PERFORMANCE ANALYSIS OF CSMA/CD
FOR IN-BAND FULL DUPLEX WIRELESS
COMMUNICATION
a thesis submitted to
the graduate school of engineering and science
of bilkent university
in partial fulfillment of the requirements for
the degree of
master of science
in
electrical and electronics engineering
By
Wardah Sarmad
January 2019
Performance Analysis of CSMA/CD for In-Band Full Duplex Wireless Communication
By Wardah Sarmad January 2019
We certify that we have read this thesis and that in our opinion it is fully adequate, in scope and in quality, as a thesis for the degree of Master of Science.
Ezhan Kara¸san(Advisor)
Nail Akar
Mehmet K¨oseo˘glu
Approved for the Graduate School of Engineering and Science:
ABSTRACT
PERFORMANCE ANALYSIS OF CSMA/CD FOR
IN-BAND FULL DUPLEX WIRELESS
COMMUNICATION
Wardah Sarmad
M.S. in Electrical and Electronics Engineering Advisor: Ezhan Kara¸san
January 2019
In-band full duplex (IBFD) wireless communication is gaining a lot of interest as it improves spectral efficiency. Due to development in self-interference cancella-tion schemes and overcoming hardware limitacancella-tions, IBFD communicacancella-tion can be realized in future generation wireless systems. IBFD introduces new problems in the MAC layer such as false collision detection and miss collision detection. The false collision detection wastes the channel usage and miss collision detection re-sults in a lost frame. These errors occur due to imperfect sensing that rere-sults from residual self-interference (RSI). In this thesis, we introduce a new Markov chain model for the performance analysis of full duplex MAC protocol which allows the nodes not only to transmit but also continuously monitor the channel and abort the transmission as soon as a collision is detected (CSMA/CD). Maximum retry limit and backoff freezing are implemented in the Markov chain model. A more realistic miss collision detection implementation is done which take care of all miss collision detection cases. Goodput and packet loss probability are derived to evaluate the performance of CSMA/CD while considering false and miss collision detection. The accuracy of the analytical model is validated by using extensive simulations for saturated traffic condition, that show a reasonable numerical error for goodput and packet loss probability.
Keywords: In-Band Full Duplex, Markov Chain Model, Maximum Retry Limit, Backoff Freezing, Goodput.
¨
OZET
BANT˙IC
¸ ˙I TAM C
¸ ˙IFT Y ¨
ONL ¨
U KABLOSUZ
HABERLES
¸ME ˙IC
¸ ˙IN CSMA/CD’N˙IN PERFORMANS
ANAL˙IZ˙I
Wardah Sarmad
Bilgisayar M¨uhendisli˘gi, Y¨uksek Lisans Tez Danı¸smanı: Ezhan Kara¸san
Ocak 2019
Aynı bant tam ¸cift y¨onl¨u (IBFD) kablosuz haberle¸sme, spektral verimlili˘gi artırdı˘gı i¸cin olduk¸ca ilgi g¨ormektedir. Oz giri¸sim baskılama sistemlerindeki¨ geli¸smeler ve donanımsal sınırlamaların a¸sılmasıyla, IBFD haberle¸smesi gele-cek nesil kablosuz sistemlerde ger¸gele-cekle¸stirilebilegele-cektir. IBFD, MAC katmanında yanlı¸s ¸carpı¸sma tespiti ve ¸carpı¸sma tespiti ka¸cırma gibi yeni problemler ortaya ¸cıkarmaktadır. Yanlı¸s ¸carpı¸sma tespiti gereksiz kanal kullanımına sebep olmakta ve ¸carpı¸sma tespiti ka¸cırma kayıp ¸cer¸ceveye sebebiyet vermektedir. Bu hatalar, artık ¨oz giri¸sim sonucu olu¸san hatalı sezimlerden kaynaklanmaktadır. Bu tezde, d¨u˘g¨umlerin sadece iletim yapmasını de˘gil, aynı zamanda s¨urekli olarak kanalı g¨ozetlemesini ve ¸carpı¸sma tespiti yapılır yapılmaz iletimi durdurmasını sa˘glayan, tam ¸cift y¨onl¨u MAC protokol¨un¨un performans analizi i¸cin yeni bir Markov zinciri modeli tanımlanılmı¸stır (CSMA/CD). Maksimum tekrar limiti ve geri ¸cekilme dondurma, Markov zinciri modelinde uygulanmı¸stır. B¨ut¨un ¸carpı¸sma tespiti ka¸cırma durumlarını g¨ozeten daha ger¸cek¸ci bir ¸carpı¸sma tespiti ka¸cırma uygu-laması yapılmı¸stır. Yanlı¸s ¸carpı¸sma tespiti ve ¸carpı¸sma tespiti ka¸cırma durumları g¨ozetilerek, CSMA/CD’nin performansının de˘gerlendirilebilmesi i¸cin, uygulama seviyesindeki veri hızı ve paket kaybı ihtimali t¨uretilmi¸stir. Analitik modelin do˘grulu˘gu, doygun trafik ko¸sulu i¸cin yapılan kapsamlı sim¨ulasyonlar sonucu, ku-ramsal en b¨uy¨uk yararlı i¸s ve paket kaybı ihtimalinde g¨or¨ulen makul n¨umerik hatalar ile g¨osterilmi¸stir.
Anahtar s¨ozc¨ukler : Aynı Bant Tam C¸ ift Y¨onl¨u Haberle¸sme, Markov Zincir Mod-eli, Maksimum Tekrar Limiti, Geri C¸ ekilme Dondurma, Uygulama Seviyesindeki Veri Hızı.
Acknowledgement
First of all, I would like to thank my advisor Prof. Ezhan Kara¸san for his con-tinuous guidance and kindness. Without his constant support, it would have not been possible for me to complete my work, and my parents for their trust and their consent for allowing me to fulfill my dream of studying abroad, my husband Mr. Syed Maaz Shahid for always encouraging me whenever I had lost hope, and also for helping me with coping up with tough times I have been through. Without his immense motivation, I would have not been able to achieve it.
I would also like to thank my all teachers, who have made me what I am today. Last but not the least, I would like to use this opportunity to give regards to my friends in Bilkent especially Mr. and Mrs. Muhammad Anjum Qureshi and Ms. Sadia for helping me in my work and the joy they had brought me throughout my stay in Bilkent.
Contents
1 Introduction 1
2 In-Band Full Duplex Wireless Communication 7
2.1 Physical Layer Issues . . . 7
2.2 MAC Layer Issues . . . 10
2.3 Contributions of the Thesis . . . 13
3 Markov Chain Model for CSMA/CD for IBFD 16 3.1 Introduction of Full Duplex MAC Protocol Design . . . 16
3.1.1 Improvements in the Markov Chain based IBFD-MAC Model 18 3.1.2 Model Assumptions . . . 19
3.1.3 IBFD-MAC Protocol Design . . . 20
3.2 Markov Chain Model and Analysis . . . 23
3.2.1 Goodput Analysis . . . 27
CONTENTS vii
4 Numerical Results of the Proposed Protocol 31
4.1 Flowchart . . . 31
4.2 Goodput Evaluation . . . 32
4.2.1 Throughput and Goodput Comparison . . . 33
4.2.2 Introducing Retry Limit and Probability of Busy Channel 35 4.2.3 Optimal Packet Length . . . 37
4.2.4 Effect of False Alarm Probability . . . 39
4.2.5 Effect of Number of Nodes . . . 40
4.3 Ploss Evaluation . . . 41
4.3.1 Ploss with Initial Contention Window . . . 42
4.3.2 Ploss with Number of Nodes . . . 45
4.4 HD CSMA/CA and IBFD CSMA/CD Comparison . . . 48
4.4.1 Comparison against Number of Nodes . . . 48
4.4.2 Comparison against Initial Contention Window . . . 49
4.4.3 Comparison against Probability of Miss Detection . . . 50
List of Figures
1.1 Bi-directional Full duplex Communication . . . 2
1.2 Self-interference in Bi-directional Full duplex Communication . . . 3
3.1 System Model for IBFD-MAC protocol in which uplink traffic is considered only . . . 17
3.2 IBFD-MAC Protocol . . . 22
3.3 Discrete Time Markov Chain Model . . . 23
4.1 Flowchart of simulation working . . . 32
4.2 Throughput and Goodput vs. Initial Contention Window for L = 1000 . . . 33
4.3 Throughput and Goodput vs. Length of Packet for CWmin = 24 . 34 4.4 Analytical results of each modification and simulation results are plotted with respect to CWmin . . . 36
4.5 Analytical results of each modification and simulation results are plotted with respect to L . . . 37
LIST OF FIGURES ix
4.7 Goodput is plotted against false alarm probability for L = 100 and
L = 1000 . . . 39
4.8 Goodput is plotted with respect to initial contention window for the different number of nodes M = 10, 50, 100, 150 . . . 40
4.9 Pmiss vs. CWmin . . . 42 4.10 Pmrl vs. CWmin . . . 43 4.11 Ploss vs. CWmin . . . 44 4.12 Pmiss vs. M . . . 45 4.13 Pmrl vs. M . . . 46 4.14 Ploss vs. M . . . 47
4.15 Goodput comparison of HD CSMA/CA with IBFD CSMA/CD with respect to number of nodes . . . 49
4.16 Goodput comparison of HD CSMA/CA with IBFD CSMA/CD with respect to Initial contention window . . . 50
4.17 Goodput comparison of HD CSMA/CA with IBFD CSMA/CD with respect to Probability of miss detection . . . 51
List of Tables
3.1 List of parameters used throughout this chapter . . . 18
3.2 List of parameters for goodput calculation . . . 27
4.1 Parameters used for simulation and anayltical results for Figures 4.4 and 4.5 . . . 35
Chapter 1
Introduction
In recent years, wireless communication technologies have been reviving owing to the significant increase in mobile traffic demand. However, radio resources are scarce and expensive. Half duplex and out-of-band full duplex schemes (trans-mit and receive over different frequencies and time slots) are utilizing the re-sources inefficiently. The increase in demands, as well as efficient utilization of available resources, has become a challenging task, and new technologies are re-quired to overcome these challenges. Multiple-input-multiple-output (MIMO) and orthogonal frequency division multiplexing (OFDM) are being utilized to improve spectral efficiency of current generation wireless communication systems but additional techniques are needed to increase the efficient use of the expensive frequency spectrum in future generation wireless systems that are required to provide substantially higher data rates to mobile users.
In-Band Full Duplex (IBFD) communication has gained a lot of interest within the last five years as it is assumed to double the capacity with zero self-interference theoretically. In IBFD communication, a node can simultaneously transmit and receive using the same frequency resource. A significant amount of work is being done on IBFD, but it has its own disadvantages. The work on full duplex com-munication includes the deployment of a full duplex physical layer and medium
access control (MAC) protocols. Many MAC protocols are also designed to over-come the shortcomings of half duplex (HD) communication such as packet loss due to network congestion and delay.
Figure 1.1: Bi-directional Full duplex Communication
The work in this thesis focuses on IBFD because of its following advantages [1]:
1. It can obtain twice the capacity of the conventional half duplex (HD) sys-tems due to simultaneous transmission and reception over the same channel owing to perfect SI cancellation.
2. IBFD communication improves spectral efficiency and resource utilization.
3. IBFD communication reduces end-to-end delay with relay nodes as they allow simultaneous reception and transmission of data.
4. It can also improve network security as eavesdroppers receive mixed signals and its difficult for them to decode it due to interference.
5. It can also reduce latency for the feedback signals such as acknowledgment, resource allocation information and channel state information. This is be-cause during transmission they can receive these signals and improve the throughput accordingly.
6. It also provides flexibility in spectrum usage as each node has an option that it can select full duplex or half duplex transmission mode.
7. Hidden node problem can also be solved in ad hoc networks. As a node transmit to access point (AP) and AP also sends a packet back to the source node simultaneously, other nodes will hear the transmission of AP and delay their own transmissions.
One of the reasons IBFD communication requires further research is that it has some drawbacks which affect its performance. The biggest problem that af-fects the performance of IBFD systems is self-interference (SI) that results due to simultaneous transmission and reception in the same frequency band. SI cannot be perfectly canceled due to hardware limitations, estimation errors etc. The hardware limitations include noise, transmitter and receiver dynamic range, non-linearities, and channel dynamics also. SI decreases the performance of full du-plex systems as compared to HD if it is not mitigated perfectly. Apart from self-interference, the signals reflected from nearby obstacles also causes interfer-ence. Inter-node interference almost gets doubles due to IBFD communication which results in an increase in aggregate interference [1] and link reliability is also degraded in IBFD. System designs are more complex for full duplex communi-cation that increases power consumption. Due to these reasons, the full duplex communication systems are implemented with WARP [2, 3] and USRP platform is used for testbeds.
Figure 1.2: Self-interference in Bi-directional Full duplex Communication
In self-interference, a node will receive its own transmitted signal and it can be 90-100 dB stronger than the signal of interest as shown in Figure 1.2. The power difference between the signal of interest and self-interference increases ex-ponentially as distance increases [1]. With the help of various self-interference cancellation techniques in propagation, analog, and digital domains, IBFD is an
emerging technology although each scheme has its own advantages and disad-vantages. Propagation domain SI suppression separates the receive antenna from transmit antenna electromagnetically, whereas analog cancellation is performed before the signal enters ADC, and digital cancellation is done after ADC to sup-press any remaining SI using the DSP scheme [4]. In the analog domain, the transmit signal can be tapped in the analog or digital domain. Digital domain cancellation is by far the weakest among the three due to the limitation of ADC dynamic range.
Other physical layer advancements are also required along with SI attenuation to get benefits of IBFD communication. Although SI limits the performance of IBFD systems, an asymptotic analysis shows that inter-link interference and spa-tial reuse also limits the capacity gain. They suggested that in order to properly utilize full duplex advantages, spatial reuse and asynchronous contention in large networks need to be managed carefully [5].
In order to fully utilize IBFD communication, all wireless communication sys-tems need to be redesigned such as cellular syssys-tems and Wi-Fi. It requires not only to upgrade the physical layer but also needs to update the upper layers, for example, the medium access control layer. The HD IEEE 802.11 MAC pro-tocol for wireless local area networks (Wi-Fi) has a standardized MAC propro-tocol which works according to distributed coordination function (DCF) [6]. It sup-ports asynchronous data transfer and it works as carrier sense multiple access with collision avoidance (CSMA/CA). DCF uses a discrete-time backoff scale and uses a two-way handshaking process called the basic access method. There also exists another mechanism in the standard called point coordination function (PCF) used for time-bounded services. We have considered DCF in this thesis.
In HD CSMA/CA, the node monitors the channel before transmission in order to avoid a collision. If the packet is successfully received at the destination, a positive acknowledgment (ACK) is sent to the transmitter. The only way to detect the collision is if the transmitting node does not receive ACK. Further, collisions are avoided by the exponential backoff process. Each node maintains a contention window CW from which it selects a random backoff time k ∈ (0, CW −
1) and waits for k time units before transmission. If another collision occurs, the node doubles the contention window in order to reduce the likelihood of further collisions.
Full duplex CSMA/CD, which is used by the IEEE 802.3 MAC protocol for Ethernet, stands for carrier sense multiple access with collision detection. In CSMA/CD, the node also performs sensing to check channel condition prior to transmission. When a node is transmitting and senses another transmission, it will immediately stop the transmission and waits for a random time interval before starting the transmission again. The advantage of CSMA/CD is that it reduces the time of a collision and also the time it has to wait before retransmission. With the full-duplex communication possible in IBFD systems, CSMA/CD can be used as the MAC protocol.
In CSMA/CA, collision detection is not possible as the transmitter cannot detect that the packet is successfully received by listening to its own transmission. By updating the MAC protocol for IBFD communication to CSMA/CD, we not only enhance the network capacity but we can also detect the collisions earlier. However, detection of a collision is not perfect due to imperfect self-interference cancellation in IBFD communication. Although a lot of work is presented over the years to cancel SI but still residual self-interference (RSI) exists. RSI affects the sensing performance and results in imperfect collision detection, i.e., false collision detection as well as miss collision detection. When a transmitting node mistakenly senses another transmission and aborts its own transmission, this is called a false alarm. When at least a single transmitting node involved in a collision is unable to detect the collision, it results in miss collision detection. Both of them affect the performance of the protocol in terms of throughput, delay, and packet loss. Due to the false alarm, the node unnecessarily backoffs and increases its contention window, which results in wastage of resources, whereas miss detection results in a lost frame as nodes are unable to recover them.
The above mentioned problems lead us to analyze the performance of the CSMA/CD MAC protocol for IBFD communication. The purpose of the anal-ysis is to measure the effect of parameters such as packet length, false alarm
probability, miss detection probability, maximum number of retransmission and the number of nodes on goodput and packet loss probability. Other than that, the parameters such as false alarm and miss collision detection cannot be controlled, therefore, other above mentioned parameters are required to be optimized for maximizing the throughput and for minimizing the packet losses and the delay in transmission. In this thesis, a modified version of the Markov chain model is proposed for the CSMA/CD based MAC protocol for IBFD [7]. We have intro-duced the maximum retry limit and backoff freezing into the Markov chain based model. Instead of the channel utilization considered in [7], a more realistic perfor-mance metric, goodput is used. Goodput only considers the time intervals used for successful transmissions. A detailed analysis of the Markov chain, goodput and packet loss probability are performed and a more realistic miss collision de-tection approach is implemented. The validity of the analytical model is carried out by comparing the simulation and the analytical results.
The rest of the thesis is organized as follows. Literature review for full duplex communication dealing with both physical and MAC layer issues are presented in Chapter 2. Chapter 3 contains the Markov chain model for CSMA/CD for IBFD, analysis of the Markov chain, and the derivation of goodput and packet loss probability. Detailed numerical and simulation results are presented in Chapter 4. The thesis is concluded in Chapter 5.
Chapter 2
In-Band Full Duplex Wireless
Communication
In this chapter, we will overview IBFD, the issues in the physical and MAC layers, and solutions proposed. Physical layer issues include self-interference mitigation, capacity enhancement, and the reliability of data transmission. While designing MAC layer protocols for full duplex communication, for both infrastructure based and ad hoc networks, many issues, such as, inter-node-interference and channel utilization are discussed.
2.1
Physical Layer Issues
The major problem in full duplex communication is self-interference and its mitigation. In [2], the authors have used analog cancellation, i.e., Balun (bal-anced/unbalanced) transformer for signal inversion. When the received signal is combined with the inverted signal (negative of the transmitted signal) after balancing the delay and attenuation of the inverted signal, self-interference is assumed to be completely canceled theoretically. The self-interference cannot be perfectly canceled as there are some limitations such as balun imperfections
and implementation difficulties in applying delay and attenuation on the inverted signal same as that of the transmitted signal. However, results show that 45dB suppression can be obtained and it results in a 73dB reduction in SI when com-bined with digital domain cancellation.
Sahai et al. proposed a real-time full duplex physical layer (FD-PHY) [3]. The FD-PHY presented is OFDM based and in order to combat self-interference, active analog cancellation is applied before the analog-to-digital converter. This suppression technique is applied on a per subcarrier basis. The authors have car-ried out experiments to find optimal antenna placement as well and showed that 80dB of self-interference can be attenuated which shows that FD communication can be implemented on mobile devices.
Ali et al. proposed an interference management and resource allocation mech-anism for full duplex infrastructure based D2D communication for cellular net-works considering residual self-interference [8]. A power control technique is used in order to reduce the interference on the base station by D2D nodes based on a threshold which ensures maximum transmit power. Interference limited area method is used to limit interference from cellular users to D2D nodes. In the end, resource allocation is done in which few or one resources are shared in order to maximize the throughput. The advantage of D2D communication is that nodes can communicate directly without routing through a base station and energy is also conserved. However, the whole mechanism is based on the assumption that BS knows the location of the users.
Bliss et al. studied the effects of channel estimation errors on full duplex MIMO radios under some assumptions for full duplex relays [9]. Authors have used adaptive transmit processing to protect the receive antenna array from SI and also calculated lower bound on self-interference mitigation due to channel estimation error. The simulation results and analytical results are demonstrated but on the other hand, the interference exists already in MIMO also needs to be explored and is neglected here.
networks in order to solve inter-node interference and obtained rate gain over half duplex communication in terms of degree-of-freedom (DoF) [10]. The full duplex network comprises of multiple-input-multiple-output (MIMO) base station which is communicating with many nodes and some assumptions are made which in-cludes perfect SI cancellation. The performance is studied for fixed node settings and spatial randomness is neglected.
Ju et al. discussed the resource problem associated with full duplex relay communication systems [11]. Authors have considered a two-way relay (TWR) and full-duplex relay (FDR) systems in order to solve resource wastage problem. In TWR, all antennas and nodes operate in half-duplex mode while in FDR, some nodes operate in half-duplex mode and some in full duplex mode. It is also assumed that each node has one antenna. They have mentioned different TWR and FDR systems, their pros and cons, frame structure and challenges. In TWR, multiple access channel (MAC) and broadcasting channel (BC) are formed which results in a change in the frame structure. For FDR, a protocol is presented classified on the basis of time resource utilization. Both relay systems increase the resource utilization efficiency and the FDR outperforms TWR in terms of sum rate. Another advantage of relay systems is that it reduces the end-to-end delay.
In [12], Zhang et al. presented a full duplex power allocation scheme for 5G mo-bile wireless networks. Authors have also analyzed the self-interference mitigation model. They have jointly applied propagation-domain interference suppression (PDIS), the analog-domain interference cancellation (AIC) and digital-domain interference cancellation (DIC) technique at various stages of the receiver. The proposed optimal FD power allocation scheme is obtained by optimizing spec-trum efficiency which is based on self-interference mitigation model. The author used FD multiplexing MIMO two-node bidirectional and three-node unidirec-tional transmission for the power allocation scheme.
Wang et al. used stochastic geometry to model full duplex communication [13]. Authors showed that in a large scale wireless network, the aggregate interference is more severe due to more transmissions in the same network density as half duplex
communication. Thus, the capacity gain is limited for full duplex. Only in low-density networks, FD outperforms HD. They also proved that self-interference cancellation does not guarantee that full duplex is extendable.
2.2
MAC Layer Issues
Kim et al. also surveyed many MAC layer protocols for IBFD communication which provides a complete insight of work done over the past years [1]. Future challenges on MAC protocol are presented. A centralized MAC protocol is also presented which uses busy tones in order to eliminate the hidden node problem in an asymmetric transmission which means no handshaking process is required [2]. The proposed protocol not only increases the fairness but also throughput. WARP V2 platform is used for its implementation. The simple case of two nodes transmitting is considered, and it is assumed that nodes do not experience interference.
Sahai et al. also proposed a centralized FD-MAC protocol which is based on three new operations shared random backoff, header snooping and virtual backoff [3]. In order to exploit full duplex transmission opportunities, FD header is added in each packet. Therefore, it increases the overhead. It also does not guarantee fairness. The throughput of the proposed protocol is 70% higher than the half-duplex communication. Authors have also made its real-time implementation for two nodes using the WARP platform.
A distributed full duplex MAC protocol is proposed for decentralized FD com-munication, which enables the nodes not only to transmit but also sense the chan-nel continuously and if a collision happens, it aborts the transmission and long collision is avoided as in CSMA/CD [7]. However, due to residual self-interference (RSI), collision detection may result in missed detection as well as false alarms. They are using Markov chain (MC) to carry out analysis and number of assump-tions are used such as ideal channel condition, unbounded transmission attempts, constant collision probability, and saturated condition. Also, backoff freezing is
not considered.
Zhang et al. also proposed full duplex MAC protocol in order to minimize the collision among FD nodes along with full duplex power allocation scheme [12]. By jointly optimizing them, throughput and spectrum efficiency is maximized. The MAC protocol is used for both bi-directional and unidirectional transmission and employs RTS/full-duplex clear-to-send (FCTS) packets for the handshaking process which also solves the hidden node problem.
Song et al. also presented a full duplex cross-layer protocol design for CSMA/CD [14] based on the same working principle as in [7]. In [7], the sens-ing error probabilities are fixed values while in [14], residual self-interference and physical characteristics were taken into account and error probabilities were de-rived based on them. As a result, throughput is improved over the conventional half duplex scheme as collision length is reduced [7, 14].
Relay Full-Duplex MAC (RFD-MAC) is proposed by Tamaki et al. [15]. The nodes use the CSMA/CA protocol and the winner becomes the primary transmit-ter. The primary transmitter than chooses the secondary transmitter from the same flow in order to prevent a collision, based on the surrounding node table. If the secondary node has a frame, it will also transmit. All the other nodes after hearing the transmissions also construct a surrounding node table. The table is built on the basis of the 1-bit information about the successive frame in the MAC header. With the help of a MAC header, all the nodes choose a communication link and thus increasing full duplex transmission opportunities. The simulation results show performance metrics, i.e., end-to-end throughput and FD ratio has increased as it supports both bi-directional and relay communication. Authors have only tested the algorithm on random network topology.
Listen-and-Talk (LAT) protocol has been introduced using the full-duplex technique for cognitive radios in which secondary users (SUs) simultaneously sense and transmit data [16]. Conventional Listen-before-talk (LBT) is also dis-cussed in it and a switching scheme is proposed for LBT and LAT which is adaptive in order to maximize the throughput because the performance of LAT
is limited by self-interference. Hence, a power-throughput tradeoff exists. Also, authors have considered one case in which all miss detection probabilities of the SUs are equal to the interference constraint of the PU. It lacked optimization of detection threshold in order to maximize the throughput and sensing results.
Goyal et al. proposed a distributed MAC protocol for full duplex communica-tion which takes care of traffic condicommunica-tions, contencommunica-tion and inter-node interference while making a simultaneous transmission between two or three full-duplex nodes [17]. Authors have introduced full duplex acknowledgment (FDA) which is a two bit signal to notify about the type of transmission. The three node full duplex transmission is decided based on signal-to-interference ratio (SIR). The proposed protocol can be used for both infrastructure and ad hoc networks and busy tones are used to notify the neighboring nodes but during header snooping, data trans-mission can not occur.
Energy efficient FD MAC protocol is presented by Al-Kadri et al. for dis-tributed wireless networks [18]. The protocol supports FD bi-directional, FD-unidirectional and also HD communication, i.e., it is backward compatible. It uses the RTS-CTS access method. The nodes involved in full duplex transmis-sion use FD clear-to-send (FD-CTS) while for HD, regular CTS packet will be used. The nodes calculate Pmin which is minimum power required for successful
transmission. They start transmission with Pmin and increase it to Pmax in order
to ensure no hidden node problem exists. The simulation results demonstrate im-proved throughput and lower energy consumption but at the same time, it highly depends on the self-interference cancellations technique adopted in the physical layer.
Kim et al. proposed a centralized full duplex MAC protocol for D2D dense traffic using spatial channel reuse algorithm [19]. The protocol is based on five phases in which AP schedules all the transmission based on the location of nodes which in returns minimize the scheduling overhead. It can also solve the hid-den/exposed node problem and throughput is also improved as compared to half duplex communication. However, a centralized algorithm is difficult to implement in a wireless network and is useful in less dense wireless networks.
Zuo et al. proposed a novel distributed medium access control protocol for full duplex communication which supports both symmetric (two nodes) and asym-metric (three nodes) dual link with the help of one channel access [20]. The mechanism not only increases IBFD transmission opportunities but also solves hidden node problem. All transmission modes, throughput, and delay analysis are explained in detail. Theoretical analysis and simulation results show that the throughput of proposed protocol almost gets double as compared to HD communi-cation using RTS/CTS process based on some assumptions which include perfect self-interference and inter-node interference cancellation, and channel status and all other node status is detected by each node.
Chan developed a full duplex MAC protocol for cellular networks in which a frame structure is proposed which consists of subframes that govern the commu-nication between user equipment and the base station [21]. An analytical model is presented for the mean and second moment of packet delay and utilization of the MAC protocol and the accuracy is measured by simulation results. The analytical model and simulation results match perfectly. Authors have not com-pared the proposed full duplex MAC protocol with already existing protocols. They concluded that with appropriate choice of parameters, utilization can be enhanced and with full duplex, the delay will be reduced notably.
2.3
Contributions of the Thesis
In this thesis, we have used the Markov chain model for full duplex CSMA/CD MAC protocol for FD communication networks [7] and made some modifications in the Markov chain model to improve it. The performance metrics considered are goodput and packet loss probability. Both simulation and analytical analysis are performed in order to validate the model.
Firstly, we have introduced a maximum retry limit in the Markov chain which already exists in half duplex IEEE 802.11 MAC protocol [22]. When a packet after continuous collisions reach the maximum retry limit value and if another
collision happens, the packet will be dropped and the next packet in the queue will be served. The reasons for adding maximum retry limit are that it complies with the existing standard and it prevents head-of-line blocking. HOL blocking occurs as the packet can be transmitted an unbounded number of times before its successful reception. Delay is reduced with the retry limit while an increase in packet loss occurs.
Secondly, we have added the backoff freezing in the Markov chain model as implemented in the IEEE 802.11 MAC protocol [22]. When a node has a packet to transmit while it is in the backoff process and it senses that the channel is busy, it freezes its counter. If the channel is free, the backoff timer of that node decrements by 1. The effect of backoff freezing will be more prominent on goodput when the number of nodes contending for the channel access increases.
In this thesis, we have investigated goodput instead of throughput which com-prises the time that the channel is occupied with a successful transmission, while throughput corresponds to simply channel utilization [7]. Goodput gives a more realistic performance metric for the MAC protocol.
In [7], miss collision detection occurs only when two nodes starts the transmis-sion in the same slot. Only two cases of miss detection were considered, which are either both nodes will detect the ongoing collision in the same slot or both of them will not detect collision for entire packet length. A more realistic miss collision detection implementation is incorporated into the Markov Chain based analytical model proposed in this thesis. Along with the first two cases, two more events can take place: any one of the two nodes detects the collision and aborts it immediately while the other node continues to transmit for the full packet length as it cannot detect the collision.
Packet loss probability Ploss is derived which means the packet cannot be
recovered although it has not been delivered. It is based on two circumstances which are when a packet cannot be transmitted in the maximum number of retransmission tries and when two nodes start transmission in the same slot and collision goes undetected by at least one of them.
We used a saturated traffic model throughout the analysis which means that each node always have a packet to transmit, i.e., the queue is non-empty. This will help us to study the maximum load that the network can support in stable condition.
Based on the contributions stated above, modified Markov chain model for IBFD-MAC protocol is presented in the next chapter. To derive and calculate the goodput and packet loss probability of the IBFD-MAC protocol, derivation of different parameters are also shown in the next chapter.
Chapter 3
Markov Chain Model for
CSMA/CD for IBFD
This chapter contains the modified Markov chain model for full duplex carrier sense multiple access with collision detection (CSMA/CD) protocol. It intro-duces the improvements made in the analytical model in [7] and also contains the analysis of model and assumptions. Calculation of the goodput and the packet loss probability are also presented here.
3.1
Introduction of Full Duplex MAC Protocol
Design
Liao introduced a medium access control (MAC) protocol for full duplex (FD) communication [7]. In this protocol, a node senses the channel while transmitting and aborts the transmission as soon as a collision is detected. The long colli-sions will be avoided as in carrier sense multiple access with collision detection (CSMA/CD). In other words, the blindness of nodes during transmission is elimi-nated. All nodes can detect the change in channel state quickly and therefore, can change their states. Channel utilization is efficient but residual self-interference
(RSI) causes more error. The detection of the collisions can never be perfect due to RSI that results in a false alarm and miss detection. Generally, these errors increase with an increase in RSI.
A false alarm occurs when a single node transmits and senses incorrectly that a collision occurs with another transmission. The node immediately stops the trans-mission, considers it a collision and increases the contention window. Whereas miss detection takes place when two nodes start the transmission in the same time slot but at least one node cannot detect the collision due to imperfect sensing. If nodes cannot detect the collision during the entire transmission, they will con-sider it a success and reduce the contention window to initial contention window. It can be concluded that false alarm wastes the channel slots and miss detection results in a lost frame. Both of these sensing errors reduces the performance of the protocol, therefore care must be taken.
Figure 3.1: System Model for IBFD-MAC protocol in which uplink traffic is considered only
Figure 3.1 shows a full duplex Wi-Fi network which consists of an access point (AP) and M nodes (U1, U2, U3, ..., UM) which are randomly and independently
distributed in the coverage area of the access point. Each node will always have a packet to transmit and only uplink data transmission is taken into consideration, i.e., data packets are transmitted from node to AP. Also, each node has two antennas and simultaneously uses one antenna for data transmission and one for sensing. At most one node can access the channel at a time, otherwise, a collision
will occur. Each node performs carrier sensing before transmission to detect channel condition and contend using the backoff procedure. The self-interference after suppression technique still exists between two antennas as shown in Figure 3.1.
The list of the parameters used in this chapter is given in Table 3.1. CWmin Initial contention window
CWmax Maximum contention window
W Number of retransmission attempts for each packet Wmax Maximum number of retransmission attempts for each packet
ps Probability that at least a successful transmission
occurs without the awareness of collision
p Probability that a certain node begins transmission in the next slot Pb Probability of a busy channel
M Total number of nodes
L Packet length
Pf Probability of false alarm per slot in case of collision
Pm Probability of miss detection per slot in case of collision
Table 3.1: List of parameters used throughout this chapter
3.1.1
Improvements in the Markov Chain based
IBFD-MAC Model
The enhancements made in the analytical model presented in [7] are:
1. In [7], there is no upper bound on transmission attempts. It means that when a packet reaches CWmax, and another collision happens, it will keep
its contention window to CWmax and will make another transmission. This
process is repeated unless the packet is successfully received to the desti-nation. It can cause head-of-line blocking, in which a line of packets is jammed by the first packet in the queue. We introduced a maximum retry limit which is also used in the half duplex IEEE 802.11 MAC protocol [22]. If a collision happens in CWmax, the packet is discarded and a new packet
2. With probability 1, the backoff counter decrements without taking into account channel condition in [7]. We introduced the probability of a busy channel Pb into the model for invoking the backoff procedure, i.e., backoff
freezing. We assume that Pb is constant and it does not depend on the
backoff procedure. The channel is free with probability 1 − Pb and it goes
to the next state while with probability Pb channel is busy and it remains
in the same state.
3. [7] considers channel usage, whether the transmission is successful or not as throughput. We consider the time that the channel is used for successful transmission, which is called Goodput. For goodput, we calculated the probability of successful transmission.
4. For modeling the collision detection in [7] when two nodes transmit, only two cases were considered: either both of them detect the collision or both of them will not be able to detect. We considered all possible cases which will be explained further in the next section. The four cases are:
(a) Both nodes detect the collision.
(b) Any of the two nodes will detect the collision, for example, Node 1 detects the collision and Node 2 continues the transmission.
(c) Node 2 detects the collision during transmission and Node 1 will con-tinue the transmission.
(d) Both nodes are unable to detect the collision for the entire packet length.
3.1.2
Model Assumptions
The assumptions considered for IBFD based MAC protocol are [7]:
• A saturated traffic condition is considered. Each node will always have a packet for transmission in the queue. With such traffic patterns, we can see the maximum load that the network can maintain in stable condition.
• Imperfect sensing is taken into account only because of residual self-interference. Effect of thermal noise on sensing is neglected for simplicity and a fair comparison with [7]. Therefore, we assume that silent nodes have perfect sensing.
• Collision detection is considered perfect when more than two nodes collide. This is because the received collided signal is much stronger than residual self-interference. Thus, sensing error exists only in two cases: when a single node transmits but a false collision is detected, or when two nodes start transmission in the same slot and the collision is not detected by at least one node.
• Channel is ideal, i.e., hidden terminal, capture effect, and channel errors are not taken into account for simplifying the model.
• The key assumption is the constant probability of a successful transmis-sion without the awareness of collitransmis-sion, ps. This is because it is assumed
that the packet gets collided with the same probability regardless of the number of transmissions and it is independent of contention window value. The probability of the busy channel Pb is also constant based on the same
approximation. This independence is assumed for the mathematical con-venience in order to be able to model the protocol as a bi-dimensional discrete-time Markov Chain.
3.1.3
IBFD-MAC Protocol Design
In this section, the detail description of IBFD-MAC protocol is presented. The protocol works as CSMA/CD. All M-contending nodes will sense the channel continuously regardless of their own actions. The process is assumed to be time slotted and energy detection is used as a sensing technique. The contending nodes will make a decision at the end of each slot time. The slot time is set equal to an appropriate time in which a node can detect transmission from any other node.
of distributed inter-frame space (DIFS), it will attempt a transmission. Before transmission, it will generate the backoff interval and the backoff process is binary exponential. The node will select a random backoff time which is uniformly distributed in [0, CWi− 1], where CWi is the contention window and it depends
on the number of unsuccessful transmissions. Each node has a fixed number of transmission attempts for each packet given as W and it varies from [0, Wmax]. For
the first transmission attempt, the contention window is set to CWmin, i.e., initial
or minimum contention window. When a collision occurs, the contention window of each node doubles unless it reaches the maximum contention window CWmax =
2WmaxCW
min, where Wmax are the maximum number of transmission attempts
and each of them gets a random backoff number for which they wait before transmission. This is the collision avoidance mechanism. If another collision happens at this stage, the packet is discarded and the new packet will start its transmission with contention window set to CWmin and same backoff process is
repeated. Also, when a packet is successfully delivered, the contention window is set to CWmin.
When a single node is transmitting, all other nodes will be able to detect its transmission, i.e., when a collision-free transmission begins, it is either successful or aborted due to a false alarm. When there are more than two transmitting nodes, the collision is detected by all of the nodes within DIFS period as they are sensing continuously along with the transmission. For consecutive packet transmission, all the nodes have to go through random backoff mechanism to avoid unfairness and channel capture. The backoff counter decrements as the channel remain idle, freezes as another transmission is detected on the channel and reactivated again as it is sensed idle for more than DIFS time. Lastly, the node will transmit when backoff counter reaches zero.
IBFD based MAC protocol is also demonstrated with the help of an example in Figure 3.2. Three nodes labeled as node 1, node 2 and node 3 are contending for channel and follow IBFD-MAC protocol. It starts with node 1 transmitting and completing it also.
Figure 3.2: IBFD-MAC Protocol
2 won. During the transmission as it is sensing the channel continuously, it will mistakenly sense a collision, i.e., a false alarm has occurred and will stop the transmission and increases its contention window. The same backoff process is repeated and both node 1 and node 3 will start transmission in the same slot. Due to residual self-interference, sensing is imperfect and both of them will not detect the ongoing collision for full packet length and considers it a success. They will set contention window to CWmin. This is an example of miss detection. Node 1, node
2 and node 3 again start transmission in the same slot after backoff procedure and at this time they are able to detect the collision within DIFS period and will stop the transmission along with increasing contention window. It highlights the main advantage of the protocol as collision does not last for the full packet length as happens in conventional half duplex communication. Lastly, after DIFS time and backoff process, node 2 and node 3 will start the transmission in the same slot and now node 3 during the transmission will sense the collision and at once it will stop the transmission. While node 2 will continue the transmission as there is no way for it to detect the collision now. Node 3 will increase the contention window while node 2 will consider it a success and set its contention window to the initial contention window. This is a more common case of miss detection.
3.2
Markov Chain Model and Analysis
The modified bidimensional discrete-time Markov chain is shown in Figure 3.3 where the state of a node is represented by {wi, Wi} and wi is the backoff
num-ber selected randomly from the interval [0, CWi − 1] and Wi is the number of
transmission attempts so far. The relation among these variables can be written as
CWi = 2Wi CWmin and wi = Random(CWi)
Figure 3.3: Discrete Time Markov Chain Model
In Figure 3.3, psis the probability that a successful transmission occurs without
the awareness of collision and CWi represents contention window. The packet
starts its transmission from state {0,0}, if a collision occurs, it will move to next state with probability (1−ps)/CWi and if its a success, it will again go back to the
states Wi = 0 with probability ps/CW0. Pb represents the probability that the
channel is busy. The backoff counter will be frozen when the channel is detected busy and it will stay in the same state with probability Pb, i.e., self-transitions.
When the channel is free, with probability 1 − Pb it will transit to next state. If a
packet gets collided continuously and reaches Wmaxstates, the packet will be lost
if the next retransmission attempt fails and the Markov chain will again transit to states Wi = 0 for the transmission of the next packet.
The one-step state transition probabilities are P (k, i|k + 1, i) = 1 − Pb k ∈ [0, CWi− 2], i ∈ [0, Wmax] P (k, i|k, i) = Pb k ∈ [1, CWi− 1], i ∈ [0, Wmax] P (k, 0|0, i) = ps/CW0 k ∈ [0, CW0− 1], i ∈ [0, Wmax] P (k, i|0, i − 1) = (1 − ps)/CWi k ∈ [0, CWi− 1], i ∈ [1, Wmax] P (k, 0|0, Wmax) = 1/CW0 k ∈ [0, CW0− 1] (3.1)
The first equation in (3.1) shows that backoff time is decremented when the channel is idle. Second equation shows the self-transition when the channel is busy. The third equation accounts the fact that a successful transmission has taken place and a new packet will now be transmitted with the backoff stage 0 and a backoff number from [0, CW0 − 1]. The fourth equation shows an unsuccessful
transmission and the fifth equation takes into account the retry limit.
Let bk,i = limt→∞ P {b(t) = k, s(t) = i} represents stationary distribution of
the chain where b(t) is backoff time counter and s(t) represents the number of transmission attempts. The global balance equation for this Markov chain which will be used to obtain a closed form expression are written as
(1 − ps)b0,i−1 = b0,i b0,i= (1 − ps)ib0,0 , f or 0 ≤ i ≤ Wmax (1 − ps)b0,Wmax−1 = b0,Wmaxps b0,Wmax = (1 − ps)Wmaxb0,0 ps (1 − Pb)bk,i = bk−1,i bk,i= 1 (1 − P )bk−1,i, f or 0 ≤ i ≤ Wmax 1 ≤ k ≤ CWi− 1 (3.2)
As the chain has regularity, so for all k ∈ (1, CWi− 1), we can write bk,i = CWi− k CWi Wmax−1 P j=0 psb0,j + b0,Wmax i = 0 (1 − ps)b0,i−1 0 < i ≤ Wmax (3.3)
By using the set of equations in (3.2) and the following fact
Wmax−1 X j=0 psb0,j+ b0,Wmax = b0,0 (3.4) We can write (3.3) as bk,i= CWi− k CWi 1 1 − Pb b0,i 0 ≤ i ≤ Wmax , 1 ≤ k ≤ CWi− 1 (3.5)
Pb is the probability that the channel is busy and it can be expressed as
Pb = 1 − (1 − p)M −1 (3.6)
p and M in (3.6) are the probability that a certain node begins transmission in the next slot and the total number of nodes contending for the channel, respec-tively. With the help of (3.2) and (3.5), all the values of states can be written in terms of b0,0. As the sum of all the state probabilities must be equal to 1 and
is given by 1 = Wmax X i=0 CWi−1 X k=0 bk,i (3.7)
By solving (3.7), we can find b0,0
b0,0 =
2(1 − Pb)(2ps− 1)ps
(2ps− 1)(1 − (1 − ps)Wmax+1) + CWmin(1 − (2 − 2ps)Wmax+1)ps
(3.8)
= 2(1 − p)
M −1(2p
s− 1)ps
(2ps− 1)(1 − (1 − ps)Wmax+1) + CWmin(1 − (2 − 2ps)Wmax+1)ps
(3.9)
Now p which is the probability that a certain node begins transmission in the next slot can be derived. The mathematical expression of p can be given as
p = Wmax X i=0 b0,i (3.10) = Wmax X i=0 (1 − ps)ib0,0 (3.11) = 1 − (1 − ps) Wmax+1 ps b0,0
Replacing b0,0 by equation (3.8), we get
p = 1 − (1 − ps)
Wmax+1
ps
2(1 − Pb)(2ps− 1)ps
(2ps− 1)(1 − (1 − ps)Wmax+1) + CWmin(1 − (2 − 2ps)Wmax+1)ps
By solving it, the final expression for p is
p = 2(1 − Pb)(2ps− 1)(1 − (1 − ps)
Wmax+1)
(2ps− 1)(1 − (1 − ps)Wmax+1) + CWmin(1 − (2 − 2ps)Wmax+1)ps
(3.12)
= 2(1 − p)
M −1(2p
s− 1)(1 − (1 − ps)Wmax+1)
(2ps− 1)(1 − (1 − ps)Wmax+1) + CWmin(1 − (2 − 2ps)Wmax+1)ps
(3.13)
Successful transmission psrequires at least one node transmits the entire packet
without the awareness of collision. It depends on the length of packet L, the num-ber of nodes M , transmission probability p, the probability of false alarm per slot Pf and probability of miss detection per slot Pm. False alarm and miss detection
are the results of imperfect sensing due to residual self-interference. When a node falsely senses a collision during the transmission is called false alarm and it aborts the transmission immediately. When two nodes start to transmit in the same slot and the collision is not detected by at least one of them is called miss detection. With these parameters, ps can be written as [7]
ps = (1 − p)M −1(1 − Pf)L+ (M − 1)p(1 − p)M −2Pm
(1 − Pf)L− Pm2L
1 − Pf − Pm2
(3.14)
ps is different from non-collision probability. It should be noted that when two
and when one collision-free node is transmitting, it may abort the transmission due to the false alarm.
Equation (3.13) and (3.14) form a fixed point problem of nonlinear equations and they can be solved iteratively to obtain p and ps which are further used
in the analysis part. We used the intersection method to solve them. For equa-tion (3.13), all terms containing p are shifted to the left-hand side of the equaequa-tion. The left-hand side is solved for all the values of p varying from 0 to 1 correspond-ing to each value of ps which will be used to solve the right-hand side. Then,
an absolute error between left-hand and right-hand side is calculated and that value of p is chosen which gives the minimum error. Solving equation (3.14) is straightforward. The intersection point is obtained by plotting both equations.
3.2.1
Goodput Analysis
Goodput is defined as the time channel is occupied for the successful transmission and it can be formulated as
G = E[ Successf ul transmission length]
E[ T ime consumed f or a successf ul transmission] (3.15)
For the numerical calculation of goodput, the parameters required are listed in Table 3.2.
PS Probability of a successful transmission occurrence
PE Probability that the channel is empty
PC Collision probability
LS The average length of successful transmission
LC The average length of collision
DIF S Distributed inter-frame space
Table 3.2: List of parameters for goodput calculation
By using the parameters defined in Table 3.2, (3.15) can be written as
G = PSL
PE + PS(LS+ DIF S) + PC(LC+ DIF S)
The parameters in Table 3.2 can also be written mathematically. PS = M p(1 − p)M −1(1 − Pf)L−1 (3.17) PE = (1 − p)M (3.18) PC = 1 − PE − PS (3.19) LS = 1 − (1 − Pf)L−1 Pf + (1 − Pf)L−1 (3.20) LC = PC + M 2 p2(1 − p)M −2 L−1 X l=1 Pm2l(1 − Pm2)l ! /PC = 1 +M 2 p2(1 − p)M −2P 2 m(1 − Pm2L−2) PC(1 − Pm2) (3.21)
Equations (3.20) and (3.21) show that LS and LC are inversely and directly
proportional to Pf and Pm, respectively. The goodput can be calculated by
solving equations (3.13) and (3.14) for PC, PE, PS, LS and LC described above
and using them in (3.16).
3.2.2
Packet Loss Probability Analysis
Ploss is the probability of packet loss. Packet loss means that the packet cannot
be recovered although it has not been delivered. Loss of a packet occurs under two circumstances:
1. Pmrl : when a packet cannot be transmitted in the maximum number of
2. Pmiss : when two nodes start transmission in the same slot and at least
collision goes undetected by one of them.
3.2.2.1 Pmrl derivation
Pmrl can be derived based on the fact that the packet will be lost when the retry
limit has reached.
Pmrl = (1 − ps)b0,Wmax b0,0 where b0,Wmax = (1 − ps) Wmaxb 0,0
By using the above expression, Pmrl can be written as
Pmrl = (1 − ps)Wmax+1 (3.22)
(3.22) shows that a packet is lost when it undergoes Wmax+ 1 collisions.
3.2.2.2 Pmiss derivation
Miss detection only occurs when there is a single interference. If the collision involves more than two nodes, it will be detected. When two nodes start the transmission in the same slot, four scenarios can take place:
1. Both nodes detect the collision.
2. Node 1 detects the collision and stops the transmission while Node 2 con-tinues to transmit.
3. Node 2 detects the collision and stops the transmission while Node 1 will continue the transmission.
For Pmiss, we will consider the cases when at least a collision by any one of them
is not detected. The derivation is explained stepwise.
Pr{exactly one node starts transmission in
the f irst slot while a node is transmitting} = (M − 1)p(1 − p)M −2 Pr{the collision is not detected by both nodes} = Pm2L
Pr{one node detects the collision and stops transmission} = L−1 X l=1 Pm2l−1(1 − Pm) = P 2 m− Pm2L P2 m+ Pm
Now by combining the above probabilities, we can write Pmiss as
Pmiss = (M − 1)p(1 − p)M −2 " Pm2L+ P 2 m− Pm2L P2 m+ Pm # (3.23)
Ploss is sum of Pmrl and Pmiss.
Ploss = Pmrl+ Pmiss Ploss= (1 − ps)Wmax+1+ (M − 1)p(1 − p)M −2 " Pm2L+P 2 m− Pm2L P2 m+ Pm # (3.24)
In order to check the accuracy of the modified analytical model, it needs to be verified by simulations. In the next chapter, both analytical and simulation results are presented which shows the accuracy of the model in terms of goodput and packet loss probability.
Chapter 4
Numerical Results of the
Proposed Protocol
In this chapter, simulation results and analytical results are evaluated for the IBFD-MAC protocol. With each enhancement made in the model which are retry limit and the probability of busy channel, analytical results are calculated and compared with the simulation results. Goodput is evaluated against many parameters which include initial contention window, packet length, false alarm probability, and the number of nodes. Packet loss probability is also examined in detail. In the end, simulation results of HD CSMA/CA and IBFD CSMA/CD are compared to see the percentage increase in goodput. The simulation environment is implemented in MATLAB. The duration of each simulation is 107 slot time
which is long enough to develop the system and to obtain smooth results.
4.1
Flowchart
The flowchart in Figure 4.1 explains the calculation of goodput and packet loss probability done in simulation in detail.
Figure 4.1: Flowchart of simulation working
4.2
Goodput Evaluation
Goodput is the portion of throughput that considers the time a packet occupies the channel for successful transmission. It demonstrates the actual output of the protocol. First, throughput and goodput comparison is shown, and then the results of each modification in the model are given. After that, the optimal packet length is obtained. In the end, the effect of false alarm and the number of nodes is also studied.
4.2.1
Throughput and Goodput Comparison
Throughput and goodput are plotted against minimum contention window and length of packet. The parameters used for analytical and simulation results are CWmax = 215, M = 100, Pf = 10−3, Pm = 10−2 and DIFS=2.
2 4 6 8 10 12 14 CW min 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1 Throughput and Goodput Comparison
Throughput:Simulation Throughput:Analytical Goodput:Simulation Goodput:Analytical
Figure 4.2: Throughput and Goodput vs. Initial Contention Window for L = 1000
Figures 4.2 and 4.3 show the comparison of throughput and goodput against initial contention window and length of packet respectively and in both cases, throughput is higher than goodput. The former takes care of all the time slots used for transmission which includes
• time for successful transmission
101 102 103 L 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1 Throughput and Goodput Comparison
Throughput:Simulation Throughput:Analytical Goodput:Simulation Goodput:Analytical
Figure 4.3: Throughput and Goodput vs. Length of Packet for CWmin = 24
• time before false alarm occurrence
• time before the collision is detected or goes undetected when two users are transmitting
while goodput considers the time for successful transmission only. It is intuitive that throughput is going to be high as compared to goodput while keeping in mind that goodput is the real output of the system. The reason behind the decrease in throughput and goodput in Figure 4.2 is as initial contention window is increased, the nodes get larger backoff numbers and they will wait longer on the average before attempting a new transmission, i.e., channel wastage has increased. Furthermore, the maximum retrial count decreases as CWmin increases.
The behavior of throughput and goodput are different in Figure 4.3. Through-put is increasing with the increase in packet length asymptotically while goodThrough-put
first increases but as packet length increases beyond 100 slots, goodput begins to decrease. This is due to the increased false alarm, as packet length increases the chances of occurrence of false alarm also increases.
4.2.2
Introducing Retry Limit and Probability of Busy
Channel
In this section, retry limit and probability of a busy channel are introduced in the analytical model to see the effect on goodput. The goodput is checked against initial contention window and packet length in Figures 4.4 and 4.5 respectively. The parameters are listed in Table 4.1 for these figures.
List of parameters
Parameters Figure 4.4 Figure 4.5 CWmax 215 215
CWmin − 24
M 100 100
Pf 10−3 10−3
Pm 10−2 10−2
DIFS 2 slots 2 slots
L 100 −
Table 4.1: Parameters used for simulation and anayltical results for Figures 4.4 and 4.5
All figures in this section have the following five curves:
• Simulation of IBFD-MAC protocol
• FD-MAC model in [7] (ORIG)
• Model after modifying the probability of a successful transmission occur-rence PS (PS)
• Model after adding retry limit (RL)
2 4 6 8 10 12 14 CW min 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Goodput/Throughput
Results after introducing RL and BF in FD-MAC
Simulation ORIG
with PS
with RL withRL-BF
Figure 4.4: Analytical results of each modification and simulation results are plotted with respect to CWmin
In Figures 4.4 and 4.5, the simulation is compared with the analytical model with each enhancement. It can be seen that with the probability of a successful transmission occurrence, i.e., PS, the change is significant in goodput in both
figures while RL and RL-BF models, does not improve it. The effect of maximum retry limit (RL) and backoff freezing (RL-BF) will be observed on p given by equation (3.13) as parameters used for goodput calculation depends on it. With RL and RL-BF models, the difference between analytical and simulation results of p reduce. As these values are small and they are nearly the same as the values of the model after modifying PS. Hence, the improvement in goodput is negligible.
In Figure 4.4, goodput falls as initial contention window increases, because more time is used in the backoff procedure. Figure 4.5 depicts effect of length, which is same as in Figure 4.3.
101 102 103 L 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Goodput/Throughput
Results after introducing RL and BF in FD-MAC
Simulation ORIG
with PS
with RL withRL-BF
Figure 4.5: Analytical results of each modification and simulation results are plotted with respect to L
4.2.3
Optimal Packet Length
Goodput is plotted in Figure 4.6 for different packet lengths against contention window to find optimal L where chosen L are 10, 100, 500, 1000. Here optimal length corresponds to the packet length that gives the maximum goodput. The parameters are CWmax = 215, M = 100, DIFS=2, Pf = 10−3 and Pm = 10−2.
The goodput increases as packet length L increases from 10 to 100, but it starts to decreases as packet length increases to L = 500 and L = 1000. It can be observed in Figure 4.6 that the goodput drops sharply for L = 10 as compared to other packet lengths after CWmin = 29. With L = 10, the number of transmission
in 107 slot time are large as compared to other lengths, i.e., the channel access
increases with L = 10. It means there will be more collisions. Also, as initial contention window is increased, goodput falls due channel wastage. Thus, the
2 4 6 8 10 12 14 CW min 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Goodput Goodput vs. CW min for L=10, 100, 500, 1000 L(sim)=10 L(RL-BF)=10 L(sim)=100 L(RL-BF)=100 L(sim)=500 L(RL-BF)=500 L(sim)=1000 L(RL-BF)=1000
Figure 4.6: Goodput vs. Initial Contention Window for different packet lengths
waterfall effect is observed on goodput for L = 10. For L = 100, goodput is highest, i.e., 90% for CWmin = 22 to 28. The number of transmissions as
compared to L = 10 reduces, which also reduces the collisions, in turn, increasing the goodput.
When packet length is increased further to L = 500, the goodput gets lower than L = 100. Transmissions are further reduced and thus collisions, but now the false alarm effect becomes more dominant. As the length of packet increases, the chances of false alarm increases. This is the reason for lower goodput as compared to L = 100. False alarm also affects L = 10 and L = 100, but is more prominent for L = 500 and L = 1000. With an increase in the contention window, the maximum number of retransmission reduces. The retransmission attempts at CWmin = 22 is 13 and reduces to 0 at CWmin = 215. At CWmin = 215, optimal
goodput is obtained with L = 500. As collisions for L = 10 and L = 100 are high as compared to L = 500, hence, at CW = 215, L = 10 and L = 100 will be
penalized more. The goodput with L = 1000 is nearly constant along the x-axis and lowest as compared to other packet lengths up to CWmin = 29due to excessive
false alarm. The number of transmissions and collision are low as compared to other packet lengths, therefore, drop in goodput is not sharp. Another important observation that can be drawn is as the packet length increases, the analytical and simulation results match better. Hence, the model is more accurate for larger packet lengths.
4.2.4
Effect of False Alarm Probability
The IBFD-MAC protocol is highly affected by the false alarm probability.
10-5 10-4 10-3 10-2 10-1 P f 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Goodput Goodput vs. P f for L=100, 1000 L(sim)=100 L(RL-BF)=100 L(sim)=1000 L(RL-BF)=1000
Figure 4.7: Goodput is plotted against false alarm probability for L = 100 and L = 1000
A false alarm occurs when a user wrongly judges another ongoing transmission and ceases its own transmission. It does not only halt its own transmission
but also considers it a collision, increases its contention window which further increases channel wastage and packet losses. In order to understand its behavior, Figure 4.7 can be seen. The parameter used for both results are CWmin = 24,
CWmax = 215, DIFS=2, Pm = 10−2 and M = 100. Two packet lengths L = 100
and L = 1000 are chosen for comparison. The goodput is higher when the Pf is
less than 10−4 even with L = 1000, but as probability starts to increase, goodput also decreases. The fall in goodput occurs earlier for L = 1000. This is because, for larger packets, the chances of false alarm appearance also increases. It can be deduced that a suitable packet length should be selected for IBFD-MAC protocol due to the false alarm which is the consequence of residual self-interference.
4.2.5
Effect of Number of Nodes
2 4 6 8 10 12 14 CW min 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Goodput Goodput vs. CW min for M=10, 50, 100, 150 M(sim)=10 M(RL-BF)=10 M(sim)=50 M(RL-BF)=50 M(sim)=100 M(RL-BF)=100 M(sim)=150 M(RL-BF)=150
Figure 4.8: Goodput is plotted with respect to initial contention window for the different number of nodes M = 10, 50, 100, 150
Goodput is also investigated with the number of nodes participating in channel contention. The graph is plotted against initial contention window for the differ-ent number of users M = 10, 50, 100, 150 and the simulation results are matched with analytical results. The parameters are CWmax = 215, DIFS=2, L = 100,
Pf = 10−3 and Pm = 10−2. The reason for selecting packet length as 100 is
that the goodput is highest, as compared to other packet lengths on parameters defined as shown in Figure 4.6. The goodput falls sharply as contention window is increased for less number of nodes. This is because the channel is not properly utilized. The channel wastage increases due to prolong backoff process. Whereas as contention window increases along with the number of nodes, the cumulative goodput also increases. It can be observed that we can accommodate even 150 nodes in the system model as shown in Figure 4.8. It highlights another major edge of the protocol. Even with false alarm and collisions, with M = 150, it gives higher goodput from CWmin = 29 as compared to M = 10, 50, 100,.
4.3
P
lossEvaluation
Packet loss probability is derived in Chapter 3 and given by equation (3.24). The two main causes considered in Ploss analysis are also derived which are Pmrl and
Pmiss. Pmrl results when a packet cannot be transmitted to the destination in
Wmax retransmission attempts whereas Pmiss occurs only when two nodes are
transmitting and any one of them or both of them are unable to detect the collision. Both of them are evaluated separately and the combined result is also given as Ploss. In this section, contention window, the number of nodes and length
of the packet are also used for numerical results.
The use of PS, i.e., the probability of a successful transmission occurrence will
not affect Ploss, so we will investigate the following four models throughout the
section:
• FD-MAC model in [7] (ORIG)
• Model after adding retry limit (RL)
• Model after implementing backoff counter freeze (RL-BF)
4.3.1
P
losswith Initial Contention Window
The parameters used for numerical results are M = 100, DIFS=2, Pf = 10−3 and
Pm = 10−2. Initial contention window is used as a base parameter in this section.
2 4 6 8 10 12 14 CW min 10-5 10-4 10-3 10-2 P miss P
miss vs. CWmin for L=100, 1000
L(Sim)=100 L(ORIG)=100 L(RL)=100 L(RL-BF)=100 L(Sim)=1000 L(ORIG)=1000 L(RL)=1000 L(RL-BF)=1000
Figure 4.9: Pmiss vs. CWmin
The increase in initial contention window means that retransmission attempts are reducing. The maximum contention window is fixed, i.e., CWmax = 215.
