Aduwati Sali
Supervisors: Prof. Barry Evans,
Dr. Guray Acar
Adaptive Approach in Reliable Multicast
Transmission over Geostationary
Satellite Networks
2
Presentation Outline
• Challenges and objectives
• Adaptive forward scheduling
• Channel State Information (CSI) collection and suppression
• Integrated channel-aware scheduling
• Multicast link adaptation
• Conclusion
3
Challenges in reliable multicast
transmission
• retransmission cycles sensitive to the actual loss experienced by the terminals
• inefficient resource utilisation in the forward link
• increase in the traffic load in the return link due to feedback implosion
• numbers of terminals increase, the problem exacerbates
System Architecture:
RAN
user link
MTMFTP
Client
MTMFTP
Client
MT
MFTP
Client
feeder link
4
Multicast File Transfer Protocol (MFTP) over BGAN
MFTP
Server
System Architecture:
RAN
Internet
Gateway GPRS
Support Node
GGSN
MCSMulticast Control Server
C-band
feeder link
L-band
user link
MTMFTP
Client
MTMFTP Client
MT
MFTP
Client
Adaptive forward scheduling: CAS, ADT (Chap. 4)
Multicast link adaptation (Chap. 7)
Change detection mechanism: CD (Chap. 5)
Integrated channel-aware scheduling:
CAS+CD (Chap.6)
5
Channel model
• 2-state model from an L-band Land Mobile Satellite (LMS) Inmarsat satellite network
• Fading duration follows lognormal distribution, LOS duration follows power law distribution
• In ns2, output is time-varying Eb/No, , or PER, p
• Propagation environments: suburban and wooded areas
2-state model Determine
current state Calculate
Eb/No,
Determine
BER,
Calculate
PER,p
BER
Sent back to the RAN
(Chapter 7)
Sent back to the RAN
(Chapters 4-6)
γ
γφ
Lp )1(1 φ−−=
γ
6
Adaptive Forward Scheduling algorithms
MFTP
Server
System Architecture:
RAN
Internet
Gateway GPRS
Support Node
GGSN
MCSMulticast Control Server
C-band
feeder link
L-band
user link
MTMFTP
Client
MTMFTP Client
MT
MFTP
Client
Adaptive forward scheduling: CAS, ADT (Chap. 4)
Aim: Increase forward resource utilisation
7
Channel Aware Scheduler (CAS) Algorithm
Probability of retransmission from reported CSI,
Threshold probabilityfrom reported CSI,
1. Concept: resume transmission when terminals are perceived in good channel conditions
2. Input: Channel State Information (CSI) from terminals: PER, p3. Scheduling parameters: prtx and prtxthresh
4. Scheduling decision: to place packet in the current slot for transmission or not
∏=
−−=N
j
jpprtx
1
)1(1
Njpprtxthresh )1(1 −−=
prtxthreshprtx ≤If true, then transmit, if not, defer transmission by one
slot
8
Adaptive Downlink Transmission (ADT) algorithm
1. Concept: adaptively set transport layer flow rate according to the estimated probability of retransmission
2. Input: CSI from terminals: PER, p3. Scheduling parameter: prtx 4. Scheduling decision: to set flow rate, η, from the MFTP
server
[bps]
Legend: L forward bearer size (in bits)
tRTT round trip time (in seconds)
BMax maximum transmission rate (bps)
( )
+
+
=
23218
312
3
2
,min
prtxprtxprtxprtx
t
LB
RTT
Maxη
9
CAS and ADT performance:
retransmitted DTUs vs. N
Worst-case scenario, math. approx.
File size: 10MBNo. of DTUs in the first cycle: 7364 DTUs
10
CAS and ADT performance: session
duration vs. N
File size: 10MBNo. of DTUs in the first cycle: 7364 DTUs
11
Conclusion – Adaptive forward
scheduling algorithms
• ADT: marginally fewer retransmitted DTUs
• CAS: significantly lower session durations
• The difference between mathematical
approximations and the simulation results are due
to:
(i) lack of scheduling impact
(ii) worst-case channel conditions representation (iii) lack of impact on how CSI is collected
12
CSI Collection and Suppression Policy
MFTP
Server
System Architecture:
RAN
Internet
Gateway GPRS
Support Node
GGSN
MCSMulticast Control Server
C-band
feeder link
L-band
user link
MTMFTP
Client
MTMFTP Client
MT
MFTP
Client
CSI Collection and Suppression Policy
using Change detection mechanism: CD (Chap. 5)
Aim: Reduce feedback implosion in the return link
due to CSI updates
13
Change Detection (CD) mechanism
for CSI collection policy
0
V seconds consists of
measurements of pj
Observation
window,
Reference
window,
t-V seconds
consists of R0
measurements of
pj
t
1. Concept: Each terminal access uplink slot to update its CSI if it decides that sufficient CSI change is observed
2. Input: CSI from terminals: PER, p, from 2 window frames: reference( ) and observation window ( )
3. Change detection decision: access uplink slot if
Legend: V observation window size (in seconds)average PER from observation windowaverage PER from reference window size
Gamma minimum CSI variation threshold to qualify for updates
0θ̂ 1θ̂
0θ̂
1θ̂
Gamma≥− 10 θθ
0θ̂
1θ̂
14
Comparison of CSI collection policy
vs. periodic CSI updates
PROP. TYPE
SESSION
DURATION
NUMBER OF
RETRANSMITTED
DTUs
CHANGE DETECTION
RATE
(ratio of CSI updates over number
of periodic CSI updates x 100%)
GAIN
(+/- %)
GAIN
(+/- %)
GAIN
(+/- %)
Suburban area
(PT1)
-0.11% -5.12% +32%
Wooded area
(PT3)
-0.94% -8.46% +20%
15
Conclusions – CSI collection policy
• Combination of CD and CAS: reduce feedback volume per
terminal; up to 32% reduction compared to the periodic CSI
collection policy.
• Stabilises return link at little penalty on forward link performance
16
Integrated Channel-Aware Scheduling
MFTP
Server
System Architecture:
RAN
Internet
Gateway GPRS
Support Node
GGSN
MCSMulticast Control Server
C-band
feeder link
L-band
user link
MTMFTP
Client
MTMFTP Client
MT
MFTP
Client
Integrated channel-aware scheduling:
CAS+CD (Chap.6)
Aim: model mathematical approx. when
both forward link and return link proposals
are simulataneously implemented
17
Integrated channel-aware scheduling1. Concept: Mathematical approach to characterise
performance metrics considering integrated CAS-CD
mechanism
2. Central idea: Suppression error, perror, due to non-
periodic CSI updates is represented in performance metrics estimation
3. Suppression error directly influences scheduling
performance
18
Performance evaluations –Integrated CAS
Number of terminals, N
- With suppression
- No suppression
File size: 10MBNo. of DTUs in the first cycle: 7364 DTUsSlot size, L: 3000 bits
19
Performance evaluations –Integrated CAS
- With suppression
- Without suppression
Number of terminals, N
File size: 10MBNo. of DTUs in the first cycle: 7364 DTUsSlot size, L: 3000 bits
20
Conclusions – Integrated channel-
aware scheduling
• Simulation and mathematical results approach agreement at large numbers of terminals
• The mathematical approximations capture error due to the suppression occurrence only.
• There are other errors which can be analysed and used in the mathematical approximations, such as
(i) collision errors which might reduce the number of CSI updates reaching the RAN, and
(ii) inaccuracy in reported CSI value due to propagation delay
21
Multicast Link Adaptation
MFTP
Server
System Architecture:
RAN
Internet
Gateway GPRS
Support Node
GGSN
MCSMulticast Control Server
C-band
feeder link
L-band
user link
MTMFTP
Client
MTMFTP Client
MT
MFTP
Client
Multicast link adaptation (Chap. 7)
Aim: to increase resource
utilisation, fairness and terminal
throughput
22
Other link adaptation technique for
multicast transmission
]})1(
1{[maxarg:1,
∏−
∗
−+=
i tj
l
RT
rlMPF
}{minarg: ljj
lMIN γγ ≥=∗
}{maxarg: ljj
lMAX γγ ≥=∗1. Based on the best terminal,
2. Based on the worst terminal,
3. Multicast Proportional Fair,
23
Proposed optimal algorithm: Multicast
Link Adaptation (MLA)
1. Output: Optimal transmission rate, l* by finding the solution to 0-1 multiple knapsack formulation:
Z = subject to and
2. Heuristic solution: transmission rate which maximises total terminal throughput
,maxarg ∑∑N
j l
ljl xr lj rr ≥
≥
=otherwise
rrifx
lj
lj,0
,1
}:{,maxarg lj
i
l jirl γγ ≥∈= ∑∗
24
Session duration distribution, N = 20
terminals
File size: 10MBNo. of DTUs in the first cycle: 7364 DTUs
25
Simulation results – retransmitted
DTUs vs N
File size: 10MBNo. of DTUs in the first cycle: 7364 DTUs
26
Simulation results – total number of
slots vs. N
File size: 10MBNo. of DTUs in the first cycle: 7364 DTUs
27
MLA vs. MIN algorithms
File size: 10MBNo. of DTUs in the first cycle: 7364 DTUs
28
Performance Comparison – Multicast
link adaptationLINK
ADAPTATION
ALGORITHM
SESSION DURATION FAIRNESS FORWARD
RESOURCE
UTILISATION
MLA Moderate. Good
especially in wooded
propagation
environment (robust
channel conditions) and
when large numbers of
terminals are considered
Excellent fairness as
the discrepancy of
session duration
performance between
the terminals with the
best and the worst
channel conditions are
small
Excellent (lowest
number of physical
bearer slots in the
forward link) in
wooded propagation
environment (robust
channel conditions) and
when large numbers of
terminals are considered
MIN Moderate. Good in
suburban propagation
environment and at low
numbers of terminals
Good Good (lowest physical
bearer slots in the
forward link) in
suburban propagation
environment and at low
numbers of terminals
MAX Excellent Very poor Very poor
MPF Poor Very poor Very poor
29
Main conclusions
• CAS algorithm increases resource utilisation in the
forward channel
• CSI collection policy - CD mechanism reduces load in
the return link
• Mathematical approximations - suppression error in agreement with simulation results
• MLA algorithm is more robust and fair performance
30
Future Work
• Better analytical study by considering
- scheduling impact
- collision in return link
- channel variation
• Adaptive slot allocation in multicast link adaptation for transmission of multiple multicast streams
• Combination with intelligent-based approach in
optimising resource allocation
31
PublicationsJournal:
1) G.Giambene, S. Giannetti, C. P. Niebla, M. Ries, A. Sali, ‘Traffic Management in HSDPA via GEO Satellite’, Special Issue on Satellite Network for Mobile Services, International Journal of Space Communications, Volume 21, Number 1-2 / 2007, pp51-68
Book Chapter:
1) G.Giambene, K. P. Niebla, V. Y. H. Kueh, A. Sali, et al. "Access Schemes and Scheduling Techniques," Adaptive Resource Management and Optimization in Satellite Networks: Optimization and Cross-Layer Design’, Chapter 5, pp. 119-175, Springer, Apr. 2007, ISBN 978-0-387-36897-9
Conferences: (8 in total)
• A. Sali, G. Acar, B. G. Evans, G. Giambene, ‘A Comparison of Multicast Adaptive Techniques in Reliable Delivery over GEO Satellite Networks’, IEEE 69th Vehicular Technology Conference VTC2009-Spring 26–29 April 2009, Barcelona, Spain
• G. Fairhurst, G. Giambene, G. Giannetti, C. Parraga, A. Sali, ‘Multimedia Traffic Scheduling in DVB-S2 Networks with Mobile Users’, International Workshop on Satellite and Space Communications 2008 IWSSC’08, Toulouse, 1st – 3rd October 2008, pp211-215
32
Publications• A. Sali, G. Acar, B. G. Evans, G. Giambene, ‘Channel-Aware Scheduling Algorithms with
Channel Prediction for Reliable Multicast Data Transmission over Geostationary Satellite
Networks’, International Workshop on Satellite and Space Communications 2008 IWSSC’08,
Toulouse, 1st – 3rd October 2008, pp85-89
• A. Sali, G. Acar, B. Evans, G. Giambene, ‘Change Detection Mechanism in Feedback
Implosion Suppression Algorithm for Reliable Transmission of Multicast Data over
Geostationary Satellite Network’, 26th AIAA International Communications Satellite Systems
Conference (ICSSC-2008), San Diego, 10th -12th June 2008
• A. Sali, G.Acar, B.Evans, G.Giambene, ‘Feedback Implosion Suppression Algorithm for
Reliable Multicast Data Transmission over Geostationary Satellite Networks’, International
Workshop on Satellite and Space Communications 2007 IWSSC’07, Salzburg, 12th-14th
September 2007, pp139-144
• A.Sali, G.Acar, B.Evans, ‘A Cross-Layer Approach for Packet Scheduling in reliable Multicast
Data Transmission over Geostationary Satellite Networks’, IEEE 65th Vehicular Technology
Conference VTC2007, Dublin, 23rd-25th April 2007, pp1395-1399
• A.Sali, G.Acar, B.Evans, ‘A Channel-Aware Scheduler with Feedback Implosion Suppression
in Reliable Multicast Data Transmission over Geostationary Satellite Networks’, 25th AIAA
International Communications Satellite Systems Conference (ICSSC-2007), Seoul, 10th -13th
April 2007
• A. Sali, A.Widiawan, S. Thilakawardana, R. Tafazolli, B. Evans, 'Cross-Layer Design
Approach for Multicast Scheduling over Satellite Networks’, Wireless Communication
Systems, 2005, 2nd International Symposium on, 5-7 Sept. 2005, pp701- 705
33
Thank you.
34
EXTRA SLIDES…
35
Background
1. 2 reliability aspects; full reliability or partial-
reliability
2. Full-reliable multicast transmission is
considered throughout this presentation where
the mobile terminals (MTs) are subject to
fluctuating channel conditions
3. Feedback-based transport protocol initiates
retransmission of lost packets until all users
receive the exact copy of the original data.
36
Performance metrics characterisation
1. Estimated retransmitted DTUs, A
2. Session duration in seconds, sd
[no. of DTUs]RTTthresh
Max
DTU tKB
LAM⋅+≤+⋅
⋅+)]Pr(1[
)(γγCASsd
=
][
)
))ˆ321(ˆ8
ˆ312
3
ˆ2(
,min(
)(
][)(
2
s
rtxprtxprtxprtxp
t
L
T
L
LAM
stKLAM
sd
RTT
Slot
DTU
RTT
DTU
ADT
++
×+=
⋅+×+
=η
( ) ( )( )
−−−⋅⋅= ∑
=
+K
k
Nk
DTUj
Nk
DTUj ppkMA0
111
37
Change Detection (CD) mechanism for CSI collection policy
0
V seconds consists
of measurements
of pj
Observatio
n window,
Reference
window,
t-V seconds
consists of R0
measurements
of pj
t
1. Concept: Each terminal access uplink slot to update its CSI if it decides that sufficient CSI change is observed
2. Input: CSI from terminals: PER, p, from 2 window frames: reference( ) and observation window ( )
3. Change detection decision: access uplink slot if
Legend: V observation window size (in seconds)average PER from observation windowaverage PER from reference window size
Gamma minimum CSI variation threshold to qualify for updates
0θ̂ 1θ̂
0θ̂
1θ̂
Gamma≥− 10 θθ
0θ̂
1θ̂
38
0.001 0.002 0.0030
200
400
600
800
Gamma
Num
ber
of M
FT
Ppa
cket
s re
tran
smitt
ed
0.001 0.002 0.0030.9
0.92
0.94
0.96
0.98
1
Gamma
Dow
nlin
k re
sour
ce u
tiliz
atio
n
0.001 0.002 0.003600
620
640
660
680
Gamma
Ses
sion
dur
atio
n (s
)
0.001 0.002 0.0030
0.2
0.4
0.6
0.8
1
Gamma
Cha
nge
dete
ctio
n ra
te
Impact of CSI variation threshold, Gamma and
observation window size, V (suburban area)
Obs. Window size, V = 0.08s Obs. Window size, V = 0.16s Obs. Window size, V = 0.4s
File size: 10MBNo. of DTUs in the first cycle: 7364 DTUs
39
Feedback implosion suppression (FIS)1. Concept: The RAN decides which terminal should
access the uplink slot for CSI updates based on CSI change, last update duration, and NACK availability.
2. Input: CSI from terminals: and , transport layer ACK/NACK, database on update timer
3. FIS decision: Based on Multiple Subset Sum Problem (MSSP) formulation:
max subject to where
4. Solution: terminals with S highest rj will be selected to access the uplink slots.
5. Only a subset of terminals will be selected:
0θ̂ 1θ̂
∑∑= =
S
i
N
j
jij xr1 1
iRxr Max
N
j
jij ∀≤∑=
,
1
jUpdate
j
j
j tr ,∆×=φ
ω
}1:{ == njxjN
(
40
FIS performance evaluations
41
Performance evaluations
• Change detection (CD) mechanism: fixed-size observation
window and a growing reference window performed the best
in terms of session duration with tolerable suppression gain.
• Combination of CD and CAS: reduce feedback volume per
terminal; up to 32% reduction compared to the periodic CSI collection policy.
• Feedback implosion suppression (FIS) mechanism: able to support not only on the proper arrival of CSI updates, but
also in the absence of suppression problem.
• Combination of FIS and CAS: use of dedicated uplink slots
for selected terminals reduce CSI inaccuracy due to collision
problem
42
Multicast link adaptation
(dB)
Equivalent
(dB)
Forward
bearer
sub-type, l
Payload
size, L
(bits)
Payload
duration
(s)
Average
transmission rate
(kbps)
3.92 -2.10 L3 2000 0.01 200.0
4.85 -1.17 L2 2320 0.01 232.0
5.81 -0.21 L1 2664 0.01 266.4
6.71 0.69 R 3000 0.01 300.0
7.84 1.82 H1 3440 0.01 344.0
8.90 2.88 H2 3840 0.01 384.0
9.88 3.86 H3 4224 0.01 422.4
10.97 4.95 H4 4640 0.01 464.0
11.92 5.90 H5 4920 0.01 492.0
12.99 6.37 H6 5120 0.01 512.0
os NEob NE
*Wideband 16-QAM bearer operating at a symbol rate of 151.2 ksym/s
Link Adaptation Table: Forward link vs. bearer type
43
Performance comparison
]})1(
1{[maxarg:1,
∏−
∗
−+=
i tj
l
RT
rlMPF
}{minarg: ljj
lMIN γγ ≥=∗
}{maxarg: ljj
lMAX γγ ≥=∗
Comparison with other multicast link adaptation techniques:
1. Proposed MLA:
2. Based on the best terminal,
3. Based on the worst terminal,
4. Multicast Proportional Fair,
}:{,maxarg lj
i
l jirl γγ ≥∈= ∑∗
44
Transmission rate vs. no. of terminals
45
Session duration vs. no. of terminals
46
Performance Evaluations