Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
RESCUE D3.2 Version 2.0
Page 1 (79)
ICT- 619555 RESCUE
D3.2 Version 2.0
Report on revised WP3 architecture including simulation results
Contractual Date of Delivery to the CEC: 15 July 2015
Actual Date of Delivery to the CEC: 17 July 2015
Delivery of Revised Version to CEC: 18 December 2015
Editor: Olayinka Adigun
Author(s): Marek Natkaniec, Marek Sikora, Katarzyna Kosek-Szott, Szymon Szott,
Łukasz Prasnal, Andrzej R. Pach, Olayinka Adigun, Nikos Karanastasis,
Nikos Pavlatos, Sebastian Kühlmorgen, Andreas Festag, Hicham Khalife,
Jawad Seddar, Matias Brulatout, Ade Irawan, Valtteri Tervo
Participant(s): AGH, UbiTech, TUD, TCS, JAIST, UOULU
Work package: WP3 – Message Transfer
Security: PU
Nature: R
Version: 2.0
Total number of pages: 79
Abstract: This deliverable reports the updates to the RESCUE message transfer architecture and presents
the results of the simulation-based performance evaluation of developed protocols in lossy link
environments. The contributions contained in this deliverable are updates to D3.1 and new simulation
results for three complementary medium access protocols for centralized and distributed networks, two
routing protocols (R-OLSR and geo-routing protocols), two complementary error control methods and a
multirate algorithm.
Keyword list: unpredictable environments, lossy links, medium access, routing, error control, cross-layer
design, rate adaptation
Disclaimer:
RESCUE D3.2 Version 2.0
Page 2 (79)
Executive Summary
This deliverable reports the updates to the RESCUE message transfer architecture (including MAC and
routing protocols) and presents the design updates with respect to D3.1. It also presents the results of the
simulation-based performance evaluation of the protocols in lossy link environments. The contributions
contained in this deliverable are updates and new simulation results for the following:
Three complementary medium access protocols (one for centralized and two for distributed
networks).
Two complementary routing protocols (based on multipath, link quality and geographical routing).
Two complementary error control methods: for ensuring end-to-end reliability and efficient
retransmissions.
Partial hybrid ARQ with BPSK modulation.
A multirate algorithm for providing improved performance by adapting to channel conditions.
The design details presented in this deliverable will serve as a basis for the implementation and performance
analysis of each component to be reported in D3.3 as well as the implementation and validation of the
integrated components to be performed in WP4.
RESCUE D3.2 Version 2.0
Page 3 (79)
Authors
Partner Name Phone / Fax / e-mail
AGH University of Science Marek Natkaniec [email protected]
and Technology Marek Sikora [email protected]
Katarzyna Kosek-Szott [email protected]
Szymon Szott [email protected]
Łukasz Prasnal [email protected]
Andrzej R. Pach [email protected]
UbiTech Olayinka Adigun [email protected]
Nikos Karanastasis [email protected]
Nikos Pavlatos [email protected]
Technical University Dresden Sebastian Kühlmorgen [email protected]
(TUD) Andreas Festag [email protected]
Thales Communications & Hicham Khalife [email protected]
Security SAS (TCS) Jawad Seddar [email protected]
Mathias Brulatout [email protected]
Japan Advanced Institute Ade Irawan [email protected]
of Science and Technology
(JAIST)
University of Oulu Valtteri Tervo [email protected]
(UOULU)
RESCUE D3.2 Version 2.0
Page 4 (79)
Table of Contents
Executive Summary ......................................................................................... 2
Authors.............................................................................................................. 3
List of Acronyms and Abbreviations .............................................................. 6
1. Introduction ................................................................................................. 7
2. Background ................................................................................................. 8
2.1 Design basics ............................................................................................................................... 8 2.2 Frame format ............................................................................................................................... 9 2.3 Toy scenarios ............................................................................................................................. 10 2.4 Large scale scenarios ................................................................................................................. 11
2.4.1 V2V ................................................................................................................................... 11 2.4.2 Public safety ...................................................................................................................... 12
2.5 PHY Black Box ......................................................................................................................... 13 2.5.1 Technical details ............................................................................................................... 14
2.6 Baselines for comparison........................................................................................................... 15 2.7 Simulation parameters ............................................................................................................... 16
3. MAC ............................................................................................................ 18
3.1 SimpleMAC ............................................................................................................................... 18 3.1.1 Protocol overview & design update .................................................................................. 18 3.1.2 Simulation results .............................................................................................................. 19
3.2 Distributed MAC ....................................................................................................................... 24 3.2.1 Protocol overview & design update .................................................................................. 24 3.2.2 Simulation results .............................................................................................................. 26
3.3 Centralized MAC....................................................................................................................... 28 3.3.1 Protocol overview & design update .................................................................................. 28 3.3.2 Simulation results .............................................................................................................. 29
4. Routing ...................................................................................................... 35
4.1 RESCUE Multipath Routing Protocol ....................................................................................... 35 4.1.1 RESCUE Routing Relay approaches ................................................................................ 35 4.1.2 Protocol overview & design update .................................................................................. 35 4.1.3 Simulation results and Performance Analysis ................................................................... 39 4.1.4 Convergence Time ............................................................................................................ 43 4.1.5 Conclusion ........................................................................................................................ 44
4.2 Geo-routing ............................................................................................................................... 44 4.2.1 Services ............................................................................................................................. 45 4.2.2 Protocol overview & design update .................................................................................. 45 4.2.3 Simulation results .............................................................................................................. 47
5. Error control .............................................................................................. 51
5.1 End-to-end ARQ ........................................................................................................................ 51 5.1.1 Protocol overview & design update .................................................................................. 51 5.1.2 Simulation results .............................................................................................................. 52
5.2 Partial ARQ ............................................................................................................................... 58 5.2.1 Protocol overview and design update ................................................................................ 59 5.2.2 Simulation Setup ............................................................................................................... 60 5.2.3 Simulation Results ............................................................................................................ 63
6. Multirate ..................................................................................................... 65
7. Conclusion ................................................................................................ 74
Appendix A. Forwarding Threshold Optimization ....................................... 75
RESCUE D3.2 Version 2.0
Page 5 (79)
Toy Scenarios ...................................................................................................................................... 75 V2V Scenario ...................................................................................................................................... 77 Summary ............................................................................................................................................. 78
References ...................................................................................................... 79
RESCUE D3.2 Version 2.0
Page 6 (79)
List of Acronyms and Abbreviations
Term Description
ACK End-to-end acknowledgement
AODV Ad-hoc on-demand distance vector routing
AP Access point
ARQ Automatic repeat request
ARR Advanced relay routing
AWGN Additive white Gaussian noise
B Beacon
BR Beacon repetition
BS Base station
CBF Contention-based forwarding
CC Control channel
CFP Contention free period
CI Confidence indicator
CP Contention period
CW Contention window
CRC Cyclic redundancy check
CSMA Carrier sense multiple access
CSMA/CA Carrier sense multiple access with collision avoidance
D Destination
DC Data channel
ETX Expected Transmission Count
Geo Geographic
GF Greedy forwarding
ID Identifier
IL Interleaver index
LIFS Long interframe space
LLR Log likelihood ratio
LOS Line of sight
LOTF Links on the fly
MAC Medium Access Control
MCS Modulation and coding scheme
MIC Message integrity check
MPR Multi Point Relay
NET Network
OLSR Optimized link state routing
PDR Packet delivery Ratio
PHY Physical
PT Processing time
QoS Quality of service
R Relay
R-OLSR RESCUE-OLSR
RR Resource reservation
RREQ Route request
RTS Routing Toy Scenario
S Sender
SDR Software Defined Radio
SFD Start Frame Delimiter
SGB Simple geo-broadcast
SIFS Short interframe space
SMS Short Message Service
SRR Simple Relay Routing
SSR Source Suggested Routing
STA Station
TC Topology control
TDMA Time division multiple access
TX Transmission
RESCUE D3.2 Version 2.0
Page 7 (79)
1. Introduction
The RESCUE project focuses on research and innovation regarding the “links on the fly” (LOTF) concept
in wireless networks. The core idea is the utilisation of diversity in communication channels and paths
between communicating devices. This takes advantage of the inherent broadcast nature of the radio
channels and the multi-route diversity of multi-hop ad-hoc networks that have the potential to increase
communication reliability and decrease outage probabilities. The RESCUE project's advancement of the
physical layer technologies and mechanisms requires modification of the message transfer methods, i.e.,
protocols and functionalities provided by the data link and network layers. The goal of WP3 is to provide
a message transfer architecture which allows the use of lossy links and the two characteristics which follow
from this idea: (i) frames with errors are not discarded but should be processed (forwarded) and (ii) due to
these errors, the information stored in typical headers (e.g., the IP destination address) cannot be trusted
(the errors may have occurred there).
In the first WP3 deliverable (D3.1), we presented the initial design of a RESCUE data link and network
architecture which can leverage the gains from using lossy links in communication taking into consideration
the two imminent characteristics above. In this deliverable (D3.2), we present a revised data link and
network layer design and provide simulation results for the proposed protocols and mechanisms. During
the design process, it became clearly evident that the LOTF concept impacts all OSI layers and an integrated
cross-layer design approach is required. However, to maintain compatibility with existing networks we
limit the required revolutionary changes to the two lowest layers (physical and data link). At the network
layer, we have opted to provide a modified routing protocol but otherwise maintain full IP compatibility.
The transport and higher layers remain intact as they are out of the scope of the project.
With these constraints in mind, this deliverable reports the updated protocol designs and architectures with
simulation based performance evaluation. The evaluations presented for the MAC protocols, routing
protocols, error control mechanisms and rate adaptation in this deliverable are all performed using a
harmonised key parameter table with their justifications in section 2.7 and simulations are all performed
using NS3. RESCUE’s modified NS3 platform used in the simulations integrates an abstraction of the
physical layer (an input from WP2), the MAC protocols and Automatic Repeat reQuest (ARQ) or routing
protocols in order to properly access the concept of RECUE’s lossy forwarding. Section 2 provides a
background and overview of fundamentals required to understand the basics of message transfer in a lossy-
link network. Section 3 presents details of the Simple MAC, Distributed MAC and Centralised MAC with
their respective simulation-based performance evaluation. Section 4 provides details of RESCUE's routing
protocols: RESCUE-OLSR (R-OLSR), based on a multipath link-quality approach using the Expected
Transmission Count (ETX) as a link quality metric, and Georouting, based on a geographical routing
approach, with their performance evaluations. Section 5 provides details of the error control mechanisms
with the end-to-end ARQ (Automatic Repeat reQuest) and the Partial ARQ with their evaluations using
simulations. In Section 6, we describe another mechanism for improving performance based on appropriate
data rate selection. Finally, Section 7 provides a conclusion and summary for deliverable D3.2.
RESCUE D3.2 Version 2.0
Page 8 (79)
2. Background
In this section we concisely report on the basics of message transfer in RESCUE, the established frame
format, the consolidated toy scenarios, the large-scale scenarios, the physical layer abstraction, and the
baselines for comparison. The information provided in this section serves as a basis for understanding the
protocol designs provided in the subsequent sections.
2.1 Design basics
As explained in D3.1, the RESCUE message transfer design considers packet-based channel access in the
form of carrier sense multiple access with collision avoidance (CSMA/CA). Since even the most basic
RESCUE scenario (TS1 in Figure 2.4) involves three communicating stations where messages are received
simultaneously, we assume that from a MAC-layer perspective all frames are broadcast. Upon message
reception, it is each station’s task to decide if they are relays, i.e., if they should decode and forward the
message. This decision is based on two factors: the quality of the received frame and the placement of the
potential relay on a path towards the destination as determined by the routing protocol. Furthermore, in
order for a message to be properly processed, it has to include error-free meta-data. We store the minimum
required amount of information in the PHY header, which is protected against errors (Section 2.2).
Figure 2.1 presents the cross-layer data flow for a single message transmission in the simplest RESCUE
toy scenario (TS1 in Figure 2.4). At the source, this message is passed down through all the OSI layers with
each layer adding its own header. In RESCUE, we define new data link and physical layers maintaining
full compatibility with existing higher layer protocols (including the Internet protocol, IP).
Figure 2.1: Cross-layer data flow in the first toy scenario. Colour coding: green – message without
errors, red – message as received by destination (with errors), blue – message as received and
forwarded by relay (with errors), grey – control messages.
We assume that the message is then received with errors at the relay and destination. Upon message
reception, two message checks are performed. First, a PHY header integrity check to determine that the
control information (meta-data) is without errors. Then, the quality of the decoded message is checked. If
it does not exceed a certain threshold, the message is dropped. Otherwise, it is provided to the higher layers.
Then, at the relay, the routing protocol performs a forwarding decision based on the addresses contained in
the PHY header. If this decision is positive, the message is passed to the lower layers and transmitted.
In parallel, the destination processes the first received message. After a successful PHY header integrity
check and distortion check, it concludes that it is the final destination. Therefore, an end-to-end message
integrity check (MIC) is performed1. We assume that in the analysed case the test fails and the message is
stored in a repository while the destination awaits further copies to arrive. When such a copy arrives from
the relay, it is processed as the first message. Since the end-to-end MIC fails for this copy as well, it is again
stored in the repository. A new message in the repository triggers an iterative decoding procedure. Based
on their PHY headers, both messages are identified as copies of the same message so joint decoding is
performed on both of them. The output message is again verified in the end-to-end message integrity check.
We assume that now this check is positive and the message itself can be provided to the higher layers. The
positive integrity check also triggers an end-to-end acknowledgement (ACK) to inform the source that this
message has been received correctly. The ACK frame is a short message which can be identified and
1 Note that intermediate relays do not need to perform this check.
Physical
Data Link
Network
Application
Laye
rs
Source Relay
Forwarding decision
Destination
Repository & Iterative decoding
AC
KMessage checks Message checks
End-to-end MIC
fail
pass
RESCUE D3.2 Version 2.0
Page 9 (79)
associated with the DATA frame that it acknowledges. To avoid the problem of flooding, the ACK frame
is sent to the source along a single best path as determined by routing.
2.2 Frame format
As specified in D3.1, each RESCUE message requires a mandatory set of meta-data attached to each
message and guaranteed to be error-free. After a RESCUE message is concatenated with a header and/or
footer at each level of the OSI model, the resultant MAC frame becomes the PHY payload. This is then
preceded by the PHY header which stores the abovementioned meta-data.
Typ
e
Dat
aR
ate
3 2
Sen
d W
indo
w
4
Blo
ck A
CK
Con
t. A
CK
1 1Bits:
MA
C
Pro
toco
l
1
Synchronization
SFD CTRL
Du
ratio
n
Source AddressSender Address
Destination Address
SEQ.
Inte
rle
aver Header
checksumMAC Frame
Preamble PHY Header
2 2 6 6 6 2 1 4Bytes:
Type: 000 (DATA); Other types: 001 (e2e ACK); 010 (partial ACK); 011 (Beacon); 100 (Resource Reservation)DataRate: MSC used;Send Window: how many unACKed frames the station can send;Block ACK: 1 – requested/supported, 0 – not allowed;Continous ACK: 1 – requested/supported, 0 – not allowed;QoS: 1 – supported, 0 – not supportedMAC Protocol: 0 – distributed, 1 – centralizedRetransmission: 0 – false, 1 – trueReserved: for future use
FCS
42324
1
Qo
S
MBF
2
Ret
ran
sm
issi
on
1
Res
erv
ed
2
Figure 2.2: The DATA frame format.
Figure 2.2 provides a description of each of the fields in the RESCUE DATA frame. The frame begins with
a preamble to facilitate asynchronous transmission. Within the preamble, the synchronization field
(containing alternating 0s and 1s) is followed by the Start Frame Delimiter (SFD) (containing a predefined
bit sequence, e.g., 0000 1100 1011 1101). The PHY header begins with a 2-byte control (CTRL) field
within which are encoded: the frame type (3 bits), the data rate used when transmitting the PHY payload
(2 bits), the Send Window, Block ACK, and Continuous ACK fields related to the end-to-end ARQ
mechanism (Section 5.1), one bit indicating support for QoS, one bit indicating the RESCUE MAC type
(distributed or centralized), one bit indicating whether the frame is being retransmitted, and 2 bits reserved
for future use.
The CTRL field is followed by the length (duration) of the PHY payload (2 bytes). Then, there are three
addresses (source, sender, and destination) – each 6 bytes long and corresponding to the MAC addresses of
the stations. This length was chosen to ensure simplify the addressing, although in practice shorter addresses
may also be used. The source and destination are the end-to-end stations of the message; the sender is the
station which has transmitted this particular frame (either the source or one of the relays). The 2-byte
sequence (SEQ) field is used to distinguish consecutive frames. The 2-byte multi-bit field (MBF) was added
as a requirement from WP1 and WP2. It describes the “frame quality” in the form of a confidence indicator
which we define as the bit error probability after decoding. It can be obtained by measuring the mutual
information (MI) of the output LLRs from the decoder. Next is a byte-long field containing the interleaver
index necessary for the operation of the decoder. The final field is the 4-byte header checksum which
ensures that the PHY header has been decoded correctly.
The PHY payload contains the MAC frame. Since we cannot guarantee that the PHY payload is error-free,
we do not specify any relevant information within the MAC frame. The last part of the MAC frame contains
a frame check-sequence (FCS) field which allows the destination to confirm that the outcome of the joint
decoding is successful and that the frame can be passed to higher layers.
The format of the end-to-end (e2e) ACK frame (Figure 2.3) is based on the previously described DATA
frame format. The frame is made as short as possible: it consists only of the PHY header from which
information describing the non-existing payload has been removed. Other unnecessary fields (such as the
sender address) have also been removed. The ACK frame is mapped to the DATA frame which it
acknowledges based on the 3-tuple of (source address, destination address, sequence number). The fields
related to the end-to-end ARQ mechanism are described in Section 5.1.
RESCUE D3.2 Version 2.0
Page 10 (79)
Synchronization
SFD CTRL Source AddressDestination
AddressSEQ.
Header checksum
Preamble PHY Header
2 6 6 2 4Bytes:
Block ACK
2
Con
t. A
CK
1
Typ
e
3Se
nd
Win
dow
4
Blo
ck A
CK
Con
t. A
CK
1 1Bits:
NACKed Frames
5N
AC
K/A
CK
1
Res
erv
ed
1
Type: 001 (e2e ACK);NACK/ACK: 1 – ACK, 0 - NACK;Send Window: number of received frames since last transmitted (N)ACK;Block ACK: 1 – used, 0 – not used;Continous ACK: 1 – used, 0 – not used;NACKed Frames: How many frames are missed (to notify when Block ACK cannot inform about all missed frames – for proper non-continous send window operation);Reserved: for future use
Figure 2.3: The ACK frame format.
2.3 Toy scenarios
S D
R
S D
R2
R1
S D
R2
R1
TS1 TS2
TS3
S1
D
S2
R
TS4
Figure 2.4: The four basic network topologies (toy scenarios, TS1-TS4) considered in the RESCUE
project.
To harmonise the analysis of basic network topologies among WP1, WP2, WP3, and WP4, we have
consolidated the four toy scenarios (TSs) which are the subject of our initial investigation (Figure 2.4). This
constitutes an update from what was presented in D3.1. The first scenario (TS1) is the simplest case of
relay-assisted message transfer and consists of only three stations: the source (S), relay (R), and destination
(D). In the second toy scenario (TS2) we add an additional relay station. Both relays (R1 and R2)
simultaneously receive the message transmission from S. The third toy scenario (TS3) is a combination of
TS1 and TS2. This scenario requires that routing addresses the issue of multiple paths to the destination.
Finally, TS4 contains two source stations (S1 and S2). Their message is combined at R before it retransmits
to D. Due to the high complexity of the relay in TS4, this scenario is not considered in WP3 and remains
defined for the theoretical work of WP1 and WP2.
RESCUE D3.2 Version 2.0
Page 11 (79)
S D
R2
R1
R3
Figure 2.5: The basic network topology considered for demonstrating routing in the RESCUE
project.
Additionally, for the purpose of demonstrating the RESCUE routing protocols, we have defined a routing
toy scenario (RTS) as shown in Figure 2.5. This topology is an extension of TS3 with the addition of an
extra relay (R3). Therefore, there are three paths between S and D, each with an increasing number of hops
(from one to three). In this scenario we expect the routing protocol to limit the transmission to the two paths
with the lowest number of hops. The routing toy scenario will be used to demonstrate the RESCUE routing
protocols in the next phase of the project when we move with the implementation to SDR devices.
2.4 Large scale scenarios
For more advanced analysis we define two large scale scenarios (V2V and public safety) relevant to the
application of the LOTF concept in practice.
2.4.1 V2V
The “links-on-the-fly” concept has been theoretically well investigated in toy scenarios in D1.2.1, D2.1.1
and D2.2.1 [4], [5], [11]. In order to validate the functionality and assess the performance, the system needs
to be considered in more realistic scenarios including multi-hop dissemination. In such multi-hop scenarios
with “links-on-the-fly” it is necessary to have lossy links, i.e., the received packets have errors after channel
decoding. With the absence of errors at the destination the LOTF scheme is not necessary anymore.
Therefore a scenario is selected where vehicles move in small groups to ensure lossy links among them and
the destination has a large distance to the last-hop relays in order to have a low reception SNR (see Figure
2.6, dashed line indicates a lossy link).
We consider a typical highway scenario with a sparse density of nodes. With a high number of nodes many
links would be available for redundant packet re-transmissions and these decrease the number of lossy
links. To show the best gain of LOTF, a representative scenario with two small groups of four vehicles are
chosen, which have at least three disjoint paths in parallel. Four vehicles are necessary following the
assumed statistical channel model, i.e., Nakagami, where the reception probability beyond a configurable
distance is less than 100%. For simplicity the vehicles drive in one direction. The source sends 1,000
packets with a size of 8,000 bits each to a destination, which is out of communication range of the source
such that the relays need to forward the packet towards the destination. The transmissions among the
vehicles can be lossy due to the distance between them. When all three parallel links are lossy it is not
possible to recover the packet without joint decoding since the packet are received corrupted in the relays.
For the simulations the RESCUE protocol stack with the Black Box abstraction for the PHY (Section 2.5),
the functionality of the MAC and NET layers was implemented. Furthermore, a facility layer is used to
send event-triggered safety messages, i.e., DENMs [2]. The parameters for the simulation are shown in
Table 2.1.
Table 2.1: Parameters for the large scale scenario for V2V communication
Parameter Values
Number of Vehicles 10
PHY/MAC/NET RESCUE PHY abstraction, RESCUE MAC, RESCUE CBF
Data Rate 6 Mbit/s
Tx Power 5 – 35 dBm
Channel Bandwidth 10 MHz at 5.9 GHz
DENM interval / size 0.5 s / 1000 bytes
Total DENMs sent 1000
RESCUE D3.2 Version 2.0
Page 12 (79)
Figure 2.6: Large scale scenario for vehicle-to-vehicle communication
2.4.2 Public safety
In order to apply the links-on-the-fly concept in a public safety deployment the following conditions should
be gathered. First, the network should be dense in terms of the number of participating nodes. This is
particularly the case in a disaster relief operation where often a large number of rescuers are deployed.
These rescuers are equipped with relaying-capable devices able to establish a dynamic multihop network.
Second, in such extreme situations the network can be partially destroyed forcing nodes to establish longer
range communications thus over less reliable and potentially lossy links. In these relatively large and dense
networks, more than a single route can exist from the source to the destination. More importantly, in such
an uncoordinated environment a decentralized MAC requiring a minimal synchronization is required.
Furthermore, the intrinsic properties of public safety traffic force these networks to support point to
multipoint communication patterns and also favour downlink traffic over uplink.
To assess the gains expected by the LOTF, we investigate a representative public safety scenario composed
of around 20 nodes operating in an outdoor relatively small area, i.e., dense outdoor scenario. For simplicity
reasons, already deployed base stations (BS) and users devices are considered to operate in the same
manner, undergoing the same relaying functions. Such public safety deployment is also characterized by
specific traffic requirements such as broadcasting regularly positions and locations as well as short message
exchanges. These traffic patterns require relatively low throughput (64 kbit/s), but demands are expected
to grow with the emergence of high resolution video applications and interactive voice communications.
We rely in these public safety simulations, as in all our ns-3 [12] simulations, on the physical layer
abstraction model called the PHY Black Box (Section 2.5). The considered simulation parameters are
shown in Table 2.2.
Table 2.2: Parameters for the large scale scenario for the public safety scenario
Parameter Values
Number of nodes (BS and devices) > 6
PHY/MAC/NET RESCUE PHY abstraction, RESCUE MAC, RESCUE
R-OLSR
Data Rate 64 kbits/s to 6 Mbits/s (future apps)
Tx Power Up to 40 dBm
Node distribution Uniformly inside cells
Traffic pattern Downlink (SMS or video stream)
Message size Short to medium
Simulation duration Few minutes
1 1 1
1
1
1
1
1
Distance to achieve lossy links
S D
RESCUE D3.2 Version 2.0
Page 13 (79)
Figure 2.7: Large scale scenario for Public safety
Note that the topology presented in Figure 2.7 is the simplest topology possible, overlapping paths can be
available offering multiple variants of the same considered topology.
2.5 PHY Black Box
In order to test the performance of the message transfer protocols designed by WP3, it became apparent
that an abstraction of the physical layer was required. With the help of other WPs, a model of the joint
decoding process, named the PHY Black Box, was created (Figure 2.). This abstraction was first
implemented in Matlab and then ported to C++ to be used in ns-3 simulations as reported in this deliverable.
A separate C++ port for GNU Radio is being implemented to facilitate SDR tests of WP3 protocols to be
reported in D3.3.
Theoreticalresults
Coding/decodingscheme design
Message transferrequirements
Abstraction
Implementations
PHY Black Box
WP1 WP2 WP3
Input
Matlab
C++(ns-3)
WP3 Simulations
(D3.2)
WP3 SDR Implementation
(D3.3)
Application
C++(GNU Radio)
Figure 2.8: The cross-WP cooperation in providing a physical layer abstraction and outcome
implementations.
Zone with
low SNR
Destination
source
relay
Lossy link
BS can be
considered as
standard relays
Path1
RESCUE D3.2 Version 2.0
Page 14 (79)
2.5.1 Technical details
The RESCUE PHY Black Box is an error rate model based on a semi-analytical approach where theoretical
results are also used to avoid an excessive amount of simulations. Figure 2.9 divides the abstraction model
into black box #1 and black box #2, which illustrate the deterministic and stochastic parts,
respectively. The idea is to match the theoretical results given by black box #1 to simulation based
results stored in black box #2. In the following we will describe the content of both parts of the
abstraction model.
Figure 2.9: The two components of the PHY Black Box abstraction model.
Black box #1 was originally described in D2.1.1. The bit error rate (BER) after joint decoding at the
destination depends on the bit error probability (BEP) at the relay i, ip , the signal to interference plus noise
ratio (SINR) of the link between relay i and the destination, iSINR , and the used modulation and coding
scheme (MCS). The sum of MIs over the parallel incoming links is calculated as i
iMM for
0,1,1min
0,
iibiib
iii
ppHRpH
pRM
where iR is the normalized channel constellation constrained capacity of the parallel link i and ib pH
denotes the binary entropy associated with probability ip . In case of TS2 (the CEO problem), an error
floor occurs. Black box #1 calculates the error floor by using the method described in Section 4.1.4.1
of D1.2.1.
It is well known that in block codes, the number of errors in the packets is not uniformly distributed. Black
box #2 is needed to produce more realistic error distributions. In the following, we will describe the
construction of simulation tables used in Black box #2 for N parallel links.
Since the BEP at the relay is already taken into account in Black box #1, we can set 0ip for all i.
Furthermore, since the simulation results and the theoretical results are compared only in terms of the sum
of the MIs over the links, we can set ̂iSINR , and sweep ̂ through an appropriate SINR range. The
simulations have to be performed separately for all MCSs considered.
Simulations were performed by sending 10000 packets (8000 bits each) through the channel of each parallel
link. The output of the simulations was the number of errors in each packet. Empirically, it was found that
the errors are distributed according to a distribution defined as 2,10 , where is the
probability that the number of errors in a packet is zero, 0 is the Dirac delta function, and 2,
denotes the normal distribution with mean and variance 2 .
The parameters 0ip and ̂iSINR are given to Black box #1 which outputs MI. Now we can fit
the simulation data to the distribution given above, resulting in numerical values of parameters , , and
2 corresponding to a value of MI given by Black box #1. As a result, we can construct a table which
contains the numerical values for MI and the distribution parameters.
Since the resolution for the MI values listed in the table is finite, linear interpolation can be used to obtain
numerical values for the distribution parameters between the quantized points. For TS1 and TS3, errors can
RESCUE D3.2 Version 2.0
Page 15 (79)
be generated directly from the distribution defined above. In the case of TS2, there exists an error floor
which is composed of evenly distributed errors. Let L be the length of the packet and let )( floorp denote
the probability of errors induced by error floor. The number of bit errors N can be generated as follows:
1. Input parameters Kppp ,...,, 21
, KSINRSINRSINR ,...,, 21
2. Calculate MI using black box #1
3. Find 2,, matching the values on the tables with MI using linear interpolation
4. Generate random number 1,0
5. if then
1. Generate N from 2,
6. else 1. N=0
7. end if 8. Calculate p(floor) using black box #1
9. if 0)( floorp then
1. )( floorpLNN
10. end if
11. end
2.6 Baselines for comparison
The MAC and routing message transfer baselines were defined in D1.2.1. Here, we briefly summarize the
choices, discuss necessary changes where applicable, and add baselines for additional protocols developed
within WP3.
For the MAC protocol baseline, IEEE 802.11 was chosen because of its widespread adoption, packet-based
operation and use of unlicensed frequency bands. The 802.11 standard defines multiple medium access
functions but the two basic ones are the distributed coordination function (DCF) and the point coordination
function (PCF). DCF is the fundamental uncoordinated MAC technique in 802.11, available in both
software and hardware implementations, and relatively simple (being based on CSMA/CA). It was used as
the baseline for the distributed MAC protocols developed in RESCUE. For the centralized MAC (Section
3.3), we adopted a coordination procedure which shares some similarities to the PCF of 802.11. However,
since PCF-compliant implementations are not commonly available, without loss of generality and to
provide a fair comparison, we used as a baseline the proposed centralized MAC but with all RESCUE
functionality turned off. As key performance indicators (KPIs), in all cases, we consider the following
metrics: throughput, delay, jitter, and packet loss.
As the baseline for the RESCUE geographic routing we consider the ETSI GeoNetworking protocol, which
provides single-hop and multi-hop communication in vehicular ad-hoc networks based on IEEE
802.11p/ITS-G5. The following metrics are used as KPIs: packet success rate2, end-to-end delay, and data
traffic overhead.
As the baseline for the RESCUE multi-path routing (R-OLSR) we consider an implementation based on
OLSR with link quality metric ETX (Expected Transmission Count) [8] [16] as the baseline. Previously,
we planned to consider multi-path OLSR (MP-OLSR) [9]. However, it is not available in ns-3 and its
implementation would reduce our focus on developing R-OLSR. We consider the following KPIs:
throughput, packet loss rate, end-to-end delay, routing overhead, and the cold start convergence time.
Regarding the RESCUE end-to-end ARQ protocol, the goal of the simulations was to show that a RESCUE
network can observe a performance gain from advanced MAC-layer end-to-end ARQ. Since the concept
of lossy forwarding is a paradigm shift with respect to message transfer design, an ARQ protocol in this
case cannot be decoupled from the underlying MAC. With this in mind, and to achieve the goal of the
simulations, we used SimpleMAC as a baseline which is based on the Stop-and-Wait (SaW) approach
similarly to other computer network standards (e.g., IEEE 802.11, IEEE 802.15.1, IEEE 802.15.4) although
with significant modifications related to ACK handling as necessitated by the specific RESCUE case. As
2 This metric was mistakenly labelled as throughput in D1.2.1, although the equation (1.1) in that deliverable does in
fact measure the packet success rate.
RESCUE D3.2 Version 2.0
Page 16 (79)
KPIs we consider throughput, end-to-end delay, and jitter (ARQ guarantees transmission without packet
loss).
In case of partial ARQ, our contribution is that of adding Forward Error Correction (FEC) on top of the
ARQ protocol, which is referred to as Partial Hybrid-ARQ (P-HARQ) in this document. Accordingly, the
main focus is not only the ARQ protocol itself, but also the design of encoding and decoding for the FEC.
Hence, we consider the following KPIs: bit-error-rate (BER), packet-error-rate (PER), and e2e throughput.
As the baseline, we consider Smart Hybrid-ARQ (SHARQ) [13].
Finally, for the multirate algorithm, we conclude that using existing algorithms as a baseline for comparison
is not applicable in the RESCUE case because existing rate selection algorithms:
have been only prepared for single input single output transmission systems and did not take
cooperative transmission into account,
have been designed for the existing 802.11 standard which does not have the constraints that
RESCUE has,
assume access to data link layer parameters,
have mostly been prepared for nomadic Wi-Fi mobility and, consequently, they assume a slow
fading channel.
As KPIs for the RESCUE rate selection we consider throughput, delay, and jitter.
2.7 Simulation parameters
To ensure good comparability between all WP3 algorithms and protocols, in addition to the common set of
scenarios (Section 2.3 and 2.4) as well as the baselines and metrics (Section 0), the simulation parameters
were set to common values. In this section, we explain the parameters and give a short justification of the
configured values.
The data rate is set to 6 Mbit/s, which is a common value in car-to-car communication and fits to the low
rate capabilities of the SDR platforms. The restriction in LOTF is that the PHY header has to be error free.
Additionally, in harsh channel conditions the whole packet should be relatively small. Therefore, we
decided to configure 1000 byte packets to have an optimal trade-off between a small packet size and a good
payload to PHY header ratio.
An appropriate channel model is a further important point for gathering realistic simulation results. In our
simulations we set a Nakagami fading model with log-distance path loss and a channel exponent of 3, which
best fits the empirical data. In some simulations we also use a fixed model, where the reception probability
is deterministic and depends on the distance among two nodes. This is done in order to simulate the specific
toy scenarios TS1 to TS3, which would be difficult with the unexpected reception behaviour of nodes in
case of a Nakagami channel. Additionally, we used the BPSK modulation because it is robust against the
assumed poor channel conditions.
The PHY Black Box operates (depending of the number of message copies) down to -6 dB of SNR. To
cover the whole working range the Frame Receive Power Threshold is decreased to -105 dBm and a
common value for the Carrier Sense Power Threshold is the half of it (-108 dBm ≙ -3dB less than the
receive power threshold).
The 2.4 GHz frequency band is most frequently used by current 802.11 networks and, therefore, it is very
crowded. For this reason we choose the 5 GHz frequency band in the majority of our simulations and the
5.9 GHz band for the car-to-car communication (as it is standardized in ITS/G5, 802.11p).
In all simulations the designed algorithms are compared with their baselines or with the LOTF functionality
turned off. The main variable in the simulations is the channel quality (SNR) which was varied by changing
either the noise level, distance or the TX power.
The used simulator is ns-3, which is an open source network simulator, accepted by the scientific
community. The common simulation parameters and their justification are summarized in Table 2.3.
RESCUE D3.2 Version 2.0
Page 17 (79)
Table 2.3: Common simulation parameters for all partners
Parameter Common value Justification
Data rate 6 Mbit/s Low rate in line with SDR device capabilities, default for car to
car communication
Packet size 1000 Byte Typical large packet size in wireless research, good PHY
payload to PHY header ratio
Path loss model LogDistance Default model in wireless communications
Fading model Fixed Nakagami Realistic model which fits empirical data, suitable model to
simulate particular toy scenarios
Path loss
exponent
3 Simulation of unpredictable environments to show the gain of
RESCUE
Modulation
scheme
BPSK Modulation robust to poor channel conditions, supported by
PHY Blackbox
Carrier Sense
Power
Threshold
-108 dBm Half (3 dB) less than the receive power threshold
Frame Receive
Power
Threshold
-105 dBm Guarantees at least -10 dB of SNR to work in the whole range
of PHY BlackBox operation
Frequency 5 GHz/5.9 GHz Transmission frequency for 802.11 and ITS/G5 (802.11p),
license-free band, less congested than the 2.4 GHz band
Forwarding
threshold
0.2 Based on WP2 results; it represents the maximum allowable
BER of a received message which is to be processed by the
receiver (the optimization of this parameter is discussed in
Appendix A).
RESCUE D3.2 Version 2.0
Page 18 (79)
3. MAC
3.1 SimpleMAC
3.1.1 Protocol overview & design update
SimpleMAC is the implementation of the basic RESCUE principles. Its design allows basic message
transfer in a RESCUE network, i.e., in a network in which at least one link on each path between the source
and the destination stations is lossy. By taking into account the RESCUE assumptions and requirements
described in D3.1, SimpleMAC defines:
the basic cross-layer data flow at the source, relay, and destination stations,
rules for medium access based on CSMA/CA,
a method for ensuring reliable end-to-end delivery of messages including
o handling transmission of acknowledgement frames over multi-hop links,
o handling DATA frame retransmissions,
methods for storing multiple copies of messages for joint decoding.
Therefore, SimpleMAC serves as
a starting point for the development of more advanced MAC protocols,
a reference for measuring improvements in the RESCUE cross-layer design,
an abstraction of the MAC layer basics to be used in the analysis of other protocols.
MA
C
PHY
TDMA MAC
AP
SimpleMAC (CSMA)
STA ADHOC
HIGHER LAYERS
CHANNEL
REM
OTE
ST
ATI
ON
M
AN
AG
ER
MULITRATE CONTROL
ARQ
CONTENTION WINDOW CONTROL
Figure 3.1: The RESCUE simulation model architecture.
Figure 3.1 illustrates the implementation of the RESCUE architecture in ns-3. SimpleMAC is the core MAC
module which can operate in a simple network without any additional mechanisms. However, it is designed
in such a way that cooperation with additional control mechanisms (e.g., multi-rate control algorithms,
advanced ARQ methods, routing protocols) is possible. These additional mechanisms are intended to
improve RESCUE performance especially in large-scale scenarios. SimpleMAC may also be used by the
Centralized MAC protocol (cf. Section 3.3) during the Contention Period. Additionally, it is also the basis
for the advanced Distributed MAC Protocol (cf. Section 3.2).
The SimpleMAC protocol is based on the CSMA/CA protocol for medium access. If a dedicated routing
protocol is not used, SimpleMAC resorts to a flooding mechanism for multipath frame transmission.
Additionally, multipath ACK forwarding is performed to provide basic end-to-end reliability. The
implemented SimpleMAC behaviour is the following:
1. All stations receiving frames with correct PHY headers provide them to the MAC layer. Based on
the addresses in the frame header, the decision about taking part in the pending transmission is
made. If routing is not configured, the flooding mechanism is used which means that all stations
which are not the source or the destination of the given transmission become relays.
2. Relays forward frames even if the frame payload is corrupted. However, frames of which BER
exceeds a specified threshold are dropped. The BER estimation is performed by the underlying
PHY layer.
RESCUE D3.2 Version 2.0
Page 19 (79)
3. The destination accumulates all corrupted frame copies until the joint decoder is able to restore
the original data. In the RESCUE ns-3 implementation the information about a frame’s quality is
based on the PER derived from the abstraction of the joint decoder (i.e., the PHY Black Box,
Section 2.5) implementation. Currently up to three frame copies can be used by the PHY Black
Box, thus up to three frame copies with the lowest BER are be stored by the destination.
4. After the reception of a correct frame, SimpleMAC may transmit an ACK frame after the SIFS
period without performing any backoff procedure. The presence of the ACK transmission depends
on the ARQ configuration (cf. Section 5.1).
5. Acknowledgment frames may be forwarded by relays. In this situation the backoff procedure is
obligatory to reduce the risk of collisions.
6. If an additional routing mechanism is not configured, a particular station tries to forward the ACK
frame only if it transmitted a copy of the DATA frame which is being acknowledged.
7. To further reduce the intensity of frame transmissions, the following additional rules of relay
behaviour are used:
a. Only one copy of a given frame is forwarded (with the exception of a DATA frame
retransmission).
b. All DATA frame copies waiting for possible transmission are deleted from the stations’
transmission queue upon receiving the ACK frame for the stored data frame.
c. Any given ACK frame is forwarded only once.
d. The forwarded ACK frames are stored by the relays. If the station receives a frame copy
coming from the retransmission of any DATA frame which should be acknowledged, it
responds with sending the stored ACK frame and drops the retransmitted DATA frame.
8. A simple ARQ protocol is implemented as part of SimpleMAC. It provides the end-to-end
acknowledgments with a configurable ACK waiting time. This mechanism does not allow for a
new data frame transmission if the ACK frame for the previous data frame is waiting for
transmission. As this behaviour may significantly reduce network performance, more advanced
ARQ protocols described in Section 5.1 can be used instead.
9. The following four independent transmit queues are used (starting from the highest priority):
a. ACK – for ACK forwarding
b. CTRL – for transmission of control frames (e.g., routing messages)
c. RELAY – for forwarding of frame copies by the relays
d. DATA – for transmission of data frames originated by a given station
10. Frames are selected with strict priority, which means that the ACK transmission always has
priority over all other frame types, which helps in fast ACK forwarding and thus is important for
the reduction of network traffic. Moreover, control frames are transmitted before the relayed
frames and frame relaying always has priority over transmission of a locally-generated DATA
frame.
11. The contention window (CW) size used by the backoff procedure may be controlled by an external
mechanism (e.g., the multi-rate control algorithm). Otherwise, CW remains constant.
The RESCUE SimpleMAC module for ns-3.21 was created on the basis of the “Simple CSMA/CA protocol
for ns-3” (available online: http://www2.engr.arizona.edu/~junseok/simple_wireless.html). Additionally,
the RESCUE frame formats, transmission modes, and multi-rate control mechanisms were implemented on
the basis of the corresponding implementations imported from the ns-3 Wi-Fi module.
3.1.2 Simulation results
The simulation analysis was performed using the parameters listed in Table 3.1. The initial study, to validate
the successful operation of SimpleMAC, was done with a fixed propagation loss model (the
MatrixPropagationLossModel in ns-3). In this model, the propagation loss is fixed for each pair of nodes
and does not depend on their relative positions. Subsequent analyses were done with a more advanced
model (the NakagamiPropagationLossModel in ns-3).
Table 3.1: Simulation parameters
Parameter Value
Warmup time 25 s
Simulation time 200 s (600 s for unsaturated
Nakagami case)
Propagation loss model Fixed or Nakagami
Path loss exponent 3.0
Reference loss (for distance = 1m) 46.667 dB
RESCUE D3.2 Version 2.0
Page 20 (79)
Nakagami shape factor 1 (for distance <
177.3)
1.0
Nakagami shape factor 2 (for distance >
177.3)
0.85
Noise level 92.515 dBm
Carrier Sense Power Threshold 108 dBm
Frame Receive Power Threshold 105 dBm
TX Power 20 dBm
PHY Data Rate 6 Mbit/s
PHY Basic Rate 3 Mbit/s
BER threshold 0.2
CW 16
Slot time 15 µs
SIFS time 30 µs
LIFS time 60 µs
Queue length 100 frames
Retry Limit under non-saturation 6
Retry Limit under saturation 2
ACK Timeout
(Baseline CSMA/CA, TS-1, stop and wait
ARQ)
3500 µs
ACK Timeout
(TS-2, TS-3, stop and wait ARQ)
7000 µs
Traffic load under non-saturation 0.16 Mbit/s
Traffic load under saturation 15 Mbit/s
Packet Size 1000 bytes
Figure 3.2: Throughput comparison between CSMA/CA (baseline) and RESCUE (SimpleMAC) for
the three toy scenarios: TS1, TS2, and TS3. The link between the R and D stations is lossless. The S-
R and S-D links are lossy and their SNR varies from -1.5 dB to -7 dB.
We begin by comparing the performance of RESCUE SimpleMAC and CSMA/CA (the baseline) when the
channel model uses the fixed propagation loss model. In the most interesting case, presented in Figure 3.2,
the link between R and D is lossless while the links between S and R as well as between S and D are lossy
and their SNR varies from -1.5 dB to -7 dB. There are two baseline results: for TS1/TS3 and for TS2. For
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
-7-6-5-4-3-2
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
TS1/TS3 CSMA/CA TS2 CSMA/CA TS1 RESCUE TS2 RESCUE TS3 RESCUE
RESCUE D3.2 Version 2.0
Page 21 (79)
TS1/TS3 the baseline throughput values are calculated on the S-D link3. In TS2 there is no direct link
between the S and D stations and, therefore, even the baseline transmission is realized through relays. In
all considered cases, the performance of RESCUE SimpleMAC is the same as or much better than that of
CSMA/CA. For each toy scenario, when SNR drops below -2.35 dB, transmission with CSMA/CA
becomes impossible but with SimpleMAC non-zero throughput is achieved up to an SNR value of -5.5 dB
(TS1), -6 dB (TS3), or -7 dB (TS2).
(a) (b)
(c) (d)
Figure 3.3: Comparison of RESCUE SimpleMAC with CSMA/CA for TS1 under non-saturation: (a)
throughput, (b) FLR, (c) delay, (d) jitter. The distances of the S-D and S-R links
In the next scenario, we replace the fixed propagation loss model with the Nakagami model. For the TS1
topology, we configure the station placement so that S and D as well as S and R are placed 300 meters
apart. We measure the SimpleMAC performance over varying distances of the R-D link. Figure 3.3 presents
the values of the different KPIs (throughput, FLR, delay, and jitter) obtained for RESCUE (Simple MAC)
in comparison with the baseline (CSMA/CA) for the TS1 under non-saturation. For small R-D distances,
the throughput improvement is up to 250%. For distances exceeding 350 meters, the RESCUE improvement
is about 150% while the FLR is improved between 15 and 50 percentage points. The delay and jitter values
are comparable for both protocols with SimpleMAC achieving a slightly higher (but slightly more stable)
delay.
3 In the considered case, adding the S-R-D link to the baseline calculation does not impact the results. This is because
of the selected channel model and the propagation loss configuration.
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
50 100 150 200 250 300 350 400 450 500
Thro
ugh
pu
t [M
b/s
]
Distance [m]
CSMA/CA RESCUE
0
10
20
30
40
50
60
70
80
90
100
50 100 150 200 250 300 350 400 450 500
FLR
[%
]
Distance [m]
CSMA/CA RESCUE
12
12.5
13
13.5
14
14.5
15
15.5
50 100 150 200 250 300 350 400 450 500
De
lay
[ms]
Distance [m]
CSMA/CA RESCUE
7.4
7.6
7.8
8
8.2
8.4
8.6
8.8
9
50 100 150 200 250 300 350 400 450 500
Jitt
er
[ms]
Distance [m]
CSMA/CA RESCUE
RESCUE D3.2 Version 2.0
Page 22 (79)
(a) (b)
(c) (d)
Figure 3.4: Comparison of the RESCUE SimpleMAC with CSMA/CA for TS1 under saturation: (a)
throughput, (b) FLR, (c) delay, (d) jitter. The distances of the S-D and S-R links and S-R links
are set to 300 m. The distance of the R-D link is variable.
Figure 3.4 presents the same results as in the previous scenario but under saturation conditions. The
throughput improvement from using SimpleMAC in this case varies from 100% (at 300 m) to 350% (at 50
m). The saturation conditions are evident in the elevated FLR and delay values although the jitter remains
low. In all cases, RESCUE is equal to or outperforms the baseline.
(a) (b)
0
0.05
0.1
0.15
0.2
0.25
0.3
50 100 150 200 250 300
Thro
ugh
pu
t [M
b/s
]
Distance [m]
CSMA/CA RESCUE
0
10
20
30
40
50
60
70
80
90
100
50 100 150 200 250 300
FLR
[%
]
Distance [m]
CSMA/CA RESCUE
990
1000
1010
1020
1030
1040
1050
1060
1070
1080
1090
50 100 150 200 250 300
De
lay
[ms]
Distance [m]
CSMA/CA RESCUE
0
1
2
3
4
5
6
7
8
50 100 150 200 250 300
Jitt
er
[ms]
Distance [m]
CSMA/CA RESCUE
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
50 100 150 200 250 300 350 400 450 500
Thro
ugh
pu
t [M
b/s
]
Distance [m]
CSMA/CA RESCUE
0
20
40
60
80
100
120
50 100 150 200 250 300 350 400 450 500
FLR
[%
]
Distance [m]
CSMA/CA RESCUE
RESCUE D3.2 Version 2.0
Page 23 (79)
(c) (d)
Figure 3.5: Comparison of RESCUE SimpleMAC with CSMA/CA for TS2/3 under non-saturation:
(a) throughput, (b) FLR, (c) delay, (d) jitter. The length of each link (S-D, S-R1/R2, R1/R2-D) is the
single variable.
In the next analysed case we considered TS2 and TS3 jointly. This is because the difference between them
is the existence or lack of the direct S-D link. As we increase the distance between all stations this link
becomes increasingly unreliable until it ceases to exist and all traffic goes through the two relays. Thus,
TS2 changes into TS3. We consider stations to be placed equidistantly from each other with the distance
between them being the single variable of the analysis. For the unsaturated case (Figure 3.5), RESCUE is
able to provide up to a 75% throughput increase (at 250 m), which corresponds with a similar decrease in
the frame loss ratio. The delay and jitter values remain low up to a distance of 300 m. Afterwards, the
number of successfully delivered messages is close to zero which causes the large increase and variation of
these metrics.
(a) (b)
(c) (d)
Figure 3.6: Comparison of RESCUE SimpleMAC with CSMA/CA for TS2/3 under saturation: (a)
throughput, (b) FLR, (c) delay, (d) jitter. The length of each link (S-D, S-R1/R2, R1/R2-D) is the
single variable.
For the saturated case of the TS2/TS3 analysis (Figure 3.6), the throughput increase is moderate – up to
50% (at 200 m) along with the FLR decrease (up to 75% improvement at 200 m). The delay and jitter values
become increasingly large for this network – saturation conditions coupled with lack of flow rate control
are the result of this. Nonetheless, the RESCUE approach brings improvements – up to 21% decreased
delay and up to 25% decreased jitter (at 200 m).
0
200
400
600
800
1000
1200
50 100 150 200 250 300 350 400 450 500
De
lay
[ms]
Distance [m]
CSMA/CA RESCUE
0
100
200
300
400
500
600
50 100 150 200 250 300 350 400 450 500
Jitt
er
[ms]
Distance [m]
CSMA/CA RESCUE
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
0 50 100 150 200 250 300 350 400
Thro
ugh
pu
t [M
b/s
]
Distance [m]
CSMA/CA RESCUE (TS2/TS3)
0
20
40
60
80
100
120
0 50 100 150 200 250 300 350 400
FLR
[%
]
Distance [m]
CSMA/CA RESCUE (TS2/TS3)
0
500
1000
1500
2000
2500
0 50 100 150 200 250 300 350 400
De
lay
[ms]
Distance [m]
CSMA/CA RESCUE (TS2/TS3)
0
100
200
300
400
500
600
0 50 100 150 200 250 300 350 400
Jitt
er
[ms]
Distance [m]
CSMA/CA RESCUE (TS2/TS3)
RESCUE D3.2 Version 2.0
Page 24 (79)
The conclusions drawn from the implementation and simulation results of SimpleMAC are the following.
1. SimpleMAC is equal to or outperforms the baseline in the considered scenarios.
2. The improvements provided by SimpleMAC are higher in non-saturation scenarios than in saturation
scenarios. This is in line with the traffic requirements specified for both public safety and V2V
communications as defined in D1.1.
3. The demonstrated SimpleMAC performance and improvement gains over the baseline (CSMA/CA)
can be further increased by optimizing for the following criteria:
optimal PHY header size,
optimal placement of relays,
optimal values of offered loads,
optimal retry limits and timeouts.
4. SimpleMAC can serve as a good baseline for comparison of further WP3 message transfer protocols.
Indeed, as the first complete cross-layer solution for lossy link communications it opens a new field of
research. With respect to the other protocols developed within WP3, further performance
improvements can be expected from the following protocols:
centralized MAC,
distributed MAC,
end-to-end ARQ,
partial ARQ,
rate adaptation algorithm.
5. The presented results are upper bound by the limitations of the PHY Black Box. This, together with
the promising results, makes SimpleMAC a good candidate for implementation and validation in SDR
devices.
3.2 Distributed MAC
In addition to the basic MAC mechanisms essential to implementing the links of the fly concept (i.e., no
CRC check and corrupted frames buffering), managing the access and controlling the level of errors in a
large ad-hoc network requires novel and advanced functionalities. For this reason, we propose in RESCUE
a distributed MAC protocol capable to schedule transmissions and ensure controllable reliability in large
ad-hoc topologies. We give a brief overview of our distributed MAC protocol in the following section. For
additional details, please refer to the D3.1 deliverable.
3.2.1 Protocol overview & design update
As thoroughly described in D3.1, in order to handle the challenges imposed by the RESCUE lossy
forwarding paradigm, the RESCUE distributed MAC protocol offers three essential features:
A busy tone control channel: This low data rate channel is used to announce an activity by a
node receiving a frame in order to extend the traditional wireless MAC carrier sensing (refer to
Figure 3.7). Indeed, this mechanism increases the reliability of the MAC protocol in an attempt to
replace the classical MAC layer acknowledgment deemed useless with our paradigm. Clearly in
this case acknowledging packets that can be corrupted makes no sense in a lossy forwarding
context. However, using a busy tone control channel does not guarantee the frame reception nor
an accepted level of distortion in the received information.
Replacing ACK by overhearing since the full reliability of frames transmission is not required.
Indeed, many researchers have advocated to alleviate the MAC layer process by making nodes
listening after transmission to detect if neighbours are forwarding the received frame. However,
their proposals did not meet the expected success for efficiency and reliability limitation. In the
case of RESCUE lossy forwarding, receiving corrupted frames can be mitigated by having
multiple copies of the message reach the final destination. Nevertheless, in order to make this
overhearing possible we also introduce the third feature of our distributed MAC detailed hereafter.
Fostering relaying over new transmissions by setting shorter timers for relayed frames
compared to those arriving from higher layers. This feature completes the overhearing process by
forcing neighbours to relay as fast as possible making it easy to the source to identify successful
transmissions. Additionally, our protocol propagates messages in the network closer to the
destination thus creating more diversity and avoiding bottlenecks close to the source node.
RESCUE D3.2 Version 2.0
Page 25 (79)
S
R1
R2
Transmission
DCCC
CC
CC
DC
DCDC: Data Channel
CC: Control Channel
Figure 3.7: Control channel and data channel interaction in the RESCUE distributed MAC.
Figure 3.8 shows the operating mode of our distributed MAC protocol through a detailed state diagram. In
summary, one can extract the following main characteristics of our mechanism:
The use of a busy control channel (CC) before any transmission in order to detect the activity of
2-hop neighbours.
The exploitation of different timer values in order to favour relaying over new transmissions.
Indeed, T_Short is employed whenever the information comes from lower layers (relayed).
The establishment of controllable reliability through neighbourhood management when CC is
occupied. In fact, a sender can decide to transmit or not when the control channel is busy depending
on the number of neighbours it possess. In practice, such decisions have a direct impact on the
reliability of the mechanism. More precisely, a source can decide to transmit when the CC is busy
while having more than N neighbours. This means that in the case that one of the neighbours is
already receiving, our distributed MAC can tolerate some losses that will be corrected with the
joint decoding mechanism. A more conservative approach would prevent transmission whenever
the CC is busy or, in other words, when N ≥ 1.
Figure 3.8: Distributed MAC state diagram
RESCUE D3.2 Version 2.0
Page 26 (79)
In summary, our MAC protocol allows to dynamically optimize the reliability based on the number of
neighbours when the CC is busy. In fact, allowing transmissions when many neighbours are available
reduces delay at the price of increasing losses, whereas forbidding any transmission whenever the CC is
busy constructs a conservative approach that increases the reliability of our protocol. Note that this
configurable reliability feature was not initially proposed in D3.1 but is now added as a system
enhancement.
3.2.2 Simulation results
In order to validate the distributed MAC protocol and characterize its key performance parameters, we have
simulated our solution (the distributed ad-hoc MAC) using the ns-3 simulator with the physical layer
abstraction designed for RESCUE. Our simulations extend the simple MAC as shown in Figure 3.1 by
adding the control channel functionality, adapted timers to favor relaying and controllable reliability.
The simulation analysis was performed using the parameters listed in Table 3.2. In order to control distances
and losses in the considered topologies we consider the simple ThreeLogDistance loss model in ns-3. The
objective here is to have multi-hop topologies with SNR decreasing with distance.
Table 3.2: Distributed (ad-hoc) MAC ns3 simulation parameters
Parameter Value
Warmup time 10 s (margin for neighbourhood
discovery)
Simulation time 15 s
Simulation runs 10
Loss model ThreeLogDistance
Path loss exponent 3
Reference loss (for distance = 1m) 46.667 dB
Carrier Sense Power Threshold -108 dBm
Frame Receive Power Threshold -105 dBm
TX Power Variable to change received SNR
PHY Data Rate 6 Mb/s
Slot time 15 µs
SIFS time 30 µs
LIFS time 60 µs
Queue length 100 frames
Packet Size 1000 bytes
N value 1 (conservative approach)
We first focus on a simple topology shown in Figure 3.9 below in order to compare the distributed MAC
to the basic MAC functionalities required for joint decoding that we called simple MAC. Note that in this
simple topology, no joint decoding is invoked since a single route exists between the sender and the
receiver. Furthermore, no ARQ protocol is implemented in our simulations. This was done in order to
isolate the MAC layer and investigate its parameters and their interplay. In fact, the main objective of these
simulations is to investigate whether the added complexity and overhead of the distributed MAC affect
negatively the performance of the system. We want to quantify in particular the impact of the added features
in the distributed MAC mainly the busy tone channel as well as the modified timer values that favor relaying
over new transmissions.
Figure 3.9: Simple topology for characterizing the overhead of the distributed MAC
Destination
source
relayPath1
RESCUE D3.2 Version 2.0
Page 27 (79)
The obtained simulation results are shown in Figure 3.10 and Figure 3.11 below.
Figure 3.10: Throughput comparison between
simple and distributed MAC for 2 input rates
Figure 3.11: Frame loss rate comparison
between simple and distributed MAC for 2
input rates
Figure 3.10 shows that at relatively high SNR (higher than 0), the simple MAC outperforms the distributed
MAC in terms of throughput when the input rate is high (3 Mbits/s). We believe that this advantage is due
to the aggressive scheduling method of the simple MAC that relies only on CSMA/CA mechanism unlike
the ad hoc MAC that implements a busy tone channel to reduce collisions. However, even if the throughput
is high, the transmissions of the simple MAC encounters losses even at high SNR as highlighted in Figure
3.11. When reducing the SNR value, we notice a decrease in the throughput of the simple MAC caused by
a high frame error rate. On the contrary, the distributed MAC is able to keep the throughput stable and the
frame loss ratio very low (almost 0) until reaching an SNR of around -2.4. Note after this value, the receiver
is not able to retrieve the initial information hence the collapse of the performance. Note here, that at low
input rate (1 Mbits/s), the load of the system is low that both MAC protocols are able to schedule their
transmissions easily (few nodes competing for resources), thus simple and distributed MAC offer similar
performance. The system continues to operate without errors until reaching the reception breaking point
(SNR of -2.4).
We now consider a more complex topology, offering multiple routes i.e. a scenario where the RESCUE
joint decoding applies. In our particular case we have considered two routes with different lengths to
increase the diversity. Note that during these simulations, no ARQ mechanism is implemented. This is done
in order to allow fair comparisons between the considered protocols. The simulated topologies simplify the
general public safety scenario described previously and are shown in Figure 3.12.
Figure 3.12: Simulated assymetric multi-route topology that simplifies the public safety scenario
As stated earlier, our objective is to characterize the influence of the added features of the ad hoc MAC on
the performance in a multihop topology. For this reason, we investigate again in our comparisons on the
two main studied metrics in this document i.e. the end-to-end throughput observed at the destination as well
as the Frame Loss Ratio (FLR). We derive these two metrics for different input application rates at the
Destination
source
relayPath1
RESCUE D3.2 Version 2.0
Page 28 (79)
source and also while varying the minimum SNR in our topology. Indeed, since we consider a multihop
topology, we plot the observed throughput and loss ratio function of the bottleneck SNR over the path. Note
that the effective total SNR obtained at the destination is higher than the one we use as it considers all the
SNRs on the route. Note also, that in order to explore different SNR values, we modified the nodes
transmission power. We have again run these simulations for 2 different input (application) rates.
Figure 3.13: Throughput comparison over mult-
route topology
Figure 3.14: Frame loss ratio over the multi-route
topology
Looking at the throughput curves, they highlight 2 important results. First, the distributed MAC
outperforms the simple MAC. Indeed, regardless of the input rate 1Mbits/s or 3 Mbits/s, the added
mechanisms in the ad hoc MAC help alleviate the collisions thus increase the throughput. In particular, the
busy tone control channel is able to extend the carrier sensing one additional hop and eliminate in some
cases the hidden node problem. Second, one can find surprising that for 6Mbits/s capacity link, the observed
throughput is higher for 1 Mbits/s input rate than for 3 Mbits/s for the simple MAC protocol. However, this
result is somehow classical in multihop wireless network using CSMA based access techniques. More
precisely, with these access methods the wireless resource is shared evenly between competing nodes,
hence the “real” capacity observed by users shall be roughly divided by 4 in the considered line topology
in our simulations. These observations were already proven and demonstrated in the literature [14] [15].
Therefore, the communication capacity cannot exceed 2Mbits/s in our topology when the theoretical link
capacity is 6 Mbits/s using classical CSMA access technique (simple MAC). For this reason in our
simulations, the throughput completely collapses for input the 3 Mbits/s input rate causing congestion and
retransmissions and yielding lower throughput than 1 Mbits/s input rate. With the help of the receiving busy
tone channel and adapted inter frame timers that favor relaying over new frames from the source, the
distributed MAC can increase the capacity for communications thus allowing higher input rates and
throughputs.
Similar conclusions can be made for the Frame Loss Ratio (FLR) curve. In particular, the simple MAC
suffers from higher losses than the proposed distributed MAC. Moreover, an input rate of 1 Mbits/s offers
better performance than 3 Mbits/s with this simple MAC strategy due to reasons detailed earlier. Besides,
the distributed MAC can maintain a low level of errors until an SNR of -4.9 dB. Note that this breaking
point is the same for simple and distributed MAC meaning that it constitutes a limit for the joint decoding
of RESCUE.
3.3 Centralized MAC
3.3.1 Protocol overview & design update
The RESCUE Centralized MAC protocol described in D3.1 is a coordinated medium access method which
was designed to minimize collisions. It works in networks consisting of an access point (AP) and user
stations (STAs) being one-hop neighbours of the AP. The operation of Centralized MAC is based on a
super-frame consisting of a contention free period (CFP) and a contention period (CP). Channel access is
realized using CSMA/CA in the CP (in which stations send resource reservation requests) and using TDMA
in the CFP. The AP sends a beacon (B) frame at the beginning of each CFP in order to inform STAs about
transmissions scheduled for the next CFP. The AP also decides which STAs should act as relays and
RESCUE D3.2 Version 2.0
Page 29 (79)
therefore no routing mechanism is required by the centralized MAC. Additionally, the source nodes reserve
TDMA slots for the planned uplink transmission in the CFP using the resource reservation (RR) frames
transmitted in the CP.
The DATA frame specific to the RESCUE Centralized MAC is similar to the one presented for the
Distributed MAC in Section 3.2, however, the MAC Protocol subfield of the CTRL field is set to 1 instead
of 0.
Figure 3.18: The Resource Reservation frame format.
Figure 3.18 provides a description of each of the fields in the RESCUE Resource Reservation frame. The
frame begins with a preamble followed by a 2-byte control (CTRL) field within which the Data Rate Next
field specifies information regarding the MCS used to transmit the next data frame. The Frame Length Next
field informs about the length of the planned DATA frame transmission. The CTRL field is followed by
the Duration Next field, which informs about the duration of the transmission of the next data frame. The
source Address is the PHY address of the sender S. The destination Address is the PHY address of the
destination. Other fields are the same as specified in Section 2.2.
Figure 3.19: The Beacon frame format.
Figure 3.19 provides a description of each of the fields in the Beacon frame. The 2-byte CTRL field includes
information on the supported data rates, QoS capability, configured channel number, assumed transmission
(TX) power and the number of entries in the scheduled list. The Timestamp field is used for network
synchronization. The source address field is the address of the AP. The beacon interval is the time duration
between two beacon frames. The length of the CFP period is specified in the 4-byte CFP period field. The
4-byte TDMA timeslot field indicates the length of a single TDMA slot. The scheduled list field has a
dynamic length and it contains the list of reserved uplink transmissions.
3.3.2 Simulation results
The simulations of the RESCUE Centralized MAC were performed for the network topology presented in
Figure 3.20 with one to 25 stations sending data to the AP. Each sending station Si had a dedicated relay
Ri, placed in the middle of the distance between the sending station Si and the destination (D) station (i.e.,
the AP).
Typ
e
Dat
aR
ate
3 2
Sen
d W
indo
w
4
Blo
ck A
CK
Con
t. A
CK
1 1Bits:
Dat
aR
ate
Nex
t
2
Synchronization
SFD CTRL Source AddressDestination
AddressSEQ.
Inte
rle
aver Header
checksum
Preamble PHY Header
2 6 6 2 1 4Bytes:
Type: 100 (Resource Reservation)DataRate: MSC used;Send Window: how many unACKed frames the station can send;Block ACK: 1 – requested/supported, 0 – not allowed;Continous ACK: 1 – requested/supported, 0 – not allowed;QoS: 1 – supported, 0 – not supportedData Rate Next: the frame length of the planned DATA transmissionReserved: for future use
1
Qo
S
MBF
2
Du
ratio
nN
ext
2
Res
erv
ed
2
Typ
e
Supp
.D
ata
Rat
es
3 2 1Bits:
Synchronization
SFD CTRLSource
Address
Preamble PHY Header
2 6Bytes:
Qo
SC
apa
bilit
y
Cha
nn
el Type: 011 (Beacon)Supp. Data Rate: Supported Data Rates;QoS Capability: 1 – supported, 0 – not supported;Channel: Channel numberTX Power: The assumed TX powerSched. list size – the numer of entries in the schedules list
Header checksum
4
2
Timestamp
4
TX P
ower
Sch
ed.
list
size
3 5
Beacon Interval
4
Scheduled list
CFP period
Dynamic4
TDMA timeslot
4
RESCUE D3.2 Version 2.0
Page 30 (79)
Figure 3.20: Network topology scenarios used for the RESCUE Centralized MAC validation.
The parameters listed in Table 3.1 were used for the simulation with the following exceptions:
Propagation loss model: fixed
Traffic load: 15 Mb/s (saturation)
Throughput improvement of the Centralized MAC over the baseline (as described in Section 2.6) is visible
when the link quality deteriorates for different metrics (throughput, FLR, delay, jitter), regardless of the
number of stations transmitting to the AP (Figures 3.213.24). In Figures 3.21-3.24 the throughput and
FLR values are similar for each configuration. For the Centralized MAC, the throughput is about 4 Mb/s at
higher SNR (from -1.5 dB down to -1.8 dB), 2 Mb/s at lower SNR (from -2.1 dB down to -5 dB), and it
drops to zero under SNR lower than -6.2 dB. For the baseline the throughput drops to zero when SNR is
lower than -2.4 dB. FLR for the Centralized MAC is 0% for SNR in the range (-1.5 dB, -5.7 dB) and for
the baseline it is 0% for SNR in the range (-1.5 dB, -1.8 dB). For the Centralized MAC and the baseline
delay and jitter values increase when the number of sending stations increases because each sender has to
wait longer for accessing the wireless channel.
(a) (b)
(c) (d)
Figure 3.21: Performance results of Centralized MAC with one sender for four metrics: (a)
throughput, (b) FLR, (c) delay, and (d) jitter under saturation.
S1R1 D
S2
R2
S3
R3
SN
RN
RiSi
0
0.5
1
1.5
2
2.5
3
3.5
4
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
Baseline Centralized MAC
0
20
40
60
80
100
120
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
FLR
[%
]
SNR [dB]
Baseline Centralized MAC
0
200
400
600
800
1000
1200
1400
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
De
lay
[ms]
SNR [dB]
Baseline Centralized MAC
0
100
200
300
400
500
600
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Jitt
er
[ms]
Title
Baseline Centralized MAC
RESCUE D3.2 Version 2.0
Page 31 (79)
(a) (b)
(c) (d)
Figure 3.22: Performance results of Centralized MAC for five sending stations in the network for
four metrics: (a) throughput, (b) FLR, (c) delay, and (d) jitter under saturation.
(a) (b)
(c) (d)
Figure 3.23: Performance results of Centralized MAC for 10 sending stations in the network for four
metrics: (a) throughput, (b) FLR, (c) delay, and (d) jitter under saturation.
0
0.5
1
1.5
2
2.5
3
3.5
4
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
Baseline Centralized MAC
0
20
40
60
80
100
120
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
FLR
[%
]
SNR [dB]
Baseline Centralized MAC
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
De
lay
[ms]
SNR [dB]
Baseline Centralized MAC
0
500
1000
1500
2000
2500
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Jitt
er
[ms]
SNR [dB]
Baseline Centralized MAC
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
Baseline Centralized MAC
0
20
40
60
80
100
120
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
FLR
[%
]
SNR [dB]
Baseline Centralized MAC
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
De
lay
[ms]
SNR [dB]
Baseline Centralized MAC
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Jitt
er
[ms]
SNR [dB]
Baseline Centralized MAC
RESCUE D3.2 Version 2.0
Page 32 (79)
(a) (b)
(c) (d)
Figure 3.24: Performance results of Centralized MAC for 15 sending stations in the network for four
metrics: (a) throughput, (b) FLR, (c) delay, and (d) jitter under saturation.
(a) (b)
(c) (d)
Figure 3.25: Performance results of Centralized MAC for 20 sending stations in the network for four
metrics: (a) throughput, (b) FLR, (c) delay, and (d) jitter under saturation.
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
Baseline Centralized MAC
0
20
40
60
80
100
120
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
FLR
[%
]
SNR [dB]
Baseline Centralized MAC
0
2000
4000
6000
8000
10000
12000
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
De
lay
[ms]
SNR [dB]
Baseline Centralized MAC
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Jitt
er
[ms]
SNR [dB]
Baseline Centralized MAC
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
Baseline Centralized MAC
0
20
40
60
80
100
120
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
FLR
[%
]
SNR [dB]
Baseline Centralized MAC
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
De
lay
[ms]
SNR [dB]
Baseline Centralized MAC
0
500
1000
1500
2000
2500
3000
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Jitt
er
[ms]
SNR [dB]
Baseline Centralized MAC
RESCUE D3.2 Version 2.0
Page 33 (79)
(a) (b)
(c) (d)
Figure 3.26: Performance results of Centralized MAC for 25 sending stations in the network for four
metrics: (a) throughput, (b) FLR, (c) delay, and (d) jitter under saturation.
When all stations around the AP are within interference range, the distributed approach (Simple MAC)
slightly outperforms the centralized approach for all metrics except PLR, regardless of the number of
stations (Figure 3.27).
(a) (b)
(c) (d)
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
Baseline Centralized MAC
0
20
40
60
80
100
120
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
FLR
[%
]
SNR [dB]
Baseline Centralized MAC
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
De
lay
[ms]
SNR [dB]
Baseline Centralized MAC
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Jitt
er
[ms]
SNR [dB]
Baseline Centralized MAC
0
0.5
1
1.5
2
2.5
0 5 10 15 20 25 30
Thro
ugh
pu
t [M
b/s
]
No of STAs
Centralized MAC Simple MAC
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
0 5 10 15 20 25 30
FLR
[%
]
No of STAs
Centralized MAC Simple MAC
0
2000
4000
6000
8000
10000
12000
0 5 10 15 20 25 30
Jitt
er
[ms]
No of STAs
Centralized MAC Simple MAC
0
2000
4000
6000
8000
10000
12000
0 5 10 15 20 25 30
De
lay
[ms]
No of STAs
Centralized MAC Simple MAC
RESCUE D3.2 Version 2.0
Page 34 (79)
Figure 3.27: Performance comparison of Centralized MAC and Simple MAC for variable number of
sending stations in the network for four metrics: (a) throughput, (b) FLR, (c) delay, and (d) jitter
under saturation.
However, in the presence of hidden nodes, the coordinated channel access of the Centralized MAC is
superior to Simple MAC since it allows avoiding collisions between data transmitted by nodes hidden from
each other. Figure 3.28 presents the results for a network composed of an AP, five senders, and five relays
(all five Sx-Rx pairs are hidden from each other).
(a) (b)
(c) (d)
Figure 3.28: Performance comparison of Centralized MAC and Simple MAC for a five STA network
with hidden nodes for four metrics: (a) throughput, (b) FLR, (c) delay, and (d) jitter under saturation.
The conclusions drawn from the implementation and simulation results of RESCUE Centralized MAC are
the following.
1. The Centralized MAC outperforms the baseline (described in Section 2.6) under all key
performance indicators (throughput, delay, jitter, and FLR).
2. The performance of the Centralized MAC is worse than the performance of the Simple MAC for
infrastructure networks if hidden nodes are not present in the network. However, for a network
with hidden nodes the performance of Simple MAC is unsatisfactory while the performance of
Centralized MAC is much better than the baseline.
0
0.5
1
1.5
2
2.5
3
3.5
-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
Simple MAC Baseline Centralized MAC
0
20
40
60
80
100
120
-5.5-5-4.5-4-3.5-3-2.5-2-1.5
FLR
[%
]
SNR [dB]
Simple MAC Baseline Centralized MAC
0
1000
2000
3000
4000
5000
-5.5-5-4.5-4-3.5-3-2.5-2-1.5
De
lay
[ms]
SNR [dB]
Simple MAC Baseline Centralized MAC
0
100
200
300
400
500
600
700
-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Jitt
er
[ms]
SNR [dB]
Simple MAC Baseline Centralized MAC
RESCUE D3.2 Version 2.0
Page 35 (79)
4. Routing
Routing protocols are fundamental to the determination of the appropriate paths over which data are
transmitted in ad-hoc networks. Ad-hoc networks are established by a group of nodes that communicate
with each other to form a multi-hop radio networks. Their autonomous topology allows the creation of non-
permanent purpose-built networks, that can be set up without infrastructure constraints and operate under
dense conditions. Current routing protocols inherently trust all participants being cooperative by nature and
depend on neighbouring nodes to route packets to the destination. Depending on their route discovery
mechanisms ad hoc routing protocols are classified into three discrete categories named proactive, reactive
and hybrid. Proactive protocols maintain all possible routes throughout the routing operation by exchanging
control messages on a regular basis, whereas reactive protocols remain in a “sleep” mode, until triggered
by a transmission request. To address the challenges that arise from the performance trade-offs, hybrid
protocols combine both approaches with the view to increase the overall routing performance.
4.1 RESCUE Multipath Routing Protocol
This section introduces a multipath routing protocol based on the Expected Transmitted Count (ETX)
instead of the traditional hop count, proposed for the purposes of the RESCUE project. RESCUE routing
protocol called RESCUE OLSR (R-OLSR), adopts the proactive routing which allows topology awareness
through the periodic exchange of control messages. The designed protocol considers a multipath approach
for the route calculation, aiming to increase throughput and transmission reliability. In addition, it utilizes
a link metric called ETX which estimates the number of successful transmission required for a packet to
traverse a path. Two variants of the routing protocol (R1-OLSR and R2-OLSR) have been designed
following different relay approaches described below and taking into account the RESCUE MAC layer
capability to perform joint decoding of the received packets via different transmission paths.
4.1.1 RESCUE Routing Relay approaches
In the RESCUE project, considering the scenarios described and plans for extending the described
implementations in dense deployments, the following two relay approaches have been considered and
adopted in the design of R-OLSR. In particular, R1-OLSR has been designed to be in line with the simple
relay routing approach while R2-OLSR follows the characteristics of the advanced relay routing approach.
Simple Relay Routing (SRR): each node upon receiving a packet performs a simple “sanity” check to
determine whether there is an available path to the destination that does not include the message sender. In
the case where no available path exists, the message is discarded. In any other case, the message is
forwarded to all available paths, regardless the quality of the link or the number of hops required to be
transmitted.
Advanced Relay Routing (ARR): each node with paths towards the destination upon receiving a packet
takes a binary decision on whether to drop or forward the received data towards the destination based on
the available paths. The ARR approach instead of flooding the network with redundant information,
forwards data to a predefined number of paths originating after a node with disjoint paths. R2-OLSR which
corresponds to the ARR approach sets a transmission path limit based on the minimum (optimal) ETX
values towards the destination.
4.1.2 Protocol overview & design update
This section presents the fundamental concepts on which the R-OLSR lays its foundation. It introduces the
theoretical concepts by discussing the multipath approach as an efficient routing solution to provide
improved performance resiliency, and outlines the theoretical foundation of the ETX link metric.
4.1.2.1 R-OLSR Specification and Design R-OLSR is a proactive OLSR based routing protocol, extended to calculate multiple routing paths based
on the ETX link metric instead of the traditional hop count. The hop count's attribute to calculate routes
based on the minimum hops increases the probability of broken links, since other parameters such as
distance, BER or bandwidth are not evaluated. The ETX metric has emerged in recent years as a promising
and intelligent alternative, which calculates the estimated number of transmissions required for a node to
successfully transmit a single packet using a link. Loss rates are calculated in both directions as illustrated
in Figure 4.1. In particular, node A to B calculates the direct link quality (dlq) as the probability of a
successful transmission within a window period. Similarly, node B to A calculates the probability of a
RESCUE D3.2 Version 2.0
Page 36 (79)
successful transmission on the reception (rlq). The ETX function used for the purposes of R-OLSR for a
single link equals to:
ETX = dlq * rlq (4.1)
Within this context, the ETX value of an investigating path equals to the sum of the ETX values
corresponding to the links of this path.
Figure 4.1: Representation of ETX functionality
The operation of R-OLSR is similar to that of OLSR, adopting the periodic exchange of control messages
named HELLO and Topology Control (TC) messages. HELLO messages are generated and transmitted to
1-hop neighbour and are not any further relayed to the rest of network nodes. The periodic exchange of
HELLO messages offers a convenient mechanism to calculate the ETX, simplifying the link quality
evaluation of the path. In the example of Figure 4.1, node A is not aware of the number of packets node B
wants to transmit. However, it is able to estimate the total number of HELLO messages through node’s B
HELLO time interval. The main tasks of HELLO messages based on the original OLSR protocol involve
the link sensing, the neighbour detection and the MPR selection. R-OLSR follows these principals and
extends them so as to include the ETX calculation. The link sensing process is responsible for the
maintenance of the node's local link information base, which stores information about the node's links to
its neighbours. In R-OLSR, the ETX computation is incorporated inside the link sensing process so as to
allow the link tuples to store information about the ETX value attributed to the link. The neighbour detection
offers 2 hop neighbours identification where addresses are stored within the 2-hop neighbour set. Similar
to the link set, the 2-hop neighbours set maintains information of the ETX corresponding to the link quality
of the 2-hop neighbour.
R-OLSR adopts the MPR computation approach of OLSR, via the periodic exchange of HELLO messages.
Unlike common flooding mechanisms where each node of the network retransmits the received
information, MPRs are composed by a set of nodes which are efficiently segmented throughout the network.
The MPR flooding procedure reduces the improvident emission of routing messages allowing elimination
of redundant transmissions. TC messages as part of the flooding procedure send diffused information
globally in the network and are used for the calculation of the routing table through the topology sensing
process. Initially, TC messages offer nodes the ability to obtain the ETX value corresponding to the
generator of the message and its own MPR selectors. The reason the TC messages are carrying the ETX
value is to offer link information to the topology set. Each time a TC message is received, the topology set
is updated. Similarly to the link set and the 2-hop neighbour set, an ETX value is added in each of the
topology tuples. It essentially evaluates the quality of the link between the MPR selector and the message
generator.
The multipath approach in R-OLSR includes two discrete parts. The first part is the topology sensing which
involves the topology awareness through the exchange of topology information by the MPRs. The second
part is the route computation, including the multi-path calculations through the use of the information
obtained during the topology sensing. Nodes determine the number and size of routes through the periodic
exchange of HELLO and TC messages. During the multipath approach, destinations utilise multiple
available routes which may result to extra computational costs that impact the energy resources of the
protocol. In the traditional OLSR, only one path is used for packet delivery between source and destination
nodes and packets are lost when nodes in the paths to destination go out of range. The path to destination
is also embedded onto the packet. In R-OLSR, R1-OLSR has been developed using the SRR approach
while R2-OLSR has been developed using the ARR approach. In both approaches, multiple paths are used
in data delivery between source and destination. There is no limit on the number of paths for concurrent
RESCUE D3.2 Version 2.0
Page 37 (79)
data transmission using the SRR while the ARR approach introduces a limit on the number of concurrent
transmissions between source and destination. To achieve that goal, disjoint paths corresponding to a
particular destination within the routing table, are sorted in an ascending order based on the minimum ETX
value, to offer efficient transmission to the two most optimal paths. At the destination, the RESCUE MAC
layer functionality of joint decoding is applied on received packets in order to increase the integrity of the
message delivery process and recover erroneous packets delivered over lossy links.
4.1.2.1 Route computation
R1-OLSR Route Computation: The route is computed following the SRR approach. In order to deliver a
packet from the source to the destination, all nodes upon receiving a packet check whether they have a path
to the destination or not. The path must not include the message sender and if so, the packet is forwarded.
Figure 4.2 shows the routing decision diagram for the SRR approach.
Figure 4.2: R1-OLSR Routing Decision Diagram for SRR Approach
On the other hand side, the ARR approach may be considered as an advancement of the SRR. According
to this approach, each node upon receiving a packet checks to see if paths for transmission towards the
destination exists. The node will decide to forward the packet or not, based on its knowledge of the global
topology. In order to reduce flooding, the number of paths to which packets are forwarded can be varied
within the ARR approach. As previously explained, for evaluation purposes in RESCUE, R2-OLSR has
limited the set of transmissions arising from disjoint nodes to two minimum (optimal) ETX values
corresponding towards the destination. In the case of having a number of paths to a destination with the
same ETX values, then a random selector is used to determine the path. Figure 4.3 presents the routing
decision diagram for the ARR approach.
Source Node
Identify all the
reachable
nodes in the
network
Identify Node A
Does Node A have a path to
the detination?
Do nothingForward the
packet to the
next node(s)
No Yes
Repeat process for next nodes
RESCUE D3.2 Version 2.0
Page 38 (79)
Figure 4.3: R2-OLSR Routing Decision Diagram for ARR Approach
Figure 4.4 presents an extension of the toy scenario for a better understanding of the SRR and ARR
approaches. For the SRR, let’s assume that the source node has to send a packet to the destination node.
The source node is aware of the topology information of the network through the topology sensing process.
Hence, it sends the packet to node 2, which has a path to the destination, and then node 2 sends the packet
to node 3. Node 3 has 5 available paths towards the destination, so it sends the packet through 5 different
routes. This process is repeated as long as the node has a path to the destination. At the destination,
RESCUE MAC layer functionality of joint decoding is applied on the packets received from different paths.
Figure 4.4: A multi-hop extension of the toy scenario
For the ARR approach, in the case that the source node has to send a packet to the destination node with
the source node being aware of the topology information of the network through the topology sensing
process, it sends the packet to node 2, which has a path to the destination, and then node 2 sends the packet
to node 3. Node 3 will as well broadcast the packet to nodes 4, 5, 6, 7 and 8. Each of these nodes upon
receiving the packets will decide whether to forward packets or not based on the two most optimal ETX
values. At the destination, RESCUE’s MAC layer joint decoding is thereafter applied and packets (either
erroneous or not) are combined as required.
RESCUE D3.2 Version 2.0
Page 39 (79)
4.1.3 Simulation results and Performance Analysis
4.1.3.1 Simulation Parameter
The proposed routing protocols following the SRR and ARR approaches are developed and simulated in
the NS-3 environment and compared with the OLSR-ETX. The project experimental setup is implemented
within a 800m by 800m area in which nodes are randomly installed with a transmission power of 20dBm.
The following table describes the overview of the simulation characteristics.
Table 4.1: Simulation Parameter for Multipath Routing Protocol
Parameter Value
Data Rate 6Mbit/s
Packet Size 1000 bytes
Topology Dimension 800m x 800m
Mobility Model Random Waypoint
Transmission Power 20dBm
Path Loss Model LogDistance
Fading Model Nakagami
Path Loss Exponent 3
Modulation Scheme BPSK
Simulation time 100sec
4.1.3.2 Evaluation Criteria
The objective of the simulations with the network simulator NS-3 is to validate our proposed routing
algorithms based on the two stated approaches by analysing the following performance metrics:
Throughput: the amount of traffic received in Mbps
Packet Loss: describes the percentage of packets that failed to arrive to the destination. Average End-to-End Delay: represents the average time required for the data packets from the
source to arrive at the destination. This includes the transmission and propagation delays and it
reflects the effectiveness of the routes. Routing Overhead: The number of routing messages generated per simulation, used for the
purposes of route discovery and maintenance. Convergence Time: the time it takes for nodes to fully populate their routing tables until stability
is reached for the first time.
4.1.3.3 Simulation Results
There are two implemented scenarios with the view to validate the performance of the proposed protocols.
In the first scenario we evaluate the effect of node mobility on the performance of the two proposed routing
protocols. The total number of nodes within the investigating topology are 15. Among them, there are 2
pairs of nodes which act as sources and destinations, while the rest of nodes are used as intermediate nodes.
Figure 4.5: Throughput versus node speed
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
4 8 12 16 20 24 28 32
Th
rou
gh
pu
t (
Mb
ps)
Speed of nodes (m/sec)
R1 OLSR
R2 OLSR
OLSR ETX
RESCUE D3.2 Version 2.0
Page 40 (79)
Figure 4.5 illustrates the throughput as the node speed increases. In each run for this scenario, the node
speeds have been set to be constant throughout the simulation period. The graph shows that all protocols
have a similar behaviour while operating between 4m/s and 8m/s. Beyond that point, throughput gradually
decreases due to the complexity of the topology. The graph shows that the R1-OLSR through its ability to
identify and transmit data via unlimited multiple paths, achieves the largest amount of data transfer
compared to the rest of the protocols. R2-OLSR also achieves better throughput throughout the range of
compared speed than OLSR-ETX. It can be deduced that both routing protocols utilizing multiple paths
provide higher throughput values when compared to OLSR-ETX with a single path between Source and
Destination.
Figure 4.6: End-to-end delay versus node speed
Figure 4.6 presents the end-to-end delay as the speed of nodes vary from 4m/s to 32m/s. OLSR ETX has
the highest end-to-end delay, followed by the two proposed RESCUE routing protocol. In particular, the
R2-OLSR experiences the least end-to-end delay while operating between 16m/s to 24m/s, whereas the R1-
OLSR gains leverage compared to R2-OLSR while operating at 12m/s, 28m/s and 32m/s. The superior
performance of the RESCUE protocols can be attributed to the fact that they efficiently distribute packets
to different paths restricting the effects of queue delay therefore they could be more suited to satisfy QoS
requirements.
Figure 4.7: Packet loss versus node speed
Similar to throughput and end-to-end delay, the RESCUE routing protocols achieves lesser packet loss as
shown in figure 4.7. It is evident that the packet loss of the two multipath routing protocol is reduced
because of the availability and usage of multiple paths from source to destination. The R1-OLSR and R2-
OLSR work in similar manner by forwarding packets to next hop nodes with paths towards destinations. In
0
10
20
30
40
50
60
70
4 8 12 16 20 24 28 32
En
d-t
o-e
nd
del
ay
(m
sec)
Speed of Nodes (m/sec)
R1 OLSR
R2 OLSR
OLSR ETX
0
2
4
6
8
10
12
14
4 8 12 16 20 24 28 32
Pa
cket
lo
ss (
%)
Speed of Nodes (m/sec)
R1 OLSR
R2 OLSR
OLSR ETX
RESCUE D3.2 Version 2.0
Page 41 (79)
the case of OLSR ETX, packets are lost once the singular path towards the destination fails. The results
also shows that the packet loss increases as the mobility of the nodes increase.
Figure 4.8: Routing overhead versus node speed
Figure 4.8 shows the number of routing messages generated for the purposes of route discovery and
maintenance for each routing protocol while increasing the speed of nodes. R1-OLSR generates the least
routing messages compared to R2-OLSR and OLSR ETX. As nodes constantly move in different directions
within the investigating area, the routing overhead is seen to gradually increase as the speed of the nodes
increase.
In the second scenario, we evaluate the effects of varying the network size on the performances of the
routing protocols. In this case the node speed remains constant (6m/s) while the network size increases
from 4 nodes to 40 nodes. The rest of the simulation parameters follow the attributes described in Table
4.1.
Figure 4.9: Throughput versus number of nodes
Figure 4.9 shows the throughput as the network size varies. R1-OLSR and R2-OLSR achieve similar
performance in terms of throughput, thereby presenting stability as the network size increases. The
multipath attribute of the R-OLSR approaches retain a more resilient behavior during operation by
maintaining multiple paths within the routing table compare to the OLSR ETX.
13050
13100
13150
13200
13250
13300
13350
13400
13450
13500
4 8 12 16 20 24 28 32
Ro
uti
ng
Ov
erh
ead
Speed of Nodes (m/sec)
R1 OLSR
R2 OLSR
OLSR ETX
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
4 8 12 16 20 24 28 32 36 40
Th
rou
gh
pu
t (
Mb
ps)
Number of nodes
R1 OLSR
R2 OLSR
OLSR ETX
RESCUE D3.2 Version 2.0
Page 42 (79)
Figure 4.10: End-to-end delay versus number of nodes
Figure 4.10 shows the end-to-end delay as the number of nodes increase in the network, Similar to the
previous scenario, the R-OLSR approaches provide significant lesser end-to-end delay compared to the
OLSR ETX. In particular, R2-OLSR which utilises the two most optimal paths for transmission generally
achieves the least end-to-end delay compared to the simple relay routing approach of R1-OLSR and OLSR
ETX. Both R-OLSR approaches indicate a higher robustness and resilience while operating in scalable
networks compared to the single path approach. This figure essentially indicates that the ETX metric
combined with the multipath approach reduces the propagation delay by allowing transmission via higher
quality paths thereby reducing the queue delay.
Figure 4.11: Packet loss versus number of nodes
In figure 4.11, the packet loss rate as the network size increases is shown. R-OLSR multipath approaches
are able to achieve lower packet loss compared to the single path approach of OLSR ETX which has only
one single path which can fail. Broken paths have usually being identified as a major cause of packet loss
in single path transmission and typical necessitate the need for a route recovery. As shown in figure 4.11,
both R1-OLSR and R2-OLSR are able to achieve significantly low packet loss in small networks (4-12
nodes) by using multiple paths, whereas in larger networks, the packet loss are still significantly lower than
that achievable by a single path routing protocol.
0
10
20
30
40
50
60
70
80
4 8 12 16 20 24 28 32 36 40
En
d-t
o-e
nd
del
ay
(m
sec)
Number of nodes
R1 OLSR
R2 OLSR
OLSR ETX
0
2
4
6
8
10
12
14
4 8 12 16 20 24 28 32 36 40
Pac
ket
loss
(%
)
Number of nodes
R1 OLSR
R2 OLSR
OLSR ETX
RESCUE D3.2 Version 2.0
Page 43 (79)
Figure 4.12: Routing Overhead versus number of nodes
Figure 4.12 shows the number of routing messages generated for the purposes of route allocation and
maintenance for each routing protocol while increasing the number of nodes. The figure introduces a trend,
where the routing overhead gradually increases as the number of nodes increase. The number of the
produced routing messages for all approaches is almost identical, with a small leverage for R1-OLSR which
manages to generate less redundant information.
4.1.4 Convergence Time
Despite the fact that topology loops provide alternative routes to destinations or concurrent alternative
routes in the case of multipath routing protocols, the Counting-to-Infinity (CTI) problem which occurs in
topology loops has to be mitigated. Topology loops are the main reason why routing loops occur and routing
protocols should be able to deal with topology and routing loops in an efficient manner in order to provide
improved convergence time and stability for the network. In the design of R1-OLSR and R2-OLSR, routing
loops are minimised by making sure that packets are not sent back to its originating node. In R-OLSR (R1-
OLSR and R2-OLSR), we do not consider recovery time after a link failure. The concept and design of R1-
OLSR and R2-OLSR are based on broadcast of data packets. Convergence time is dependent on network
topology diameter and the number of nodes. In order to compare the convergence time of routing protocols,
according to the approach in [10], coldstart tests are used to measure the convergence time from the point
when all routers are started until the network reaches a convergence state for the first time. In this case, no
CTI is incited in the network. This specific test scenario can provide the propagation speed for the routing
updates. R1-OLSR and R2-OLSR have been developed based on standard OLSR, the control messages and
message structure used by all three protocols are similar and the frequent neighbour discovery help to speed
up the coldstart convergence time. The approximated convergence time from coldstart is calculated as:
𝑇𝑎𝑝𝑝𝑟𝑜𝑥 = 2.5𝑑 + 0.16𝑛 − 5 ± 1.89 (4.2)
Where Tapprox is the convergence time, d is the diameter of network topology and n is the number of nodes
in the network topology.
Figure 4.13 shows the convergence time for the three routing protocols. It shows that OLSR ETX, R1-
OLSR and R2-OLSR all have the same coldstart convergence time for the topology considered as the
number of nodes varies. This can be associated with the fact that OLSR ETX, R1-OLSR and R2-OLSR are
built on OLSR and the control message (Hello and TC messages) all function the same with the same
frequency. Figure 4.13 also show that the convergence time increases gradually as the number of nodes in
the network increases.
0
10000
20000
30000
40000
50000
60000
70000
80000
90000
4 8 12 16 20 24 28 32 36 40
Ro
uti
ng
Ove
rhea
d
Number of nodes
R1 OLSR
R2 OLSR
OLSR ETX
RESCUE D3.2 Version 2.0
Page 44 (79)
Figure 4.13: Convergence time
4.1.5 Conclusion
This section discusses the results regarding the performance of RESCUE multipath routing protocol which
utilizes the ETX as link metric instead of the tradition hop count, based on two different relay approaches
(R1-OLSR and R2-OLSR). The designed protocols were compared with the single path OLSR ETX. The
simulation affirmed that the use of multipath for transmission of data packet improves the robustness and
increases the resilient of performance against lossy links. Simulations were carried out on the basis of two
main scenarios within a MANET. The first simulation scenario used a fixed number of nodes to assess the
R-OLSR behaviour under a series of experiments with increasing node mobility. The second scenario
investigated the routing protocol scalability by measuring its efficiency in a series of experiments with
increasing set of nodes, constantly moving within a fixed area. For the purposes of the experimentation, the
NS-3 simulator was used for the design, implementation and evaluation of the protocol's performance.
As described in section 4.1.3.2, the acquired results in both scenarios proved that R1-OLSR and R2-OLSR
multipath routing approaches combined with an intelligent link metric such as the ETX, offered a significant
reduction of end-to-end delay. In addition, more packets were able to successfully reach the destination
compared to the single path approach of OLSR ETX, increasing throughput and reducing the packet losses.
R-OLSR offered improved performance in scenarios with high mobility, eliminating the effects of link
instabilities while operating in unreliable and lossy environments. Furthermore, the performance of R-
OLSR under heavy network load scenarios indicated that the multipath approach combined with an
intelligent link metric can efficiently address scalability issues that occur during transmission, increasing
the protocol's robustness.
However, a set of issues need to be taken into consideration. The use of ETX within the routing messages
increases the protocol's complexity which can result to extra bandwidth utilization by increasing the size of
control messages. In addition, both relay routing mechanisms introduced for the purposes of R-OLSR
consequently transmit multiple copies of the same packet towards the destination. This specific attribute
more inherent in the simple relay approach (R1-OLSR) lead to heavy transmission of redundant
information. In this light, the advanced relay routing approach aims to alleviate this phenomenon by
reducing the number of transmission paths originating after nodes with disjoint paths by the continued
forwarding of packets from nodes with the two most optimal ETX values. In any case, both SRR and ARR
generate duplicate packets that negatively affect the bandwidth utilization and energy consumption of the
network.
4.2 Geo-routing
This section describes the concept and advantage of geographic routing (geo-routing) in unpredictable and
harsh radio environments, in particular systems such as vehicle-to-vehicle networks. Geo-routing is a
network-layer protocol for mobile ad-hoc networks, which are characterized by a decentralized
organization of the network without the need of an infrastructure to manage the communication resources
and the reachability among the nodes to distribute packets in the network.
1995.641996.28
1996.921997.56
1998.21998.84
1999.482000.12
2000.762001.4
1995.641996.28
1996.921997.56
1998.21998.84
1999.482000.12
2000.762001.4
1995.641996.28
1996.921997.56
1998.21998.84
1999.482000.12
2000.762001.4
4 8 1 2 1 6 2 0 2 4 2 8 3 2 3 6 4 0
CO
NV
ERG
ENC
E TI
ME
IN M
SEC
NUMBER OF NODES
R2-OLSR R1-OLSR OLSR ETX
RESCUE D3.2 Version 2.0
Page 45 (79)
Geo-routing supports unicast (point-to-point) and broadcast (point-to-multipoint) communication. For
unicast geo-routing, the source sends a packet to a geographic position, which may need to be acquired
from a location service (such as based on a location request/reply scheme or a database that maintains
position information). For broadcast geo-routing, which is the more relevant scenario for vehicular safety
applications as well as disaster recovery scenarios, geo-routing can send a packet to a geographical region
to reach all nodes located inside the area. Due to the knowledge of the destination node or area position and
the fact that every node also knows its own position, the network topology does not to be known a priori,
which represents an important difference to existing routing protocols. There is no complete route needed
before the packet is sent, hence the network is robust against failing nodes while the packet is forwarded
through the network.
In CBF the decision whether a packet is forwarded or not is made by the relay, using its own geographic
position, the position of the sender, and the position of the destination. The node with the most progress to
the destination forwards it. If this node fails the second one takes on this part and so on, therefore it has an
intrinsic reliability if the network topology changes and some nodes fail to forward a packet.
4.2.1 Services
For the RESCUE extensions of geo-routing, the RESCUE physical layer offers a function that provides a
confidence value of the packet quality to the network layer. This value is used in the calculation for the
timeout of the forwarding algorithm.
4.2.2 Protocol overview & design update
4.2.2.1 Normal CBF
CBF is used to forward a packet to a dedicated destination or to a geographic area. It is a receiver-based
forwarding algorithm where the source broadcasts the packet to all neighbours and they decide to forward
the packet or not. Those receivers with a positive geographical progress of the packet towards the
destination buffer the packet in a buffer and start a timer with a timeout value that depends on the distance
to the destination (see (4.3) and Table 4.2). The position of the destination is obtained from a location
service. A node with a shorter distance has a shorter timer. When the timer expires the packet is re-
broadcasted to all neighbours. If a node receives the same packet a second time before the timer expires, it
removes this packet from the buffer and stops the timer.
𝑡𝑡𝑖𝑚𝑒𝑟 = 𝑡𝑚𝑎𝑥 − 𝑡𝑚𝑎𝑥−𝑡𝑚𝑖𝑛
𝑑𝑚𝑎𝑥𝑝𝑟𝑜𝑔 for 𝑝𝑟𝑜𝑔 ≤ 𝑑𝑚𝑎𝑥 (4.3)
𝑡𝑡𝑖𝑚𝑒𝑟 = 𝑡𝑚𝑖𝑛 for 𝑝𝑟𝑜𝑔 > 𝑑𝑚𝑎𝑥
𝑝𝑟𝑜𝑔 = 𝑑𝑖𝑠𝑡(D, Se ) − 𝑑𝑖𝑠𝑡(D, 𝑅𝑖)
Table 4.1 Parameter description of the CBF timer calculation
Unit Description
ttimer Timeout of the timer for buffering the packet
tmax Maximal timeout
tmin Minimum timeout
dmax Maximum theoretical communication range
S, Se, D, Ri Geographic position of the source, sender, destination and relays
prog Forwarding progress of the relay towards the destination
4.2.2.2 Extensions for RESCUE
In the standardized CBF algorithm [1] a packet is dropped if it is received a second time, hence there is
only one route to the destination. In RESCUE, multipath distribution of the packets is required to merge
erroneous packets, which arrive from different disjoint paths in the destination, to an error-free packet,
using distributed joint source and channel coding as well as specific signal processing algorithms. For this
reason, CBF is extended by retransmission counters and a mechanism that enables the transmission of
erroneous packets. Every time when a relay receives a copy of the packet in the contention period the
retransmission counter is incremented and the packet is also stored in the buffer and not dropped. When the
CBF timer expires, the packet is forwarded unless the retransmission counter exceeds a threshold. With the
threshold it is possible to control the load in the network and select the number of routes that should be
used. After the packet is sent every further packet will be treated as duplicate and dropped.
RESCUE D3.2 Version 2.0
Page 46 (79)
In general, packets transmitted over lossy links can be erroneous. In order to take into account the quality
of a received packet, the PHY layer provides a confidence value of every received packet, which is
incorporated into the calculation of the timeout, meaning that packets with a lower confidence value have
to wait longer. If the quality of a packet exceeds a threshold (in our case if there are more than 20% errors
inside) it is dropped and not retransmitted anymore.
Another requirement for links-on-the-fly is to have disjoint links in order to guarantee uncorrelated errors
inside the packets. However, in vehicle-to-vehicle scenarios the vehicles move close together (roads are
narrow compared to the communication range) and if one relay retransmits the packet it is received by
every neighbour. Then the neighbours might have a set of packets which are originated from the same
sender. To enable disjoint links and avoid forwarding of correlated errors the relay needs to know the last
two relays and forwards only packets with different last two senders. One exception are the packets from
the source which will be forwarded every time.
Table 4.3 Functions of the CBF algorithm
Function Description
Buffer Stores a packet until it will be retransmitted
Timer Calculated depending on the distance to the destination and the
confidence value from the physical layer. A small distance and a high
confidence value means a short timer.
Location Service (LS) Provides the geographic position of the destination.
Retransmission Counter (RC) Counts how often the same packet is received.
Retransmission Threshold
(RT)
If the retransmission counter exceeds the retransmission threshold,
the packet is removed from the buffer and the timer is stopped.
4.2.2.3 Protocol operations for CBF
Figure 4.13 illustrates a toy scenario (TS2) to explain the functionality of the CBF algorithm in detail.
S D
B
A
Figure 4.13: The toy scenario (TS2) relevant to geo-routing
This network consists of a source S, a destination D and two relays A, B. Node S wants to send a packet to
D but there is no LOS. S broadcasts the packet to the relays A and B, which starts a timer depending on the
distance to D. The links between S-A and S-B are lossy, meaning that the packets received in A and B
cannot be completely decoded error free but they are still forwarded to D. In this scenario, A is
geographically closer to D, therefore its timer is shorter. After expiration of the timer in A the packet is re-
broadcasted to every neighbour which are S, B and D in this scenario. Even though B receives the packet a
second time it waits until the timer expires and also re-broadcasts the packet to every neighbour, instead of
dropping it as in the standard CBF algorithm. Now the destination is able to apply joint decoding to merge
the different copies of the packet. In this scenario it is not necessary to know the last two senders since the
packets from the source will be forwarded every time.
4.2.2.4 Flow diagram for the CBF algorithm
Figure 4.14 illustrates the CBF algorithm including the extensions for RESCUE requirements.
RESCUE D3.2 Version 2.0
Page 47 (79)
Source Node
Neighbor
Broadcast packet to all
neighbors
Packet
Is P in bufferNo
Calculate progress
Remove P from buffer
Stop timer
Discard P
Progress > 0 yes
Add P to buffer
no
Discard P
Calculate timeout
Start timer
Neighbor
Timer expires
Take P from buffer
Send P to MAC
Input confidence
value
Yes
Is RT threshold reached
No
Increment retransmission
counterYes
Guarantee disjoint paths
Store copy in buffer
Figure 4.14: Flow diagram of the CBF algorithm.
4.2.3 Simulation results
This section describes the simulation results of the new and extended CBF (RESCUE CBF) compared to
the normal standardized CBF specified in ETSI EN 302 636 [1]. The evaluation scenario is described in
Section 2.4.1. The simulations were executed with 20 different seeds to achieve a statistical significance
and error bars are shown in the figures.
4.2.3.1 KPIs
To evaluate the performance and compare the algorithms the KPIs packet success rate (PSR), end-to-end
delay (E2ED) and data traffic overhead (DTO) were chosen. For safety and traffic efficiency applications
it is important that every vehicle receives the message reliably and in-time, which is measured by the PSR
and E2ED. The PSR is defined as the number of packets successfully received at the destination divided by
the source transmitted packets:
𝑃𝑆𝑅 = 𝑁𝑜.𝑜𝑓 𝑅𝑥 𝑝𝑎𝑐𝑘𝑒𝑡𝑠 𝑎𝑡 𝑑𝑒𝑠𝑡𝑖𝑛𝑎𝑡𝑖𝑜𝑛
𝑁𝑜.𝑜𝑓 𝑇𝑥 𝑝𝑎𝑐𝑘𝑒𝑡𝑠 𝑓𝑟𝑜𝑚 𝑠𝑜𝑢𝑟𝑐𝑒.
The E2ED is defined as the difference of the timestamp when the message is generated by the source and
the timestamp when it is received by the destination:
𝐸2𝐸𝐷 = 𝑡𝑅𝑥 − 𝑡𝑇𝑥.
In order to measure the network load the DTO is introduced. It enables the measurement of the
retransmissions in the network and is defined by the number of transmissions in the PHY divided by the
number of sent packets in the facility layer:
RESCUE D3.2 Version 2.0
Page 48 (79)
𝐷𝑇𝑂 = 𝑁𝑜.𝑜𝑓 𝑓𝑟𝑎𝑚𝑒𝑠 𝑠𝑒𝑛𝑡 𝑓𝑟𝑜𝑚 𝑃𝐻𝑌 𝑙𝑎𝑦𝑒𝑟
𝑁𝑜.𝑜𝑓 𝐷𝐸𝑁𝑀𝑠 𝑠𝑒𝑛𝑡 𝑓𝑟𝑜𝑚 𝑓𝑎𝑐𝑖𝑙𝑖𝑡𝑦 𝑙𝑎𝑦𝑒𝑟.
4.2.3.2 Results
Figure 4.15 presents the packet success rate of the normal CBF (LOTF-Off) and the RESCUE CBF (LOTF-
On) for the RESCUE case. The PSR reflects the reliability of the packet transport in the network. We
observe that the RESCUE CBF outperforms the normal CBF over the whole range. Up to 23 dBm LOTF-
On has a PSR of 100% which is very important for V2V application, e.g., hazardous location notification
where already one uninformed car can cause a big disaster. LOTF-Off is less than 100% up to 35 dBm.
With constant transmit power the largest improvement is at 16.4 dBm by 37.5 percentage points in packet
success rate. On the other hand if we claim a minimum PSR of 90% we have a gain of 6 dB in transmit
power.
Figure 4.15: Packet success rate of the large scale V2V-scenario
Figure 4.16 shows the end-to-end delay (E2ED) for both algorithms. LOTF-On has a shorter E2ED for all
transmission powers since in links-on-the-fly every packet is forwarded even there are errors inside or not.
In the normal CBF if a packet is not received correctly it is dropped and we have to wait until further timers
of other nodes expire and they rebroadcast the packet again which leads to a longer E2ED by a factor of
three. If the transmit power decreases under 10 dBm just a few packets receive the destination and therefore
the statistics becomes worse and the error bars increase. With further decreasing transmit power no packets
arrive at the destination which can be decoded correctly. The second reason for the shorter E2ED is that
also the link quality is taken into account. A good link (high confidence value) has a shorter timer which
also reduces the E2ED.
Figure 4.16: End-to-End-Delay of the large scale V2V-scenario
RESCUE D3.2 Version 2.0
Page 49 (79)
Figure 4.17 presents the data traffic overhead (DTO). For the RESCUE CBF the DTO is much higher due
to the permission of retransmissions after overhearing. Every packet is forwarded until a predefined
retransmission threshold is exceeded (Section 4.2.2.2). In the normal CBF the other nodes refrain from
retransmissions if a neighbour already forwarded the same packet. In very dense scenarios a high DTO
would lead to a congested data channel. In our case we consider a sparse distribution of nodes, which does
not affect the channel. Furthermore, in a scenario with a dense node distribution a sufficient number of
links for correct transmission of the packet are available – in this case LOTF would be unnecessary.
Figure 4.17: Data Traffic Overhead of the large scale V2V-scenario
Figure 4.18 presents the probability of reception over the transmit power for a different number of packets
used for joint decoding, which are required for error-free decoding: green – 1 packet, blue – 2 packets,
black – 3 packets, and the red line indicates that the packet could not be decoded correctly. The sum of all
four graphs represents the overall number of sent packets. We can see that for a high transmit power beyond
33 dBm over 80% of packets were decoded correctly with a single copy only (green line). With decreasing
Tx power also the capability to decode a packet with just one copy decreases and the blue graph where two
copies are necessary grows. At approximately 25 dBm also the black graph (three packets are needed for
decoding) becomes important and increases up to 20% at 17.5 dBm. It is worth noting that the green line
can be treated as baseline, and the blue and black graphs can be regarded as the gain achieved by the links-
on-the-fly concept.
Figure 4.18: Use of joint decoding in the large scale V2V scenario
RESCUE D3.2 Version 2.0
Page 50 (79)
4.2.3.3 Conclusion
We have shown that RESCUE CBF outperforms the normal CBF in reliability (PSR) and latency (E2ED)
aspects over all regions of Tx power. It improves the PSR up to 37.5 percentage points and the E2ED by a
factor of three. As expected, the DTO is worse for RESCUE CBF since erroneous forwarding is allowed
and necessary to apply the RESCUE functionality.
Other metrics, i.e., link stability and time of convergence when routes change, were not investigated since
they are not applicable to geographic routing for a V2V scenario. As it is described in Section 4.2, in
contrast to link-state-routing protocols, in Geo-routing the information about the network topology is not
distributed. This is because the effort is too high for getting all information of every link when the topology
changes frequently. In our scenario we have multiple vehicles moving on a highway. The links change for
every new packet and therefore the packets search a way through the network.
RESCUE D3.2 Version 2.0
Page 51 (79)
5. Error control
There are two independent ARQ techniques defined in the RESCUE network. The end-to-end (e2e) ARQ
(Section 5.1) is responsible for lossless communication between the source and destination in both single-
hop and multi-hop case. The partial ARQ (Section 5.2) works only on single-hop links and cannot guarantee
lossless communication in the RESCUE network. Both ARQ protocols are complimentary to provide
efficient and reliable transmission in distributed lossy link communication.
5.1 End-to-end ARQ
End-to-end (e2e) ARQ assures reliability and lossless communication between the source and destination
nodes. It uses a modified sliding window protocol to tell the source node to determine which DATA frames
need to be retransmitted. This e2e ARQ protocol operates in the Data Link Layer.
5.1.1 Protocol overview & design update
The e2e ARQ protocol uses ACK and NACK frames (messages sent by the destination node indicating that
it has correctly or incorrectly received a DATA frame) as well as timeouts (specified periods of time
allowed to elapse before a positive or negative acknowledgment is to be received) to achieve reliable data
transmission over multi-hop links. In comparison to the description of e2e ARQ presented in D3.1, the
Block NACK mechanism was added as a new feature. End-to-end ARQ exploits functionalities of the
Selective Repeat ARQ protocol but has been enhanced with a non-continuous transmit buffer mechanism
(the source node sends a number of DATA frames specified by a window size even without the need to
wait for individual ACK from the destination node). Moreover, the transmit window can be separated by
frames which were correctly acknowledged, so the distance between the first and last unacknowledged
frame in the window can be large. This mechanism can be especially useful for multi-hop networks with
high BER ratio on the transmission links, a large number of relays, and high propagation delays between
relay nodes. The operation of the end-to-end ARQ in simplest possible (basic) configuration is presented
in Figure 5.1.
The e2e ARQ protocol is equipped with additional mechanisms that should improve its performance
assuming the transmission distributed over lossy links. The first mechanism is Block ACK. Its operation is
as follows: the destination node after successfully joint decoding the first DATA frame can wait a
predefined time interval to receive more DATA frames that could be acknowledgement with a single Block
ACK frame. Moreover, if DATA frames are received at the destination node in the wrong order, the
destination node should wait a predefined time interval to receive missing DATA frames and after that the
Block ACK should be sent to the source node. The Block NACK mechanism is the functional opposite to
Block ACK. It informs the source node that a number of frames were received with errors at the destination
node. The second used mechanism is known as continuous acknowledgement. It is used to signalize the
relay nodes which are on the ACK frame path that the destination node received and correctly decoded
previous DATA frame(s). The continuous ACK number indicates the first frame that can be accounted to
the continuous frame chain of correctly decoded frames at the destination node. The relay node, after
receiving a continuous ACK frame (note that ACK frames can arrive through different relays), knows which
frames have been correctly received at the destination node and their copies can be removed from the relay’s
buffer.
A detailed description of the e2e ARQ protocol with examples of operation with different mechanisms
enabled and disabled is presented in D3.1.
The updated list of configurable parameters of the end-to-end ARQ protocol is presented below:
Window size – the general window size of the e2e ARQ (i.e., the maximum sequence number
used, e.g., eight in Figure 5.1).
Send window size – defines the upper bound on the number of unacknowledged frames that the
source node can transmit (e.g., four in Figure 5.1).
ACK timeout – the time interval the source node waits for the ACK frame before initiating
retransmission of the DATA frame.
NACK timeout – the time interval the destination node waits for other copies of the DATA frame
before sending the NACK frame.
Block ACK timeout – the time interval the destination node waits for other DATA frames before
sending the Block ACK frame.
RESCUE D3.2 Version 2.0
Page 52 (79)
Figure 5.1: End-to-end ARQ based on Selective Repeat with non-continuous transmit buffer
mechanism.
5.1.2 Simulation results
The simulation analysis of the e2e ARQ protocol was performed by implementing this protocol in a
modified version of the ns-3 simulator which included the MAC implementation of Section 3.1. The
parameters listed in Table 3.1 were used for the simulation with the following exceptions:
Propagation loss model: fixed
Traffic load: 15 Mb/s (saturation)
ACK timeout (t_ACK_base): 3500 s
Furthermore, the validation was performed for various configurations of e2e ARQ (Table 5.1) and for two
network topology scenarios (Figure 5.2). The different configuration options allow to validate how each
individual mechanism impacts the performance of the whole protocol. The two network topology scenarios
allow us to first assess the basic performance of the protocol (in the single-path case) and then assess its
performance in a more realistic, multi-path scenario (in the two-path case). The number of additional hops
(relays) N was an important parameter as this protocol is designed for (and more useful in) cases where the
transmission is done over multiple links (unlike the toy scenarios described in Section 2.3).
RESCUE D3.2 Version 2.0
Page 53 (79)
Table 5.1: Configuration scenarios for e2e ARQ where where N – number of additional hops, SW –
Send Window, t_ACK_base = MAX (SW×(PATHS+1), N+2), blockNACK( OFF) – the basic
sequence number in a blockACK frame refers to the last correctly received frame, blockNACK(ON)
– this sequence number refers to the last received frame (no necessarily correctly).
Scenario NACK continou
s ACK
block
ACK
block
NAC
K
ACK timeout
[s]
NACK
timeout
[s]
block ACK
timeout [s]
Basic OFF OFF OFF OFF t_ACK_base ×
2000
N/A N/A
NACK ON OFF OFF OFF 4 × t_ACK_base ×
2000
t_ACK_ba
se × 1600
N/A
ContACK OFF ON OFF OFF t_ACK_base ×
2000
N/A N/A
NACK+Cont
ACK
ON ON OFF OFF 4 × t_ACK_base ×
2000
t_ACK_ba
se × 1600
N/A
BlockACK OFF OFF ON OFF 4 × t_ACK_base ×
2000
N/A t_ACK_bas
e × 1600
BlockNACK OFF OFF ON ON 4 × t_ACK_base ×
2000
N/A t_ACK_bas
e × 1600
BlockACK+
ContACK
OFF ON ON OFF 4 × t_ACK_base ×
2000
N/A t_ACK_bas
e × 1600
BlockNACK+
ContACK
OFF ON ON ON 4 × t_ACK_base ×
2000
N/A t_ACK_bas
e × 1600
S R1 R2 RN D
Additional hops
Single-path end-to-end ARQ validation
Two-path end-to-end ARQ validation
S
RN-1
D
Additional hops
RN
Figure 5.2: Two network topology scenarios for end-to-end ARQ validation: single-path and two-
path.
To assess the performance improvement gained from using e2e ARQ we used the stop-and-wait ARQ
protocol as the baseline. This is the simplest ARQ protocol possible. Its operation in RESCUE scenarios is
described in Section 3.1.1 (i.e., it was used in the SimpleMAC protocol). The baseline results are presented
in each figure for comparison.
In general, the performance was measured with respect to three metrics (throughput, delay, and jitter), three
link quality values (SNR equal to -6.5, -4.5, or -1.5 dB), between 1 and 10 additional hops (relays) and
varying SW sizes (2, 4, 8, and 16). The link quality values describe the link quality of each individual link
between each transmitter and receiver in the network.
First, we present the results for the single path case in order to establish a basic understanding of the e2e
ARQ protocol operation in lossy-link scenarios. Figure 5.3 contains a complete set of results obtained for
the basic configuration of the e2e ARQ protocol. In this case, throughput improvement is visible for lower
RESCUE D3.2 Version 2.0
Page 54 (79)
link qualities. When the links are of better quality (-1.5 dB) the throughput improvement is only for longer
paths (with more than four relays). Similarly, using e2e ARQ decreases delay at the cost of elevated jitter
values. For longer paths, delay can be stabilized at fixed values with the use of e2e ARQ. Furthermore, we
observed that SW sizes above two do not give significant gains in terms of throughput or delay but do
increase jitter.
SNR = -6.5 dB SNR = -4.5 dB SNR = -1.5 dB
Figure 5.5.3: Performance results of the basic end-to-end ARQ configuration in the single-path
topology for three link quality values (SNR equal to -6.5, -4.5, or -1.5 dB) for three metrics:
throughput (first row), delay (second row), and jitter (third row).
For the other configuration cases (NACK, BlockACK, BlockNACK) we present selected results below
(i.e., only for SNR = -4.5 dB). The results for the configuration cases which included the Continuous ACK
mechanism were omitted as this mechanism did introduce any significant changes in the results for the
single-path scenario.
Enabling the NACK mechanism also increases throughput and decreases delay but its effects are positive
for paths with at least three relays (Figure 5.4). Now, the impact of the SW value is evident with the best
throughput and delay performance achieved for SW equal to at least eight. However, this is offset by
increased jitter values.
Figure 5.4: Performance results of the NACK end-to-end ARQ configuration in the single-path
topology for links with SNR equal to -4.5 dB for three metrics: throughput (left), delay (middle), and
jitter (right).
0
0.1
0.2
0.3
0.4
0.5
0.6
1 2 3 4 5 6 7 8 9 10
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 2 3 4 5 6 7 8 9 10
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
0.5
1
1.5
2
2.5
1 2 3 4 5 6 7 8 9 10
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
2000
4000
6000
8000
10000
12000
1 2 3 4 5 6 7 8 9 10
Del
ay [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
1000
2000
3000
4000
5000
6000
7000
1 2 3 4 5 6 7 8 9 10
Del
ay [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
500
1000
1500
2000
2500
1 3 5 7 9
Del
ay [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
50
100
150
200
250
300
350
400
1 2 3 4 5 6 7 8 9 10
Jitt
er [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
50
100
150
200
250
1 2 3 4 5 6 7 8 9 10
Jitt
er [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
5
10
15
20
25
30
35
40
45
50
1 2 3 4 5 6 7 8 9 10
Jitt
er [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1 3 5 7 9
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
1000
2000
3000
4000
5000
6000
7000
1 3 5 7 9
De
lay
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
50
100
150
200
250
300
350
400
1 3 5 7 9
Jitt
er [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
RESCUE D3.2 Version 2.0
Page 55 (79)
For the BlockACK configuration the impact of the SW size is even more evident (Figure 5.5). In this case,
not all SW configurations can improve over the baseline in terms of throughput and delay. Only a large SW
(equal to 16) improves throughput regardless of the path length.
Figure 5.5: Performance results of the BlockACK end-to-end ARQ configuration in the single-path
topology for links with SNR equal to -4.5 dB for three metrics: throughput (left), delay (middle), and
jitter (right).
As can be expected, the BlockNACK mechanism performs better (than BlockACK) in lossy-link scenarios
(Figure 5.6). Improvements over the baseline are evident for all analysed path lengths. The SW value
slightly impacts throughput and delay and SW = 16 causes a significant increase in jitter.
Figure 5.6: Performance results of the BlockNACK end-to-end ARQ configuration in the single-path
topology for links with SNR equal to -4.5 dB for three metrics: throughput (left), delay (middle), and
jitter (right).
To assess the performance of e2e ACK in the single-path topology in each configuration, we provide a
direct comparison with the stop-and-wait baseline in terms of relative gain in throughput, relative decrease
of delay, and absolute change of jitter. For fixed link quality and SW values (Figure 5.7), it is evident that
not all mechanisms provide the same performance in the single-path case. In terms of throughput and delay
the BlockACK mechanism stands out with the worst performance. With regard to these metrics, the basic
and BlockNACK configurations are the best with the latter much better in achieving a low jitter. For fixed
path length and SW values (Figure 5.8) it is clear that the RESCUE e2e ARQ protocol provides the best
improvement for the target low link quality scenarios (i.e., lossy links). This is in particular true for the
basic and BlockNACK configurations. Conversely, the BlockACK mechanism performs better for higher
quality links. Another observation is that the BlockNACK mechanism can significantly outperform NACK
in terms of jitter.
Figure 5.7: Comparison of end-to-end ARQ performance with baseline in the single-path topology
for three metrics (throughput, delay, and jitter) for links with SNR equal to -4.5 dB and for a constant
send window size (SW=8).
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1 3 5 7 9
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
1 3 5 7 9
Del
ay [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
20
40
60
80
100
120
140
160
180
1 3 5 7 9
Jitt
er [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 3 5 7 9
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
1000
2000
3000
4000
5000
6000
7000
1 3 5 7 9
De
lay
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
20
40
60
80
100
120
140
160
180
200
1 3 5 7 9
Jitt
er [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
-40%
-20%
0%
20%
40%
60%
80%
100%
120%
140%
160%
1 3 5 7 9
Thro
ugh
pu
t ch
ange
Additional hops
Basic NACK BlockACK BlockNACK
-70%-60%-50%-40%-30%-20%-10%
0%10%20%30%40%
1 3 5 7 9
Del
ay c
han
ge
Additional hops
Basic NACK BlockACK BlockNACK
20
40
60
80
100
120
140
160
1 3 5 7 9
Jitt
er c
han
ge [
ms]
Additional hops
Basic NACK BlockACK BlockNACK
RESCUE D3.2 Version 2.0
Page 56 (79)
Figure 5.8: Comparison of end-to-end ARQ performance with baseline in the single-path topology
for three metrics (throughput, delay, and jitter) for a constant send window size (SW=8) and constant
number of additional hops (N=8) and varying SNR values.
Next, we continue presenting the results with the two-path topology (Figure 5.2) which constitutes a more
realistic scenario for the operation of the e2e ARQ protocol in lossy-link scenarios. Figure 5.9 contains the
results obtained for the basic configuration of the e2e ARQ protocol in this topology. Surprisingly, the
throughput improvement is visible for all link qualities. When the links are of better quality (-1.5 dB) the
throughput improvement is higher. This is especially visible for longer paths (with more than three relays).
Indeed, the proposed protocol outperforms the baseline for all SW values and all path lengths. Again,
increasing SW improves throughput and delay at the cost of elevated jitter. For longer paths, delay can be
stabilized at almost fixed values with the use of e2e ARQ and large SW sizes (SW=8 and SW=16). In
contrast to the single path case, we observed that large SW sizes have a significant gain in terms of
throughput for all link qualities. The better the link quality, the higher is gain, especially for longer paths.
The same conclusions are for the delay but not for the jitter which increases with the growth of SW size.
SNR = -6.5 dB SNR = -4.5 dB SNR = -1.5 dB
Figure 5.9: Performance results of the basic end-to-end ARQ configuration in the two-path topology
for three link quality values (SNR equal to -6.5, -4.5, or -1.5 dB) for three metrics: throughput (first
row), delay (second row), and jitter (third row) under saturation.
The results for the configuration cases which included the Continuous ACK mechanism were again omitted
as this mechanism did not introduce any improvement in the network performance for the two-path
scenario. Enabling the NACK mechanism provides significantly worse results (Figure 5.10). The stop-and-
0%
20%
40%
60%
80%
100%
120%
140%
160%
-6.5 -4.5 -1.5
Thro
ugh
pu
t ch
ange
SNR [dB]
Basic NACK BlockACK BlockNACK
-60%
-50%
-40%
-30%
-20%
-10%
0%
-6.5 -4.5 -1.5
Del
ay c
han
ge
SNR [dB]
Basic NACK BlockACK BlockNACK
0
50
100
150
200
250
-6.5 -4.5 -1.5
Jitt
er c
han
ge [
ms]
SNR [dB]
Basic NACK BlockACK BlockNACK
0
0,1
0,2
0,3
0,4
0,5
0,6
1 2 3 4 5 6 7 8 9 10
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1 2 3 4 5 6 7 8 9 10
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
0,2
0,4
0,6
0,8
1
1,2
1,4
1,6
1,8
1 2 3 4 5 6 7 8 9 10
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
500
1000
1500
2000
2500
3000
3500
4000
4500
1 2 3 4 5 6 7 8 9 10
De
lay
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
500
1000
1500
2000
2500
3000
3500
1 2 3 4 5 6 7 8 9 10
Del
ay [
ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
500
1000
1500
2000
2500
3000
1 2 3 4 5 6 7 8 9 10
De
lay
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
20
40
60
80
100
120
140
160
1 2 3 4 5 6 7 8 9 10
Jitt
er
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
10
20
30
40
50
60
70
80
90
1 2 3 4 5 6 7 8 9 10
Jitt
er
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
2
4
6
8
10
12
14
16
18
20
1 2 3 4 5 6 7 8 9 10
Jitt
er
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
RESCUE D3.2 Version 2.0
Page 57 (79)
wait baseline outperforms the e2e ARQ protocol in this case for all three metrics: throughput, delay, and
jitter.
Figure 5.10: Performance results of the NACK end-to-end ARQ configuration in the two-path
topology for links with SNR equal to -4.5 dB for three metrics: throughput (left), delay (middle), and
jitter (right) under saturation.
The BlockACK mechanisms provides only limited improvements in comparison with the baseline (Figure
5.11). In terms of throughput, a large SW size is required and the path length contain at least 2-3 additional
hops for this mechanism to prove useful. However, using BlockACK in the two-path topology increases
both delay and jitter.
Figure 5.11: Performance results of the BlockACK end-to-end ARQ configuration in the two-path
topology for links with SNR equal to -4.5 dB for three metrics: throughput (left), delay (middle), and
jitter (right) under saturation.
The BlockNACK mechanism provides further improvements (Figure 5.12). In almost all cases the
throughput is improved in comparison with the baseline with the SW=2 configuration closely following the
stop-and-wait results. Similarly, the delay is also improved. However, as in previous cases, jitter is
increased by the selective repeat mechanism used in the e2e ARQ protocol.
Figure 5.12: Performance results of the BlockNACK end-to-end ARQ configuration in the two-path
topology for links with SNR equal to -4.5 dB for three metrics: throughput (left), delay (middle), and
jitter (right) under saturation.
Figure 5.13 provides a comparison of the performance of the various mechanisms in the two-path topology.
It reveals the following conclusions:
as in previously studies topologies, the continuous ACK mechanism does not bring any
improvements,
the NACK mechanism performs much worse than other mechanisms,
the highest throughput and lowest delays are achieved by the e2e ARQ protocol in the basic
configuration,
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
1 2 3 4 5 6 7 8 9 10
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
1000
2000
3000
4000
5000
6000
7000
8000
1 2 3 4 5 6 7 8 9 10
De
lay
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
100
200
300
400
500
600
700
1 2 3 4 5 6 7 8 9 10
Jitt
er
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
1 2 3 4 5 6 7 8 9 10
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
1000
2000
3000
4000
5000
6000
7000
8000
1 2 3 4 5 6 7 8 9 10
De
lay
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
100
200
300
400
500
600
700
1 2 3 4 5 6 7 8 9 10Ji
tte
r [m
s]Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
1 2 3 4 5 6 7 8 9 10
Thro
ugh
pu
t [M
b/s
]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
500
1000
1500
2000
2500
3000
3500
1 2 3 4 5 6 7 8 9 10
De
lay
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
0
10
20
30
40
50
60
1 2 3 4 5 6 7 8 9 10
Jitt
er
[ms]
Additional hops
baseline SW=2 SW=4 SW=8 SW=16
RESCUE D3.2 Version 2.0
Page 58 (79)
enabling blockNACK can slightly reduce jitter.
Figure 5.13: Comparison of end-to-end ARQ performance mechanisms for three metrics
(throughput, delay, and jitter) in the two-path topology for links with SNR equal to -4.5 dB and for
a constant send window size (SW=16), under saturation.
Furthermore, if the obtained results are compared with the baseline for various SNR values (Figure 5.14),
it becomes evident that the additional complexity of the e2e ARQ protocol is most beneficial in networks
with links of higher quality (SNR = -1.5 dB). If the link quality is even lower, the best performance is
achieved with the basic or the blockNACK configurations.
Figure 5.14: Comparison of end-to-end ARQ performance with baseline in the two-path topology for
three metrics (throughput, delay, and jitter) for a constant send window size (SW=16) and constant
number of additional hops (N=8) and varying SNR values, under saturation.
Surprisingly for all the considered scenarios, the continuous ACK mechanism (ContACK) in all possible
combinations (NACK+ContACK, BlockACK+ContACK, and BlockNACK+ContACK) did not improve
the overall network performance. The authors think that this mechanism can behave much better for larger
networks composed of tens of nodes and when there are larger propagation delays between relays. The
analysis of such scenarios is left for future studies.
In summary, the simulation analysis of end-to-end ARQ has provided the following conclusions:
using a complex end-to-end ARQ protocol can significantly improve RESCUE network
performance in terms of increasing throughput and decreasing delay, especially for longer
transmission paths;
the best consistent performance improvements were provided by e2e ARQ in its basic and
blockNACK configurations and these mechanisms should be the focus of future study;
scenarios in which the continuousACK mechanism can provide impact have yet to be determined;
the other proposed configurations have not proven to be as useful and can therefore be omitted
from future study.
5.2 Partial ARQ
Lossy forwarding technique is effective in reducing the end-to-end latency and increasing the throughput
of multi-hop multi-relay systems, where erroneously decoded frames at the relay node can be forwarded by
re-interleaving and re-encoding them in such a way that the destination node can exploit the correlation of
the information parts of the frames received via all links. In fact, the correlation exists because the frames
are generated from the same source. Multiple copies of correlated frames are received not only from parallel
paths but also from retransmitted frames following an ARQ protocol. Every time a frame is retransmitted,
the receiving node will increase the amount of information. Hence, by accumulating sufficient information,
the node will be able to decode the message. In this case, ARQ technique, based on frame combining with
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1 2 3 4 5 6 7 8 9 10
Thro
ugh
pu
t [M
b/s
]
Additional hops
basic/contACK NACK/NACK+contACK blockACK
contACK+blockACK blockNACK contACK+blockNACK
0
1000
2000
3000
4000
5000
6000
1 2 3 4 5 6 7 8 9 10
Del
ay [
ms]
Additional hops
basic/contACK NACK/NACK+contACK/blockACK
contACK+blockACK blockNACK/contACK+blockNACK
0
100
200
300
400
500
600
700
1 2 3 4 5 6 7 8 9 10
Jitt
er [
ms]
Additional hops
basic/contACK NACK/NACK+contACK/blockACK
contACK+blockACK blockNACK/contACK+blockNACK
-100%
-50%
0%
50%
100%
150%
200%
250%
-6.5 -4.5 -1.5
Thro
ugh
pu
t ch
ange
SNR [dB]
Basic NACK contACK NACK+contACK
BlockACK contACK+blockACK BlockNACK contACK+blockNACK
-100%
-50%
0%
50%
100%
150%
-6.5 -4.5 -1.5
De
lay
chan
ge
SNR [dB]
Basic NACK contACK NACK+contACK
BlockACK contACK+blockACK BlockNACK contACK+blockNACK
-
100
200
300
400
500
600
700
800
900
-6.5 -4.5 -1.5
Jitt
er
chan
ge [
ms]
SNR [dB]
Basic NACK contACK NACK+contACK
BlockACK contACK+blockACK BlockNACK contACK+blockNACK
RESCUE D3.2 Version 2.0
Page 59 (79)
FEC in a Hybrid-ARQ (HARQ) scheme, can achieve not only coding gain but also the time diversity
through retransmissions, as well as the spatial diversity through the relays which retransmit the source
information.
Forwarding frames having large distortions with e2e protocol may invoke continuously retransmission
request to the source node, resulting in reduced end-to-end throughput. Moreover, the capability of
correcting errors in a parallel network systems at the destination node is made possible regardless the
qualities of each path, so far as there is at least one connection where errors are not introduced in the relay
before re-encoding. Therefore, P-HARQ introduces a confidence indicator (𝐶𝐼) as a threshold by which a
relay node decides either forwarding the erroneous frame or requesting retransmission. In contrast with
CRC, 𝐶𝐼 has capability of identifying the reliability of the entire frame. Furthermore, the 𝐶𝐼 calculation is
beneficial since the receiving nodes do not need to know the original information sequence. It is given by
𝐶𝐼 = 1 −1
𝑁∑ 𝐻𝑏(
1
1+𝑒−|𝐿𝑝|)𝑁𝑛=1 ,
where 𝐻𝑏 and 𝐿𝑝 denote binary entropy function and the a posteriori log-likelihood ratio (LLR) of channel
decoder, respectively. The probability of error corresponding to the 𝐶𝐼 value can be calculated by
𝑃𝑏 ≈1
2𝑒𝑟𝑓𝑐(
𝐽−1(𝐶𝐼)
2√2 ),
where 𝐽−1(∙) is the inverse of function 𝐽(∙). For BPSK, 𝐽−1(𝐶𝐼) can be approximated by
𝐽−1(𝐶𝐼) = −1
𝐻1𝑙𝑜𝑔2(1 − 𝐶𝐼
1
𝐻3)1
2𝐻2 , 𝐻1 = 0.3073, 𝐻2 = 0.8935, 𝐻3 = 1.1064.
5.2.1 Protocol overview and design update
P-HARQ with the stop-and-wait protocol works for a two-hop multi-relay network, which corresponds to
TS2. The simulation results are shown in Section 5.2.4. The proposed algorithm is given as follows. At the
very beginning of each HARQ rounds for multiple information sequence to be transmitted, the frame from
the source node is forwarded to the destination node through the relay(s) even though it still contains errors.
It is due to the 𝐶𝐼 threshold not set yet. Each time receiving frame, the receiving nodes calculate its 𝐶𝐼 by
equation given in the previous sub-section.
The receiving nodes send NACK to their previous node to indicate unsuccessful decoding, and hence
requesting retransmission. There are two types of NACK in P-HARQ: NACK_1 indicating a retransmission
required from the node in one-hop back, and NACK_2 indicating retransmission required from the node in
two-hop back. Therefore, if a transmitting node receives NACK_1, it will retransmit the frame to the next
node. On the other hand, if a transmitting node receives NACK_2, it will transmit NACK_1 to the previous
node.
The destination node evaluate 𝐶𝐼 values of frames transmitted from all paths, before frame combining. The
destination node transmits NACK_1 whenever the frames transmitted for the first time (not retransmitted
version) by the relay are unsuccessfully recovered. This is to avoid the excessive end-to-end latency. In this
case, the 𝐶𝐼 is used as the threshold. Additionally, the destination node transmit NACK_2 whenever the
have-retransmitted frames are unsuccessfully recovered. In this case, the destination node uses the 𝐶𝐼 value,
which are larger than the previous 𝐶𝐼 as the threshold. As for the relay node, the threshold is set equal to
𝐶𝐼 of the very beginning of the HARQ rounds and update it whenever receiving NACK_2. The more details
of the algorithm is given in the table below.
RESCUE D3.2 Version 2.0
Page 60 (79)
5.2.2 Simulation Setup
We consider a TS2 topology where a source node 𝑆𝑁 aims to transmit information sequence to a destination
node 𝐷𝑁 through two relay nodes 𝑅𝑁1 and 𝑅𝑁2 that are located physically separate in parallel links. There
are three time slots in one transmission cycle. In the first time slot, the node 𝑆𝑁 broadcasts its coded
sequences 𝒙𝑆 to the node 𝑅𝑁1 and 𝑅𝑁2. The rest time slots, both relays transmit their coded
sequence 𝑥𝑅𝑁𝑙, 𝑙 ∈ {1, 2} to the destination 𝐷𝑁, sequentially.
We consider a static channel within one block but varying link-by-link as well as transmission-by-
transmission during HARQ rounds. We use the terminologies transmitting nodes and receiving nodes for
RESCUE D3.2 Version 2.0
Page 61 (79)
referring to the source node and the relay nodes on transmit phase, and relay nodes and destination node on
receive phase, respectively.
5.2.2.1 Transmit Phase
Figure 5.15: Block diagram of the source and the relay node. Both relays in TS2 have the same
structure.
Figure 5.15 depicts the transmitter structure of the source node and the relay node 𝑅𝑁1. The structure of
𝑅𝑁2 is similar to 𝑅𝑁1. With 𝑚 (re)transmissions, 𝑚 ∈ {1, 2, . . . , 𝑀}, where the maximum number of
retransmissions is 𝑀 − 1, the binary information sequence 𝑢𝑚 is first encoded by the channel encoder 𝐶𝑚.
For the retransmit phases, at relay node, 𝑢𝑚 is first random-interleaved by inner interleaver Π0,𝑚+1 before
encoding. It should be noticed that the use of the different interleaver for each round of transmission by the
relays converts the system into a distributed Turbo Code. The relay node discards the old packet whenever
receiving a new packet. Combining the packets at the relay node should further improve the performance,
but it is out of the scope of this study for the sake of simplicity. The same process is performed at the other
relay of the first transmission. The encoded bit sequence is then randomly interleaved by the outer
interleaver Π1,𝑚 followed by doped-accumulator 𝐷𝐴𝑚 with doping ratio 𝜌 = 𝜌𝑚. The outer interleaver
enables extrinsic LLR exchanging of systematic bits via vertical iteration at the receiver side, likewise, the
inner interleaver plays important function of extrinsic LLR exchanging via horizontal iteration. The doped-
accumulated bits are mapped in the Map box in Figure 5.15 to constellation points for modulation. In this
paper, we use binary phase shift keying (BPSK), which follows the mapping rule 0 → −1, 1 → +1, and
then transmitted over frequency flat block Rayleigh fading channel with the complex channel gain ℎ𝑞, 𝑞 ∈ {𝑆𝑁 − 𝑅𝑁1, 𝑆𝑁 − 𝑅𝑁2, 𝑅𝑁1 − 𝐷𝑁, 𝑅𝑁2 − 𝐷𝑁}. The transmitter transmits the modulated signal having N
symbols, as
𝒙𝑖𝑚 = [𝒙𝑖
𝑚(1), 𝒙𝑖𝑚(2), … , 𝒙𝑖
𝑚(𝑁)]𝑇 , 𝑖 ∈ {𝑆𝑁, 𝑅𝑁1, 𝑅𝑁2}
5.2.2.2 Receive Phase
The received signal of the transmitted packet can be formulated as
𝒚𝑗𝑚 = ℎ𝑖𝑗𝒙𝑖
𝑚 + 𝝊𝑗𝑚 , 𝑗 ∈ {𝑅𝑁1, 𝑅𝑁2, 𝐷𝑁},
Where 𝝊 is a zero mean complex additive white Gaussian noise (AWGN) vector with variance 𝜎2 (double
sided). The average signal-to-noise power ratio (SNR) is ⟨|ℎ𝑖𝑗|2
⟩ 𝜎2⁄ since 𝐸[𝒙𝑖𝑚] = 1. The conditional
log-likelihood ratios (LLRs) based on the probability that the receiver’s matched filter output 𝑦𝑗𝑚(𝑛) for
the 𝑛-th bit in 𝒙𝑖𝑚 is defined as
𝐿 (𝑦𝑗𝑚(𝑛)|ℎ𝑖𝑗 , 𝑥𝑖
𝑚(𝑛)) = 𝑙𝑛Pr (𝑦𝑗
𝑚(𝑛)|ℎ𝑖𝑗 , 𝑥𝑖𝑚(𝑛) = +1)
Pr (𝑦𝑗𝑚(𝑛)|ℎ𝑖𝑗 , 𝑥𝑖
𝑚(𝑛) = −1),
where 𝑛 = {0,1,2, … , 𝑁}. Hence, with Pr (𝑦𝑗𝑚(𝑛)|ℎ𝑖𝑗 , 𝑥𝑖
𝑚(𝑛)) =1
√𝜋𝜎2exp [−
|𝒚𝑗𝑚(𝑛)−ℎ𝑖𝑗𝒙𝑖
𝑚|2
𝜎2 ], it is
straightforward to calculate the soft output of the channel 𝑳𝑐, for BPSK over a block fading AWGN channel
as
𝑳𝑐,𝑗(𝑛) =4
𝜎2 ℜ(ℎ𝑖𝑗∗ ∙ 𝒚𝑖
𝑚(𝑛)), (1)
where ℜ(∙) is the real value of its argument.
RESCUE D3.2 Version 2.0
Page 62 (79)
Figure 5.16: Block diagram of the destination node.
Figure 5.16 depicts the structure of the destination node. The block diagram only shows the structure of
receiving the packet from the same relay node. However, a similar structure also can be used for decoding
the packets coming from the other relay. The combiner Σ combines all the extrinsic LLRs, being the output
of channel decoder of each HARQ round. The soft output vector of the channel 𝑳𝑐𝑚 , the LLRs obtained by
(1), is first input to the demapper 𝐷𝑒𝑀 followed by 𝐷𝐴𝑚 decoder (𝐷𝐷𝐴𝑚), and its output extrinsic LLR is
forwarded to the inner-deinterleaver prior to the channel decoder 𝐷𝑚. The subtraction of a priori LLR 𝑳𝑎
from a posteriori LLR 𝑳𝑝 is not shown in the figure for the sake of simplicity. The CI of the received packet
is then calculated, which is equivalent to the mutual information between the a posteriori LLR and the
uncoded systematic bits. If the value of the CI is lower than the predetermined threshold, the extrinsic LLRs
obtained as the output of 𝐷𝑚 are exchanged via horizontal iteration (HI) between the 𝐷𝑒𝑀 + 𝐷𝐷𝐴𝑚and 𝐷𝑚.
The threshold is set adaptively while determining the threshold value of the a posteriori LLR. The
incorrectly decoded packet is stored in order to first update it to reflect the correlation knowledge and then
to combine with the retransmitted packet(s) in the following time slots.
When the retransmitted packet is received, the HI is performed independently, as in the first transmission,
and then the obtained extrinsic LLRs of the systematic information bits, 𝑳𝑒𝑢,𝑚
, are propagated crosswise
between the soft-input soft-output (SISO) channel decoders; this process is referred to as vertical iteration
(VI). VI can be seen as an iterative decoding process of parallel concatenated code. 𝑳𝑒𝑢,𝑚
is updated by the
function 𝑓𝑐 defined by (2). The function 𝑓𝑐 is utilized to help the decoder eliminate the errors in the packets
received by the relays, by exploiting the correlation knowledge between the relays. The correlation value
is indicated by the error probability 𝑝𝑒, block-by-block, which can be estimated by using a pair of a
posteriori LLRs, 𝑳𝑝,𝐷ℐ𝑢 and 𝑳𝑝,𝐷𝒥
𝑢 , ℐ ≠ 𝒥, the uncoded (systematic) bits output
from the decoders 𝐷ℐ and 𝐷𝒥, respectively, as
�̂�𝑒 =1
𝐾∑
exp(𝑳𝑝,𝐷ℐ𝑢 ) + exp (𝑳𝑝,𝐷𝒥
𝑢 )
(1 + 𝑳𝑝,𝐷ℐ
𝑢 ) (1 + 𝑳𝑝,𝐷𝒥
𝑢 ),
𝐾
𝑘=1
where 𝐾 denotes the number of the a posteriori LLR pairs obtained by the decoders for the packets
transmitted from 𝑅𝑁1 and 𝑅𝑁2, with sufficient reliability. The updated extrinsic LLR of 𝑳𝑝,𝐷ℐ𝑢 can then be
obtained by
𝑳𝑝,𝐷ℐ ,𝑢𝑝𝑑𝑎𝑡𝑒𝑑𝑢 = 𝑓𝑐(�̃�𝑝,𝐷ℐ
𝑢 , �̂�𝑒) = 𝑙𝑛(1−𝑝𝑒)∙exp(�̃�𝑝,𝐷ℐ
𝑢 )+𝑝𝑒
(1−𝑝𝑒)+exp(�̃�𝑝,𝐷ℐ𝑢 )∙𝑝𝑒
, (2)
where �̃�𝑝,𝐷ℐ𝑢 = 𝑳𝑝,𝐷ℐ
𝑢 for the first transmission, and �̃�𝑝,𝐷ℐ𝑢 = Π0,𝑚
−1 (𝑳𝑝,𝐷ℐ𝑢 )for the retransmissions. The a priori
LLR 𝑳𝑎𝑢,𝑚
is then
𝑳𝑎𝑢,𝑚 = ∑ 𝑳𝑝,𝐷ℐ ,𝑢𝑝𝑑𝑎𝑡𝑒𝑑
𝑢
𝑞∈𝜔\𝑚
,
with 𝜔 = {1, 2, . . . , 𝑀} being the set of retransmission number. Finally, by performing sufficient rounds
of iterative HI-VI-HI-VI decoding processes, the final hard decisions, �̂�𝑚 is made on the a posteriori LLR
originated by summing up all the deinterleaved and 𝑓𝑐-updated versions of the a posteriori LLR 𝑳𝑝,𝐷𝑚
𝑢,𝑚.
RESCUE D3.2 Version 2.0
Page 63 (79)
5.2.3 Simulation Results
We evaluate BER, PER and e2e throughput performance by simulations that consider the transmission of
100,000 packets with 1,000 bytes per packet. The maximum number of retransmissions per node is set to 4
(M = 5). For fair comparison, we set the same total number of retransmission for all evaluated systems with
each SNR. All nodes use the same channel coding, where a half-rate non-systematic non-recursive
convolutional coding (NSNRCC) with a generator polynomial G = [7, 5] is considered. They all also use
the same varying doping ratio ρ (re)transmission-by-(re)transmission, where ρ ∈ {2, 10, 15, 20, 25}.
We assume no delay constraints for the overall transmission of information from the source to the
destination nodes, and hence the relay nodes can decode the packet before they forward. We also assume
an ideal medium access control protocol, where each node can transmit and receive a packet independently.
Each node is allowed to transmit and receive only one packet simultaneously, and every packet transmitted
from nodes is received without collisions.
We compare P-HARQ with the SHARQ [13] as the conventional scheme. In the conventional scheme, the
relay nodes forward only error free packets and the retransmissions occur either between the source node
and the relay nodes or between the relay nodes and the destination node. As stated previously, we perform
no packet combining at the relay nodes for all the schemes.
Figures 5.17-5.18 show that P-HARQ outperforms the conventional scheme in terms of BER and PER
performances, respectively. Figure 5.19 shows that the conventional scheme fails in combining all
transmitted packets to achieve a large diversity gain while it can be achieved by P-HARQ. This is because
P-HARQ is able to carefully combine the most reliable packets by employing the CI. P-HARQ has 6.2 dB
degradation from Maximal Ratio Combining (MRC) because the relay nodes forward even erroneous
packet to the destination while MRC assumes no error at the relay nodes.
Figure 5.17: BER performance.
RESCUE D3.2 Version 2.0
Page 64 (79)
Figure 5.18: PER performance.
The throughput is defined as the ratio of the average number of successfully recovered packets at the
destination node per time slot to the total number of transmitted packets per time slot. We normalized the
throughput over two time slots, which means that the throughput of one is achieved whenever the packet is
successfully recovered within two time slots. Hence, the packet loss rate can be calculated as 𝑃𝑙𝑜𝑠𝑠 = 1 − 𝑇, with 𝑇 being the throughput of the system. Figure 5.19 shows that for the same ratio of end-to-end
packet loss, the BER performance of the proposed P-HARQ is lowest then the conventional scheme. For
the example, the 60% of end-to-end packet loss corresponds to BER of 0.0026 for P-HARQ.
Figure 5.19: Corresponding BER of throughput performances.
RESCUE D3.2 Version 2.0
Page 65 (79)
6. Multirate
Comparing to the proposal presented in D3.1 the data rate selection (multirate) algorithm for the CSMA/CA
RESCUE network had to be redesigned4. Similarly to the initial design, the network under study uses
transmission of a broadcast type which means that transmitted frames can be received and processed by all
stations within the radio coverage. The process of frame retransmission is controlled by the partial ARQ
mechanism which keeps the number of retransmitted frames at a minimal level. The proposed design
assumes that every network node uses only single input single output (SISO) radio transceivers. The data
rate selection algorithm works in cooperation with the routing protocol, however, the requirements on
providing the list of relaying stations for each connection can be relaxed due to the distributed procedure
of the best relay selection. Nonetheless, as in the initial design no information concerning position of the
relays is given to the source station. For each network station, the data rate selection algorithm decision
variable was changed from slowly converged FER to the SNR value. The behaviour of the source and relay
stations has been differentiated. For the relaying stations distributed elements such as selection of the best
relay station have been added.
The design and evaluation of the data rate selection algorithm assumed a mobile station supporting two
data transmission modes abbreviated as Ofdm12Mbps and Ofdm36Mbps. The parameters of the
transmission modes used during the experiments are listed in Table 6.1. Both transmission modes were
chosen to facilitate a comparison between the RESCUE solution with existing IEEE 802.11a/g networks.
Table 6.1 Transmission modes used in evaluation of the data rate selection algorithm
Transmission mode name Ofdm12Mbit/s Ofdm36Mbit/s
Data throughput Mbit/s 12 36
Transmission technique OFDM OFDM
Subcarrier modulation QPSK 16 QAM
Convolution code rate 1/2 3/4
Transmission bandwidth [MHz] 20 20
All the simulation experiments presented in this chapter were conducted using the ns-3 network simulator.
Each of the simulated network stations was equipped with an error model based on the RESCUE Black
Box implemented in the ns-3 simulator. The position of network stations was assumed to be fixed during
all experiments, which means that the nomadic mobility model was used. The propagation model reflecting
the used mobility pattern consisted of a large scale fading model based on log-distance propagation loss
model with an exponential damping factor of 3.5 and a small scale fading based on Jakes frequency
spreading function. Since a transmission frequency band of 2.4 GHz was assumed the maximum Doppler
frequency shift was set to 4 Hz [7]5.
The goal of the first experiment was to evaluate the performance of a RESCUE station in a point-to-point
scenario. A network consisting of two stations operating with a fixed data rate was used. For each of the
mean SNR values the simulation experiment covered 105 seconds of network operation, the first five of
which were treated as a warm up period. The results of the evaluation of the system performance in the
point-to-point scenario with a zero collision probability are presented in Figure 6.1. According to them, the
SNR threshold for switching between transmission modes for a point-to-point link should be set to 11.9 dB.
This guarantees the highest data throughput and the lowest end-to-end delay value.
4 The final design is presented after the simulation results as it is a direct consequence of the performed analysis.
5 Had the 5 GHz band been used the change would not be significant. What would change is the spatial scaling of the
model and the width of the Doppler spectrum. However, this would not have an qualitative impact on the simulation
results.
RESCUE D3.2 Version 2.0
Page 66 (79)
(a)
(b)
Figure 6.1: Performance of the RESCUE station in a point-to-point scenario: (a) throughput and (b)
end-to-end delay as a function of SNR between S and D.
The goal of the subsequent experiment conducted in the point-to-point setup was to determine the influence
of the transmission mode used for transmitting control frames on network performance. The experiment
was performed for an SNR value of 15 dB and data frames sent at a fixed data rate of 36 Mb/s. In the first
experiment control frames were transmitted with a fixed data rate of 12 Mb/s while in the second
experiment the 36 Mb/s data rate was used. The difference in the end-to-end data throughput values for
both experiments were within the statistical error bound. As a conclusion from the experiment the
Ofdm12Mbps transmission mode was selected to be the transmission mode used for transmitting control
frames. This should improve the reliability of control information transfer without significant impact on the
performance of the network throughput.
The purpose of the next experiment was to evaluate the performance of a RESCUE network operating in
the TS1 scenario. The network configuration consisted of three fixed stations: source (S), relay (R) and
destination (D). The relay station was placed on the straight line between S and D close to the optimal
location (i.e., on 0.48 of the distance between S and D [5]). In the subsequent experiments the distance
between S and D was increased while maintaining the S-R-D geometry. Each of the simulation experiments
lasted 105 seconds of the network operation, the first five of which were treated as a warm up period.
-5 0 5 10 15 20 25
0
2
4
6
8
10
12
SNR between source and destination [dB]
Thro
ugh
pu
t [M
bit
/s]
Ofdm12Mbps
0
50
100
150
200
250
-10 -5 0 5 10 15 20 25
End
-to
-en
d d
ela
y [m
s]
SNR between source and destination [dB]
Ofdm12Mbps
Ofdm36Mbps
RESCUE D3.2 Version 2.0
Page 67 (79)
(a)
(b)
Figure 6.2: Performance of the RESCUE network for TS1: (a) throughput and (b) end-to-end delay
as a function of SNR between S and D
The results presented in Figure 6.2 show that even in the case of direct visibility between S and D a station
can still benefit from using RESCUE coding. However, the increase in datarate is gained at the expense of
increased end-to-end delay. Additionally, the RESCUE gain increases with the desrease of the SNR
between S and D. The results of the experiment presented in Figure 6.2 represent the upper bound of the
gain provided by using the RESCUE scheme in a CSMA/CA network for TS1.
For the experiment presented next, the data rate selection strategy for the S station in TS1 was examined.
The spatial configuration of the network was chosen in order to maximize the gain provided by the
RESCUE coding. According to this rule the network was configured so that the SNR value between S and
D was set to -5 dB. The position of the R station was changed in subsequent runs of the experiment in the
area where the R station remained within the radio coverage of both S and D stations while the average
SNR between the S and R stations always exceeded the threshold of 11.9 dB. Two different data rate
selection strategies were examined. The S station selected its data rate according to the SNR value between
the S and D stations (red curve in Figure 6.3) or according to the SNR value between the S and R stations
(blue curve in Figure 6.3).
-10 -5 0 5 10 15 20 25
0
2
4
6
8
10
12
SNR between source and destination [dB]
Thro
ugh
pu
t [M
bit
/s]
Max throughput withoutRescueMax throughput TS1
0
100
200
300
400
500
600
700
-10 -5 0 5 10 15 20 25
End
-to
-en
d d
ela
y [m
s]
SNR between source and destination [dB]
delay TS1
delay without Rescue
RESCUE D3.2 Version 2.0
Page 68 (79)
(a)
(b)
(c)
Figure 6.3: Cumulative distribution functions of (a) end-to-end throughput, (b) end-to-end delay, and
(c) end-to-end jitter for two different data rate selection strategies in TS1.
Figure 6.3 presents a cumulative distribution function of the end-to-end data throughut, delay and jitter as
a funtion of the varying position of station R. Examining the results presented in Figure 6.3 the preffered
strategy is to adjust the datarate according to the SNR value between S and D even in the presence of relays.
The next experiment evaluates the performance of a multi-rate CSMA/CA RESCUE network for TS2,
which assumes no direct link between S and D and the presence of two relay stations (R1, R2). Additionally
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.5 1 1.5 2
Pro
bab
ility
End-to-end throughput [Mb/s]
relay oriented
destination oriented
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 200 400 600 800 1000 1200
Pro
bab
ility
End-to-end delay [ms]
relay oriented
destination oriented
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20
Pro
bab
ility
End-to-end jitter [ms]
destination oriented
relay oriented
RESCUE D3.2 Version 2.0
Page 69 (79)
the end-to-end communication is possible only when using the information provided by both relay stations,
meaning that one or both of the tandem links S-R1-D and S-R2-D are lossy. No information concerning the
position of the R stations is given to the S station.
Two basic strategies of data rate selection at the S station were examined. The goal of the first strategy was
to set the data rate at S according to the minimum value of the SNRs measured between S and the relay
stations (the weakest relay strategy). The second strategy was based on the higest value of the SNR
measured between S and the relaying stations (the strongest relay strategy).
In order to evaluate the performance of both aforementioned strategies series of experiments were
performed for TS2, consisting of S and D stations without a direct link and two relay stations forming two
lossy tandem links. The position of the relay stations was changed in subsequent runs of the experiment.
Each experiment was run for 105 seconds with 5 seconds of warm up. A thousand different relay positions
were examined and as a result a cumulative distribution of data rates achieved for both strategies was
obtained. The experiment results are presented in Figure 6.4.
Figure 6.4: Cumulative distribution functions of two data rate selection strategies for the TS2
scenario.
Analysing the CDF presented in Figure 6.4 one can observe that both strategies exhibit almost the same
probability of end-to-end transmission failure. For the majority of network configurations the data rate
selection strategy oriented on the SNR measured for the R station with a stronger signal offers higher data
throughput values.
In a configuration with more than two relays both presented data rate selection strategies may lead to a
suboptimal solution due to the lack of information of the RX-D SNR, which means that the capacity of the
S-RX-D tandem link is unknown. In the CSMA/CA network each station has an equal transmission
probability, which means that, in the case of the examined network links with a low SNR value, operating
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.5 1 1.5 2 2.5 3
Pro
bab
ility
End-to-end throughput [Mbps]
weakest relay
strongest relay
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
200 400 600 800 1000
Pro
bab
ility
End-to-end delay [ms]
weakest relay
strongest relay
RESCUE D3.2 Version 2.0
Page 70 (79)
with lower data rates consumes three times more time than links with SNR above 11.9 dB. As a result, a
wrongly placed R station can decrease the performance of the whole network. Additionally, it is worth
mentioning that the gain provided by the RESCUE coding decreases with an increasing number of
combined frames. As an example, the dependency of the RESCUE gain with the number of combined
frames received with similar SNR values is presented in Figure 6.5.
Figure 6.5: RESCUE coding gain as a function of the number of combined frames
The remedy for both phenomenas is to increase the transmission probability for the relays with a higher
tandem S-R-D link quality in favour of badly placed relay stations. In CSMA/CA networks the medium
access probability can be varied by changing the contention window (CW) value. Since complete
information (SNRSR and SNRRD) about the quality of the S-R-D tandem link is available only at the relay
stations it is reasonable to set the probality of the chanel access in a distributed way.
In the proposed algorithm the S-R-D link capacity will be used as a cost function. Since only simple
retransmission is performed at the relaying station the SNR in the whole tandem link can be estimated as:
𝑆𝑁𝑅𝑆𝑅𝐷 =1
𝑆𝑁𝑅𝑆𝑅−1 + 𝑆𝑁𝑅𝑅𝐷
−1 (6.1)
where 𝑆𝑁𝑅𝑆𝑅 is the signal to noise ratio for the S-R link and 𝑆𝑁𝑅𝑅𝐷 is the signal to noise ratio for the R-D
link
The capacity of a tandem S-R-D link as a function of the mean SNR value was determined by simulation.
The experiment was performed for a number of tandem links with a variable position of the R station. Each
subsequent simulation experiment covered 105 seconds of network operation, the first five of which were
treated as a warm up period. The resulting tandem link capacity is presented in Figure 6.6 as a function of
the average SNRSRD.
Figure 6.6: Capacity of a tandem link as a function of the average 𝑺𝑵𝑹𝑺𝑹𝑫
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 2 4 6 8 10
RES
CU
E co
din
g ga
in [
dB
]
Number of combined frames
-5 0 5 10 15 20
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Average SNR of a tandem link [dB]
End
-to
-en
d t
hro
ugh
pu
t [M
bit
/s]
Throughput of S-R-D link
1.19+0.28*SNR1.19+0.28×SNR
RESCUE D3.2 Version 2.0
Page 71 (79)
The capacity of a tandem link for a CSMA/CA RESCUE network can be estimated as:
𝑐𝑎𝑝 = 1.19 + 0.28 𝑆𝑁𝑅𝑆𝑅𝐷. (6.2)
Based on the simulation results the maximum data throughput achieved by a tandem link is determined to
be approximately 𝑅0 = 4 Mb/s. The goal of the proposed algorithm is to give the same time share instead
of the same probability of accessing the wireless channel for each of the relays. The relays capable of
achieving throughput close to 𝑅0 should be given a transmission probability equal to the probability
assigned to the source station (standard priority) other relaying stations will be given priorities according
to their transmission capabilities.
In case of a tandem link with a high SNRSRD value the transmission time of an 𝑛-bit long frame is
𝑇0 =𝑛𝑏𝑖𝑡𝑠
𝑅0
(6.3)
For a particular capacity (cap) of a tandem link the transmission time of an 𝑛-bit long frame can be
calculated as follows:
𝑇𝑐𝑎𝑝 =𝑛𝑏𝑖𝑡𝑠
𝑐𝑎𝑝=
𝑛𝑏𝑖𝑡𝑠
𝑅0
∙𝑅0
𝑐𝑎𝑝= 𝑇0 ∙
𝑅0
𝑐𝑎𝑝 (6.4)
In order to give the same time share for all the relaying stations the following term should apply:
𝑇0𝑝0 = 𝑇𝑐𝑎𝑝𝑝𝑐𝑎𝑝, (6.5)
where 𝑝0 is the reference transmission probability and 𝑝𝑐𝑎𝑝 is the probability of accessing the wireless
channel for a tandem link with capacity cap.
If for the sake of simplicity a zero frame collision probability is assumed, then the dependency of the
medium access probability and the value of the contention window 𝐶𝑊𝑖 can be expressed as [6]:
𝑝𝑖 =2
𝐶𝑊𝑖 + 1 (6.6)
Assuming that a perfectly placed relay is given a transmission priority equal to the priority possessed by a
regular source station the reference transmission probability can be expressed as:
𝑝0 =2
𝐶𝑊0 + 1 (6.7)
and based on (6.5):
𝑝0 =𝑅0
𝑐𝑎𝑝𝑝𝑐𝑎𝑝 =
𝑅0
𝑐𝑎𝑝
2
𝐶𝑊𝑐𝑎𝑝 + 1 (6.8)
after substituting (6.8) to (6.7): 𝑅0(𝐶𝑊0 + 1)
𝑐𝑎𝑝− 1 = 𝐶𝑊𝑐𝑎𝑝 (6.9)
finally combining (6.9) with (6.2) gives: 𝑅0(𝐶𝑊0 + 1)
1.19 + 0.28 𝑆𝑁𝑅𝑆𝑅𝐷
− 1 = 𝐶𝑊𝑐𝑎𝑝 (6.10)
The differentiation of the channel access probability for the relaying stations enables another data rate
selection strategy for the source station. According to the new strategy the source station adapts its data rate
to the SNR value measured for the most frequently transmitting relay station.
The performance of the new data rate selection strategy was tested for two variants of the TS2 scenario.
The results achieved by the networks implementing the new data rate selection algorithm were compared
to those obtained for a network using the data rate selection strategy oriented on the SNR measured for a
relay station with the strongest signal. In the first experiment a minimalistic TS2 scenario with only two
relaying stations was examined. The positions of the relaying stations were changed in subsequent runs of
the experiment. For each of the network configurations 105 seconds of network operation was simulated
while the first 5 seconds were treated as a warp up period. Thousands of relay positions were examined and
as a result a cumulative distribution of data rates achieved for both strategies was calculated. The results
presented in Figure 6.7 show that the new strategy does not deteriorate the end-to-end network performance.
It is worth mentioning that for a minimalistic TS2 scenario both relaying stations must be involved in the
process of a frame transmission and in this case an incorrectly selected CW value could impact the network
performance. Since both relays take part in the frame forwarding procedure no improvements in comparison
to the strongest relay oriented data rate seletion strategy were expected for this scenario.
RESCUE D3.2 Version 2.0
Page 72 (79)
(a)
(b)
Figure 6.7: Cumulative distribution functions of the (a) end-to-end throughput and (b) end-to-end
delay calculated for the TS2 scenario. Comparison of two strategies of data rate selection by the
source: strongest relay (highest measured SNR) or most frequently transmitting relay (assuming CW
control).
The second experiment evaluated the performance of the new relaying algorithm for an extended version
of the TS2 scenario with the number of potentially relaying stations increased to ten. The experiment was
conducted according to the same scenario as for the case of the minimalistic TS2 except for the simulation
time for each run, which was increased to 1005 seconds of network operation. The results of the experiment
are presented in Figure 6.8:
In case of the TS2 topology with an increased number of potentially relaying stations, the differentiation of
the medium access probability for the relaying stations leads in most cases to the increase of the end-to-end
throughput comparing to the strongest relay strategy. An additional benefit from using the distributed
selection strategy of the best relaying stations is increasing the reliability of frame delivery. The distributed
selection strategy of best relaying stations can also play a role in the simplification of the routing algorithm
which now can be freed from selecting the set of relaying stations for a particular connection.
0
0.2
0.4
0.6
0.8
1
0 0.5 1 1.5 2 2.5 3
Pro
bab
ility
End-to-end throughput [Mb/s]
most frequent +CW
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
200 400 600 800 1000
Pro
bab
ility
End-to-end delay [ms]
most frequent + CWstrongest relay
RESCUE D3.2 Version 2.0
Page 73 (79)
(a)
(b)
Figure 6.8: Cumulative distribution functions of the (a) end-to-end throughput and (b) end-to-end
delay calculated for the TS2 scenario with more than two relays. Comparison of two strategies of
data rate selection by the source: strongest relay (highest measured SNR) or most frequently
transmitting relay (assuming CW control).
In conclusion, the final decision block diagram of the distributed data rate selection algorithm for RESCUE
CSMA/CA networks has been updated to reflect the lessons learned from the simulation results (Figure
6.9). The first step is to determine the stations role in the transmission of a given frame (source or relay).
In the former case, the S-D link quality is checked. If it exceeds the minimum (-5 dB) threshold, the data
rate is selected based on the results shown in Figure 6.1. Otherwise, data rate is determined based on the
SNR of the S-R link quality for the most frequently transmitting relay. In the case of the relay, the model
based on equations (6.1) and (6.10) is used to improve performance (airtime usage) and select the optimum
data rate.
Start of message processing
Check role
SNRSD<-5 dBNY
source
SNRSD>11.9 dB
relay
Y
Return Ofdm12Mbps
N
Return Ofdm36Mbps
Determine the SNRSR for the most
frequently transmitting relay
SNRSR>11.9 dB
Y
Return Ofdm12Mbps
N
Return Ofdm36Mbps
Calculate the SNRSRD for tandem
link (6.1)
Calculate new CW (6.10)
SNRRD>11.9 dB
Y
Return Ofdm12Mbps
N
Return Ofdm36Mbps
Figure 6.9: Diagram of the RESCUE data rate selection algorithm
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Pro
bab
ility
End-to-end throughput [Mb/s]
most popular relay…
0
0.2
0.4
0.6
0.8
1
0 500 1000 1500 2000 2500
Pro
bab
ility
End-to-end delay [ms]
most popular relay + CW…
RESCUE D3.2 Version 2.0
Page 74 (79)
7. Conclusion
In this deliverable, we have presented the design updates of the protocols previously defined in D3.1.
Furthermore, we have provided the simulation-based performance evaluation of these protocols, using
RESCUE’s unified NS3 platform which integrates an abstraction of the physical layer (an input from WP2),
various MAC mechanisms &protocols, Automatic Repeat reQuest (ARQ) or routing protocols in order to
properly access the concept of RECUE’s lossy forwarding which validate their usefulness in facilitating
RESCUE message transfer. The simulations, results and evaluations presented for the MAC protocols,
routing protocols, error control mechanisms and rate adaptation in this deliverable have all been produced
using key common parameter tables which are relevant to the verification of the RESCUE’s concepts in
lossy communication networks. This deliverable provided updates to the background on message transfer,
toy scenarios, and baselines for comparison. The full design details and simulation-based results which take
into consideration lossy wireless communication links to show functionalities proposed as benefits of
RESCUE project have also been presented for the following: (i) three complementary medium access
protocols for centralized and distributed networks, (ii) two complementary routing protocols (R-OLSR and
geo-routing protocols), (iii) two complementary error control methods for ensuring end-to-end reliability
and efficient retransmissions, and (iv) a multirate algorithm for providing improved performance by
adapting to channel conditions. More details on the forwarding threshold optimization for the considered
toy scenarios (TS1, TS2 and TS3) and the V2V scenario are presented in the appendix of this document.
The design details and simulation results provided in this deliverable will serve as a basis for the
implementation and performance analysis of each component to be reported D3.3 as well as the
implementation and validation of the integrated components to be performed in WP4.
RESCUE D3.2 Version 2.0
Page 75 (79)
Appendix A. Forwarding Threshold Optimization
There are many parameters governing the behaviour of the RESCUE message transfer protocols which can
be subject to further optimization. One such parameter of interest is the forwarding threshold p which
represents the maximum allowable BER of a received message which is to be processed by the receiver
(i.e., passed to higher layers) as noted in Table 3.1. In the particular case of p=0, the resulting message
transfer system can be considered as being a selective decode and forward (SDF) approach. SDF is a
particular case of lossy forwarding (LF) where only correct packets are forwarded and joined decoding is
applied at the destination. While in general there are different SDF methods in the state of the art, in this
case we apply the RESCUE-developed joint decoding in SDF. Therefore, both approaches (LF and SDF,
for p=0.2 and p=0, respectively), which have been evaluated below, are contributions from the project.
Toy Scenarios
Simulation results of SimpleMAC for TS1-TS3 (Figures A.1 to A.3) show that in the basic RESCUE
topologies indeed the LF approach is better than SDF, especially in terms of throughput and end-to-end
delay in the range -3 to -6 dB of SNR. The forwarding of erroneous frames in LF together with the joint
decoding at the decoder can provide an increase in throughput, in comparison to SDF, of up to 84%.
However, this may not always be the case, as can be seen in the next analysed scenario.
Figure A.1: Throughput, end-to-end delay, jitter, and frame loss ratio for TS1 for two forwarding
thresholds. The link between the R and D stations is lossless, while the S-R and S-D links are lossy
and their SNR is the variable. For other lossy/lossless link configurations the performance of the two
forwarding thresholds was similar.
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
p = 0 p = 0.2
0
200
400
600
800
1000
1200
1400
1600
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
De
lay
[ms]
SNR [dB]
p = 0 p = 0.2
0
100
200
300
400
500
600
700
800
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Jitt
er
[ms]
SNR [dB]
p = 0 p = 0.2
0
20
40
60
80
100
120
-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
FLR
[%
]
SNR [dB]
p = 0 p = 0.2
RESCUE D3.2 Version 2.0
Page 76 (79)
Figure A.2: Throughput, end-to-end delay, jitter, and frame loss ratio for TS2 for two forwarding
thresholds. The S-R2 and R1-D links are lossless, while the other two links are lossy and their SNR
is the variable. For other lossy/lossless link configurations the performance of the two forwarding
thresholds was similar.
Figure A.3: Throughput, end-to-end delay, jitter, and frame loss ratio for TS3 for two forwarding
thresholds. The relay to destination links are lossless, while the other links are lossy and their SNR is
the variable. For other lossy/lossless link configurations the performance of the two forwarding
thresholds was similar.
0
0.5
1
1.5
2
2.5
-7.5-7-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
p=0 p=0.2
0
200
400
600
800
1000
1200
1400
-7.5-7-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
De
lay
[ms]
SNR [dB]
p=0 p=0.2
0
100
200
300
400
500
600
-7.5-7-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Jitt
er
[ms]
SNR [dB]
p=0 p=0.2
0
20
40
60
80
100
120
-7.5-7-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5FL
R [%
]SNR [dB]
p=0 p=0.2
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
-7.5-7-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Thro
ugh
pu
t [M
b/s
]
SNR [dB]
p=0 p=0.2
0
200
400
600
800
1000
1200
-7.5-7-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
De
lay
[ms]
SNR [dB]
p=0 p=0.2
0
0.5
1
1.5
2
2.5
3
3.5
-7.5-7-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
Jitt
er[m
s]
SNR [dB]
p=0 p=0.2
0
20
40
60
80
100
120
-7.5-7-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5
FLR
[%]
SNR [dB]
p=0 p=0.2
RESCUE D3.2 Version 2.0
Page 77 (79)
V2V Scenario
For the V2V scenario (Section 2.4.1), Figure A.4 shows the simulation results and it turns out that SDF
(RESCUEp=0.0) performs better than LF (RESCUEp=0.2) in this large-scale V2V scenario. The reason lies in
the particular scenario and the V2V-specific requirements. With SDF, the CBF algorithm forwards error-
free packets only. Then, in the given scenario, where the SNR is too small to decode the packet as error
free at the last hop to the destination, joint decoding shows better performance in terms of packet success
ratio. With LF, the CBF algorithm forwards also packets with bit errors, which are used at the destination
to perform joint decoding. However, the packet success ratio of LF performs better compared to the
baseline. The intuitive explanation for the worse performance of LF to SDF is that CBF prefers forwarding
of packets with bit errors over error-free packets when the relay node is considered as the appropriate
forwarder (i.e., provides more spatial progress towards the destination). RESCUE CBF takes the confidence
value into account when the relay calculates the forwarding delay. This feature addresses the trade-off
between spatial progress and confidence value and would favour a relay which has less bit errors in a packet
over a relay which provides larger spatial progress. In this context SDF can be regarded as an extreme case,
where the importance of having a very small number of bit errors (actually zero bit errors) is much higher
than spatial progress, which then leads to SDF being a corner case of LF.
Due to the Nakagami channel also the relays in the second group (see Section 2.4.1) receive packets with
a high probability of bit errors inside the packets. In the LF approach relays forward the packet with errors
inside, in SDF they drop it. One constraint is the limit that each node is allowed to forward a packet at most
once. The advantage of SDF in VANETs is the shape of the network. The vehicles in a group are close
together and in direct communication range. If at least one vehicle receives the packet error free, it
distributes the packet in its group and generates three new potential forwarders. The probability that at least
one packet of the four forwarders in the first group is received by at least one node in the second group is
very high. Now, in the second group we have another four potential forwarders, which can send the packet
error free to the destination (but with low SNR). The situation now is as follows: in both cases (SDF/LF)
we receive four packet copies at the destination. In SDF all four packets were error free before the last
transmission whereas in LF errors can occur in the packets before. Therefore when the PHY Black Box
applies joint decoding SDF has an advantage compared to LF. However, the comparison of SDF and LF
can be seen as unfair because in SDF the nodes wait until an error-free packet arrives whereas in LF also
erroneous packets are forwarded. The Black Box supports just three packet copies and in this scenario the
probability that three error-free packets arrive is high. If all arriving packets would be used at the destination
LF should show better results. The end-to-end delay is worse in SDF since the vehicles in the second group
have to wait for subsequent error-free packets, where in LF erroneous packets are immediately forwarded.
RESCUE D3.2 Version 2.0
Page 78 (79)
Figure A.4: Packet success rate and end to end delay of the large scale V2V-scenario
Summary
In conclusion, we can state that the forwarding threshold parameter is important to the performance of a
RESCUE network. Its optimal value is topology-specific and should be carefully considered during the
implementation and validation tests. Indeed, in some cases, as exemplified by the V2V scenario, adopting
SDF can be more beneficial. The presented results may lead to the formulation of a hypothesis that LF is
more appropriate in small-scale scenarios while SDF – in larger scenarios. Further research is required to
validate this hypothesis.
RESCUE D3.2 Version 2.0
Page 79 (79)
References
[1] ETSI EN 302 636: „Intelligent Transport Systems (ITS); Vehicular Communications;
GeoNetworking”, Part 1 – 6, 2013
[2] ETSI EN 302 637-3 V1.2.2: “Intelligent Transport Systems (ITS); Vehicular Communications;
Basic Set of Applications; Part 3: Specifications of Decentralized Environmental Notification
Basic Service”, 2014
[3] S. Kühlmorgen, S. Llatser, A. Festag, and G. Fettweis: "Performance Evaluation of ETSI
GeoNetworking for Vehicular Ad hoc Networks", VTC Spring 2015, Glasgow, Scotland,
May 2015
[4] RESCUE Project, “D1.2.1 – Assessment on Feasibility, Achievability, and Limits”, Version 1.0,
April 2015
[5] RESCUE Project, “D2.2.1 Interim Report of Detailed Optimisation Algorithm and Performance
Comparisons”, Version 1.0, April 2015
[6] G. Bianchi, "Performance analysis of the IEEE 802.11 distributed coordination function," Selected
Areas in Communications, IEEE Journal on, pp. 535-547, 2000.
[7] H. Hashemi, "A study of temporal and spatial variations of the indoor radio propagation channel,"
in Personal, Indoor and Mobile Radio Communications, 1994. Wireless Networks - Catching the
Mobile Future, 5th IEEE International Symposium on, Sep. 1994, pp. 127-134vol1.
[8] T. Clausen, and P. Jacquet, "Optimized Link State Routing Protocol (OLSR)", RFC 3626, October
2003.
[9] Y. Jiazi, C. Eddy, H. Salima, and P. Benoît, “Implementation of Multipath and Multiple
Description Coding in OLSR”, IEEE WCNC 2008, Las Vegas, USA 2008
[10] F. Bohdanowicz, M. Jakobs and C. Steigner “Statistical Convergence Investigation of Routing
Protocols” International Journal on Advances in Systems and Measurements, vol 3 no 3 & 4, 2010
[11] RESCUE Project, “D2.1.1 Interim Report of Detailed Coding/Decoding Algorithms and
Correlation Estimation Techniques”, Version 1.0, October 2014
[12] http://www.nsnam.org/
[13] A. Bhamri, F. Kaltenberger, R. Knopp, J. Hamalainen, “Smart Hybrid-ARQ (SHARQ) for
Cooperative Communication via Distributed Relays in LTE-Advanced”, in Signal Processing
Advances in Wireless Communications (SPAWC), 2011 IEEE 12th International Workshop on,
pp.41-45, 26-29 June 2011.
[14] Capacity of Ad-hoc Network, J. Li, C. Blake et al., Proceedings of the 7th ACM International
Conference on Mobile and Networking, 2001
[15] The capacity of wireless networks, P. Gupta and P.R. Kumar, IEEE transactions on information
theory, 2000
[16] Rogge, H., Baccelli, E., Kaplan, A., "Packet Sequence Number based ETX Metric for Mobile Ad
Hoc Networks", Internet-Draft, Sept. 2010