Decentralized congestion control in etsi intelligent transport systems and its effect on safety applications

Resum

This thesis evaluates the effect of Decentralized Congestion Control (DCC) mechanisms in Intelligent Transport Systems (ITS) safety applications, which fall within the scope of the European Telecommunications Standard Institute (ETSI). The main objectives of this thesis are: 1. Study and characterize different Decentralized Congestion Control mechanisms in the ETSI Framework. 2. Propose improvements to the standardized DCC and the services that are affected by it. The motivation behind this work is the imminent appearance of Cooperative, Connected and Automated Mobility (CCAM) in real world scenarios, and the fact that there is already a standardized framework to abide by. The limits set by this framework do not only apply to standards, but also to the wide range of devices that run ITSs, from computers to embedded systems. The main guideline for this work is to imagine novel ways to tackle identified obstacles, while considering the current structure that will support and power these innovations. This principle drove the applied research work we carried out. This thesis stays within the scope of current standards, and any proposals or additions we developed must comply with the existing architecture. Hence, the contributions that involved additions to existing standards in the ETSI ITS framework had to consider the limitations listed below: • Compliance: the proposals for improvement have to adhere to the current ETSI ITS architecture. This means that, while we can propose new algorithms, the base they build upon should not change. One example is Generate-on-Time (GoT), the mechanism we proposed for the Cooperative Awareness basic service, which leverages the existing cross-layer interfaces between entities in the ETSI ITS architecture [1]. • Feasibility: one of the substantial difficulties we faced during the development of our contributions was the suitability of our proposal for its intended use. Vehicular networks mostly run in compact appliances, which most of the time have limited computing power and storage capacity. The ETSI set of standards for ITS conform a framework that is known as ETSI ITS-G5, since vehicular communications occur in the 5 GHz band. The current framework provides specifications for applications that enable safety and traffic efficiency [2]. The framework consists of entities that can be mapped to layers in the OSI model. The Access layer corresponds to OSI layers 1 and 2. It relies on access technologies based on IEEE 802.11 Outside the Context of a Basic Service Set (IEEE 802.11 OCB), a protocol specialized on Vehicular ad hoc Networks (VANETs), which means that nodes do not need infrastructure to communicate with each other. The Networking & Transport layer maps to layers 3 and 4. It is at this layer that GeoNetworking protocols reside. The Facilities and Applications layers represent OSI layers 5, 6 and 7. It is at the Facilities layer that services that support the applications layer resides. The Applications layer consists of applications that support road safety and traffic efficiency, and which provide an interface to the end user (i.e., Human-machine interface – HMI) or trigger functions through automated processes (e.g., an application that detects adverse road conditions). The first part of this thesis focuses on the Access layer, specifically on Decentralized Congestion Control (DCC) mechanisms [3]. These mechanisms are an integral part of Intelligent Transport Systems and VANETs in general. At the Access layer, the DCC component senses channel conditions (e.g., channel occupation), to which the station adapts or reacts. The specific actions that the DCC component takes to prompt the station to keep channel conditions depend on the specific mechanism used at this layer, and the Access layer uses its interfaces to other entities to make other layers aware of channel conditions and usage allowances by using a cross-layer component at the Management Entity. DCC mechanisms have one main objective: to maximize throughput while avoiding congestion. This means that a DCC mechanism should allocate a fair medium usage to ITS stations (ITS-Ss) such as vehicles and road-side units (RSUs) in order to keep network conditions below a certain level of channel usage – Channel Busy Ratio (CBR). CBR is the fraction of time the medium is occupied with information. DCC mechanisms in the ETSI framework have specific CBR targets, in the region of 60% to 68% for the Reactive and Adaptive mechanisms, respectively [3]. DCC mechanisms control medium usage through several techniques: 1) controlling the number of messages a vehicle can send – Transmission Rate Control (TRC); 2) controlling the transmission power used to transmit messages – Transmission Power Control (TPC); and 3) controlling the data rate at which stations communicate – Transmission Datarate Control (TDC). ETSI has specifications for these techniques, but its specification [3] that describes its DCC mechanism in the Access layer and its two approaches (Reactive and Adaptive) fall into the TRC category. In this regard, this thesis provides an overview of the ETSI Reactive DCC mechanism, while providing an in-depth study of the ETSI Adaptive DCC algorithm. ETSI Adaptive DCC stems from LIMERIC [4] [5], a linear message rate control algorithm that was proposed after the multiple shortcomings from the reactive approach were studied. However, LIMERIC and ETSI Adaptive DCC have a different set of parameters that affect characteristics such as throughput, number of serviceable ITS-Ss in the system, and convergence speed. The first contribution of this thesis is a thorough experimental evaluation of ETSI Adaptive DCC, comparing it to LIMERIC, and to a benchmark in which no DCC mechanism is present in order to assess the real effect of DCC mechanisms. One of the first conclusions of this thesis is that DCC mechanisms are crucial in order to keep system stability. In the absence of DCC, and even with a single type of traffic, successful communication between vehicles is affected in case of congestion. The Packet Delivery Ratio (PDR) – the ratio of successful message receptions over attempted transmissions – drops below 70% at distances shorter than 200 meters. On the other hand, when DCC mechanisms are present, PDR stays over 80% for distances around 400 meters. When multiple types of traffic are present, DCC mechanism manage to keep its performance, while systems without DCC show a drastic decrease in PDR even at short distances. However, not all Adaptive DCC mechanisms are equal. During our evaluation of ETSI Adaptive DCC and LIMERIC from [5] (identified as LIMERIC 0.79 in this work) we concluded that, while both mechanisms outperform the absence of DCC and keep channel usage within a limit, they do not behave the same way regarding the full set of objectives of a DCC mechanism. The CBR target is the maximum channel occupation which maximizes throughput, and this target must be met by also maximizing the number of vehicles sharing the medium. LIMERIC 0.79 has shortcomings in this area, since it causes medium underutilization when fewer vehicles are present, and it goes above the optimal usage level when too many stations share the medium. On the other hand, ETSI Adaptive DCC maximizes usage at lower vehicle densities, while showing a steady performance when a large number of vehicles share the medium. As an example, LIMERIC can support up to a thousand vehicles before going unstable, but starts overutilizing the medium with 300 vehicles, and underutilizes it below 200, whereas ETSI Adaptive DCC can handle optimally up to 1133 vehicles, and is still usable above that quantity, while allowing vehicles to maximize medium usage even at low densities. After assessing the strengths and weaknesses of ETSI Adaptive DCC, we confirmed its efficacy in steady state situations. However, we found out that its parameters cause slow convergence speeds in transitory situations. This is where the second contribution of this thesis emerged. We proposed Dual-, a simple modification of the ETSI Adaptive DCC mechanism which improves convergence speed. We evaluated Dual- against ETSI Adaptive DCC and confirmed that it improves its performance in rapidly changing scenarios while maintaining its behavior in steady state situations. The transitory situations simulated in this thesis are: 1) A road with a traffic light that controls the entrance into a freeway, typical of major motorways beginning in or crossing urban areas, such as interstate roads I-35 and I-78 in the United States. In this scenario, speed of convergence and its effect are measured. A group of cars wait in the red light before a curve that will head them into the freeway. While they are waiting in the red light, their dynamic CAM message generation rate drops to a minimum. When the group starts moving, vehicle dynamics change, and the message rate allowance has to adapt rapidly and keep channel usage within boundaries. 2) An intersection where a small group of 25 cars in a two-lane motorway incorporates into a larger group of 300 vehicles circulating in a six-lane highway, in which fairness is evaluated. Cars in the small group initially sense a lower channel occupation, which allow them to transmit most of the messages they generate, while the larger group senses a higher occupation which causes them to modulate their transmission rate accordingly. The objective of this scenario is to evaluate how fast the two groups adjust their transmission and generation rates after merging. Simulation results in these scenarios show that using Dual- increases convergence speed significantly. Experiments in transitory situations that measure convergence speed from 40-50 seconds in ETSI Adaptive DCC to 10-20 seconds with Dual-. Furthermore, the slow reaction time exhibited by ETSI Adaptive DCC in some experiments caused the system to act as if there were not a congestion control mechanism acting for a long stretch of time (i.e., in the range of tens of seconds). Furthermore, Dual- improves fairness in the system, allowing ITS-Ss to converge faster and use the medium more equitably. The impact of DCC is not restricted to the Access layer. The ETSI framework defines cross-layer entities (i.e., Management and Security Entities), which allow different layers to exchange information. It is through these entities that DCC mechanisms influence services in the Facilities layer. One of the main services at the Facilities Layer is the Cooperative Awareness (CA) basic service [6], which makes possible for ITS-Ss to exchange information about its position and status through Cooperative Awareness Messages (CAMs). The rate at which CAMs are generated is limited by DCC through a DCC cross-layer mechanism at the Management Entity. This is a desirable behavior, since it prevents stations to generate more messages than the amount that can be sent given the current network conditions, and it is included into the specification for the CA basic service. However, we found another effect from DCC in CAMs when multiple types of traffic are present. While the cross-layer DCC mechanism provides the time interval at which CAMs can be generated, it does not provide the exact time at which messages are to be transmitted. Thus, if multiple types of traffic are present (even with lower priority than CAMs), and CAM generation is not synchronized with message transmission, CAMs may wait at the DCC queues, which increases latency. This causes a gap between the information contained in the CAM waiting at the queue (i.e., position, status, heading) and the current status of the vehicle. Following the rules established by ETSI [6], the CAM generation interval is limited within a range from 100ms to 1s by four parameters: • Vehicle dynamics (i.e. shifts in position, acceleration, and heading), • T_Elapsed (i.e. period since the last CAM was generated), • T_GenCam_DCC (i.e. lower limit of the CAM generation interval given by the DCC mechanism, expressed as tdcc in this thesis), and • T_GenCam (i.e. upper limit of the CAM generation interval). A CAM generation is triggered after T_Elapsed > T_GenCam_DCC by either of two conditions: 1) if vehicle dynamics have exceeded certain thresholds, or 2) if T_Elapsed > T_GenCam. Where, T_GenCam is T_Elapsed for the last CAM triggered by condition 1. After three CAMs have been triggered by condition 2, T_GenCam is set to 1 second. When an outgoing CAM message reaches the Access layer, it is placed at the DCC queue corresponding to its traffic class (TC2). In the presence of a single type of traffic, the time dequeuing occurs and tdcc will usually coincide, which means that a generated CAM will not have to wait in the DCC queues and it will only be delayed by the medium access control mechanism (i.e., EDCA) until it is finally transmitted. But when the system is out of synchrony, delays at the DCC queues follow a uniform distribution from zero to tdcc, with an average of tdcc/2. This becomes an issue when tdcc is high. Our experimental evaluation confirms these values. End-to-end delays influence other metrics, such as Information Age. This metric, as defined in this thesis, is the period that passed between the time information contained in the last received CAM was generated and the moment a new CAM is received. End-to-end delay and inter-packet gaps (IPG) (i.e., the time between two consecutive CAM receptions) add up to conform information age. As the third contribution from this thesis, we propose Generate-on-Time (GoT), an addition to the generation rules of the CA basic service that synchronizes CAM generations and DCC dequeuing instants, leveraging the cross-layer DCC mechanism by adding a second parameter to the information exchanged between the Access and Facilities layers. With GoT, a CAM will only be generated if the time for transmission is near, allowing to include the most recent information served by the vehicle data provider. This mechanism reduces delays at the DCC queues, but keeps transmission instants equal, this means that messages generated either by the ETSI CA basic service or GoT are transmitted exactly at the same time and at the same rate, but those generated by GoT contain more recent information. The real advantage of delaying CAM generation depends on the ability of obtaining updated location information in the vehicle. Global Navigation Satellite Systems (GNSSs) have a minimum time between measurements around 100 ms. However, because the CAM generation is not synchronized with the GNSS measurements, even small delays can allow obtaining an updated location position. Moreover, the use of additional sensors (inertial sensors) for tracking functionality enables a continuous reading of vehicle position between GNSS measurements. The improvement in information age that GoT brings is, in average, in the region of 20%. In absolute numbers, this can be reflected in real-life scenarios. For a vehicle density of 50 vehicles/km per lane, the reduction in delay due to GoT is of 302 milliseconds on average. This can be translated to errors that affect safety, since at this density the average speed is of 14.27 m/s, that reduction represents 4.3 meters, above the required position accuracy for many safety applications. For example, 4.3 meters can be the difference between being in one lane or in another one, or being inside or outside an intersection. Besides the three contributions from this thesis, its main value resides in the fact that it stays within the scope and specifications in the ETSI ITS framework. Our proposals, Dual- and GoT, can be easily implemented within the ETSI ITS architecture. Dual- only adds an extra step to decide whether to use a higher value for the  parameter, and GoT uses the same interfaces that allow the exchange of information between the Access and Facilities layers. Dual- is already a contribution for the research community. The developers of Artery, an ETSI-compliant simulation toolkit for vehicular networking, have added an implementation of Dual- as an option along with ETSI Adaptive DCC. The final takeaway from this thesis is that the ETSI ITS Framework is a strong base for reaching the final goal of Cooperative, Connected and Automated Mobility (CCAM): safer and more efficient roads. As future lines of work that can stem from this thesis are the study of DCC in safety applications that use high priority and multi-hop messages that use ETSI-defined forwarding mechanisms. Additionally, the study of DCC in other technologies such as C-V2X is also a relevant object of study, since connected communications already consider the use of different access technologies. References [1] European Telecommunications Standards Institute (ETSI), Intelligent Transport Systems (ITS); Communications Architecture EN 302 665 - V1.1.1, 2010. [2] European Telecommunications Standards Institute (ETSI), Technical Report. Intelligent Transport Systems (ITS);. Vehicular Communications;. Basic Set of Applications;. Definitions TR 102 638 - V 1.1.1, 2009. [3] European Telecommunications Standards Institute (ETSI), Intelligent Transport Systems (ITS); Decentralized congestion control mechanisms for Intelligent Transport Systems operating in the 5 GHz range; Access layer part, EN 102 687 - V1.2.1, 2018. [4] G. Bansal, J. B. Kenney and C. E. Rohrs, "LIMERIC: A Linear Adaptive Message Rate Algorithm for DSRC Congestion Control," IEEE Trans. Veh. Technol., vol. 62, no. 9, pp. 4182-4197, 2013. [5] A. Rostami, B. Cheng, G. Bansal, K. Sjöberg, M. Gruteser and J. B. Kenney, "Stability Challenges and Enhancements for Vehicular Channel Congestion Control Approaches," Trans. Intell. Transp. Syst., vol. 17, no. 10, pp. 2935-2948, 2016. [6] European Telecommunications Standards Institute (ETSI), Intelligent Transport Systems (ITS); Vehicular communications; Basic set of applications; Part 2: Specification of Cooperative Awareness basic service - EN 102 687 - V1.2.1, 2018.