-based
Semi- orthogonal Scheduling (one based Scheduling
orthogonal Scheduling per beam) Scheduling (one per beam)
Scheduling
Sum Rate
8.08 11.8 11.91 6.52 9.35
[Gbps]
Jain’s Fairness
0.98 0.12 0.11 0.99 0.72
Index

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 47 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
7 TECHNOLOGY ROADMAP

The FlexPreDem technology roadmap has been reported in [13]. Here we summarize the main
outcomes.

FlexPreDem project has focused on time flexibility and Beam Hopping, as a technique employing the
beam scheduling in order to match the supplied throughput with the requested beam demands. The
concept of conventional beam-hopping and the FlexPreDem vision are illustrated in Figure 39. Note
however that FlexPreDem has basically assumed beam-clusters of fixed size for the experimental
campaign.

Figure 39. (a) Conventional Beam-Hopping instance - subset of beams being illuminated with the same spectral resource;
(b) FlexPreDem concept, clusters of beams can be illuminated together using the same spectral resource and
implementing precoding.

Both precoding and beam-hopping have been well-studied independently from each other.

In what follows, we highlight the minor obstacles that both techniques are facing or may face before
its transition to operational systems.

7.1.1 TIME-FLEXIBILITY

▪ Dynamic Resource Orchestration: The aforementioned advances in technology flexibility open a


door to advanced resource management strategies for flexible satellite systems, but at the same
time bring new challenges, which have to be addressed so that the full potential of flexible
payloads is unleashed. These include algorithmics for dynamic resource orchestration capable of
exploiting the flexibility of future satellite payloads to maximize the system efficiency by
continuously matching the distribution of the satellite capacity on ground to the geographic
distribution of the traffic demand and following its variations in time. In this context, Machine
Learning (ML) can be foreseen to better predict the resource assignment and prompty react to
changes on the customers’ demands.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 48 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
▪ Security: Satellite services are much more dependent on decisions made on-ground, since flexible
payload are orchestrated from ground segment. Furthermore, a software-defined payload is much
more threaten by cyber-attacks than a non-flexibile analog payload.
▪ Payload Standardization: Thanks to the softwarization of the payload component, equipment or
specific components may be subject to standardization. This can reduce the overall satellite
procurement delays and lower the production costs. However, a large volume of orders is needed
in order to fully benefit from standardization.
▪ Active antenna must be a smart antenna: The future of BH is on the new payload architectures
with active antenna systems. Active on-board antenna systems are currently under development
and are subject to improvements in the coming years. Different aspects need to be further
studied, some listed in the following: (i) the size of the antenna array and payload should be
limited to keep the overall mass within affordable limits, otherwise the launching cost scales fast;
(ii) the issue of the on-board power consumption; where fully-digital architectures are known to
be power hungry.

7.1.2 PRECODING

▪ Feeder Link Congestion: One of the main practical challenges of precoded satellite systems is how
to deal with the bandwidth requirements of the feeder link, which increases proportionally with
the frequency reuse in the user link. Clearly, everything that is transmitted in the user downlink
(i.e. satellite to Earth), has to be previously up-transmitted via the feeder link. The latter
represents a bottleneck for precoded systems, as the feeder link has currently limited capacity.
Some solutions have been proposed to deal with the feeder link spectrum shortage. There is a
current tendency towards Q/V band, where there is a larger bandwidth available than in the Ka
band (currently used for this service). Although this new frequency band can support the required
feeder link capacity, rain attenuation decreases severely the performance with respect to the Ka
band. Another alternative is to consider on-board beam processing at the payload so that the
feeder link shall transfer lower signal rate, but this has been proven not to be enough to satisfy
the feeder link bandwidth requirements, besides adding considerable payload complexity which
is not always possible. Finally, the deployment of several gateways seems to be the most
promising alternative. With a multi-GW deployment, the available spectrum for the feeder link
can be reused among spatially separated gateways through very directive antennas. The multi-
GW deployment not only relaxes the feeder link capacity constraint but can also achieve the best
precoding performance when there is one central coordinating entity in the network in charge of
precoding all user signals. ESA is already investigating the feasibility of such centralized GW
approach [14]. The main drawback of such approach is that increasing the number of GW is a cost
for operators.
▪ Computational Complexity: The next generation of GEO satellite systems is expected to operate
with high number of beams (>200) and implement large antenna arrays composed of multiple
feeds. Therefore, the dimensions of the precoding matrix may scale fast translating into significant
computational complexity when calculating the precoding weights. Note that the complexity of
the precoding matrix with MMSE technique is N3 (N being the number of beams), and this

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 49 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
operation has to be completed within the frame period. Clearly, this is alleviated when considering
cluster-hopping, by selecting a subset of beams to be precoded (i.e. cluster).
▪ Sensitivity to scheduling: The analysis of precoding works is usually done by assuming an
independent upper layer which randomly selects users at each time instance. However, it has
been shown that the user scheduling significant impacts the system performance. In particular,
given the ACM implemented at frame-by-frame, the performance of the encapsulated users
within a frame is dictated by the user with the lowest SINR. As a consequence, and in order to
minimize the performance loss, all the users embedded in the same frame must show similar SINR
values. As expected, assuming a certain user terminal's density, the more users encapsulated in a
single frame, the larger the performance loss [15]. In MEO, the multicast scheduling is somehow
avoided by the fact that MEO systems are usually designed such that each user is centred in its
own beam. Furthermore, MEO traffic types are usually aggregation of multiple terminals such as
maritime or aeronautical, or backhauling traffic, where there is less need for multiplexing multiple
users within a frame.
▪ UT modem: One of the main advantages of precoding is that it requires very little modifications
on the receiver point of view to be applied, most of the complexity is in fact on the gateway side.
The main modification that is required on the UT side the addition on a “precoding-ready” modem
of a logical block responsible for computation of the Channel State Information (CSI) from the
dfferent beams simulaneusly. Currently, in fact, there is no commercial modem that perform CSI
calculation for multiple beams, but several modem manufacturers are already looking at how to
integrate these functionalities in their next generation of chipset. In addition, since precoding
relies on the alignment of the pilots field between the signals directed to different beams that
may use different ModCods, the SuperFrame structure, as described in Annex E of DVB-S2X
standard, would be necessary and this require this new framing format to be integrated on both
the UT and the GW side.

7.1.3 JOINT PRECODING AND TIME-FLEXIBILITY

In the combination of precoding and beam-hopping, all the previously highlighted gaps apply.

In general, beam-hopping reduces the number of active beams and generally assumes the activation
of geographically separated beam-clusters. This translates into a reduced precoding matrix
dimensions, thus reducing complexity of the calculation. Feeder link congestion is also alleviated by
the fact that fewer beams are served at each time instance.

Most of the challenges of a joint precoding and beam hopping architecture come from the
synchronization aspects. These were detailed in Section 4.2 of FlexPreDem TN1 [5].

Besides the synchronization aspects natural from beam-hopped transmission, the main challenge is
the CSI adquistion process in brusty forward link. In fact, during an illumination period, a user can
estimated the CSI that can be used only in the next illumination period. In this case, the blind period
has a non-negligible effect on the error given by imperfect outdated CSI. In particular, at the moment
that a user passes from the non-illuminated to the illuminated phase, it has to rely initially on the CSI
calculated in the previous illumination phase, which in general will be outdated depending on the
structure of the illumination pattern (e.g. assuming non-overlapping clusters, CSI will always be
The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 50 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
outdated at the initial frame). Having said so, we do not foresee the outdated CSI to have a strong
impact. This is because a single super frame is enough to obtain a good channel estimation and,
therefore, the outdated CSI will only be needed for the initial (single) super-frame.

In terms of the illumination pattern design, there are still many improvements to be considered in
future works:

1) Relaxing the pre-definition of beam-clusters: In FlexPreDem, pre-defined clusters have been


assumed and its size and shape have been shown to have a significant impact on the final
performance. It is a natural step to consider next more relaxed cluster definition and adapt it
to the expected average beam demands.
2) Move from pre-defined beam pattern towards full-flexible beamforming: Following the
trend in active antenna systems on-board the satellite, it becomes a critical to re-evaluate the
beam illumination pattern design considering the beamforming weights (applied on-board)
on top of the precoding weights (applied on-ground). This is aligned with the previous point
on relaxed definition of beam-clusters, where the beamforming capabilities allow us to define
each cluster with any size and shape.

7.2 NEXT STEPS

To bring the FlexPreDem concept closer to a reality, the below list of required developments is
suggested:
1) Step 1: Extend FlexPreDem for next generation satellite payloads equipped with (1)
beamforming capabilities; and (2) frequency flexibility.
a. The activities of FlexPreDem have so far assumed a fixed multi-beam pattern. This is
likely to change with the forthcoming new satellite architectures with beamforming
capabilities. The beamforming functions provides a new level of flexibility in adapting
the illuminated area size and shape. This is a simple and natural extension of the
current activity.
b. Furthermore, FlexPreDem has considered time flexibility in terms of beam-hopping.
A dual scenario may be interesting to analyze where beams are always illuminated
but the beam-frequency assignment changes over time, requiring in some cases the
use of precoding.
2) Step 2: HW-based testbed
a. To bring the flexible payloads and the precoding combination closer to a reality, the
next step would be to consider a real-time SDR testbed with real physical layer (fully
compatible with DVB-S2X standard), with the aim to accurately assess the
synchronization / impairment aspects and bring the technology to higher TRL.
3) Step 3: Over-the-Air Demo
a. Final step after successful competition of in-lab demo, would consider a live
demonstration of the technology using real satellite capacity, assuming that suitable
payloads are operational (e.g. SESNext).

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 51 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
8 CCN: FLEXIBLE CLUSTER HOPPING TECHNIQUES

The FlexPreDem main project has focused on the combination of two disruptive techniques:
linear precoding and beam hopping (BH). As a result, the concept of CH was proposed. Let us
review the CH concept and the expected improvements versus conventional BH.

▪ Conventional BH: With BH, all the available satellite resources are employed to provide
service to a certain subset of beams, which is active for some portion of time, dwelling just
long enough to fill the demand in each beam. The conventional beam hopping illumination
pattern is illustrated in Figure 3, where the active spot beams are designed to have a border
area formed by inactive beams such that degree of isolation exists between each sector (i.e.
active beams do not interfere with each other).

Figure 40. Typical illumination pattern of a BH system

▪ Proposed CH: The introduction of precoding in BH systems is required when a set of


adjacent beams is illuminated at the same time with the same frequency resources. We
define such set of adjacent beams as cluster and we refer to the concept of “Cluster
Hopping”. An example of CH illumination pattern is illustrated in Figure 4.

Figure 41. Proposed CH illumination pattern

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 52 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
The design and experimental campaign showed in Section 3-6 of the present document
assumes compact shaped, non-overlapping and equal size cluster due to the following reasons:

▪ In the case of overlapping clusters, there will be a very large number of possible clusters
making a huge search space for the proposed problem and rendering a high complexity
solution. Furthermore, it is difficult to regulate the supplied capacity.
▪ The shape of the clusters has an impact on the precoding performance. Furthermore,
compact clusters are preferred as they will allow to share resources from neighboring
beams.
▪ The size of the clusters has an impact on the complexity and flexibility of the CH
operation. When the clusters are small, it approaches the conventional BH solution
where precoding does not provide any gain. On the other hand, if the size of the clusters
is very high, we will have very less flexibility is supplying capacity, as we will have very
less snap-shots. For the sake of complexity minimization, during FlexPreDem we
tested two different cluster size of 4 beams/cluster and 6 beams/cluster.

After the testing, the results evidenced that the integration of precoding within the so-called
CH concept is beneficial when extended high demand areas cannot be covered with a single
beam.

In fact, we noticed that cluster size and shape should be as close as possible to the expected
demand clusters in order to reach the best demand-matching performance.

The motivation for the FlexPreDem CCN was therefore the following: A more flexible CH
illumination pattern design allowing clusters of different shapes and sizes would improve the
gains achieved with the pre-defined clustering used in FlexPreDem. An example of the
illumination pattern achieved with a more flexible CH is illustrated in Figure 42.

Figure 42. Illumination pattern achieved by Flexible CH

Summarizing, while FlexPreDem considered a pre-defined set of fixed beam-clusters as input


for the illumination pattern design block, the CCN proposes to develop new algorithms for the
illumination pattern design such that the best beam-clusters are determined by the algorithm
itself (see Figure 43).

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 53 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 43. CH illumination design (a) for the main FlexPreDem activity, (b) As proposed in the CCN

FlexPreDem CCN contains 2 Technical Notes, i.e. [17] and [18]. Both are summarized in the
following sections.

8.1 FLEXIBLE CH TECHNIQUES

During FlexPreDem CCN, 3 techniques were proposed for flexible CH. These are described in
the subsequent sections.

8.1.1 EXTENSION OF BH-SNT SOLUTION

The first natural step was to find a solution between the conventional BH and the proposed
CH, where adjacent beams are activated (and precoded) only if needed. Therefore, we first
propose a simple extension of the BH benchmark proposed by the team (so called BH-SnT
solution) in the main FlexPreDem activity.

In the main FlexPreDem activity we formulated a fair demand-satisfaction problem making


use of the max-min of the ratios (supplied capacity / demand). We first express the achievable
data transmission of beam 𝑛 during all 𝑇 time slots as

𝑅𝑛 = ∑𝑇𝑡=1 𝑥𝑛 [𝑡] Δ 𝑇 𝑓(𝑆𝐼𝑁𝑅𝑛 [𝑡]).

Where the variables 𝑥𝑛 [𝑡] are the binary variable which are used to define the beam
illumination solution, i.e.,

1, 𝑖𝑓 𝑏𝑒𝑎𝑚 𝑛 𝑖𝑠 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑒𝑑 𝑖𝑛 𝑡𝑖𝑚𝑒𝑠𝑙𝑜𝑡 𝑡,


𝑥𝑛 [𝑡] = {
0, 𝑖𝑓 𝑏𝑒𝑎𝑚 𝑛 𝑖𝑠 𝑡𝑢𝑟𝑛𝑒𝑑 𝑜𝑓𝑓 𝑖𝑛 𝑡𝑖𝑚𝑒𝑠𝑙𝑜𝑡 𝑡.

Then, the extension of the conventional BH-SnT design problem is stated as in


(Max_Min_Rate), by using the binary optimization vector 𝒙=
𝑇
[𝑥1 [1], … , 𝑥𝑁 [1], … , 𝑥1 [𝑇], … , 𝑥𝑁 [𝑇]] :

𝑅 𝑅 𝑅
(Max_Min_Rate) max min (𝐷1 , 𝐷2 , … , 𝐷𝑁 )
𝑥 1 2 𝑁

s.t. 𝑳𝒙 ≤ 𝒌

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 54 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
where the constraint that ensured activation of non-adjacent beams has been removed. Note
that 𝑳 = 𝑰𝑇×𝑇 ⊗ 𝟏1×𝑁 , which is introduced to account for the number of active beams at each
time-slot instance. Obviously, to make sure this scheme works, a final step needs to be
incorporated where adjacent beams are precoded together to mitigate inter-beam
interference.

By introducing the variable 𝜂, this problem is equivalent to the following one

(P_max_eta) max 𝜂
𝑥,𝜂

s.t. 𝑳𝒙 ≤ 𝒌, or 𝑳𝒙 = 𝒌,

∑𝑡 𝑥𝑛 [𝑡]Δ 𝑇 𝑓(𝑆𝐼𝑁𝑅𝑛 [𝑡]) ≥ 𝜂𝐷𝑛 , ∀𝑛

Problem (P_max_eta) is non-convex and non-linear mainly due to the presence of a binary
variable (𝒙) and the interference term in the SINR expression.

Regarding the interference, the interference coming from adjacent beams can be assumed
mitigated by the use of precoding, while the interference coming from non-adjacent beams is
generally low and can be assumed negligible. Therefore, the first simplification step is to re-
formulate (P_max_eta) using SNR instead of SINR,

(P_max_eta’) max 𝜂
𝑥,𝜂

s.t. 𝑳𝒙 ≤ 𝒌,

∑𝑡 𝑥𝑛 [𝑡]Δ 𝑇 𝑓(𝑆𝑁𝑅𝑛 [𝑡]) ≥ 𝜂𝐷𝑛 , ∀𝑛

We can further simplify problem (P_max_eta’) by considering the demand vector 𝒅 =


[𝐷1 , … . , 𝐷𝑁 ]𝑇 and writing the offered capacity vector as follows,

𝒔 = ∑ 𝒙[𝑡] ⊙ 𝑹
𝑡

𝑇
where 𝒙[𝑡] = [𝑥1 [𝑡], … , 𝑥𝑁 [𝑡]] , ⊙ denotes the element-wise product, and 𝑹 = [𝑟1 , … , 𝑟𝑁 ]𝑇 , is a
vector that contains the interference-free offered throughput per beam (which is independent
of the time-instance as it ignores the interference term).

Therefore (P_max_eta’) transforms into,

(P_max_eta’’) max 𝜂
𝑥,𝜂

s.t. 𝑳𝒙 ≤ 𝒌, or 𝑳𝒙 = 𝒌,

𝒔 ≥ 𝜂𝒅,

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 55 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Although (P_max_eta’’) still contains a binary variable, some state-of-the-art solvers, e.g.,
MOSEK3, can be applied to obtain a solution.

8.1.2 PENALTY-BASED MINIMIZATION TECHNIQUES

Herein we present a different approach where the objective is generalized to a minimization


of a given penalty function subject to demand constraint. This approach is casted by the
following problem

(Min_Pen) min 𝒙𝑇 Ω𝒙
𝒙

s.t. 𝑳𝒙 ≤ 𝒌,

∑𝑡 𝑥𝑛 [𝑡]Δ 𝑇 𝑓(𝑆𝐼𝑁𝑅𝑛 [𝑡]) ≥ 𝐷𝑛 , ∀𝑛

where Ω ∈ R(NT×NT) stands for the penalty matrix whose element represents a penalty cost
when the corresponding two beams are activated/illuminated in timeslot 𝑡, 𝑳 = 𝑰𝑇×𝑇 ⊗ 𝟏1×𝑁 ,
𝑇
𝒌 = [𝐾1 , 𝐾2 , … , 𝐾𝑇 ]𝑇 , and 𝑥 = [𝑥1 [1], … , 𝑥𝑁 [1], … , 𝑥1 [𝑇], … , 𝑥𝑁 [𝑇]] .

As can be seen, MIN_PEN problem is formulated in a general form where the penalty matrix
can be flexibly determined according to the design goals. In FlexPreDem CCN, we have focused
on the interference minimization and, as a consequence, we have use a penalty matrix based
on the interference generated by the activated beams.4

In particular, the design aims to reduce the cross interference among the illuminated beams
in an attempt to avoid the use of precoding whenever possible. The potential solution for this
design can be achieved if we consider the penalty matrix which is decided based on estimating
the cross interference between two beams if they are illuminated at the same time. In
particular, Ω can be defined as follows

Ω = 𝐼𝑇×𝑇 ⊗ Φ,

where ⊗ stands for the Kronecker product, Φ ∈ 𝑅𝑁×𝑁 and its elements are estimated as follows
(assuming a super-user in the beam centre representing the set of users within the beam):

ℎ𝑗𝑖𝑝𝑖
Φ𝑖,𝑗 = 𝑗 .
ℎ 𝑗 𝑝𝑗

The (Min_Pen) problem has a non-convex constraint related to the demand satisfaction:

∑ 𝑥𝑛 [𝑡]Δ 𝑇 𝑓(𝑆𝐼𝑁𝑅𝑛 [𝑡]) ≥ 𝐷𝑛 , ∀𝑛


𝑡

The non-convexity comes in part from the interference in the denominator of the SINR. The
strong interference from the adjacent beams can be eliminated while the interference from

3
https://docs.mosek.com/modeling-cookbook/mio.html

4
Other penalties can be considered, e.g. computation cost of precoding or the number of time of switching on/off beams

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 56 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
non-adjacent beams is exceptionally low thanks to the narrow beam pattern design of satellite
systems. Then, the SINR can be approximated by SNR as

𝑆𝐼𝑁𝑅𝑛 [𝑡] ≈ 𝑆𝑁𝑅𝑛 [𝑡] = ℎ𝑛𝑛 [𝑡]𝑝𝑛 [𝑡]⁄𝜎 2 (interference-free model)

Then, the data-demand constraint can be transferred to a linear form:

𝑪𝒙 ≥ 𝒅

where 𝒅 = [𝐷1 , … . , 𝐷𝑁 ]𝑇 is the demand vector and

𝑪 = Δ 𝑇 [𝑑𝑖𝑎𝑔 (𝑓(𝑆𝑁𝑅1 [1]), … , 𝑓(𝑆𝑁𝑅𝑁 [1])) … 𝑑𝑖𝑎𝑔 (𝑓(𝑆𝑁𝑅1 [𝑇]), … , 𝑓(𝑆𝑁𝑅𝑁 [𝑇]))].

Where f(.) denotes the function to translate SNR into capacity (involving the DVB-S2(X) ACM
function).

The approximated problem can be stated as

(Min_Pen) min 𝒙𝑇 Ω𝒙
𝒙

s.t. 𝑳𝒙 ≤ 𝒌

𝑪𝒙 ≥ 𝒅

𝑥𝑛 [𝑡] ∈ {0,1}, ∀𝑛, 𝑡

The main challenging issue for solving Min-Pen problem comes from the binary constraint.
To overcome this issue, we apply feasible point pursuit successive convex approximation (FPP-
SCA), which not only linearizes the non-convex parts of the problem as conventional SCA does,
but also adds slack variables to sustain feasibility, and a penalty to the objective function to
ensure the slacks are sparingly used. More details can be found in TN8.1 Section 3.2.3 [17].

8.1.3 QUEUE-BASED TECHNIQUES

So far, the objective has been to obtain the full illumination pattern design with a single shot
optimization. While this has been shown to provide good results, sometimes the
computational complexity may become significant as that the size of the variable set increases
linearly with the number of time slots.

To overcome this, transferring a window-model optimization problem into a sequence of


snapshot-model problems using the coupling weights which are defined relatedly to the buffer
lengths or remaining data in the queues has been proposed.

Employing this approach, we first model the demand in each beam as a queuing process where
𝑞𝑛 [𝑡] is the queuing length at beam 𝑛 in time slot 𝑡. On another hand, 𝐴𝑛 [𝑡] denotes the length
of data arriving for transmission at beam 𝑛 in time slot 𝑡. Then, the queuing length at beam 𝑛
in time slot 𝑡 + 1 can be written as

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 57 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
𝑞𝑛 [𝑡 + 1] = 𝑞𝑛 [𝑡] + 𝐴𝑛 [𝑡] − ∆ 𝑇 𝑓(𝑆𝐼𝑁𝑅𝑛 [𝑡])

The throughput-optimal beam hopping policy for time slot 𝑡 can be casted by the following
problem

(Throughput-Optimal) max ∑𝑁
𝑛=1 𝑥𝑛 [𝑡]𝑔(𝑞𝑛 [𝑡])𝑓(𝑆𝐼𝑁𝑅𝑛 [𝑡])
𝒙[𝑡]

s.t. ∑𝑁
𝑛=1 𝑥𝑛 [𝑡] ≤ 𝐾𝑡

where 𝑔(. ) is an effective function which is employed to manage the impact of queuing lengths
in the optimization problem.

Two different weighting functions 𝑔 have been used to define the queuing methods used for
the tests. One proportional to the queuing length 𝑞𝑛 [𝑡] called Linear-Weight Queuing-based
method (LWQ), and another inversely proportional to 𝑞𝑛 [𝑡], called Hyperbolic-Weight
Queueing-based method (HWQ)

For the model with uncertain window length or 𝑇 ≫ 1, the CH and BH design can be optimized
by employing this approach and modelling the demand in each beam as a queuing process.
For the sake of simplicity, the initial queue status 𝑞𝑛 [𝑡 = 0] is set based on the beam demand
𝐷𝑛 . Next the algorithm executes (Throughput-Optimal) at each time instance “t” and updates
the queue status based on the supplied throughput per beam.

8.2 NUMERICAL EVALUATION (SUMMARY)

This section gathers an overview of the results obtained with the techniques that have been
shortlisted and presented in Section 8.1 of the present document. The complete numerical
evaluation can be found in TN8.2 [18].

8.2.1 FLEXIBLE CH VERSUS NON-FLEXIBLE CH

Objective: Observe the benefits of the proposed flexible cluster hopping techniques versus
the cluster hopping techniques proposed in the main FlexPreDem activity, where a fixed and
predefined clustering was considered.

Test Parameters:
▪ Per-Beam BW = 500 MHz
▪ Payload RF power = 6.000 Watts. The per-beam power depends on illumination ratio (or
on the number of active beams per snapshot) and can be expressed as,
𝑃total
𝑃beam [𝑑𝐵] = 10 ∙ log10 ( ) − Loss [dB]
Num. Active Beams per Snapshot

Where 𝑃total = 6000 W in this case, and Loss denote the power loss considering Output
Power Backoff (OBO) and additional losses in the RF chain. Unless otherwise mentioned,
we take Loss=5 dB.
▪ Single user per-beam located in beam center.
▪ Operating Frequency = 19.5 GHz
▪ CH window fixed to 256 time-slots (of 1.3ms each)
The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 58 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
▪ The results are obtained in terms of DVB-S2x capacity and considering interference from
all active beams after eventual application of precoding
▪ For fixed-cluster hooping we consider the clustering option (1 or 2)
▪ For a fair comparison with the fixed-cluster hooping, we assume the following illumination
ratios (1/4, 1/6 or 1/8)
o Note that we have 67 beams in total, and 4, 6 and 8 are not common denominators
of 67:
▪ 67 / 4 = 16.75
▪ 67 / 6 = 11.17
▪ 67 / 8 = 8.38
o To address this issue, for each illumination ratio, we propose the configurations
given in Table 24. In the same table we indicate the maximum number of beams
that can be activated with the flexible clustering.

Table 24. Number of clusters and number of clusters per snapshot for two different clustering options.
Number
Illumination Ratio
Configurati of beams
on per
1/4 1/6 1/8
cluster
Number of active clusters
4 3 2
Clustering per snapshot
4
Option 1 Average number of active
16 12 8
beams per snapshot
Number of active clusters 3 2 1
Clustering per snapshot
6
Option 2 Average number of active 18 12 6
beams per snapshot
Flexible Number of active beams 17 11 8
Not fixed
Clustering per snapshot

▪ It is also recommended to keep in mind the number of possible snapshots for each
illumination ratio and clustering option.

Table 25. Number of possible snapshots


Illumination Ratio
Configurations
1/4 1/6 1/8
Clustering
304 263 101
Option 1
Clusturing
36 35 11
Option 2
Flexible Adaptive depending on the
Clustering demand

KPI to be measured:
The results will be provided in terms of:
- System Supplied Capacity
- System Demand
- System Unmet Capacity, defined as

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 59 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
𝑁𝑏𝑒𝑎𝑚𝑠

C𝑢𝑛𝑚𝑒𝑡 = ∑ max(𝑑𝑖 − 𝑠𝑖 , 0)
𝑖=1
where 𝑑𝑖 and 𝑠𝑖 denote the demand and the supplied capacity of i-th beam,
respectively.
- System Unused Capacity, defined as
𝑁𝑏𝑒𝑎𝑚𝑠

C𝑢𝑛𝑢𝑠𝑒𝑑 = ∑ max(𝑠𝑖 − 𝑑𝑖 , 0)
𝑖=1

- Another interesting figure is the beam demand satisfaction percentage, which is


defined as,
𝑠𝑏
Satisfaction beam "𝑏" (%) = 100 ∙ min ( , 1)
𝑑𝑏

For the sake of clarity, we show here results for the demand instance at 1am from the SnT
Traffic Emulator. Other demands were tested in TN8.2 [18].

Results: Non-Flexible CH proposed in main FlexPreDem activity:


The CH was already shown to have difficulties in following the demand was not fitting the pre-
defined cluster design (see Section 6.1.1.2). This is again confirmed by the two figures below,
showing the demand-matching performance of the non-flexible CH design.

Figure 44. Cluster-Hopping Option 2 Results with Illumination 1/4: Beam demand-matching

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 60 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 45. Cluster-Hopping Option 2 Results with Illumination 1/4: Beam demand-satisfaction

Results: Flexible CH – Extension of BH-SnT Solution


With the extension of BH-SnT solution, the figures below show a much better demand
matching, confirming that the flexible design is able to adapt to any input of demand.

Figure 46. Beam demand matching for Flexible Beam-Hopping with illumination 1/6 (Demand 1am)

Figure 47. Beam demand satisfaction percentage for Flexible Beam-Hopping with illumination 1/6 (Demand 1am)

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 61 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Results: Flexible CH – Penalty-based Minimization
While evaluating the same demand instance with the proposed penalty-based SCA method,
we can observed a more accurate demand-matching (see Figure 48 and Figure 49).

Figure 48. Beam demand matching for SCA Penalty Method with illumination 1/4 (Demand 1am)

Figure 49. Beam demand satisfaction percentage for CSA Penalty Method with illumination 1/4 (Demand 1am)

Results: Flexible CH – Queue-based Techniques


Finally, the queue-based technique is shown to provide a very good performance in terms of
demand satisfaction as observed in Figure 50 and Figure 51.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 62 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 50. Beam demand matching for Queuing-based Method with illumination 1/4 (Demand 1am)

Figure 51. Beam demand satisfaction percentage for Queuing-based Method with illumination 1/4 (Demand 1am) for
LWQ.

Detailed comparison of Non-Flexible CH and the queue-based method:

Table 26 shows the system unmet and unused capacity as well as the supplied capacity for the
non-flexible CH solution; while Table 27 shows the same for the (LWQ) queueing-based
method. One can easily identify the superior performance of the flexible CH queueing-based
method, providing much lower unmet and unused capacity in all illumination ratios.

Table 26. Cluster-Hopping Results for Demand 01:00am: Unmet and Unused Capacity
Illumination Ratio
Demand Instance 1am
¼ 1/6 1/8
System Unmet
10.93 Gbps 11.34 Gbps 12.01 Gbps
Clustering Capacity
Option 1 (17 System Unused
18.11 Gbps 13.43 Gbps 7.00 Gbps
clusters) Capacity
Supplied Capacity 31.10 Gbps 26.10 Gbps 19.02 Gbps
System Unmet
Clustering 9.08 Gbps 9.14 Gbps 13.00 Gbps
Capacity
Option 2 (11
System Unused
clusters) 21.95 Gbps 12.32 Gbps 3.83 Gbps
Capacity

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 63 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Supplied Capacity 36.90 Gbps 27.20 Gbps 14.85 Gbps

Table 27. LWQ Queuing Method Results in Gbps for demand instance 1am
Illumination Ratio
Demand Instance 1am
¼ 1/6 1/8
System Unmet Capacity 2.18 Gbps 1.974 Gbps 4.469 Gbps
System Unused Capacity 1.872 Gbps 2.779 Gbps 1.057 Gbps
Supplied Capacity 23.712 Gbps 24.825 Gbps 20.608 Gbps

8.2.2 COMPUTATIONAL COMPLEXITY

Focusing on the penalty-based SCA design and the queue-based approach, Table 28 shows
their Matlab implementation running time for one demand realization which is estimated by
calling Tic-toc Matlab function on PC with Intel Core i7-1085H CPU @ 2.7GHz. As can be
seen, the implementation time sue to queueing method is extremely short comparing to that
of the SCA-based method, making it the preferable option.

Table 28. The Matlab implementation running time of SCA and Queueing-based methods for one demand realization.
Method Illumination ratio Illumination ratio Illumination ratio
1/4 1/6 1/8

SCA penalty 2.1355 * 1e4 secs 2.0684 * 1e4 secs 1.9804 * 1e4 secs

Queueing-based 1.5276 secs 1.4345 secs 1.4413 secs


(LWQ)

Queueing-based 1.49057 secs 1.4231 secs 1.4389 secs


(HWQ)

8.2.3 COMPARISON WITH 4 COLOR SCHEME

For comparison purposes, in this section we present the numbers obtained when assuming
the conventional 4 color scheme (4CR). The power and bandwidth allocation of beam hopping
(for certain illumination ratios) and 4CR is illustrated in Table 29.

Table 29. Power and Bandwidth allocation


Illumination Ratio (FFR)
4CR 1/4 1/6 1/8
Beam Power 14.5208 dBW 20.4770 dBW 22.3676 dBW 23.7506 dBW
Max. Active 67 17 11 8
Beams
Max. Power 6000 W 6000 W 6000 W 6000 W
Satellite

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 64 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Beam 125 MHz 500 MHz 500 MHz 500 MHz
Bandiwdth
(*) Note that each beam considers a 5 dB loss, composed of 3dB OBO and 2dB of additional
losses
(**) We assume a single polarization for all schemes

The 4CR is illustrated in Figure 52.

Figure 52. 4CR beam pattern

As an example, we consider the demand instance at 01:00am from the traffic emulator. The
results below are obtained:
(*) Unmet = 9.5881 Gbps
(*) Unused = 19.9173 Gbps
(*) Supplied = 34.3491 Gbps
(*) Satisfaction Rate = 93.2102 %
(*) Min Satisfaction Rate = 12.7658 %
(*) Demand = 24.02 Gbps
(*) Met = 14.4319 Gbps
(*) Average Beam Availability = 100 %
(*) Min Beam Availability = 100 %
(*) Satisfaction Rate 4CR = 93.2102 %

Figure 53 and Figure 54 show the offered per-beam capacity versus the demand per-beam. As
expected, the 4CR scheme provides a uniform offered capacity per-beam, which clearly does
not adapt to the demand. Finally, Figure 55 shows the beam demand satisfaction, which is not
bad as there are many beams with low demand for this case. However, the 4CR struggles in
meeting high demands and adapting to the uneven beam demands.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 65 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 53. Beam offered capacity versus demand (4CR scheme)

Figure 54. Beam offered capacity versus Demand (beams ordered according to decensing demand- 4CR scheme)

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 66 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
Figure 55. Beam Demand Satisfaction (beams ordered according to decensing demand- 4CR scheme)

To facilitate the comparison, Table 30 shows the comparison of 4CR results with the queue-
based technique (results presented in Test 1 of this document). Clearly, the flexible technique
is able to adapt the supplied capacity to the demand, minimizing the unmet and unused system
capacity and achieving a higher satisfaction rate.

Table 30. Comparions of queue-based technique and 4CR


4CR Results Queue-Based Technique

System Unmet Capacity 9.5881 Gbps System Unmet Capacity 1.974 Gbps

System Unused Capacity 19.9173 Gbps System Unused Capacity 2.779 Gbps

Supplied Capacity 34.3491 Gbps Supplied Capacity 24.825 Gbps

Satisfaction Rate 93.2102 % Satisfaction Rate 98.8%

Furthermore, it shall be highlighted that the amplified bandwidth per time-slot is drastically
reduced when compared with 4CR as shown in Table 31.

Table 31. On-Board Amplified BW comparison


Any flexible beam hopping with
4CR Results
illumination 1/6

67 beams x 125 11 beams x 500 MHz


On-Board Amplified BW MHz = On-Board Amplified BW = 5.5 GHz

8.38 GHz

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 67 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
8.2.4 EVALUATION OF TRANSMIT POWER PER TIME-SLOT

Another interesting figure of merit is the total transmit power that is used on each time-slot,
which generally increases with the number of beams that are simultaneously activated in that
particular time-slot.

Focusing on the SCA-based interference minimization methods and both the LWQ and HWQ
queueing-based methods, Figure 56 shows the transmit power (in Watts) per time-slot
considering the demand instance at 1am from the traffic emulator. When comparing the
results in Figure 56 with the so-called CH benchmark, we notice that HWQ method consumes
less than 50% of the transmit powerfor approximately half of the beam hopping window time.
This is because the CH benchmark considers a fixed number of beams to be activated at each
time slot, while the new methods developed within the FlexPreDem CCN are able to reduce
the number of active beams whenever possible.

Figure 56. Transmit power per time-slot for different methods (traffic demand 1am from traffic emulator)

8.2.5 CONCLUSIONS

After evaluating the new flexible CH techniques with different demand instances, we can
confirm that the new proposed techniques are able to provide very good demand
satisfaction with any type of input demand. In particular, we observed that, with
appropriate illumination ratio, the techniques can provide an average beam demand
satisfaction >95% in most of the cases (only exception occurs when demand is extremely high
over the whole coverage area and there are not enough satellite resources to satisfy this).

Among the 3 proposed techniques, the extremely low computational time of the queueing-
based technique (which is based on a time-slot by time-slot decision-making process)
combined with its good performance make it the most attractive for practical systems.

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 68 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP
9 REFERENCES

[1] Freedman, A. and Rainish, D. and Gat, Y., “Beam Hopping – How To Make it Possible”, Ka and
Broadband Communication Conference, Bologna, Italy, Oct. 2015.
[2] C. Rohde, H. Stadali, J. Perez-Trufero, S. Watts, N. Alagha, and R. De Gaudenzi, “Implementation
of DVB-S2X Super-Frame Format 4 for Wideband Transmission,” in International Conference on
Wireless and Satellite Systems, 2015.
[3] C. Rohde, N. Alagha, R. De Gaudenzi, H. Stadali, and G. Mocker, “Super-framing: a powerful
physical layer frame structure for next generation satellite broadband systems,” International
Journal of Satellite Communications and Networking, vol. 34, no. 3, 2016.
[4] X. Giraud, G. Lesthievent, H. Méric, “Receiver synchronisation based on a single dummy frame for
DVB-S2/S2X beam hopping systems” International Conference on Telecommunications (ICT), Saint
Malo, France, June 2018.
[5] Freedman, A. and Rainish, D. and Gat, Y., “Beam Hopping – How To Make it Possible”, Ka and
Broadband Communication Conference, Bologna, Italy, Oct. 2015.
[6] C. Rohde et al. “Beam-hopping systems for next-generation satellite communications systems”,
Chapter 14 of IET book “Satellite Communications in the 5G Era”, Edited by S.K. Sharma, S.
Chatzinotas and P.D. Arapoglou, 2018.
[7] ESA FlexPreDem Technical Note 1, TN1 – Mission, system, payload and techniques definition, April
2019.
[8] ESA FlexPreDem Technical Note 2.1, TN2.1 – Flexible precoding demonstrator definition, October
2019.
[9] ESA FlexPreDem Technical Note 2.2, TN2.2 – Flexible precoding demonstrator validation (including
validation plan and validation campaign), April 2020.
[10]ESA FlexPreDem Technical Note 3.2, TN3.2 Flexible precoding demonstrator Experimental Plan,
April 2020.
[11]ESA FlexPreDem User Manual, November 2020.
[12]ESA FlexPreDem Technical Note 4.1, TN4.1 – FlexPreDem Experimentation Results, November
2020.
[13]ESA FlexPreDem Technical Note 4.2, TN4.2 – FlexPreDem Technology Roadmap, November 2020.
[14]ESA CGD “Prototype of a Centralized Broadband Gateway for Precoded Multi-beam Networks”,
https://wwwfr.uni.lu/snt/research/sigcom/projects/esa_cgd
[15]Vázquez MÁ, Pérez-Neira A, Christopoulos D, Chatzinotas S, Ottersten B, Arapoglou P-D, Ginesi A,
Taricco G. Precoding in multi-beam satellite communications: present and future challenges. IEEE
Wirless Commun Mag. 2016;23(6):88-95
[16]H. Al-Hraishawi, E. Lagunas and S. Chatzinotas, "Traffic Simulator for Multibeam Satellite
Communication Systems," 2020 10th Advanced Satellite Multimedia Systems Conference and the
16th Signal Processing for Space Communications Workshop (ASMS/SPSC), Graz, Austria, 2020
[17] FlexPreDem TN8.1 “Flexible Cluster Hopping Techniques”, December 2021.
[18] FlexPreDem TN8.2 “Evaluation of Flexible Cluster Hopping Techniques”, May 2022

The copyright in this document is vested in University of Luxembourg. This document may only be
reproduced in whole or in part, or stored in a retrieval system, or transmitted in any form, or by any means Page 69 of
electronic, mechanical, photocopying or otherwise, either with the prior permission of the University of 69
Luxembourg or in accordance with the terms of ESTEC Contract no 4000123383/18/NL/CLP

You might also like