2018 9th Advanced Satellite Multimedia Systems Conference and the 15th Signal Processing for Space

Communications Workshop (ASMS/SPSC)

Geographical Scheduling for Multicast

Precoding in Multi-Beam Satellite Systems
Alessandro Guidotti and Alessandro Vanelli-Coralli
Department of Electrical, Electronic, and Information Engineering, University of Bologna
Bologna, 40134, Italy. Email: {a.guidotti,alessandro-vanelli}@unibo.it

Abstract—Current State-of-the-Art High Throughput Satellite to fully exploit the available licensed bandwidth and target
systems provide wide-area connectivity through multi-beam ar- much larger throughput levels. In this context, the overall
chitectures. Due to the tremendous system throughput require- system performance is strongly limited by the interference
ments that next generation Satellite Communications (SatCom)
expect to achieve, traditional 4-colour frequency reuse schemes between adjacent beams. It is, thus, of paramount importance
are not sufficient anymore and more aggressive solutions as full to implement advanced interference mitigation techniques at
frequency reuse are being considered for multi-beam SatCom. the receiver, e.g, Multi-User Detection (MUD), [5], or at the
These approaches require advanced interference management transmitter, i.e., precoding, [6]-[9].
techniques to cope with the significantly increased inter-beam The success of multi-user Multiple-Input Multiple-Output
interference both at the transmitter, e.g., precoding, and at the
receiver, e.g., Multi User Detection (MUD). With respect to (MU-MIMO) techniques in terrestrial communications, to-
the former, several peculiar challenges arise when designed for gether with the introduction of the super-frame structure in
SatCom systems. In particular, multiple users are multiplexed the DVB-S2X standard, [10], led the Satellite Communications
in the same transmission radio frame, thus imposing to con- community to assess and implement precoding techniques in
sider multiple channel matrices when computing the precoding multi-beam HTS systems. In [6]-[8], the authors proposed
coefficients. In previous works, the main focus has been on the
users’ clustering and precoding design. However, even though both Zero-Forcing (ZF) and Minimum Mean Square Error
achieving significant throughput gains, no analysis has been (MMSE) algorithms and some considerations on the main
performed on the impact of the system scheduling algorithm implementation challenges as the impact of partial Channel
on multicast precoding, which is typically assumed random. In State Information (CSI) at the transmitter (CSIT). In [11], [12],
this paper, we focus on this aspect by showing that, although practical challenges related to the implementation of precoding
the overall system performance is improved, a random scheduler
does not properly tackle specific scenarios in which the precoding to DVB-S2X HTS systems are discussed, e.g., framing and
algorithm can poorly perform. Based on these considerations, multiple gateways. In [13], [14], the authors provide a review
we design a Geographical Scheduling Algorithm (GSA) aimed at of several precoding techniques and propose an optimisation
improving the precoding performance in these critical scenarios of the linear precoding design, with linear and non-linear
and, consequently, the performance at system level as well. power constraints, [13]. The authors of [15] implemented the
Through extensive numerical simulations, we show that the
proposed GSA provides a significant performance improvement Tomlinson-Harashima precoding (THP) by also taking into
with respect to the legacy random scheduling. account the beam gain. In [16], the authors assessed the
performance of linear beamforming in terms of satisfying
I. I NTRODUCTION specific traffic demands.
During the last years, Satellite Communication (SatCom) Building on these works, the SatCom community has been
systems evolved from the traditional single-beam architec- intensifying the work on precoding-based satellite systems. In
ture to a more advanced multi-beam deployment, aiming particular, since SatCom systems provide large single-link data
at improving the system throughput. State-of-theArt (SoA) rates to the users, multicast precoding techniques have gained
High Throughput Satellite (HTS) systems in Ka-band provide momentum. In this case, in order to enhance the overall system
connectivity to single regions, e.g., Europe, by means of a very level efficiency, multiple users are multiplexed into the same
large number of beams, e.g., more than 100, [1]. The basic PHY codeword. Within this framework, traditional single-user
principle behind this system architecture is the frequency reuse (unicast) precoding is not always applicable and multicast pre-
concept, which is well-known in the terrestrial community, coding techniques can provide an efficient solution as shown in
with SoA HTS systems typically exploiting 4-colour frequency several recent works, [17]-[23]. In [17], the authors proposed
reuse schemes. However, when aiming at providing terabit the design of multicast precoding for satellite systems based
connectivity through satellite systems, different solutions are on a regularised channel inversion and a geographical-based
needed to increase the system throughput, i.e., increasing grouping of users in the same FEC codeword. The computation
either the available spectrum or the spectral efficiency at of the precoding matrix is based on the pragmatic approach
system level. On the one hand, several activities focused on the proposed in [18], in which the authors jointly design the linear
former by means of advanced spectrum sharing technologies, precoding and ground-based beamforming at the gateway.
[2]-[4]. On the other hand, more aggressive frequency reuse The authors of [19] discuss on the implementation of linear
schemes, as full frequency reuse, can be targeted in order precoding techniques to multi-beam broadband fixed satellite

978-1-5386-6035-5/18/$31.00 ©2018 IEEE

 2018 9th Advanced Satellite Multimedia Systems Conference and the 15th Signal Processing for Space
Communications Workshop (ASMS/SPSC)

ment. Section III outlines the traditional random scheduling

algorithm. In Section IV, we introduce and describe the
proposed Geographical Scheduling Algorithm. In Section V,
numerical results are provided comparing the performance of
the random and GSA algorithms. Finally, Section VI concludes
this paper.

II. S YSTEM M ODEL AND P ROBLEM S TATEMENT

A. System model
We consider a HTS GEO multi-beam satellite system op-
erating with a full frequency reuse scheme, [23]. The satellite
Fig. 1. Channel elements for the generic i-th user in the b-th beam.
payload is assumed to be transparent and equipped with NB
colocated transmitting antennas, generating NB on-ground
communications by also also introducing some preliminary beams. A single system Gateway (GW) manages the Channel
consideration on the issues related to users grouping. In [20], State Information (CSI), which in the following is assumed
the optimisation problem of multicast precoding with per- to be ideal, obtained from the Return Link; in order to
antenna power constraints is addressed. In [21], the authors cope with the aggressive frequency reuse scheme, the CSI is
focus on framing multiple users per transmission and on exploited to compute the precoding weights by means of linear
the per-antenna transmit power limitations and propose a precoding techniques. In particular, Minimum Mean Square
solution for frame-based precoding based on the principles Error (MMSE) precoding is considered in the following. Time
of physical layer multicasting to multiple co-channel groups Division Multiple Access (TDMA) is implemented to serve the
under per-antenna constraints. In [22], a two stage linear users in the NB on-ground beams in each time frame, as, for
precoding is proposed to lower the complexity in the ground instance, in DVB-S2 and DVB-S2X, [25], [10].
segment, under the presence of non-ideal CSI as well. In [23], a) User deployment and channel model: focusing on
(b)
we extensively addressed the grouping of users in multicast the generic b-th beam, we assume NU = ⇢Ab uniformly
precoding as a clustering problem, providing an insight on the distributed users with Ab being the beam area and ⇢ the
most critical design aspects to be considered with different user density, assumed to be the same across all beams. The
(b)
power constraints on MMSE precoding, similarity metrics, and NU users are further assumed to be in fixed locations. In
algorithm complexity. this context, the channel coefficient between the generic i-th
The valuable contributions provided in the literature showed user of the b-th beam and the j-th transmitting antenna, with
(b)
significant gains in terms of average spectral efficiency at i = 1, . . . , NU and i, j = 1, . . . , NB , is given by:
system level. However, these analyses did not take into ac- q
(i)
count aspects related to the system scheduler, which have a (i)
GR Gloss Gbj 2⇡ (i)
hbj = e | db e |#b (1)
strong impact on the precoding performance. In particular, db p
(i)
4⇡ PZ,b
as highlighted above, the research efforts have been mainly
focused on the implementation of more advanced precoding where, as shown in Fig. 1: i) GR is the receiver antenna gain;
(i)
algorithms and/or different grouping strategies for multicast ii) Gloss models the overall antenna losses; iii) Gbj is the
precoding. While these aspects are relevant for the precoding multi-beam antenna gain between the j-th antenna feed and
(i)
implementation, they do not capture the whole picture since i-th user in the b-th receiving beam; iv) db is the distance
also the choice of the users to be served in the same time between the satellite and the considered i-th user in the b-
frame has a strong impact on the system performance. th beam, which is the same for all colocated transmitting
In this paper, we move from the work performed in [23] antennas; v) the carrier wavelength; and vi) #b ⇠ U [0, 2⇡)
with respect to the users’ clustering and provide an analysis is the random phase offset that depends on the transmitting
of the system scheduler. In the literature, the users are always antenna only. In addition, PZ,b = Tb B is the noise power
assumed to be randomly selected across the system beams. at the b-th receiving antenna, in which  is the Boltzmann
However, as we will show in the next sections, this choice constant, Tb the clear-sky noise temperature, and B the user’s
can lead to specific scenarios in which the precoded system bandwidth. In this context, the channel between the i-th user
has a worse performance with respect to the non-precoded of the b-th ⇣beam and the N⌘B transmitting antennas is given
(i) (i) (i)
one. Moving from this analysis, we propose and design a by hb = hb,1 , . . . , hb,NB and the signal received on the
Geographical Scheduling Algorithm (GSA), which aims at Additive White Gaussian Channel (AWGN) can be written as:
improving the performance of multicast and unicast precoding q
in these critical scenarios, which results in a significant overall (i) (i) (i) (i)
yb = pb hb x + zb , i = 1, . . . , NU
(b)
(2)
gain at system level as well.
In Section II, we introduce the multi-beam satellite system where: i) x qis the NB ⇥ 1 vector of complex transmitted
(i)
and multicast precoding model, as well as the problem state- symbols; ii) pb is the power allocated to the i-th user in the
 2018 9th Advanced Satellite Multimedia Systems Conference and the 15th Signal Processing for Space
Communications Workshop (ASMS/SPSC)

(i)
b-th beam1 ; and iii) zb is a complex circularly-symmetric in-
dependent and identically distributed (i.i.d.) Gaussian random
variable with zero-mean and unit variance, since the noise term
is included in theqchannel coefficients in (1). In the following,
(i) p
we assume that pb = PT X , 8b, i.
b) MMSE precoding: when traditional unicast precoding
is implemented, one user per beam is selected based on the
scheduling algorithm and a NB ⇥ NB ✓ MIMO channel matrix ◆ T
e is built based on the CSI as H
e = (i) T (i) T
H h1 , . . . , h NB ,
where, since we have one user per beam, we dropped the user
index i for the sake of clarity. This channel matrix is used to
compute the MMSE precoding matrix as follows:
⇣ ⌘ 1
W= H e HHe + diag (↵) IN eH
H (3)
B

Fig. 2. SINR with and without precoding for NB = 71 receiving beams,

where diag (↵) is the diagonal matrix of regularisation factors, MMSE precoding, and random scheduling.
with ↵b = PZ,b /PT X . The precoding matrix W is changed
at each time frame, since the users to be served (i.e., the cor-
responding channel vectors) are different. The signal received (b)
by a feature vector ui , which depends on the similarity
in the precoded system can thus be written as: metric used in the clustering algorithm: i) its location in the
e x + z = HWPx
y = He e +z (4) Euclidean space when Euclidean distance is used as similarity;
(i)
or ii) its vector of channel coefficients, hb , when users are
where: i) xe = WPx is the NB ⇥ 1 vector of precoded trans- grouped based on their distance in the 2NB -dimensional space
mitted symbols; and ii) P is the diagonal matrix p of allocated of channel coefficients. Two algorithms were proposed, one
p p
power levels P = diag p1 , . . . , pNB = PT X INB . with a fixed cluster size and another with variable cluster
In SatCom systems, data sent to different users is multiplexed size, which showed improved performance with respect to
into a single codeword in each time frame and, in addition, SoA solutions. In the following, we assume that the MaxDist
their information bits are also interleaved together, making algorithm is implemented to cluster users together, which is
traditional symbol-level unicast precoding solutions unfeasi- mainly based on the following two steps, [23]: i) at each
ble. To circumvent this issue, multicast precoding has been time frame, a reference user is selected as the farthest from
proposed in which the same precoding matrix is applied to the barycentre of the available users Q(b)2 in the considered
all of the symbols in the same codeword. In this context, similarity space, g(b) ; and ii) a cluster is formed by taking the
the precoding matrix W is constant over a time frame, and K 1 users in Q(b) that are the closest to the reference user.
multiple users, denoted by K in the following, are to be (b)
This algorithm provides a NK l-partition mof the NU users
(b)

properly selected and multiplexed in the same codeword. (b) (b)

in the b-th beam, with NK = NU /K . This partition is
Two main technical challenges arise: i) how to select the
the set of all of the clusters
⇢ in the considered beam and it is
K users to be multiplexed together in each beam, i.e., the (b) (b)
clustering algorithm; and ii) how to process the users’ channel represented by C = C1 , . . . , C (b) .
(b)
NK
vectors in each beam so as to have a single vector per
beam. With respect to the latter, in [17], average precoding B. Problem statement
has been proposed in which an equivalent channel matrix In general, as shown in [17]-[23], precoding provides a
is built by means of a simple arithmetic. In particular, in significant improvement in the overall system performance in
the generic b-th beam, an equivalent channel vector is built terms of average spectral efficiency. However, even though the
as he b = (1/K) PK h(i) , which yields to the average
i=1 b ⇣ ⌘T system average spectral efficiency is significantly improved,
estimated channel matrix H e = h
1
eT
eT , . . . , h . The more
NB
there are scenarios in which the Signal-to-Interference plus
e b is for the single Noise Ratio (SINR) in the precoded system is below that of
representative the equivalent channel vector h
the non-precoded one for specific users. A situation like this is
users’ channel vectors, the more adapted the precoding vector
shown in Figure 2 where in beams 1, 36, and 71 (the details of
will be to the actual channel conditions.
the numerical simulations are provided in Section V) users are
c) Clustering algorithm: in [23], the authors extensively
experiencing a loss in the system performance when precoding
discussed the problem of the user selection for multicast
is implemented. This kind of situations occur when users in
precoding by formulating it as a clustering problem. In par-
adjacent beams that are too closely located are scheduled
ticular, the generic i-th user in the b-th beam is represented
in the same time frame; this happens due to the fact that
1 It shall be noted that this value is inherently different from the power
emitted by each antenna, which is assigned by the precoder. 2 This represents the set of not yet clustered users.
 2018 9th Advanced Satellite Multimedia Systems Conference and the 15th Signal Processing for Space
Communications Workshop (ASMS/SPSC)

the precoder cannot properly allocate the transmission power 8). This condition is required so as to guarantee that, while
across the NB antennas so as to improve the intended signal still serving clusters in the more populated beams, we have
and reduce the interfering ones, when users are too close. users also in the beams with less clusters. If there are still
In particular, this is due to the better precoding performance unserved clusters in the considered beam, the pool of available
when the channel matrix has rows as linearly independent as clusters for the next iteration is updated based on eq. (5)
possible and, as exploited in [24] for user clustering, users (Step 6). The overall scheduling sequence SNffor the random
closely located typically show correlated channel coefficients. algorithm is finally given by SRAN D = n=1 rame
S[n], with
Typically, scheduling is implemented by means of a random S[n] = {s1 [n], . . . , sNB [n]} being the set of clusters scheduled
algorithm , i.e., users are randomly selected across the system in the n-th time frame across the NB beams.
NB beams to be served in the same time frame, and scenarios
in which the precoding performance is poor are quite frequent. Algorithm 1 Random scheduling algorithm
⇢
In particular, approximately in 70% of the transmitted time (b) (b) (b)
frames numerical results have shown at least one user for Input: NK -partition C = C1 , . . . , C (b)
(b)
NK
which the precoded SINR is below the non-precoded one. SNf rame
Output: SRAN Dn= n=1 S[n]
o n o
A similar behaviour can be observed also when multicast (b) (b)
1: Set Ab [1] = 1, . . . , NK , Nf rame maxb NK
precoding is implemented. Based on these observations, in the
2: for n = 1 to Nf rame do
following sections we introduce a Geographical Scheduling
3: for b = 1 to NB do do
algorithm aimed at circumventing this issue.
4: Cluster
⇣ random ⌘ selection: sb [n] ⇠ U (Ab [n])
(b)
III. R ANDOM S CHEDULING 5: if n < NK then
In the considered SatCom multicast precoding system, 6: Update available clusters: Ab [n+1] = Ab [n]\sb [n]
one 7: else
⇢ cluster per beam is selected from the partition C
(b)
=
(b) (b) 8: Re-initialise available clusters: Ab [n + 1] = Ab [1]
C1 , . . . , C (b) at each time frame to be served with 9: end if
NK
TDMA. Let us denote by n, with n 1, the generic time frame 10: end for
index. The set of clusters, one from each beam, served in the 11: return S[n] = {s1 [n], . . . , sNB [n]}
n-th time frame is denoted by S[n] = {s1 [n], . . . , sNB [n]}, in 12: end for
which the generic b-th element sb [n] represents the index of
the cluster selected for the b-th beam. At the first time frame
n = 1, in each beam, all of the clusters are available for IV. G EOGRAPHICAL S CHEDULING
selection, due to the fact that noneh of themi has been previously In this section, we propose a Geographical Scheduling
(b)
served. This means that sb [1] 2 1, NK , 8b. In the generic Algorithm (GSA) aimed at circumventing the issues identified
time frame n > 1, the system scheduler shall select a new in Section III. In particular, the algorithm design principle
cluster to be served from each beam, with the constraint that is that of scheduling clusters in adjacent beams in the same
those already served in previous time frames are not eligible. time frame only if they are not too close. A possible solution
Thus, in the generic time frame n 1, the scheduler shall would be that of selecting these two clusters so as to have the
select the cluster to be served in the generic b-th beam from maximum distance between them. However, this would lead
the following set: to the same critical scenario as the one we are trying to avoid,
as these two cluster might then be too close to the clusters
n[1
in the other adjacent beams. To circumvent this issue, while
Ab [n] = {1, . . . , Kb } \ sb [m] (5)
scheduling together clusters that are not too close to each other,
m=1
we propose to define a certain number of zones within each
It is straightforward to note that Ab [1] = {1, . . . , Kb }, 8b. beam, which will be referred to as scheduling sectors, and only
When focusing on precoded SatCom systems, the schedul- schedule together clusters that belong to the same scheduling
ing algorithm is typically assumed to be random. In particular, sector in their respective beams.
in the generic n-th time frame the cluster to be served,
the index of which was denoted as sb [n], in the b-th beam A. Sectorisation
is randomly chosen from those that are still available, i.e., Focusing on a generic beam, the aim is to define NS
sb [n] ⇠ U (Ab [n]). This scheduling algorithm is referred to scheduling sectors to be exploited by the GSA. The first zone
as Random Scheduling in the following and it is outlined in that we identify is the Beam Center sector, ZBC . This is
Algorithm 1 for the generic b-th beam. It shall be noticed defined as the locus of points the distance of which from the
(b)
that each beam has been partitioned in NK clusters, i.e., the beam center is not exceeding a pre-defined beam center radius.
number of clusters is not constant across the system beams. Since the beams are not circular and, actually, show a different
To cope with this aspect, if the randomly chosen cluster (Step geometry one from another, we introduce a normalised polar
4) is the last one available in the beam, the scheduler shall coordinate system centered in the beam center. In particular,
(b)
re-initialise the pool of available clusters Ab to Ab [1] (Step denoting by vi the location of the generic i-th user, we
 2018 9th Advanced Satellite Multimedia Systems Conference and the 15th Signal Processing for Space
Communications Workshop (ASMS/SPSC)

Algorithm 2 Geographical Scheduling Algorithm

(b) (b)
Input: Zq , NK -partition C (b)
Output: SGSA from
n eq. (10) o
(q) (b) (q)
1: Set Ab [1] = i vi 2 ZBC
(q)
2: for nq = 1 to Nf rame do
3: for b = 1 to NB do do
(q)
4: Cluster
⇣ selection
⌘ in sector q: sb [nq ] ⇠
(q)
U Ab [nq ]
⇣ ⌘
(q)
5: if Ab [nq ] > 1 then
(q)
6: Update available clusters in sector q: Ab [n+1] =
(q) (q)
Ab [n] \ sb [n]
Fig. 3. Sectorisation example for beam 42 with NS = 13. Setup: rBC = 0.2,
7: else r1 = 0.6, r2 = 0.8, r3 = 1; '0 = 0, '1 = ⇡/2, '2 = ⇡, '3 = 2⇡.
(q)
8: Re-initialise available clusters: Ab [n + 1] =
(q)
Ab [1]
9: end if the locus of points having a normalised radius in the range
10: end for n o (rq 1 , rq ] and falling in an angular range ('q 1 , 'q ]:
(q) (q) n o
11: return S (q) [nq ] = s1 [nq ], . . . , sNB [nq ] (b) (b) (b)
12: end for Zq(b) = vi 2 Bb rq 1 < rei  rq , 'q 1 < 'i  'q
n o (8)
(b) S SNZ (b)
with r0 = rBC , ZBC q=1 Z q = B b , ' 0 = 0, and
identify
⇣ ⌘its location by means of its: i) angular⇣coordinate,
⌘ 'NS 1 = 2⇡. It shall be noted that the sectors boundaries,
' vi
(b) (b) (b)
= 'i ; and ii) its normalised radius re vi , i.e., and in particular their radius identified by the range (rq 1 , rq ],
are not depending on the considered beam, since they are
its distance from the beam center normalised to the distance
adapted to its geometry by construction being normalised
between the beam center and⇣the point ⌘ on the beam edge in
(b) (b) values. Finally, it is worth noting that the above identification
the angular direction 'i , R 'i : of the sector in which a user is falling can be easily extended
to the considered multicast scenario. In particular, each cluster
(b)
⇣ ⌘ vi is assumed to be represented by its centroid, i.e., barycenter,
(b) (b)
re vi = rei = ⇣ ⌘ (6) which is then used to identify its sector based on (7)-(8).
(b)
R 'i Figure 3 shows an example of beam sectors for beam 42,
in which the sector barycentres are highlighted.
(b)
By construction, it can be easily verified that 0  rei  1:
(b) B. Geographical Scheduling Algorithm (GSA)
when the selected point vi coincides with the beam center,
(b) (b)
the distance vi = 0 in (6) and rei = 0, while, if the point The proposed GSA scheduling is outlined in Algorithm 2.
⇣ ⌘ Once the beam sectors have been defined as described above,
(b) (b)
is located at beam edge, then we have vi = R 'i and the scheduling algorithm only schedules together clusters
(b)
rei = 1. Based on the normalised radius, we can define the across the NB system beams that are located in the same sector
beam center sector ZBC as the locus of points with distance of their beam. Without any loss of generality, let us assume
from the beam center that is lower than or equal to a predefined that the geographical scheduler serves the users in increasing
normalised radius, rBC : orders of scheduling sectors, i.e., first we serve the clusters
(b) (b)
n o belonging to ZBC and then Zq for q = 1, . . . , NS 1.
(b) (b)
ZBC = vi 2 Bb rei  rBC
(b)
(7) The GSA first identifies the cluster indexes that are related
to clusters in the considered q-th sector (Step 1), denoted by
(q)
where Bb is the locus of points belonging to the b-th beam. Ab , and uses it as the initial pool of available clusters to
In order to define the other scheduling sectors, we identify: be scheduled. At each time frame, the scheduler randomly
i) NR normalised radius values, in addition to rBC , to discern select a cluster to be served from the pool⇣ of those
⌘ that have
(q) (q)
among sectors based on their distance from the beam center; not been served yet, i.e., sb [nq ] ⇠ U Ab [nq ] (Step 4).
and ii) N' angles to discern among sectors based on the The time frame index nq ranges from 1 to the maximum
direction we are observing from the beam center. Thus, the number of frames to be transmitted for that sector, which is
number of sectors identified with the proposed approach is given by the maximum number ofnclusterso in the q-th sector
given by NS = NR · N' + 1, in which we must sum 1 to (q)
across all beams: Nf rame = maxb Zq
(b)
. This condition is
(b)
include the previously defined beam center sector ZBC . The necessary so as to guarantee that all of the clusters belonging
(b)
q-th scheduling sector Zq in the b-th beam is defined as to the q-th sector are served at least once in all beams.
 2018 9th Advanced Satellite Multimedia Systems Conference and the 15th Signal Processing for Space
Communications Workshop (ASMS/SPSC)

Figures 4-5, an example of cluster scheduling with the

geographical and the random scheduler, respectively, for the
first three time frames is provided. When the random scheduler
is implemented, Figure 4, it can be noticed that users are
indeed selected independently from their location. In partic-
ular, during the third time frame in beam 32 the selected
user is extremely close to the user selected for beam 31.
Consequently, as discussed in Section III, these two users will
experience a performance degradation due to the fact that the
precoding algorithm will try to balance the power transmitted
towards them so as to balance the system performance. This
critical scenario is avoided thanks to the novel geographical
scheduler as shown in Figure 5, where users are always
(b)
selected from the beam center sector ZBC and, thus, their
Fig. 4. Example of served users with random scheduling for the first 3 time relative distance is kept under control.
frames. Setup: K = 1. (unicast).
V. N UMERICAL R ESULTS
In this section, we compare the performance of the proposed
GSA with the random scheduling algorithm in terms of
average spectral efficiency. The main simulation parameters
are listed in Table I and the considered multi-beam satellite
system covers the whole of Europe through NB = 71 beams
operating with a full frequency reuse scheme. As for the user
densities across the beams, the largest value is 10 2 users/km2 ,
which is a low density. While this choice is motivated by the
memory and time computational complexity, numerical results
show that it is sufficient to understand the trend for the overall
system performance. Based on this density, during each Monte
(b)
Carlo iteration, NU = [⇢Ab ] users are randomly deployed
in fixed locations in each beam with a uniform distribution.
Finally, we assume a uniform traffic request for the users,
Fig. 5. Example of served users with the proposed GSA for the first 3 time
frames. Setup: K = 1. (unicast).
i.e., all users are requesting the same amount of traffic and no
priorities are present or requested, and the multicast precoding
clusterisation based on the channel coefficient space.
Moreover, this upper bound on nq implies that we transmit For the generic c-th cluster in the b-th beam and n-th time
PNS 1 (q) (BC) (b)
frame, we obtain a rate ⌘c,n (K, ⇢), which is a function of the
Nf rame = q=1 Nf rame + Nf rame frames when serving
clusters in all sectors and beams. In addition, please note that, minimum SINR among the cluster members, since all of them
(b)
for the beam center sector ZBC , we can simply substitute the shall be in condition to decode the received data, i.e.:
⇣ ⌘ n o
index q with BC in Algorithm 2. (b)
⌘c,n (K, ⇢) = f ec(b) , ec(b) = min
(b)
(11)
c,i
Similarly to the Random algorithm, the set of clusters available (b)
i2Cc
for scheduling is continuously updated (Step 6), which leads The function f (·) models the considered Modulation and
to the following definition of the q-th sector3 set of available Coding (ModCod) scheme, which in the following is assumed
users in the generic time frame nq : to be the one provided by the DVB-S2X standard with 64800
n o [nq bits FEC codewords, [10]. Thus, the average spectral efficiency
(q) (q) (q)
Ab = s1 [nq ], . . . , sNB [nq ] \
(q)
sb [m] (9) is computed by averaging over all time frames, clusters, and
m=1 beams: n o
(b)
⌘(K, ⇢) = En,c,b ⌘c,n (K, ⇢) (12)
By repeating Algorithm 2 for all sectors, we obtain the overall
scheduling sequence: Figure 6 shows the average spectral efficiency with users’
8 9 clustering based on the channel coefficient similarity with the
(BC) (q)
Nf rame
[ [ >
< N[S 1 Nf rame
[ >
= random and GSA scheduling algorithms. The overall trend
SGSA = S (BC)
[nBC ] (q)
S [nq ] related to the increase in the cluster size is the same for both
nBC=1
>
: q=1 nq =1 >
; scheduling algorithms and reflects the analysis provided in
(10) [23]. In particular:
• The average spectral efficiency decreases for increasing
3q = BC for the beam center. values of the cluster size. This is due to the fact that the
 2018 9th Advanced Satellite Multimedia Systems Conference and the 15th Signal Processing for Space
Communications Workshop (ASMS/SPSC)

TABLE I
N UMERICAL SIMULATION PARAMETERS

Parameter Value
Carrier frequency 19.5 GHz
Receiving antenna diameter 0.6 m
Receiving antenna efficiency 0.6
Antenna losses 2.55 dB
GEO satellite longitude 30
Satellite transmitted power Psat = 90W
⇢ 2.5 · 10 4 , . . . , 10 2 users/km2
K 1 (unicast), 2, 4, . . . , 12
Target Bit Error Probability 10 5

Sector radii rBC = 0.2, r1 = 0.6, r2 = 0.8, r3 = 1

Sector angles '0 = 0, '1 = ⇡/2, '2 = ⇡, '3 = 2⇡

Fig. 7. Gain in average spectral efficiency with the GSA w.r.t. the random
scheduling algorithm as a function of K for varying user densities ⇢.

Fig. 8. Average SINR per user with the GSA algorithm. Setup: beam 42,
Fig. 6. Average spectral efficiency with the random and GSA algorithms as ⇢ = 2.5 · 10 3 users/km2 .
a function of the cluster size K for varying user densities ⇢.

other, which leads back to the critical scenarios in the random

precoding matrix in (3) is built by averaging the cluster clustering. On the other hand, for larger user densities, the
users’ channel coefficients and, thus, the precoding matrix average gain is quite stable and included between 0.25 and 0.3
will be less representative. This effect is more evident in bit/s/Hz. It can still be noticed that for an increasing cluster
low density scenarios, since users are more spread in the size the gain tends to decrease, with the same motivation as
considered similarity space. for the lowest density. However, in this case the gain would
• Typically, the channel coefficient similarity provides bet- go to zero only for extremely large cluster sizes.
ter performance. This is due to the fact that users that This gain is not only obtained thanks to avoiding critical
are close in the Euclidean space do not necessarily have scenarios, i.e., scenarios in which users from adjacent beams
similar channel coefficients. and close to each other are scheduled in the same time frame.
With respect to the proposed GSA solution, it can be Figures 8-9 shows the average SINR for beam 42 with the
noticed that the performance gain with respect to the random GSA and random algorithms, respectively. It can be noticed
scheduling is significant. This aspect is highlighted in Figure 7, that even users that were distant from the beam edge are
where the gain in average spectral efficiency is shown, i.e., experiencing a significantly improved SINR with the proposed
the average spectral efficiency with the GSA algorithm with GSA. Moreover, the average SINR is more uniform with the
respect to the one obtained with the random scheduling. It GSA, which shows that this solution not only provides an
can be noticed that with the lowest user density, the gain improved average spectral efficiency at system level, but it
rapidly decreases to almost zero for increasing cluster sizes. also improves the overall fairness for the users.
This is due to the fact that we are clustering together more
users in a scenario in which they are very sparse in the beam VI. C ONCLUSIONS
area. Consequently, while the barycentre of each cluster will In this paper, we designed a Geographical Scheduling
denote its sector, the cluster users will be far away from each Algorithm (GSA) for multicast precoding in multi-beam HTS
 2018 9th Advanced Satellite Multimedia Systems Conference and the 15th Signal Processing for Space
Communications Workshop (ASMS/SPSC)

[6] G. Gallinaro, G. Caire, M. Debbah, L. Cottatellucci, R. Mueller, and

“Perspectives Of Adopting Inteference Mitigation Techniques In The
Context Of Broadband Multimedia Satellite Systems,” in Proc. 23rd
AIAA Int. Commun. Sat. Syst. Conf. (ICSSC 2005), Sep. 2005.
[7] L. Cottatellucci, M. Debbah, E. Casini, R. Rinaldo, R. Mueller, M. Neri,
and G. Gallinaro, “Interference mitigation techniques for broadband
satellite system,” in Proc. 24th AIAA Int. Comm. Sat. Syst. Conf. (ICSSC
2006), Jun. 2006.
[8] N. Zorba, M. Realp, and A. Perez-Neira, “An improved partial CSIT
random beamforming for multi-beam satellite systems,” in Proc. 10th
Int. Work. on Sig. Proc. for Sp. Comm. (SPSC 2008), Oct. 2008.
[9] P.-D. Arapoglou, K. Liolis, M. Bertinelli, A. Panagopoulos, P. Cottis,
and R. De Gaudenzi,“MIMO over Satellite: A Review,” IEEE Commun.
Surv. & Tut., vol. 13, no. 1, pp. 27-51, 2011.
[10] ETSI TR 102 376-2, “Digital Video Broadcasting (DVB); Implemen-
tation guidelines for the second generation system for Broadcasting,
Interactive Services, News Gathering and other broadband satellite
Fig. 9. Average SINR per user with the random scheduling algorithm. Setup: applications; Part 2: S2 Extensions (DVB-S2X),” Nov. 2015.
beam 42, ⇢ = 2.5 · 10 3 users/km2 . [11] P.-D. Arapoglou, A. Ginesi, S. Cioni, S. Erl, S. Andrenacci, and A.
Vanelli-Coralli, “DVB-S2x Enabled Precoding for High Throughput
Satellite Systems,” Int. J. of Sat. Commun. and Net., vol. 34, pp. 439-
satellite systems. This algorithm is based on the concept that, 455, Jun. 2015.
[12] D. Christopoulos, P.-D. Arapoglou, S. Chatzinotas, and B. Ottersten,
by scheduling together users belonging to similar locations “Linear precoding in multi-beam satcoms: Practical constraints,” in Proc.
in their respective beams (denoted as beam sectors), the pre- 31st AIAA Int. Commun. Sat. Syst. Conf. (ICSSC 2013), Oct. 2013.
coding performance are significantly improved. In fact, when [13] G. Zheng, S. Chatzinotas, and B. Ottersten, “Generic optimization of
linear precoding in multi-beam satellite systems,” IEEE Trans. Wir.
users in adjacent beams and closely located are scheduled Commun., vol. 11, no. 6, pp. 2308-2320, Jun. 2012.
in the same time frame, the precoding performance is poor [14] D. Christopoulos, S. Chatzinotas, G. Taricco, M. Angel Vazquez, A.
since the precoder cannot find a proper power allocation to Perez-Neira, P.-D. Arapoglou and A. Ginesi, “Multi-beam Joint Precod-
ing: Frame Based Design,” book chapter in Cooperative and Cognitive
transmit their information. With the proposed GSA, these Satellite Networks, Elsevier, 1st ed., Elsevier Academic Press, 2015.
situations are avoided and, in addition, also users that were [15] M. Poggioni, M. Berioli, and P. Banelli, “BER performance of multi-
experiencing already good average spectral efficiency levels beam satellite systems with Tomlinson-Harashima precoding,” in IEEE
Int. Conf. on Commun. (ICC 2009), Jun. 2009.
can see a significant performance benefit. In addition, it has [16] S. Chatzinotas, G. Zheng, and B. Ottersten, “Joint precoding with
been shows that the proposed GSA tends to uniform the flexible power constraints in multi-beam satellite systems,” in IEEE
SINR levels throughout the beam, which leads to an improved Glob. Commun. Conf. (GLOBECOM 2011), Dec. 2011.
[17] G. Taricco, “Linear Precoding Methods for Multi-Beam Broadband
fairness in the system performance. Satellite Systems,” in Proc. 2014 European Wireless Conf., p.. 542-547
May 2014.
ACKNOWLEDGMENT [18] B. Devillers, A. Pérez-Neira, and C. Mosquera, “Joint Linear Precoding
and Beamforming for the Forward Link of Multi-Beam Broadband Satel-
This work has been supported by European Space Agency lite Systems,” in Proc. 2011 IEEE Global Commun. Conf. (GLOBECOM
(ESA) funded activity SatNEx IV CoO2-PART 1 WI4 “For- 2011), pp. 1-6, Dec. 2011.
[19] M. Á. Vázquez, A. Pérez-Neira, D. Christopoulos, S. Chatzinotas, B.
ward Packet Scheduling Strategies for Emerging Satellite Ottersten, P.-D. Arapoglou, A. Ginesi, G. Taricco, “Precoding in multi-
Broadband Networks.” Opinions, interpretations, recommen- beam Satellite Communications: Present and Future Challenges,” IEEE
dations and conclusions presented in this paper are those of Wir. Commun. Mag., vol. 23, no. 6, pp. 88-95, Dec. 2016.
[20] D. Christopoulos, S. Chatzinotas, B. Ottersten, “Weighted Fair Multicast
the authors and are not necessarily endorsed by the European Multigroup Beamforming Under Per- antenna Power Constraints,” IEEE
Space Agency. Trans. Sig. Proc., vol. 62, no. 19, pp. 5132-5142, Oct. 2014.
[21] D. Christopoulos, S. Chatzinotas, and B. Ottersten “Multicast Multi-
R EFERENCES group Precoding and User Scheduling for Frame-Based Satellite Com-
munications,”IEEE Trans. Wir. Commun., vol. 14, no. 9, pp. 4695-4707,
[1] EU FP7 ICT Project BATS, “Broadband Access via integrated Terrestrial Sep. 2015.
& Satellite Systems,” [online] Available: http://www.batsproject.eu/. [22] V. Joroughi, M. Á. Vázquez, and A. Pérez-Neira, “Generalized Multicast
[2] V. Icolari, A. Guidotti, D. Tarchi, and A. Vanelli-Coralli, “An interfer- Multibeam Precoding for Satellite Communications,”, IEEE Trans. on
ence estimation technique for Satellite cognitive radio systems,” in Proc. Wir. Commun., vol. 16, no. 2, pp. 952-966, Feb. 2017.
IEEE Int. Conf. on Commun. (ICC 2015), Jun. 2015. [23] A. Guidotti and A. Vanelli-Coralli, “Clustering Strategies for Multicast
[3] S. Chatzinotas, B. Evans, A. Guidotti, V. Icolari, E. Lagunas, S. Precoding in Multi-Beam Satellite Systems,” submitted to IEEE Trans.
Maleki, S. K. Sharma, D. Tarchi, P. Thompson, and A. Vanelli-Coralli, Aer. and Electr. Syst., April 2018. Available at: https://arxiv.org/abs/
“Cognitive Approaches to Enhance Spectrum Availability for Satellite 1804.03891
Systems,” Int. J. of Sat. Commun. and Net., Nov. 2016. [24] M. Á. Vázquez and G. S. Granados, “Cross-Layer Packet Scheduler
[4] A. Vanelli-Coralli, A. Guidotti, D. Tarchi, S. Chatzinotas, D. Design of a Multibeam Broadband Satellite System with Adaptive
Kapetanovic, S. K. Sharma, N. Chuberre, B. Evans, M. Lopez-Benitez, Coding and Modulation,” Trans. on Wir. Commun., vol. 6, no. 1, pp.
W. Tang, J. Grotz, and K. Liolis, “Cognitive Radio Scenarios for Satellite 248-258, Jan. 2017.
Communications: The CoRaSat Project,” book chapter in Cooperative [25] ETSI TR 102 376-1, “Digital Video Broadcasting (DVB); Implemen-
and Cognitive Satellite Networks, Elsevier, 1st ed., Elsevier Academic tation guidelines for the second generation system for Broadcasting,
Press, 2015. Interactive Services, News Gathering and other broadband satellite
[5] G. Colavolpe, A. Modenini, A. Piemontese, and A. Ugolini, “Multiuser applications; Part 1: DVB-S2,” Nov. 2011.
Detection in Multibeam Satellite Systems: Theoretical Analysis and
Practical Schemes,” IEEE Trans. Commun., vol. 65, no. 2, pp. 945-955,
Feb. 2017.