Report

802.11ec :Collision AvoidanceWithout Control Messages

Submitted in partial fulfilment of the requirementsfor the degree of

Bachelor of Engineering(Computer Engineering)

by

Anish Kumar Jha

Roll no- 3310Exam no. T120224211

Under the Guidance ofProf. Nikita Gupta

Department of Computer EngineeringArmy Institute of Technology

(Savitribai Phule Pune University)Pune 411 015

2014-15

Army institute of Technology

Department Of Computer Engineering

CertificateThis is to certify that the seminar titled

802.11 ec Collision AvoidanceWithout Control Messages.

has been prepared and presented by

Anish Kumar Jha

Roll no.- 3310Exam no.- T120224211

of Third Year (Computer Engineering)in partial fulfilment of requirement for the award of

Degree of Engineering in Computer Engineering under theSavitribai Phule Pune University

during academic year 2014-15

SEMINAR GUIDE HOD(Prof. Nikita Gupta) (Prof. Sunil Dhore)

i

Acknowledgements

I would like to express my deep gratitude to my Guide, Prof. Nikita Gupta,Faculty of the Department of Computer Engineering, Army Institute of Technology,Pune for all the valuable guidance and intellectual stimuli that she provided duringthe progress of this work. It was a unique privilege to work under her valuable guid-ance and supervision.

I am grateful to other faculty members of Department of Computer Engineer-ing, Army Institute of Technology, Pune, for timely guidance and encouragement tocomplete this work.

I am indebted to Prof. Dhore, HOD of the Department of Computer Engineering,AIT for giving me an opportunity for my intellectual pursuit.

I also wish to thank the library staff and Computer Center for all the help andfacilities that they extended to me during this dissertation work.

ii

Abstract

In this paper, we studied the design, and evaluated 802.11ec (Encoded Control),an 802.11-based protocol without control messages: Instead, 802.11ec employs cor-relatable symbol sequences that, together with the timing the codes are transmitted,encode all control information and change the fundamental design properties of theMAC. The use of correlatable symbol sequences provides two key advantages: 1)efficiency, as it permits a near order of magnitude reduction of the control time; 2)robustness, because codes are short and easily detectable even at low signal-to-interference-plus-noise ratio (SINR) and even while a neighbor is transmitting data.We implement 802.11ec on a field programmable gate array (FPGA)-based softwaredefined radio. We perform a large number of experiments and show that, comparedto 802.11 (with and without RTS/CTS), 802.11ec achieves a vast efficiency gain inconveying control information and resolves key throughput and fairness problemsin the presence of hidden terminals, asymmetric topologies, and general multihoptopologies.

iii

Contents

Acknowledgements ii

Abstract iii

1 Introduction 11.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.3 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.4 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.4.1 802.11ec Key Techniques . . . . . . . . . . . . . . . . . . . . . 21.4.2 Corelatable Symbol Sequences . . . . . . . . . . . . . . . . . . 3

2 Literature Survey 52.1 Background and Recent Research . . . . . . . . . . . . . . . . . . . . 52.2 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.2.1 MACAW: A media access protocol for wireless LANs . . . . . 52.2.2 WiFi-Nano: Reclaiming WiFi efficiency through 800 ns slots . 62.2.3 E-MiLi: Energy-Minimizing Idle Listening in Wireless Networks 72.2.4 CRMA: collision-resistant multiple access . . . . . . . . . . . . 72.2.5 DOF: a local wireless information plane . . . . . . . . . . . . . 82.2.6 SAM: Enabling practical spatial multiple access in wireless LAN 82.2.7 Zig-zag decoding: Combating hidden terminals in wireless net-

works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.2.8 CSMA/CN: Carrier sense multiple access with collision notifi-

cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 Algorithm 103.1 MAC Protocol Design . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.1.1 Coded Control Information Versus Message Control Information 103.1.2 Public CSSs Versus Private CSSs . . . . . . . . . . . . . . . . 123.1.3 Control During Data . . . . . . . . . . . . . . . . . . . . . . . 12

3.2 11ec Channel Reservation Primitives . . . . . . . . . . . . . . . . . . 123.2.1 Initiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133.2.2 Reservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

iv

3.2.3 Deferral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133.2.4 CSS Association . . . . . . . . . . . . . . . . . . . . . . . . . . 143.2.5 Primitive Extensions . . . . . . . . . . . . . . . . . . . . . . . 14

3.3 Contending Flows and Vulnerability Interval . . . . . . . . . . . . . . 153.4 Coexistence With Legacy 802.11 . . . . . . . . . . . . . . . . . . . . . 16

4 Experimentation and Results 174.1 Experimental evaluation of CSS . . . . . . . . . . . . . . . . . . . . . 17

4.1.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . 174.1.2 Channel Emulator Validation . . . . . . . . . . . . . . . . . . 184.1.3 CSS Length Versus Robustness Tradeoff . . . . . . . . . . . . 204.1.4 CSS Detection Versus Control Message Decoding . . . . . . . 204.1.5 Discussion on Signal Correlation . . . . . . . . . . . . . . . . . 21

4.2 Experimental Evaluation of 11ec . . . . . . . . . . . . . . . . . . . . . 224.2.1 Basic Topologies . . . . . . . . . . . . . . . . . . . . . . . . . 224.2.2 Large Network Simulation . . . . . . . . . . . . . . . . . . . . 26

5 Conclusion and Future work 285.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

v

List of Figures

1.1 Timeline of a packet exchange with Coded Control versus MessageControl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1 Effects of the use of the feature of Control during Data in 11ec onInformation Asymmetry topologies. . . . . . . . . . . . . . . . . . . . 6

3.1 Timeline of the vulnerability interval of 802.11ec (time indications arein microseconds). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4.1 Example of 127-symbol CSS correlation at (b) Channel emulation. . . 194.2 Robustness versuss length tradeoff for different CSS lengths. . . . . . 204.3 Probability of missing CSS detection versus missing message decoding. 214.4 Low cross correlation of CSSs different from the one transmitted. . . 214.5 Basic topologies at the origin of throughput losses and/or imbalances. 234.6 Throughput of 11ec, 802.11 with/without RTS/CTS in basic topolo-

gies. (a) Symmetric Hidden Terminals. (b) Asymmetric Hidden Ter-minals. (c) Infor- mation Asymmetry. (d) Fully Connected WLANs. . 24

4.7 Total airtime utilization in the case of Asymmetric Hidden Terminals. 254.8 Throughput of 11ec, 802.11 with/without RTS/CTS in 20-node simu-

lated topologies. (a) Entire CDF. (b) Zoom in. . . . . . . . . . . . . . 264.9 Throughput distribution for a five-flow topology. . . . . . . . . . . . . 27

vi

Chapter 1

Introduction

1.1 Overview

MAC CONTROL messages are essential: For example, ACKs convey correctly re-ceived data, and RTS/CTS exchange can significantly mitigate hidden terminal col-lisions. However, even though the information conveyed in MAC control messagesis small, their duration can be quite long, as in addition to the control information,they also need to include source/destination address, message type, etc., all of whichare transmitted at base rate to improve the likelihood that they can be successfullydecoded. For example, an ACK message is 14 B plus physical-layer encapsulation,but contains only one bit of relevant information (that DATA was successfully re-ceived). Likewise, RTS/CTS is rarely used in practice precisely due to excessiveoverhead despite its important role in mitigating collisions

1.2 Motivation

Many random access MAC protocols have been proposed to mitigate collisions andaddress hidden terminals. In contrast to tones and messages used by such protocols,we design a new primitive of correlatable symbol sequences that, by virtue of beingshort and robust, increases throughput and fairness with minimal overhead. Morerecently, while several papers have addressed 802.11 throughput overhead reduction.They completely neglect the case of hidden terminals and, because of their more ag-gressive contention mechanisms, suffer severe throughput penalties in their presence.Ongoing standardization efforts, namely 802.11ah, target overhead reduction for sub-gigahertz communications, with application to smart grids, surveillance systems, etc.The strategy adopted consists in removing or compressing some information fieldsof the control messages, for a total of few bytes; compared to 11ec, this has a mi-nor effect on the control message overhead, e.g., due to the preser- vation of thecumbersome preamble structure. Other techniques have been proposed to improve

1

802.11 throughput, but address collision resolution rather than collision avoidance.For this reason, they are complementary to (and can be used in combination with)11ec. Furthermore, our physical-layer model is significantly more simple than manycollision avoidance algorithms. since it relies only on the replication of components(correlators) that are already present in common 802.11 cards. Finally, other appli-cations of signal correlation have been recently shown in various algorithms

1.3 Problem Definition

In this paper, we study and evaluate 802.11ec (Encoded Control) as a control-message-free MAC. Instead of control messages, 11ec employs correlatable symbolse- quences (CSSs) that, together with their transmission timing, convey all controlinformation and change the fundamental design properties of the MAC. For example,11ec replaces an 802.11 ACK message with a predefined ACK CSS that can be cor-related instead of decoded, thereby vastly reducing its duration and dramaticallyimproving its robustness by enabling its reception at low signal-to-interference-plus-noise ratio (SINR).

1.4 Objectives

1.4.1 802.11ec Key Techniques

802.11ecs key techniques are twofold: First, we use a dictionary of correlatablesymbol sequences to convey control information that can be represented by a lim-ited dictionary. For example, instead of CTS that contains physical-layer preamble,frame control sequence, type field, frame check-sum, destination address, durationfield (as well as it incurs a delay), we transmit a short (e.g., 127 symbols) CSS froma small dictionary to convey that it is a CTS. For an 802.11a physical layer, weshow that this reduces the time to convey the control information by nearly anorder of magnitude, from 60 to 6.35 s. Second, we show that the information thatcannot be represented by a limited dictionary can be conveyed via CSS timing. Forexample, nodes overhearing the 802.11 CTS message need to defer for an amount oftime as specified by the CTS duration field contained in the message. We show that11ec nodes can instead simply defer until a channel-clear CSS is transmitted by thereceiver (or until a timeout).

802.11ecs second technique is to distinguish between public and private informa-tion. Namely, 11ec only uses public CSSs for information that is required to bepublic, such as conveying channel reservation and channel clear. On the other hand,address fields need not be public, as the identity of the sender and receiver need not

2

be known by other nodes. 11ec ensures that private control information, includingaddresses and ACKs, is not correlated by other nodes. This has the potential tothwart eavesdroppers not only from decoding data (as data can be en- crypted), buteven from knowing which nodes are communicating with each other; we show howall private control information, including addresses can only be correlated by thein- tended receiver.

1.4.2 Corelatable Symbol Sequences

802.11ec enhances robustness in two ways. First, control information is more likelyto be received in 11ec because control information is conveyed in short CSSs thatare correlatable even at low SINR. For example, because 802.11ec replaces an ACKmessage with a CSS, 11ec ACKs are more robust and can be received even in thepresence of transmitting interferers. Second, 802.11 is fragile to topological factorsin that while 802.11 DCF without RTS/CTS yields high performance in fully con-nected wireless LANs , hidden terminals, asymmetric topologies, and general multi-hop topologies can yield severe throughput degradation and unfairness. These lattertopolo- gies are becoming increasingly common because of device power asymme-tries, e.g., between access points (APs), laptops, and popular smartphones, and ofthe wider coverage achievable with the adoption of subgigahertz frequencies, includ-ing TV white spaces. While use of RTS/CTS can significantly improve throughputin such challenged topologies, the additional overhead of RTS/CTS can sometimesoverwhelm this improvement. Moreover, in fully connected topologies, RTS/CTSdegrades throughput due to its unnecessary overhead. In contrast, 11ec over-comes these limitations through robust and short-duration control signals, i.e., 11ecminimally penalizes station throughput thus allowing to enable channel reservationindependently of the network topology. Conse- quently, 11ec stations have vastlyincreased opportunities to obtain channel access thereby dramatically improving thenetworks fairness in throughput distribution.

We implement correlatable symbol sequences in a software defined radio, performa large set of experiments, and study issues that have not been experimentallyinvestigated previously. We find a correlatable symbol sequence length that simul-taneously:1) provides sufficient physical-layer robustness;2) limits communication overhead and3) supports large networks.

Specifically, we first investigate the tradeoffs between sequence length and physical-layer robustness and show that even short sequences, e.g., 127-symbol or 6.35 s long,can be detected at 6 dB SINR with only 5% false negatives. We demonstrate thatour encoded sequences can be detected at an SINR 10 dB lower than 802.11 control

3

messages. Second, we show that 127-symbol code lengths can support more than 50co-located nodes, with minimal penalty on detection errors.

Figure 1.1: Timeline of a packet exchange with Coded Control versus Message Con-trol

Finally, we study 11ec in a measurement-driven emulator, whose inputs arechannel measurements collected in a real deployment and real card performanceparameters (e.g., bit error rate (BER) and multiple supported modulations). Wecompare 11ecs performance to 802.11 with and without RTS/CTS. We examinea wide set of basic topologies that are at the origin of throughput losses and/orimbalances in 802.11-based networks in order to provide an insight in understandingthe performance of larger networks. Our finding is that 11ec can dramatically reducethroughput imbalances by improving Jains index by up to 90%. Moreover, while sucha fairness improvement can often decrease total utilization, 11ec increases channel uti-lization by more than 10% via the use of short encoded control that simultaneouslydecreases vulnerability intervals and control overhead. We also study a largertopology and show that 11ec can improve the throughput of an under-served flow bya factor of over 22 folds. Over all flows, we improve Jains index by up to 173% whilealso improving the channel utilization by up to 46%. Finally, we simulate larger20-node topologies and show that 11ec highly benefits the lowest throughput links,by at least doubling the throughput of more than 49% of them.

4

Chapter 2

Literature Survey

2.1 Background and Recent Research

Many random access MAC protocols have been proposed to mitigate collisions andaddress hidden terminals. In contrast to tones and messages used by such protocols,we design a new primitive of correlatable symbol sequences that, by virtue of beingshort and robust, increases throughput and fairness with minimal overhead. Morerecently, while several papers have addressed 802.11 throughput overhead reductionthey completely neglect the case of hidden terminals and, because of their more ag-gressive contention mechanisms, suffer severe throughput penalties in their presence.Ongoing standardization efforts, namely 802.11ah , target overhead reduction forsubgigahertz communications, with application to smart grids, surveillance systems,etc.

The strategy adopted consists in removing or compressing some information fieldsof the control messages, for a total of few bytes; compared to 11ec, this has a mi-nor effect on the control message overhead, e.g., due to the preser- vation of thecumbersome preamble structure. Other techniques have been proposed to improve802.11 throughput but address collision resolution rather than collision avoidance.For this reason, they are complementary to (and can be used in combination with)11ec. Furthermore, our physical-layer model is significantly more simpler since itrelies only on the replication of components (correlators) that are already presentin common 802.11 cards. Finally, other applications of signal correlation have beenrecently shown in various standards

2.2 Literature Survey

2.2.1 MACAW: A media access protocol for wireless LANs

In recent years, a wide variety of mobile computing devices has emerged, includingportables, palmtops, and personal digital assistants. Providing adequate network

5

Figure 2.1: Effects of the use of the feature of Control during Data in 11ec onInformation Asymmetry topologies.

connectivity for these devices will require a new generation of wireless LAN technol-ogy. In this paper we study media access protocols for a single channel wireless LANbeing developed at Xerox Corporations Palo Alto Research Center. We start withthe MACA media access protocol first proposed by Karn and later refined by Bibawhich uses an RTS-CTS-DATA packet exchange and binary exponential back-off.Using packet-level simulations, we examine various performance and design issuesin such protocols. Our analysis leads to a new protocol, MACAW, which uses anRTS-CTS-DS-DATA-ACK message exchange and includes a significantly differentbackoff algorithm.

2.2.2 WiFi-Nano: Reclaiming WiFi efficiency through 800ns slots

The increase in WiFi physical layer transmission speeds from 1 Mbps to 1 Gbps hasreduced transmission times for a 1500 byte packet from 12 ms to 12 us. However,WiFi MAC overheads such as channel access and acks have not seen similar reduc-tions and cumulatively contribute about 150 us on average per packet. Thus, theefficiency of WiFi has deteriorated from over 80% at 1 Mbps to under 10% at 1 Gbps.

In this paper, we propose WiFi-Nano, a system that uses 800 ns slots to signif-icantly improve WiFi efficiency. Reducing slot time from 9 us to 800 ns makesbackoffs efficient but clear channel assessment can no longer be completed in oneslot since preamble detection can now take multiple slots. Instead of waiting mul-

6

tiple slots for detecting preambles, nodes speculatively transmit preambles as theirbackoff counters expire while continuing to detect premables using self-interferencecancellation. Upon detection of preambles from other transmitters, nodes simplyabort their own preamble transmissions, thereby allowing the earliest transmitter tosucceed. Further, receivers speculatively transmit their ack preambles at the endof packet reception, thereby reducing ack overhead. We validate the effectivenessof WiFi-Nano using an implementation on an FPGA based software defined radioplatform and through extensive simulations, demonstrating efficiency gains of up to100%.

2.2.3 E-MiLi: Energy-Minimizing Idle Listening in WirelessNetworks

WiFi interface is known to be a primary energy consumer in mobile devices, and idlelistening (IL) is the dominant source of energy consumption in WiFi. Most existingprotocols, such as the 802.11 power-saving mode (PSM), attempt to reduce the timespent in IL by sleep scheduling. However, through an extensive analysis of real-worldtraffic, we found more than 60 percent of energy is consumed in IL, even with PSMenabled. To remedy this problem, we propose Energy-Minimizing idle Listening (E-MiLi) that reduces the power consumption in IL, given that the time spent in IL hasalready been optimized by sleep scheduling. Observing that radio power consumptiondecreases proportionally to its clock rate, E-MiLi adaptively downclocks the radioduring IL, and reverts to full clock rate when an incoming packet is detected or apacket has to be transmitted. E-MiLi incorporates sampling rate invariant detection,ensuring accurate packet detection and address filtering even when the receiverssampling clock rate is much lower than the signal bandwidth. Further, it employs anopportunistic downclocking mechanism to optimize the efficiency of switching clockrate, based on a simple interface to existing MAC-layer scheduling protocols. Wehave implemented E-MiLi on the USRP software radio platform. Our experimentalevaluation shows that E-MiLi can detect packets with close to 100 percent accuracyeven with downclocking by a factor of 16. When integrated with 802.11, E-MiLi canreduce energy consumption by around 44 percent for 92 percent of users in real-worldwireless networks.

2.2.4 CRMA: collision-resistant multiple access

Efficiently sharing spectrum among multiple users is critical to wireless networkperformance. In this paper, we propose a novel spectrum sharing protocol calledCollision-Resistant Multiple Access (CRMA) to achieve high efficiency. In CRMA,each transmitter views the OFDM physical layer as multiple orthogonal but sharablechannels, and independently selects a few channels for transmission. The transmis-sions that share the same channel naturally add up in the air. The receiver extractsthe received signals from all the channels and efficiently decodes the transmissions

7

by solving a simple linear system. We implement our approach in the Qualnet simu-lator and show that it yields significant improvement over existing spectrum sharingschemes. We also demonstrate the feasibility of our approach using implementationand experiments on GNU Radios.

2.2.5 DOF: a local wireless information plane

The ability to detect what unlicensed radios are operating in a neigh borhood, theirspectrum occupancies and the spatial directions their signals are traversing is a fun-damental primitive needed by many applications, ranging from smart radios to coex-istence to network management to security. In this paper we present DOF, a detectorthat in a single framework accurately estimates all three parameters. DOF buildson the insight that in most wireless protocols, there are hidden repeating patterns inthe signals that can be used to construct unique signatures, and accurately estimatesignal types and their spectral and spatial parameters. We show via experimentalevaluation in an indoor testbed that DOF is robust and accurate, it achieves greaterthan 85% accuracy even when the SNRs of the detected signals are as low as 0 dB,and even when there are multiple interfering signals present. To demonstrate thebenefits of DOF, we design and implement a preliminary prototype of a smart radiothat operates on top of DOF, and show experimentally that it provides a 80% in-crease in throughput over Jello, the best known prior implementation, while causingless than 10% performance drop for co-existing WiFi and Zigbee radios.

2.2.6 SAM: Enabling practical spatial multiple access in wire-less LAN

Spatial multiple access holds the promise to boost the capacity of wireless networkswhen an access point has multiple antennas. Due to the asynchronous and uncon-trolled nature of wireless LANs, conventional MIMO technology does not work effi-ciently when concurrent transmissions from multiple stations are uncoordinated. Inthis paper, we present the design and implementation of a crosslayer system, calledSAM, that addresses the challenges of enabling spatial multiple access for multi-ple devices in a random access network like WLAN. SAM uses a chain-decodingtechnique to reliably recover the channel parameters for each device, and iterativelydecode concurrent frames with misaligned symbol timings and frequency offsets. Wepropose a new MAC protocol, called CCMA, to enable concurrent transmissions bydifferent mobile stations while remaining backward compatible with 802.11. Finally,we implement the PHY and MAC layer of SAM using the Sora high-performancesoftware radio platform. Our evaluation results under real wireless conditions showthat SAM can improve network uplink throughput by 70% with two antennas over802.11.

8

2.2.7 Zig-zag decoding: Combating hidden terminals in wire-less networks

This paper presents ZigZag, an 802.11 receiver design that combats hidden terminals.ZigZags core contribution is a new form of interference cancellation that exploitsasynchrony across successive collisions. Specifically, 802.11 retransmissions, in thecase of hidden terminals, cause successive collisions. These collisions have differentinterference-free stretches at their start, which ZigZag exploits to bootstrap its de-coding.

ZigZag makes no changes to the 802.11 MAC and introduces no overhead whenthere are no collisions. But, when senders collide, ZigZag attains the same through-put as if the colliding packets were a priori scheduled in separate time slots. Webuild a prototype of ZigZag in GNU Radio. In a testbed of 14 USRP nodes, ZigZagreduces the average packet loss rate at hidden terminals from 72.6% to about 0.7%.

2.2.8 CSMA/CN: Carrier sense multiple access with colli-sion notification

A wireless transmitter learns of a packet loss and infers collision only after com-pleting the entire transmission. If the transmitter could detect the collision early[such as with carrier sense multiple access with collision detection (CSMA/CD) inwired networks], it could immediately abort its transmission, freeing the channel foruseful communication. There are two main hurdles to realize CSMA/CD in wirelessnetworks. First, a wireless transmitter cannot simultaneously transmit and listen fora collision. Second, any channel activity around the transmitter may not be an in-dicator of collision at the receiver. This paper attempts to approximate CSMA/CDin wireless networks with a novel scheme called CSMA/CN (collision notification).Under CSMA/CN, the receiver uses PHY-layer information to detect a collisionand immediately notifies the transmitter. The collision notification consists of aunique signature, sent on the same channel as the data. The transmitter employsa listener antenna and performs signature correlation to discern this notification.Once discerned, the transmitter immediately aborts the transmission. We show thatthe notification signature can be reliably detected at the listener antenna, even inthe presence of a strong self-interference from the transmit antenna. A prototypetestbed of 10 USRP/GNU Radios demonstrates the feasibility and effectiveness ofCSMA/CN.

9

Chapter 3

Algorithm

3.1 MAC Protocol Design

11ec collision avoidance realizes and improves on the collision avoidance mechanismof IEEE 802.11 DCF with RTS/CTS, reduces the overhead by nearly an order ofmagnitude, and practically eliminates collisions, even in hidden terminal topologies.Specifically, 11ec retains the four-way handshake suggested by 802.11, where controlmessages are replaced by very short correlatable symbol sequences.It is importantto note that:1) the duration of each correlatable symbol sequence is nearly zero;2) the duration between the start of the reservation signal until the data transmissionis negligible, hence practically invulnerable to collisions.In this section, we define CSSs and explain how control messages can be turnedinto CSSs. Furthermore, we show that CSSs are a key element for the realization of11ec efficiency and robustness. The second part of the section provides a detailedde- scription of 11ec including protocol primitives and an analysis of hidden terminalvulnerability leading to novel collision reduction opportunities.

3.1.1 Coded Control Information Versus Message ControlInformation

CSS

Correlatable symbol sequences are predefined pseudo-noise binary codewords; namely,while codewords are deterministically generated, they retain the statistical prop- er-ties of a sampled white noise. For this reason, the cross correlation of any suchsequence with a matching copy obtains spike values, while it appears random toa listener without prior knowledge of the codeword. An example of a CSS is the802.11 preamble used for packet detection, symbol synchronization, and radio pa-rameter tuning.

10

The CSS detection process via cross correlation enjoys three key advantages overdata decoding. First, cross correlation obtains a large processing gain even forsmall codewords (e.g., the 802.11 preamble used for detection is 64 symbols), whichper- mits reliable detection even at low SINR. Second, differently from decoding, de-tection is highly robust to imperfect radio pa- rameter tuning, and thus a codeworddoes not need to be pre- ceded by a preamble. For these reasons, a CSS can be short;for example, in our implementation, 11ec utilizes 127-symbol codewords that can betransmitted in 6.35 s. Third, detection is almost instantaneous as no decoding isneeded. For example, 802.11 including about 14 s of data processing, while in 11ecno control message processing is required; hence, 11ec reduces substan- tially theshort inter-CSS time.

Encoded Control

While consuming a significant amount of airtime, 802.11 control messages usuallyconvey little information. For example, an ACK occupies the medium for up to60 s, i.e., an airtime sufficient to transmit 3240 bits at 54 Mb/s, while it contains asingle bit of relevant information. 11ec replaces control messages with CSSs, whichpermit to shorten the transmission duration of nearly an order of magnitude to6.35 s, while retaining the information content.

According to the 802.11 standard, RTS, CTS, and ACK control messages may includeup to four information fields: destination address, sender address, duration, andframe control (a fifth field is the frame check-sum that protects the other four). Inparticular, the frame control field is a 2-B-long sequence of bits representing specificcontrol parameters. The values of most control bits are fixed for control messages;only frame subtype (4 bits) and station power management flag (1 bit) can assumedifferent values. However, the latter does not convey novel information when usedin control messages.

In order to represent the information content of the control messages as describedabove, 11ec considers the size of the dictionary needed to represent such information.Information that can be expressed by a small dictionary is conveyed using CSSs, whileinformation that needs large dictionaries is conveyed with timing codes. First, in802.11 data exchange, the type field that distinguishes the control messages mayassume only three values, i.e., RTS, CTS, and ACK; 11ec conveys the type by as-sociating each message with distinct CSSs. Second, 802.11 control messages includeaddresses (sender and/or receiver). Since the number of nodes a station commu-nicates with at a time is generally small, i.e., can be represented by a small dic-tionary, 11ec integrates the addresses in CSSs, i.e., a single CSS may represent thecombination of a control type and a specific address. For example, in 802.11 the RTSincludes the address of the intended receiver; accordingly, in 11ec, RTS addressed todifferent receivers are represented by distinct CSSs. Third, some control messages

11

include a duration field that cannot be represented by a small dictionary. For thisinformation, 11ec utilizes a combination of time codes and new control types. Forexample, 11ec nodes reserve the channel for the duration of a data reception, bytransmitting two CSSs corresponding to channel reservation (immediately before thereception) and release (immediately after). The potential interferers do not accessthe channel during the interval between the two CSSs (or before a timeout expires),i.e., effectively implement a form of virtual carrier sensing.

3.1.2 Public CSSs Versus Private CSSs

A second dimension of control messages, and of the corresponding CSSs, is whetherthey can be private, i.e., carry information relevant only to a specific destination(e.g., acknowledgments), or are necessarily public, i.e., meant to be heard by allneighbors of a node (e.g., channel reservation/release). Accordingly, in the case of apri- vate CSS, only the intended receiver possesses a copy of the correlatable symbolsequence, and thus can correctly detect it; con- versely, all nodes possess copies ofthe public CSSs and can detect them. For example, only the data sender needs tocross-cor- relate its private acknowledgment CSS, while all nodes must cross-correlatepublic channel reservation/release CSSs.

3.1.3 Control During Data

Because correlatable symbol sequences can be detected even at subnoise SINR (e.g.,11ec 127-symbol CSSs can be detected at 6 dB with high relia- bility), 11ec nodesattempt detection even while receiving data. This technique provides a signalingmechanism effective even in cases in which the receiver is subject to long periodsof noisy channel, or undesired data overhearing. In 802.11, if a node receives anRTS while also overhearing data, the node cannot decode the RTS and thereforecannot respond. In contrast, 11ec uniquely enables a node to receive a CSS signalingthat another node is requesting to communicate. Therefore, the receiver can senda Request for RTS (RRTS) CSS to reserve the medium when it becomes free andinitiate a data exchange. Likewise, because ACKs are also encoded, they can becorrectly received even if an interfering terminal is simultaneously transmitting data.

3.2 11ec Channel Reservation Primitives

Wireless MAC protocols perform collision avoidance by si- lencing the medium inthe vicinity of a transmitting link via channel reservation. Channel reservation fun-damentally hinges on three key mechanisms:1)initiation, performed by the node that initiates the exchange to request the coop-eration of the other endpoint to reserve the channel;2)reservation, performed to inform nodes potentially hindering the exchange;

12

3)deferral, performed by the surrounding terminals in order to avoid disturbing ongo-ing transmissions. 802.11 implements the three mechanisms via: a) RTS; b) CTS anddata packetthe latter realizes channel reservation in the vicinity of the sender; and c)NAV and carrier sensing. When RTS/CTS is disabled, during the data transmissionthe medium is reserved exclusively in the vicinity of the sender. In the following, weshow how 11ec implements these three mechanisms via CSSs and timing codes. Akey concept of 11ec channel reservation is very short channel reservation negotiationfor near immunity to interrup- tions, e.g., collisions and capturing by other nodes.Specifically, 11ec channel reservation is based on three basic primitives (the subscriptc indicates CSSs).

3.2.1 Initiation

In 11ec, a sender wishing to start a data exchange performs virtual, and optionallyphysical, carrier sensing. If the medium is free, the sender waits for a backoff in-terval similar to 802.11 and then transmits a sender side channel request primitive, inshort I, to request the receiver to reserve the channel.Initiation need only be detectedby and not necessarily by the neighboring nodes; thus, we implement it as a privateCSS. In order to convey the identity of the receiver (as 11ec does not transmit thetraditional MAC address), 11ec implements initiation via several CSSs and associatesa distinct CSS with each receiver; i.e., when a sender needs to contact a receiver, ituses the receivers. Nodes in the vicinity of the sender do not detect the initiation.

3.2.2 Reservation

A node receiving an initiation checks if other nodes are communicating in its vicinityand may hinder its reception. If that is not the case, immediately transmits a chan-nel reservation primitive reservation to notify potential interferers. In order to realizethe reservation, should be detected by all nodes in the vicinity of the receiver,and itis therefore transmitted via a public CSS.

In addition to channel reservation that forces neighbors to defer, implicitly com-municates to the transmitter that the channel is available and that data can betransmitted. Instead of providing a distinct CSS to convey the sender address, asin the previous case of the , we employ a simple temporal code: Since a receivertransmits immediately after, the sender, in contrast to the other neighboring nodes,interprets the reception of a as an authorization to begin a data transmission.

3.2.3 Deferral

11ec implements the deferral and conveys its duration via a combination of CSSsand a simple time code. Specifically, after data reception and acknowledgment, thereceiver explicitly releases the channel with a channel-free primitive . Thus, nodes

13

receiving need to wait to receive an (or wait a predefined timeout) before accessingthe channel; practically, this procedure represents a form of virtual carrier sensing.Because all neighboring nodes need to receive , 11ec implements as a public CSS.11ec further increases the robustness of messages by pairing them. Our technique isbased on associating a small number of CSS pairs to distinct pairs; a receiver ran-domly picks and transmits any such pair to reserve and free the channel. This fea-ture is particularly useful for a node located in the neighborhood of several receiversthat may be active simultaneously, i.e., the receivers may send and that denoteoverlapping intervals. In that case, the node can still correctly etermine the state ofthe medium in each moment by associating each overheard reservation to its corre-sponding .

Finally, we define an acknowledgment primitive, and we implement it as a privateCSS associated to each sender . Few recently proposed packet forwarding schemesleverage the ACK exchange for additional purposes other than data acknowledg-ment, e.g., network coding and routing. In such cases, 11ec can revert to 802.11ACKs with small overhead penalty (the major gain of 11ec both in throughput androbustness derives from the novel channel reservation scheme). Note that CSMA/CNuses signatures for acknowledgment, which are similar to our CSSs.

3.2.4 CSS Association

The Initiation and Acknowledgment primitives and require association of uniqueCSSs to specific nodes. While a detailed investigation of the issue is beyond thescope of this paper, we suggest two simple mechanisms that can be used for thispurpose. The first leverages the solution already existing in the 802.11 standard,in which the AP assigns an identifier (AID) upon station association; AIDs can beeasily mapped to CSSs. In order to initiate the association, the stations may use areserved CSS. The second mechanism utilizes multiple hash functions based on thestation MAC address. In case of assignment conflicts, which can be easily detectedusing the address fields in the header of the data frames, the stations can switch thehash function utilized.

3.2.5 Primitive Extensions

As was briefly mentioned, 11ec can support a signaling mechanism to alleviate star-vation similar to RRTS via control during data and can highly reduce the occurrenceof exposed terminals via robust CSS acknowledgment. For reason of space, thesecases are not covered in this report.

14

3.3 Contending Flows and Vulnerability Interval

The shortness and robustness of correlatable symbol sequences dramatically reducesvulnerability to collisions. The vulnerability interval of a packet exchange is twicethe time delay from the beginning of a transmission until all potential interferers areprevented from corrupting the exchange, i.e., it includes the whole interval, beforeand after the beginning of the intended exchange, during which interfering transmis-sions may start and corrupt the exchange. In the case of hidden terminals in 802.11with RTS/CTS, the vulnerability interval is twice the delay from the moment a nodesends an RTS to the detection of CTS by the hidden terminals. Considering thecase of 802.11a/g and 6-Mb/s control packets, the total duration of the vulnerabilityinterval can be computed as follows. First, node transmits the RTS, for a total du-ration of 52 s including preambles. Second, the RTS propagates to the receiver forup to 1 s. Third, the receiving node waits s before sending the CTS, due to practicalcommunication issues such as RTS decoding. Fourth, the CTS propagates to hiddenterminals for up to 1 s. Fifth, the hidden terminals need up to 4 s to detect thepacket. Last, additional 2 s account for potential radio turnaround (all temporalindications are taken from section 17 of the 2007 version of the standard, for 20MHz bandwidth). The total amounts to 152 s, i.e., twice 76 s. In 802.11 withoutRTS/CTS, the vulnerability interval can be considerably larger, spanning twice thedata transmission duration and as large as 4 ms.

Figure 3.1: Timeline of the vulnerability interval of 802.11ec (time indications are inmicroseconds).

802.11ecs CSSs shorten the vulnerability interval and practically reduce the setof potential hidden terminals. The vulnerability interval of 11ec can be computed

15

as follows (see Fig. 3.1, where the intervals below are denoted by numerical timeindi- for 6.35 s. Second, the cations). First, node transmits receiving node waits forup to 2 s, in order to detect potential overlapping hidden terminal transmissions asexplained below (this interval is a design choice and includes the propagation de-lay from the transmitter). Third, needs a turnaround time to prepare its radio fortransmission, i.e., 2 s (according to 802.11 for 6.35 s. Finally, the hidden standard).Fourth, transmits terminals need a propagation delay of up to 1 s to receive theand an additional 2 s to account for potential radio turnaround. 11ec vulnerable in-terval is twice 19.7 s, i.e., 39.4 s or about 25.9% of the vulnerability interval of 802.11.

CSSs also nearly eliminate the collisions of nodes starting in the same slot whenthe SINR allows it. 5 In fact, the receiver can detect simultaneous and overlap-ping transmissions of multiple because of the CSS processing gain and request re-transmission. Specifically, waits for a round-trip propagation time in order to detectpotentially overlapping transmissions and, in that case, sends a negative acknowl-edgment that prompts the contending nodes to undergo a quick backoff repetition.However, in order to take advantage of this technique, we need to enlarge the slotsize to encompass the half vulnerability the duration, i.e., 20 s (note that this issufficient, as the receiver synchronizes all of its transmitters). Also in cases of nohidden terminals, this choice induces only minor throughput penalties at high datarates.

3.4 Coexistence With Legacy 802.11

For backward compatibility with 802.11, 11ec exactly follows the standard except forthe techniques described in this section. For example, a key element for coexistenceis the arbitration of the medium, which leverages carrier sensing based on the correla-tion of the data preambles and backoff mechanism. Accordingly, 11ec uses the samedata preamble format as 802.11 and sets the contention window size following thesame binary exponential backoff scheme. A more complete discussion of coexistenceis beyond the scope of this report.

16

Chapter 4

Experimentation and Results

4.1 Experimental evaluation of CSS

In this chapter, we present an experimental evaluation of correlatable symbol se-quences using software defined radios. Specifically, our evaluation covers the follow-ing issues, none of which has been previously experimentally studied in the literature.

We explore the tradeoff between length of the sequences, i.e., overhead, andprocessing gain, i.e., robustness.

We contrast the performance of CSS detection with control message decoding. We determine the codebook size that 11ec can support, i.e., the number of

distinct CSSs that can be practically used, by studying the cross correlationbetween different CSSs and its effect on the probability of false positives.

The experimental results in this section show that the design selection of 127-symbolCSSs via Gold codes, which can support more than 50 co-located nodes, providesa good tradeoff between overhead and robustness without any penalty on false posi-tives.

4.1.1 Experimental Setup

1. Tools: WARP and WARPLab: Our reference software defined radio is theWARP platform. WARP is a field programmable gate array (FPGA)-basedplatform, including custom designed radios based on the MAX2829 chipset.WARPLab is a programming environment that permits to drive WARP froma host computer. Relevantly to our experiments, WARPLab supports the exe-cution of micro experiments, each one of approximately 400 s duration [2 sam-ples at the 40-MHz frequency of digital-to-analog converter/analog-to-digitalconverter (DAC/ADC)], and ac- cesses analog sample send/receive buffers and

17

RSSI recordings collected during each experiment. RSSI is measured by theMAX2829 circuit and digitized by a dedicated 10-bit ADC.

Azimuth Channel Emulator

In order to perform experiments under controllable and repeatable conditions,we used an Az- imuth ACE MX channel emulator.The channel emulator per-mits creation of different network topologies by tuning the attenuation alongeach path independently and predictably.

2. Implementation: We implement CSS transmission/detection and OFDMpacket transmission/decoding on WARPLab. Specifically, CSSs are BPSK se-quences filtered, upsampled, and transmitted via standard wideband methods.This solution enjoys a practical advantage over alternative solutions, e.g.,OFDM-modulated BPSK sequences, due to the lower peak-to-average-powerratio. Finally, in order to repro- duce 802.11 as closely as possible, we imple-ment all types of OFDM 802.11a/g modulation and convolutional code pairsin WARPLab.

4.1.2 Channel Emulator Validation

The results in this section are obtained using a channel emulator. In order to validatethe emulator setting, our method- ology includes a preliminary validation contrast-ing results of a cross-correlation experiment performed over the air, with an identicalexperiment conducted with the emulator.

Specifically,our experiment consists of exchanging CSSs between two WARP nodesand , under the interference generated by random OFDM transmissions of a thirdinterfering node . In the first part of the experiment, we deploy the three nodesinside an office building. In an over-the-air setting, it is difficult to control SINRgiven interference from 802.11 networks operating in the building. For this reason,we perform the experiment late at night and measure the SINR on links and a fewseconds before and after the experiment. Then, we repeat the experiment under con-trollable and repeatable conditions using the channel emulator, where we can controlthe SINR with high accuracy. We repeat both experiments several times and for dif-ferent SINR and show a representative sample result. In both experiments, nodetransmits seven repetitions of a 127-symbol CSS. For each newly acquired sample(i.e., at 40 MHz frequency of the WARP platform ADC), node com- putes the signalcorrelation with a local copy of the transmitted CSS. Fig. 4.1(a) and (b) shows arepresentative outcome of a realization in which the SINR on link carrying the CSS is6 dB. The -axis is the temporal progression of collected samples, while the -axis is thecorrelated value. The thick crossing line in the plots represents a possible choice ofthe detection threshold; such a threshold is strategic in determining the robustness of

18

Figure 4.1: Example of 127-symbol CSS correlation at (b) Channel emulation.

CSSs by balancing false positives and false negatives. In the experiments we conductin this section, the threshold is chosen in order to obtain a false positive probabilityof 10 . While we defer more details on how to tune the threshold to Section III-F,here we observe that because of the threshold design, the correlation value on the-axis is normalized according to the magnitude of correlated I/Q samples. In thefigures, correlation spikes are clearly identifiable in coincidence with the reception ofeach single CSS as all and only marks exceeding the threshold. Thus, the detectionof 127-symbol CSSs is possible at 6 dB with few errors. By comparing the two plots,we conclude that controllable emulator experiments and over-the-air experimentsprovide similar results for identical SINR values. However, because of the difficultyto constantly control the SINR in over-the-air settings, we perform the remainingexperiments in this section using the channel emulator, thereby also ensuring theirrepeatability.

19

4.1.3 CSS Length Versus Robustness Tradeoff

The first issue is the trade-off between the CSS length and robustness under differentSINR; we quantify robustness in terms of the probability of false positives and falsenegatives. The outcome of this assessment is important, as robustness to SINR is oneof the two key elements in the choice of CSS length (the other element guiding thischoice is the number of CSSs, discussed in Section III-E). In this section, we deter-mine a length that can tolerate significant interference without high communicationoverhead. In this experiment, we deploy the three-node topology above and use thechannel emulator to vary the SINR on link . Specifically, the link between nodetransmitting the CSS, and the receiving node is maintained fixed to 82 dBm, whilethe attenuation on the interfering link is set in order to obtain the desired SINR. Weiterate the experiment for different combinations of SINR and CSS lengths. Eachexperiment consists in the detection of at least 100 CSSs of lengths ranging from 63to 511 symbols. We vary the SINR between 0 and 10 dB.

Figure 4.2: Robustness versuss length tradeoff for different CSS lengths.

4.1.4 CSS Detection Versus Control Message Decoding

Our second experiment aims to show that control CSSs are more robust to noise than802.11 control messages, i.e., that CSSs can be reliably detected at considerably lowerSINR than control messages. The metrics we use are false positives for the case ofCSS detection, and packet error rate for control message decoding. We consider127-symbol CSSs versus 160-bit messages transmitted via BPSK modulation, with

20

Figure 4.3: Probability of missing CSS detection versus missing message decoding.

1/2 rate convolutional coding, corresponding to an RTS packet transmitted at baserate of 6 Mb/s in 802.11 OFDM.

4.1.5 Discussion on Signal Correlation

Figure 4.4: Low cross correlation of CSSs different from the one transmitted.

The choice of the detection threshold is strategic in balancing the tradeoff betweenfalse positives and false negatives. For example, as mentioned above,the thresholdis denoted by the thick crossing line; corresponding to the chosen threshold, the

21

figures show 0 false positives and 0 false negatives. In general, a higher value ofthe threshold decreases the probability of false positives at the expense of a largeprobability of false negatives; vice versa, a lower threshold increases the occurrenceof false positives. The theory of correlation in Gaussian noise provides an optimalthreshold for a target detection SINR, minimizing the probability of false negativesfor a given probability of false positives.

4.2 Experimental Evaluation of 11ec

In this section, we present an experimental validation of 11ec using a measurement-based emulator that we design and implement. We perform the following experi-ments.

We investigate the performance of 11ec in basic topologies that typically incurloss or imbalance for 802.11.

We investigate the performance of 11ec in a larger 5-flow network topology. We complete the evaluation of 11ec simulating its performance in a 20-node

network topology.

4.2.1 Basic Topologies

In this set of experiments, we evaluate the performance of 11ec in a few basic topolo-gies (mostly including two flows) that are characterized by symmetries or asymme-tries in link signal strength differences and carrier-sensing relationships. This studyis important because these topologies are at the origin of throughput losses and/orimbalances in 802.11-based networks. We show that 11ec largely overcomes theproblems of 802.11 with and without RTS/CTS.

1. Symmetric Hidden Terminals include topologies in which a link is formed be-tween each of two transmitters that are not able to carrier-sense each otherand a common receiver; moreover, the links have similar reception power atthe receiver. Specifically, these topologies include links with signal strengthsat the receiver differing by no more than 4 dB; we choose this threshold toexclude capture at any 802.11 modulation.

2. Asymmetric Hidden Terminals include topologies in which two non-carrier-sensing transmitters share a common receiver, and one of the formed links hasa significant power advantage. Specifically, these topologies include links withsignal strengths at the receiver differing by more than 5 dB; this thresholdis chosen to permit one of the two links to capture over the other at BPSKmodulation.

22

Figure 4.5: Basic topologies at the origin of throughput losses and/or imbalances.

3. Information Asymmetries include link pairs a-b and c-d, with transmitters aand c not carrier-sensing each other, and with different receivers; moreover,one of the two links c-d interferes with the other link a-b, but not vice versa.

4. Fully Connected WLANs include topologies in which all nodes carrier-senseeach other and transmit to a common receiver.

Symmetric Hidden Terminals

We consider four instances of symmetric hidden terminals: Three of them are selectedfrom our deployed network and use modulation rates corresponding to 6/12/24 Mb/s.The fourth case (54 Mb/s) is artificially generated in order to explore the effectof using higher modulation schemes. Fig. 4.7(a) shows that all solutions assignsimilar throughputs to both flows. However, 11ec and 802.11 with RTS/CTS achieveconsiderably higher total throughput than 802.11 for most rates. Furthermore, 11ecoutperforms 802.11 with RTS/CTS due to the smaller duration of CSSs with respectto control messages (11ec control CSS exchange lasts 19.7 s, with respect to 128 sof 802.11 control messages). Besides reducing the control overhead, this entails thedecrease of the collision probability from 13% for 802.11 with RTS/CTS to 6.5%for 11ec regardless of the rate used to transmit the actual data. The effect of the

23

overhead reduction becomes more and more evident as the packet data rate increases;at 54 Mb/s, the total throughput gain of 11ec over 802.11 with RTS/CTS is about34%. Finally, at 54 Mb/s, the data packets are sufficiently short to permit lowcollision probability to 802.11 without RTS/CTS (19% as opposed to 66% obtainedat 6 Mb/s); nonetheless, 11ec still shows 5% throughput gain.

Figure 4.6: Throughput of 11ec, 802.11 with/without RTS/CTS in basic topologies.(a) Symmetric Hidden Terminals. (b) Asymmetric Hidden Terminals. (c) Infor-mation Asymmetry. (d) Fully Connected WLANs.

Asymmetric Hidden Terminals

We consider four instances of asymmetric hidden terminals, all based on actual linkpower measurements. In these topologies, packets sent by the sender with high

24

signal-to-noise ratio (SNR) at base rate, e.g., control packets, capture over packetssent by the sender with low SNR. However, packets sent by the sender with high SNRat data (i.e., higher) modulation rate get corrupted when overlapping with packetssent by the sender with low SNR. This is because we choose the data modulationrates of the links based on the SNR in the absence of interference. Fig. 9(b) showsthat the capture effect has disastrous consequences for the flow with low SNR in802.11 with and without RTS/CTS, while it has no effect on 11ec.For example, in the first instance presented (i.e., modulation rates 24/6), 11ec im-

Figure 4.7: Total airtime utilization in the case of Asymmetric Hidden Terminals.

proves the throughput of the underserved flow with respect to 802.11 with (without)RTS/CTS by about 13-fold (2.5-fold). In 802.11 with RTS/CTS, the imbalance isdue to the fact that in the case of overlapping RTS, the RTS of the stronger link iscorrectly decoded, while the other is ignored. Thus, while the stronger link enjoys ahigh successful probability, the RTS collision probability of the weaker link is over50% in all instances. A similar consideration also applies to 802.11ec; in fact, linkSNRs can be sufficiently different to permit the CSSs from one of the links to cap-ture over the CSSs from the other in case of overlapping. However, the probabilityof CSSs overlapping is so low for 802.11ec that even in such cases (not shown inthe figure), the resulting throughput imbalance remains confined within about 15%.Since 802.11 without RTS/CTS does not use the base rate, but transmits all packetsat data rate, the throughput imbalance originates only from the shorter durationof the data packet of the dominant link, which permits it to enjoy higher successprobability. The packet collision probability of the weaker link is again over 49%

25

for all cases of heterogeneous rates. The result for the fourth instance (i.e., last twobars in the graph) supports this claim: When both links operate at 24 Mb/s, theirthroughputs do not depend on any SNR imbalance, and the collision probability isreduced to 30%. Finally, because of the greater number of data packet collisions(that have a longer duration than RTS or CSS collisions), 802.11 without RTS/CTShas a lower total throughput. In contrast to 802.11, 11ec reduces the imbalance byreducing the packet loss to about 11%.

4.2.2 Large Network Simulation

This experiment aims to evaluate the performance gain of 11ec in larger networktopologies including 20 nodes.

Figure 4.8: Throughput of 11ec, 802.11 with/without RTS/CTS in 20-node simu-lated topologies. (a) Entire CDF. (b) Zoom in.

For each node, the data flow is determined by randomly choosing one of itsneighbors as the destination; the traffic is fully backlogged, and the packet lengthsare chosen according to the experimental distribution above. Fig. 4.8(a) showsthe cumulative distribution function (CDF) of all link throughputs achieved by thethree protocols through all 10 topologies of our experiment; Fig. 12(b) is a zoomed-inversion. In the figures, the -axes represent the fraction of flows, while the -axes arethroughput values (in Mb/s). Accordingly, the plotted values indicate the fractionof flows with a throughput smaller than or equal to the corresponding abscissa; e.g.,the point at (0.4, 0.444) means that 40% of the flows have a throughput lower thanor equal to 444 kb/s. Note that the flows are represented in the graphs according tothe sorted values of their throughputs, i.e., neither the -axis nor the -axis are relatedto the specific flow identities.

26

Figure 4.9: Throughput distribution for a five-flow topology.

The left plot represents the entire CDF, while the right plot magnifies the re-sults for the flows that achieve less than 1 Mb/s. Clearly, the right plot shows that11ec highly benefits the underserved links, i.e., the bottom 70.5% (85.5%) with re-spect to 802.11 with (without) RTS/CTS. Specifically, the average throughput ofthe 70% lowest-throughput flows is 192% (632%) greater for 11ec than for 802.11with (without) RTS/CTS. More- over, at the 20/40/60th percentiles, 11ec achievesa throughput value of 7.7 /3.5 /1.4 (25 /10 /4.8 ) greater than 802.11 with (without)RTS/CTS. As discussed above, the throughput gain of weak links in 11ec occursat the expense of the strong links; in fact, the left plot shows that the top 9%29%flows achieve best performance with 802.11 with RTS/CTS, while the top 9% obtainhighest throughput with 802.11 without RTS/CTS.

27

Chapter 5

Conclusion and Future work

5.1 Conclusion

In this report we present 802.11ec, an 802.11-based protocol without control mes-sages. 802.11ec introduces control correlatable symbol sequences that provide ro-bustness and efficiency. Through a wide set of experiments on a software definedradio, we show that 6.35- s correlatable symbol sequences can be detected at anSINR of 6 dB, i.e., 10 dB lower than 802.11 control messages. Finally, we implement802.11ec on a measurement-based network emulator and show that it improves net-work fairness by up to 90%, channel utilization by up to 90%, and throughput ofunderserved flows by over 22 folds, with respect to 802.11.

5.2 Future Work

In the last decade, 802.11 wireless devices data-rates have increased by three ordersof magnitude, while communications experiencing low throughput are still largelypresent. Such throughput loss is a fundamental problem of wireless networking that isdifficult to diagnose and amend.So more research is needed to increase the throughputin 802.11 ec.

28

References

[1] V. Barghavan, A. Demers, S. Shenker, and L. Zhang. MACAW: A media accessprotocol for wireless LANs . in Proc. ACM SIGCOMM, 1994, pp. 212225.

[2] G. Bianchi. Performance analysis of the IEEE 802.11 distributed co- ordinationfunction. IEEE J. Sel. Areas Commun., vol. 18, no. 3, pp. 535547, Mar. 2000.

[3] J. Camp, E. Aryafar, and E. Knightly. Coupled 802.11 flows in urban channels:Model and experimental evaluation . in Proc. IEEE INFOCOM, 2010, pp. 19.

[4] D. Chu. Polyphase codes with good periodic correlation properties . IEEE Trans.Inf. Theory, vol. 18, no. 4, pp. 531532, Jul. 1972

[5] T. Cui, L. Chen, and T. Ho. Energy efficient opportunistic network coding forwireless networks . in Proc. IEEE INFOCOM, 2008, pp. 10221030.

[6] D. S. J. De Couto, D. Aguayo, J. C. Bicket, and R. Morris. A high throughputpath metric for multi-hop wireless routing . in Proc. ACM MobiCom, 2003, pp.134146

[7] E. O. Elliot. Estimates of error rates for codes on burst-noise channels . BellSyst. Tech. J., vol. 42, no. 5, pp. 19771997, Sep. 1963.

[8] S. Gollakota and D. Katabi. Zig-zag decoding: Combating hidden terminals inwireless networks . in Proc. ACM SIGCOMM, 2008, pp. 159170.

[9] S. Hong and S. Katti. DOF: A local wireless information plane. in Proc. ACMSIGCOMM, 2011, pp. 230241.

[10] R. K. Jain, D. M. W. Chiu, and W. Hawe. A quantitative measure of fairness anddiscrimination for resource allocation in shared computer systems. DEC Research,Tech. Rep. TR-301, 1984.

[11] P. Karn. MACAA new channel access method for packet radio . in Proc. ARRLComput. Netw. Conf., 1990, pp. 134140.

[12] S. M. Kay. Fundamentals of Statistical Signal Processing . Upper Saddle River,NJ, USA: Prentice-Hall, 1998, vol. 2.

29

[13] X. Zhang and K. Shin. E-MiLi: Energy-minimizing idle listening in wirelessnetworks . in Proc. ACM MobiCom, 2011, pp. 205216.

[14] X. Zeng, R. Bagrodia, and M. Gerla. GloMoSim: A library for par- allel simu-lation of large-scale wireless networks . in Proc. IEEE PADS, 1998, pp. 154161.

30

AcknowledgementsAbstractIntroductionOverviewMotivationProblem DefinitionObjectives802.11ec Key TechniquesCorelatable Symbol Sequences

Literature SurveyBackground and Recent ResearchLiterature SurveyMACAW: A media access protocol for wireless LANsWiFi-Nano: Reclaiming WiFi efficiency through 800 ns slotsE-MiLi: Energy-Minimizing Idle Listening in Wireless NetworksCRMA: collision-resistant multiple accessDOF: a local wireless information planeSAM: Enabling practical spatial multiple access in wireless LANZig-zag decoding: Combating hidden terminals in wireless networksCSMA/CN: Carrier sense multiple access with collision notification

AlgorithmMAC Protocol DesignCoded Control Information Versus Message Control InformationPublic CSSs Versus Private CSSsControl During Data

11ec Channel Reservation PrimitivesInitiationReservationDeferralCSS AssociationPrimitive Extensions

Contending Flows and Vulnerability IntervalCoexistence With Legacy 802.11

Experimentation and ResultsExperimental evaluation of CSSExperimental SetupChannel Emulator ValidationCSS Length Versus Robustness TradeoffCSS Detection Versus Control Message DecodingDiscussion on Signal Correlation

Experimental Evaluation of 11ecBasic TopologiesLarge Network Simulation

Conclusion and Future workConclusionFuture Work

Report

Documents

Transcript of Report