A Performance Visualization and Fine-tuning 
Tool for Arterial Traffic Signal Systems 
 
 
A THESIS 
SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL 
OF THE UNIVERSITY OF MINNESOTA 
BY 
 
Jianfeng Zheng 
 
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR 
THE DEGREE OF MASTER OF SCIENCE 
 
Henry X. Liu (Adviser) 
 
April 2014  
 
 
 
 
 
 
 
 
 
 
 
 
© Jianfeng Zheng 2014 
All Right Reserved
 
 
i 
 
ACKNOWLEDGEMENTS 
I would like to express my sincere gratitude to my adviser, Prof. Henry X. Liu, for his 
constant support, patient guidance and many constructive suggestions throughout this 
research. I would also like to thank Prof. Gary Davis and Prof. Zongxuan Sun, for 
taking interests in this research and serving in my committees. Many courses from Prof. 
Gary Davis were greatly beneficial and inspiring for me. 
My special thanks go to my fellow colleagues in Prof. Liu's research group, in 
particular, Dr. Heng Hu, Mr. Jie Sun, Dr. Xinkai Wu from Cal Poly Pomona, for many 
advices and assistances. Special thanks also go to Mr. Steve Misgen, Mr. Timothy 
Bangsund and Mr. Curt Krohn for assistance with field implementations of the data 
collection units. Mr. Kevin Schwartz and Mrs. Nicole Flint also helped with the field 
experiments for offset changes. Their helps are also acknowledged. 
Last but not least, my gratitude goes to the unconditional love and supports from 
my families, my parents, Mr. Quanshun Zheng and Mrs. Zhujiao Lin, my sisters, Mrs. 
Surong Zheng and Mrs. Suting Zheng. My gratitude and love to my fiancée, Miss 
Hanyue Li, who walked all the way with me for past 6 years, are beyond words.  
  
 
 
ii 
 
A Performance Visualization and Fine-tuning Tool for Arterial Traffic 
Signal Systems 
By Jianfeng Zheng 
ABSTRACT 
Maintaining an efficient traffic signal operation is a challenging task for many traffic 
management agencies. Due to the intensive labor cost required, most of the traffic 
signals in the US are retimed once every 2-5 years. However, it has been shown in the 
past that traffic delay increases 3-5% per year simply because the timing plans are not kept up 
to date. For many resource-constrained agencies, it would be desirable to reduce the signal re-
timing costs by automating all or portion of the manual process. The research makes one-step 
forward towards this direction. In this research, we developed a performance monitoring 
and visualization tool for arterial traffic signal systems, aiming at reducing the labor 
cost for signal retiming, and helping to identify signal parameter adjustment 
opportunities. Specifically, an automated data collection unit (DCU) was developed to 
collect high-resolution event-based data from signal controller cabinets. Using the high-
resolution data, two parameter fine-tuning algorithms were proposed, one for offset and 
another for green splits. To fine-tune signal offsets, a practical procedure to construct 
the time space diagram (TS-Diagram) to visualize the progression quality on arterials 
was proposed. The TS-Diagram was calibrated and validated using the field data 
collected from the DCUs and the probe vehicle runs. Reasonable agreements between 
the field observations and the generated TS-Diagrams were found. A field experiment 
was then carried out, to illustrate how decisions of changes could be made by intuitively 
evaluating the TS-Diagram. For green splits, an adjusted measure of effectiveness 
(MOE), the utilized green time (UGT), was proposed for performance evaluation. The 
information was further tabulated in the form of ring-and-barrier diagram to facilitate 
evaluation. Field examples were also illustrated to demonstrate implementation 
potentials for green split evaluation and fine-tuning.  
Keywords: Traffic signal operation, High-resolution data collection, Performance 
evaluation, Data visualization, Arterial coordination, Fine-tuning.  
 
 
iii 
 
TABLE OF CONTENTS 
 
ACKNOWLEDGEMENTS ............................................................................................... i 
ABSTRACT ...................................................................................................................... ii 
LIST OF FIGURES ........................................................................................................... v 
LIST OF TABLES............................................................................................................ vi 
CHAPTER 1 INTRODUCTION ...................................................................................... 1 
1.1 Research Motivation ................................................................................................ 1 
1.2 Thesis Summary and Contribution .......................................................................... 2 
1.3 Organization of This Thesis ..................................................................................... 3 
CHAPTER 2 BACKGROUND AND RELEVANT WORK ............................................ 4 
2.1 Traffic Signal Control at Intersections ..................................................................... 4 
2.2 Data Collection for Signalized Intersections ........................................................... 6 
2.3 Event-based Data Collection ................................................................................... 7 
2.4 Optimization for Traffic Signal System ................................................................... 9 
2.5 Summary ................................................................................................................ 10 
CHAPTER 3 SYSTEM ARCHITECTURE AND DATA COLLECTION UNIT (DCU)
 ......................................................................................................................................... 12 
3.1 SYSTEM ARCHITECTURE ................................................................................ 12 
3.2 Development of Data Collection Unit ................................................................... 14 
3.2.1 Hardware Design of DCU ............................................................................... 15 
3.2.2 Firmware Design ............................................................................................. 16 
3.2.3 Installation and In-House Lab Testing ............................................................ 19 
3.2.4 Field Deployments .......................................................................................... 22 
CHAPTER 4 OFFSET EVALUATION AND FINE-TUNING ...................................... 23 
4.1 Background about Time Space Diagram ............................................................... 23 
4.2 Constructing Link TS-Diagram ............................................................................. 25 
4.3 Validation of TS-Diagram ...................................................................................... 31 
4.3.1 Case 1: TH 55 .................................................................................................. 31 
4.3.2 Case 2: TH 13 .................................................................................................. 33 
4.4 Fine-Tuning Using Ts-Diagram with Offset and Lead-Lag Sequence .................. 34 
CHAPTER 5 GREEN SPLITS EVALUATION AND FINE-TUNING ......................... 41 
5.1 Measure of Effectiveness for Green Splits ............................................................ 41 
5.1.1 v/c ratio ........................................................................................................... 41 
5.1.2 Queue Service Time (QST) ............................................................................. 42 
 
 
iv 
 
5.1.3 Utilized Green Time and Slack Green ............................................................ 43 
5.2 Green splits rebalancing based on UGT ................................................................ 45 
5.3 Split Fine-tuning Example ..................................................................................... 45 
CHAPTER 6 CONCLUDING REMARKS AND FUTURE RESEARCHES ............... 49 
6.1 Concluding Remarks ............................................................................................. 49 
6.2 Limitations and Future Researches ........................................................................ 50 
BIBLIOGRAPHY ........................................................................................................... 51 
 
  
 
 
v 
 
LIST OF FIGURES  
 
Figure 1.1 2012 National Traffic Signal Report Card ....................................................... 1 
Figure 2.1 Typical intersection layout (left) and ring-and-barrier diagram (right) at a 
typical four-arm intersection ............................................................................................. 4 
Figure 2.2 Illustration of Offset ........................................................................................ 6 
Figure 3.1 Typical layout and detector setting of an arterial ........................................... 12 
Figure 3.2 System architecture of the fine-tuning tool ................................................... 13 
Figure 3.3 Procedure for traffic signal fine-tuning ......................................................... 14 
Figure 3.4 Architecture of the traffic data collection components .................................. 15 
Figure 3.5 General architecture and snapshot of the DCUs ............................................ 16 
Figure 3.6 Data sample and format ................................................................................. 17 
Figure 3.7 Flow chart of the firmware for the DCUs ...................................................... 18 
Figure 3.8 Installations of the TS1 DCU (A) and TS2 DCU (B) .................................... 20 
Figure 3.9 TS1 (A) and TS2 (B) Testing Platform .......................................................... 21 
Figure 4.1 Example of TS-Diagram ................................................................................ 23 
Figure 4.2 Screenshot of Synchro TS-Diagram .............................................................. 24 
Figure 4.3 Illustration of queuing at intersection with QOD .......................................... 28 
Figure 4.4 Example of CFP ............................................................................................. 29 
Figure 4.5 Illustration of car following and dummy vehicle. ......................................... 31 
Figure 4.6 Layout and TS-Diagram for TH 55 ............................................................... 32 
Figure 4.7 Estimated CFPs VS Observed CFPs for EB(Up) and WB (Down) traffic .... 33 
Figure 4.8 TS-Diagrams for selected intersections on TH 13 ......................................... 34 
Figure 4.9 Offset reference point and combinations of lead lag sequence ..................... 35 
Figure 4.10 TS-Diagrams for selected intersections on TH 13 with average (left) and 
relatively congested conditions (right) ............................................................................ 36 
Figure 4.11 Before-After comparisons of TS-Diagrams with relatively congested 
condition .......................................................................................................................... 39 
Figure 4.12 Before-After comparison of maximum queue length from SMART-Signal 
System ............................................................................................................................. 40 
Figure 5.1 Example of traffic events from long detector at stop bar .............................. 42 
Figure 5.2 Illustration of QST ......................................................................................... 43 
Figure 5.3 Example of QST and UGT calculation .......................................................... 44 
Figure 5.4 Intersection layout and ring-and-barrier diagram of the signal timing .......... 46 
Figure 5.5 Green Time VS Utilized Green Time ............................................................ 47 
Figure 5.6 Ring-and-barrier diagrams with UGTs .......................................................... 48 
 
  
 
 
vi 
 
LIST OF TABLES 
 
Table 4.1 Before-After comparison of delay from SMART-Signal System ................... 40 
 
 
1 
 
CHAPTER 1 INTRODUCTION 
1.1 Research Motivation  
Maintaining an efficient traffic signal operation is a challenging task for many traffic 
management agencies. In the current state-of-the-practice, the design and update of 
traffic signal timing parameters, the so-called signal retiming, are mostly based on data 
collected manually from the field. During the retiming process, the updated signal 
timing needs to be validated by field observations and minor changes would be made if 
a discrepancy is observed. Such process is time-consuming and costly. Due to the 
intensive labor cost involved, most of the traffic signals in the U.S. are only retimed 
every 2-5 years, despite of a high benefit/cost ratio of 40:1 (Gordon, 2010). Under such 
circumstances, it is not surprising to see an overall D+ grade for traffic signal systems 
from the 2012 National Traffic Signal Operation Self-Assessment Survey, as shown in 
Figure 1.1. As indicated in National Transportation Operations Coalition (2012), 
“improvement and investment in traffic signal operations remain critical”. 
 
Figure 1.1 2012 National Traffic Signal Report Card 
[Source: National Transportation Operations Coalition, 2012] 
 Currently, many intersections in the U.S. are operated with vehicle-actuated 
 
 
2 
 
signal controller using different vehicle detection systems installed in the field. 
However, few agencies have archived or analyzed the operational data from the existing 
signal control systems to improve signal operation. One of the main impediments is that 
traditional signal controllers normally aggregate data into 5-60 minutes bins, which is 
unsuitable for performance evaluation of traffic signal system (Liu et al., 2008). 
Moreover, without standardized data logging and aggregation capability, it is rather 
inconvenient and difficult for the engineers to access the data within the controllers.  
 To improve traffic signal operation, researchers in the University of Minnesota 
have developed the SMART-SIGNAL (Systematic Monitoring of Arterial Road Traffic 
and Signals) system, for the purpose of performance monitoring of arterial traffic signal 
system (Liu et al., 2009). The SMART-SIGNAL system can automatically collect high-
resolution event-based traffic data at signalized intersections and generate real-time 
performance measures e.g. queue length and travel time. However, implementation 
issue exists for the SMART-SIGNAL system to interface with NEMA (National 
Electrical Manufacturers Association) TS2 cabinets using the SDLC (Synchronous Data 
Link Control) interface (Liu et al., 2012). More importantly, even though the algorithms 
to estimate various performance measures have been developed in previous researches 
(Liu et al., 2008; Liu et al., 2009; Liu & Ma., 2009), how to utilize the archived data to 
readily fine-tune the traffic signal system has not been addressed yet. This study 
attempts to fill in these gaps. 
1.2 Thesis Summary and Contribution 
In this research, we further develop for the SMART-SIGNAL system, with a 
focus on the development of a systematic fine-tuning tool for the arterial traffic signal 
system. Our goal is to provide a convenient tool to assist agency engineers to evaluate 
and fine-tune the arterial traffic signal systems, thus to maintain efficient signal 
operations. The objective of the fine-tuning tool is twofold: first, to reduce the efforts 
for manual validation and fine-tuning during the signal retiming process; second, to 
assist traffic engineers to identify adjustment opportunities over time. 
In addition, to address the technical issue with data collection, two types of data 
collection units (DCUs) are developed based on embedded system technology, to 
 
 
3 
 
interface with existing traffic signal cabinets. Building on high performance 
microcontrollers, the newly developed DCUs are compact and easy-to-install in the 
field, and thus could serve as a cost-effective and vendor independent solution for 
agencies to collect high-resolution event-based traffic data. 
 Based on the event-based data, two modules are proposed for performance 
visualization and fine-tuning of offsets and green splits respectively. For offsets, we 
propose a practical procedure to generate the time-space diagram (TS-Diagram), for 
visualization and evaluation of the progression quality for two-way coordinated arterials. 
For green splits, an adjusted measure of effectiveness (MOE) extended from queue 
service time (QST), the utilized green time (UGT), is proposed for performance 
evaluation. The ring-and-barrier diagram is adopted to tabulate the information, in order 
to facilitate evaluation. 
1.3 Organization of This Thesis 
The rest of the thesis is organized as follows: A brief review of the research background 
and relevant works are stated in Chapter 2. Chapter 3 describes the system architecture 
of the tool, as well as the details of the development of the DCUs. Chapter 4 introduces 
the offset fine-tuning module, with a focus on generating the TS-Diagram. The 
validation of the TS-Diagram, and illustration of fine-tuning application are described 
as well. The split fine-tuning module is described in Chapter 5. Finally, the concluding 
remarks and future research directions are summarized in Chapter 6. 
  
 
 
4 
 
CHAPTER 2 BACKGROUND AND RELEVANT WORK 
This chapter will start with a brief introduction of traffic signal control. Then, relevant 
research works regarding data collection and signal optimization will be reviewed. 
2.1 Traffic Signal Control at Intersections 
To set the stage, we introduce briefly the NEMA ring-and-barrier signal controller, 
which are widely used throughout the U.S. For more details, we refer to (Koonce et al., 
2008).  
 The NEMA signal controllers operate a sequence of signal phases in a repetitive 
manner to assign right-of-way to different traffic movements. The operation and setting 
of the controller are typically tabulated in the ring-and-barrier diagram, showing the 
phase sequence , movement direction, and green durations for different signal phases. 
One illustration for a typical four-arm intersection is shown in Figure 2.1.  
 
Figure 2.1 Typical intersection layout (left) and ring-and-barrier diagram (right) 
at a typical four-arm intersection 
[Reproduced from: Signal Timing Manual, 2008] 
 In the example intersection, there are eight traffic movements, controlled by 
signals with associated signal phase numbers from one to eight. Two rings and barriers 
are featured in the ring-and-barrier diagram, with phases 1-4 in ring 1 and phases 5-8 in 
the ring 2. Phases on the two rings are executed concurrently, under certain restrictions: 
any two phases on the same ring are associated with two conflict movements and cannot 
 
 
 
5 
 
be in green concurrently, so as for any two phases in different sides of the barriers. For 
each phase, the duration intervals for green, yellow, and red clearance signal are 
assigned, with the sum as the phase duration. The controller will execute each phase 
based on its order on the ring repeatedly. Every one complete sequence of all phases is 
called a cycle, with the time elapsed during a cycle as the cycle length. The percentage 
of a cycle length allocated as a green phase is the green split of the phase. Because of 
the cyclic operation, there are typically two clocks running in a controller, the system 
clock tracking the time of day, and a cycle clock tracking the time within a cycle. 
 Depending on different types of signal controls, the phase duration can be also 
different. Due to safety concerns, the yellow time and red clearance time are typically 
fixed regardless of control mode. The green time of each phase is constant for fixed 
signal, but could be different from cycle to cycle for actuated control, in response to the 
detector actuations. 
 On arterials, coordinated-actuated control is normally deployed, with two 
through movements on the major road prioritized by the coordinated phases, phase 2 
and phase 6 by convention. The coordinated phases will be synchronized with a 
specified reference point to facilitate vehicle platoon progression on the main road. To 
maintain the synchronization, the cycle length will be fixed for each intersection and a 
common cycle length will be shared by several adjacent intersections. Typically, a 
master intersection is indicated, running a master cycle clock. The rest of the 
intersections will run local cycle clock individually, with a specified shift from the 
master cycle clock. The difference between a local cycle clock with the master cycle 
clock, given a specified reference point, is the offset of the intersection.  
 One illustration of the offsets is shown in Figure 2.2, with the green end of the 
coordinated phase as the reference point. The intersection (Int.) B is the master 
intersection, and local clock at Int. A is shifted with 30 sec to the right from the master 
clock at Int. B. Accordingly, the offset of Int. B is 0 sec, and offset of Int. A will be 30 
sec. 
 
 
6 
 
 
Figure 2.2 Illustration of Offset 
[Source: Signal Timing Manual, 2008] 
 To keep a fixed cycle length for the coordinated actuated control, only some of 
the phases are vehicle-actuated. One common choice is to set the non-coordinated 
phases as actuated, with coordinated phases as non-actuated. Thus, the unused green 
time of the non-coordinated phases will revert to the coordinated phases. Vice versa, the 
coordinated phases can be also set as actuated, with non-coordinated phases as non-
actuated, and the unused green time of coordinated phases will revert to non-
coordinated phases. 
2.2 Data Collection for Signalized Intersections 
While real-time performance measures for freeway are commonly available in the forms 
of speeds and travel time nowadays, they are relatively new for signalized intersections 
(Day et al., 2010a). According to 2012 National Traffic Signal Operation Self-
Assessment Survey, arterial data collection remains at F grade from 2007 to 2012, 
showing significant needs and potentials for improvements (National Transportation 
Operations Coalition, 2012). 
Int.A 
Int.B 
 
 
7 
 
 In general, many existing controllers for traditional traffic signal systems have 
the capability to generate basic measure of effectiveness (MOE) reports. For instance, 
Eagle® EPAC 300 controller can generate MOE reports with volume, stops, delays and 
utilization of green time (Balke et al., 2005). Volume, occupancy and average speed (if 
speed detectors are available) aggregated in 5-60 minutes can be also obtained from 
Econolite ASC/3 family controller (Econolite ASC/3 Controller Specification, 2012). 
However, the MOEs generated within controller have not been standardized yet, and is 
not utilized by most of the engineers operating the traffic system (Balke & Herrick, 
2004; Balke et al., 2005). 
 One of noteworthy work of data collection system for signalized network is the 
Arterial Performance Measurement Systems (A-PeMS) developed by University of 
California, Berkeley in cooperation with the Los Angeles Department of Transportation 
(Petty et al., 2005). A-PeMS aggregates traffic volume and occupancy data in 30-second 
intervals, and can generate performance measures such as link travel time and 
intersection delays. 
 Recently, emerging interests can be found in using high resolution event-based 
data for performance measure of the traffic signal system, and encouraging results were 
achieved (Smaglik et al., 2007; Day et al., 2009; Day et al., 2010a; Day et al., 2010b; 
Liu & Ma, 2009; Liu et al., 2009; Wu et al., 2010). The "events" here are referred as 
vehicle-detector actuation and de-actuation, as wells as traffic signal phase changes, all 
in high-resolution time intervals. The advantages of the event-based data over 
aggregated data are: (a) performance indexes based on large (hour or day) or medium 
time interval (phase or cycle) can be easily calculated and (b) investigation on 
microscopic level is feasible and could help for more accurate and detailed evaluation, 
particularly during congested traffic condition. 
2.3 Event-based Data Collection 
To collect high-resolution event-based traffic data, additional adjustments are needed 
for most of existing traffic signal systems. Several solutions have been reported so far, 
whereas there is a lack of solutions that can be readily deployed.  
 
 
8 
 
 To address the issue of non-standardized interface of controller MOE output, 
Bullock and Catarella (1998) developed the controller interface device (CID) that can 
interface signal controller with PC regarding signal events and detector events in real 
time. With CID, the hardware-in-the-loop simulation (HILS) can be established 
incorporating traffic controllers from different vendors. Advanced Traffic Analysis 
Center (ATAC) of North Dakota State University (Smadi & Birst, 2006) has developed 
CID for NEMA TS2 controllers. Since the CIDs are designed for interfacing controller 
with simulation package in personal computer, they cannot be used to collect detector 
data independently. 
 TTI’s TSPMS extends CID with I/O data acquisition card to obtain signal and 
detector event data from TS1 cabinet. The data is further processed by an industrial 
computer within cabinet. For TS2 cabinet, they use enhanced BIU by Naztec Inc. to 
interface with SDLC bus, and the industrial computer is replaced by a rugged laptop for 
easier user interaction (Balke et al., 2005; Sunkari et al., 2011). This architecture for 
data collection system for TS1 cabinets is adopted and extended by the first version of 
the SMART-SIGNAL system. 
 National Institute for Advanced Transportation Technology (NIATT) in 
University of Idaho also developed a data logging device for high-resolution event-
based data collection, adopting similar architecture of data collection system in TSPMS 
for TS1 cabinets (Ahmed & Brian, 2008). Although Synchronous Data Link Control 
(SDLC) interface is recognized, the connection of data logging device for TS2 cabinet 
is still based on harness cable connecting the connection matrix on the back panel. 
 One trial version of Econolite Controller Software, Econolite ASC/3 Data 
Logger, features a logging capability that can retrieve event-based data (Smaglik et al., 
2007). Events are recorded with 0.1 sec time stamping, and stored as binary-formatted 
files. The raw data can be stored within controller for 30 hours, and downloaded via 
FTP. The data logger is now available within the newly released CENTRACS system 
(CENTRACS System Datasheet, 2013). 
 Major limitations of existing solutions can be summarized as: (1) solutions with 
multiple off-the-shelf components incur unnecessary cost and space within cabinets. It 
 
 
9 
 
further increases the difficulty of installation in the field. (2) vender dependent solution 
e.g. Econolite ASC/3 Data Logger, is far from standardization and is not convenient for 
agencies with controllers from different vendors. While efforts can seen from vendor 
side to enhanced event-based data collection within controllers (Duncan, 2011), for a 
perceived future, the progress for standardization could be gradual and time consuming. 
This calls for an alternative, vendor independent solution for high resolution event-
based data collection. 
2.4 Optimization for Traffic Signal System 
Since Webster’s classical work (Webster, 1952), extensive researches have been 
dedicated to the optimization of cycle length, split, and offset of a traffic signal system. 
Many of the well-established methodologies, however, are concentrated on optimization 
during design stage of signal timing.  
 Early studies were primarily concentrated on parameter optimization for fixed 
time signal control. In Webster's seminal work, analytical formulas were derived for 
split and cycle length optimization for fixed time control at isolated intersections 
(Webster, 1958). Considering intersections in coordinated networks, Gartner & Little 
(1975) proposed the generalized combination method, which uses a simplified link 
based delay calculation to optimize offset, split and cycle length simultaneously. Later, 
MAXBAND and its derivatives were proposed, in that the offset optimization problem 
is formulated as a mix integer linear programming (MILP) problem with green 
bandwidth as the optimization objective (Little et al., 1981; Gartner et al., 1990). 
Another important work was done by Robertson (1969), in which the platoon dispersion 
model was developed for more comprehensive estimation of performance index. The 
platoon dispersion model was then used in the popular signal optimization package 
TRANSYT, and later adopted in the SCOOT adaptive signal control system (Robertson 
& Bretherton, 1991). 
 With increasing implementations of vehicle-actuated controllers, attentions were 
then paid to optimize the parameters of actuated control. Lin (1982a) proposed models 
to estimate the average green for fully actuated signals assuming Poisson distribution 
for the arrivals. The model was further extended to estimate cycle length and green split 
 
 
10 
 
for both semi-actuated and fully actuated signal control (Lin, 1982b). Extension of the 
work can be also found in Akcelik (1994) by assuming M3 model by Cowan (1975) for 
headway distribution of the arrivals. To convert fixed signal plan to actuated/semi-
actuated signal plan, Skabardonis (1996) proposed adjustments with focus on force off, 
yield point and max green time of the coordinated-actuated signals. The essence of the 
adjustments is to better estimate the green start and end times of the coordinated phases.  
 More recently, with the increasing data availability, several researches have been 
conducted to optimize coordinated-actuated traffic signals based on signal and detector 
data. For instance, Yin et al (2007) developed an offset refiner utilizing historical signal 
data only. In addition to bandwidth, the red-meeting probability was proposed as 
another metric, to minimize the occurrence of early return to green. Based on the similar 
concept, Zhang and Yin (2008) proposes a robust synchronization approach to extend 
the MAXBAND with uncertain green start times. 
 Some other researches intend to investigate signal and detector data jointly. 
Abbas et al. (2001) introduce the cyclic volume and occupancy profiles and develops a 
continuous adjustment algorithm for offset transitioning. Based on time-stamped phase 
and detector state changes collected by a logger in the Econolite ASC/3 controller, 
Smaglik et al. (2007) demonstrates procedures to calculate popular performance 
measures such as v/c ratio, arrival types and vehicle delays, for actuated traffic signals. 
The same type of data is also used in Day et al (2010), which develops a visualization 
tool, namely the Purdue Coordination Diagram (PCD), for evaluating coordination 
qualities on arterials. Day & Bullock (2011) further extends the combination method, 
and propose the link pivoting combination method (LPCM) for offset optimization.  
2.5 Summary 
Over the past decades, the advances in computer and communication technologies make 
it feasible to collect and archive massive amount of high-resolution data for the 
signalized intersections. This is particularly attractive and beneficial for performance 
monitoring and evaluation, the precursors to effective and efficient operations. While 
some approaches have been proposed for event-based data collection, a general and 
readily implementable solution is still lacking. Regarding the optimization of the traffic 
 
 
11 
 
signal systems, on one hand, many well-established methods are only suitable in design 
stage of signal timing, requiring manually collected data. On the other hand, recent 
approaches either adopt simplified optimization objective e.g. green signal bandwidth, 
or rely on simplified assumptions that only adjacent intersections are considered, which 
could fail to address various concerns raised by many associated factors for signal 
coordination on arterials (Koonce et al., 2008). 
In this study, we aim to develop a systematic tool for event-based data collection 
and performance evaluation and fine-tuning, for signalized arterials. A compact, easy-
to-install, and vendor independent data collection unit (DCU) was developed for high-
resolution event based data collection. Using event-based data, two modules were 
developed for fine-tuning of offset and green splits, respectively. Different from 
previous approaches in the literature that focuses on optimization algorithms, the 
development of fine-tuning methodologies here was concentrated on visualizing the 
performance of the arterial traffic signal system. Hence, various concerns for signal 
timing parameters could be considered and evaluated easily. The details of development 
and potential applications of the developed tools are presented in the following chapters. 
  
 
 
12 
 
CHAPTER 3 SYSTEM ARCHITECTURE AND DATA 
COLLECTION UNIT (DCU) 
This chapter will describe the system architecture of the tool, followed by the details of 
the development of the DCUs for high-resolution event-based data collection.  
This research is focus on signalized arterials with two through directions 
coordinated. In the investigated scenarios, we assume that advance detectors are 
available for detection of through traffic, with the stop bar detectors available for the 
left turn and side street traffic. This is a simple yet quite common setting for the 
arterials. An illustration of the setting is shown in Figure 3.1 
 
Figure 3.1 Typical layout and detector setting of an arterial 
3.1 SYSTEM ARCHITECTURE 
For compatibility, the system architecture of the tool is inherited from the SMART-
SIGNAL system. This is shown in Figure 3.2. The first box indicates the data collection 
system implemented in the field. Two types of data, the signal events and detector 
events are collected, stored in log files, and transmitted to the central database, indicated 
in the second box. Then, evaluation report will be generated in the central server, and 
will be made available to the traffic engineers. Note that, the tool will serve as a 
decision support system, while the final change and implementation decisions still need 
to be made and executed by engineers. 
 
 
13 
 
 
Figure 3.2 System architecture of the fine-tuning tool 
[Extended from SMART-SIGNAL System architecture (Liu et al., 2009)] 
 Two fine-tuning modules are proposed, for offset and splits respectively. While 
the details of each fine-tuning modules will be described in chapter 4 and 5, the 
procedures for fine-tuning process are illustrated in Figure 3.3. As shown in the figure, 
the fine-tuning procedure is a periodical and iterative process, with implementation of 
changes in the field. Additionally, the fine-tuning of green splits and offsets are 
proposed to be executed in a parallel manner. Since many agencies have special 
concerns over the progression quality on main road, the proposed fine-tuning of green 
splits is restricted to non-coordinated phases only, to ensure minimal impact on the 
progression. 
 
 
 
14 
 
 
Figure 3.3 Procedure for traffic signal fine-tuning 
3.2 Development of Data Collection Unit 
The DCUs serve as the units to collect the high-resolution event-based traffic data. 
Figure 3.4 illustrates data collection components in the field. Data collection elements 
include the DCUs in each intersection and a data server in the remote center.  
 
 
15 
 
 
Figure 3.4 Architecture of the traffic data collection components 
[Adopted from Balke et al., 2005] 
3.2.1 Hardware Design of DCU 
Previous development of data collection system in SMART-SIGNAL system is based on 
the prototype of the TSPMS system, using industrial computer with data acquisition 
card (Liu et al., 2008). With increasing replacements of NEMA TS1 cabinets by the 
NEMA TS2 cabinets, compatibility issues occur for the data collection system to 
interface with SDLC bus in the NEMA TS2 cabinets. The design with multiple 
components also occupies cumbersome space and further complicates wiring within 
cabinets. The re-designed DCUs will overcome these issues.  
 To be compatible with both TS1 and TS2 cabinets, we designed two types of 
DCUs: TS1 DCU with interface based on contact closures on back panel of the cabinet, 
and TS2 DCU with interface using SDLC bus. Both types of the DCUs are designed as 
industrial single board computers, providing media card slot to store log files and an 
Ethernet port for data transmission. A serial port is also provided for trouble shooting of 
the DCUs. To avoid interrupting the signal operation, the interfaces of DCUs with the 
 
 
16 
 
cabinets are designed to be unidirectional with inputs only, and are not capable to output 
electrical signals. The architecture of the DCUs is shown in Figure 3.5, with snapshots 
of sample devices. 
 
Figure 3.5 General architecture and snapshot of the DCUs 
3.2.2 Firmware Design 
Data Format 
The data regarding every change of detector actuation status and signal status will be 
recorded and stored. For detector events, the start time, detector ID and actuation 
duration length are stored. For signal events, signal start time, signal status, phase ID 
and signal duration length, are stored. The data sample and the format are shown in 
Figure 3.6. 
 
 
 
17 
 
 
Figure 3.6 Data sample and format 
Program Flow Chart of the Firmware 
The flow chart of firmware for the DCUs is shown in Figure 3.7. When started, the 
DCUs first initialize the system clock, and attempt to establish the Ethernet connection 
with the remote data server. A watchdog timer is then enabled to automatically restart 
the DCUs if the program stops and idles for a long period. After Ethernet connection is 
established, the program will enter the main loop to enumerate input channels for on/off 
change. Each channel is associated with one detector or one signal phase. Once enough 
number of events, say, 10 events, are stored in the buffers, the data will be stored in log 
files and sent to central server. 
 Due to the different interfaces of the TS1 and TS2 DCUs, the firmware for the 
two is also different. The main difference is indicated in the red box in Figure 3.7. The 
TS1 DCUs check for voltage signal directly from the back panel pins, say, 24V for 
“detector on” and 0V for “detector off”. For the TS2 cabinets, the signal status and 
detector status information is packaged within the SDLC frames transmitted through the 
SDLC bus (NEMA TS2 standard, 2002). Hence, instead of checking voltage signals, the 
TS2 DCUs check the associated data bits associated with detector and signal status from 
the SDLC frames, in order to identify change events. 
 
 
18 
 
 
Figure 3.7 Flow chart of the firmware for the DCUs 
Timing and Synchronization of DCUs 
For the TS1 DCUs, a millisecond timer and a real time clock with 1 sec resolution are 
used, to time stamp the events. The clocks of the TS1 DCUs are synchronized with the 
central server via Ethernet, during midnight every day. For the TS2 DCUs, the time 
stamping mechanism is quite different. In the TS2 cabinet, the controller will broadcast 
its system time every sec, and signal status every 0.1 sec, to the SDLC bus. Every 0.1 
sec, the detector bus interface units (BIUs) will transmit detector status with duration 
length of last detector actuation/de-actuation in millisecond resolution, to the SDLC bus. 
 
 
19 
 
Such mechanism is utilized in the TS2 DCUs. In details, the time stamping at 1 sec 
resolution of the TS2 DCUs is based on the controller broadcasting time. For signal 
events, the time stamping resolution beyond 1 sec remains at 0.1 sec, based on the time 
instant when signal frame data being transmitted by the controller. For the detector 
events, the time stamping resolution beyond 1 sec remains at one millisecond, based on 
the duration time from the BIUs. The synchronization of the data with central server is 
ensured automatically, as the controllers are synchronized with the central server in real 
time.  
3.2.3 Installation and In-House Lab Testing 
The installations of both DCUs are straightforward. For TS1 DCU, the procedures are:  
1. Connect harness cable pin-to-pin to the contact closures on the back panel;  
2. Plug in power adapter;  
3. Connect the DCUs with Ethernet switch;  
4. Turn on the power switch.  
The procedures for TS2 DCUs are similar, except that the step 1 is to connect to 
SDLC interface instead of the contact closures. The procedures are illustrated in Figure 
3.8. 
 
 
20 
 
 
(A) 
 
(B) 
Figure 3.8 Installations of the TS1 DCU (A) and TS2 DCU (B) 
 Before deployments in the field, the DCUs were tested in the lab, using two 
 
 
 
21 
 
different platforms. The platform for TS1 cabinet is based on Hardware-In-the-Loop 
Simulation with VISSIM simulator, and the testing platform for TS2 is based on a 
commercial tester for TS2 controller, the TVC 3500 TS2 Virtual Cabinet by Athens 
Technical Specialists, INC. (ATSI). The snapshots and architectures of both platforms 
are shown in Figure 3.9. 
 
(A) 
 
(B) 
Figure 3.9 TS1 (A) and TS2 (B) Testing Platform 
 The testing platforms are aiming at simulating field intersections to test the 
performance of the DCUs. The vehicle detectors calls are generated by the software, 
 
 
Personal Computer 
w/Virtual Cabinet soft
Virtual Cabinet
TS2 Controller
TS2 SDLC 
Bus TS-2 DCU
Controller
Computer
TS2 DCU
SDLC Bus
Virtual 
Cabinet
  
a
s
e
T
t
t
 
o
T
p
a
3
S
H
o
(
F
nd sent to t
tore change
vents inform
S1 testing 
raffic lights
esting platfo
Tests
f data entri
S2 DCUs 
rocess SDL
pplications 
.2.4 Field D
ince Decem
ighway (T
ptics for co
RTMC). A 
igure 3.10
Figu
he controlle
s of the det
ation is sto
platform,  s
 in the sim
rm due to t
 with 2-hou
es for TS1 D
could be m
C frames. N
for perform
eployments
ber 2011, 
H) 13, Burn
mmunicatio
snapshot of
. 
re 3.10 Sna
rs that run f
ector status
red separat
ignal status
ulation pro
he lack of ap
r duration w
CUs, with 
ainly due to
evertheless
ance evalua
 
TS2 DCUs
sville, Min
n between i
 DCU insta
pshot of TS
22 
ully actuate
 and signal 
ely in the co
 is sent bac
gram. Such 
plication pr
ere conduc
a 2% avera
 the additio
, 2% data lo
tion purpose
 have been
nesota. The
ntersections
lled in field
2 DCU inst
d signals. Th
status. At th
mputers for
k to the sim
feedback i
ogram inter
ted for both
ge lost for T
nal comput
ss should b
. 
 installed at
se impleme
 and regiona
 (Int. Lynn 
alled at Int
e DCUs wi
e same tim
 validation. 
ulation pro
s not availa
face (API).  
 DCUs, resu
S2 DCUs. 
ation load o
e in accepta
 12 interse
ntation sites
l traffic ma
Ave & TH 
. Lynn Ave 
ll "listen to"
e, the simu
Note that, in
gram to co
ble for the
lting in 0%
The data los
f TS2 DCU
ble range in
ctions on T
 are using 
nagement c
13) is show
 
& TH 13
 
 and 
lated 
 the 
ntrol 
TS2 
 loss 
t for 
s to 
 our 
runk 
fiber 
enter 
n in 
 
 
23 
 
CHAPTER 4 OFFSET EVALUATION AND FINE-TUNING 
This chapter will introduce the evaluation and fine-tuning module for the offsets, with 
focus on constructing TS-Diagram to visualize progression quality for arterials. The 
proposed procedure is demonstrated with field cases and validations were performed 
using detector data and probe vehicle trajectories data collected from fields. An 
experiment is then carried out to illustrate how signal parameter changes could be made 
by intuitively evaluating the generated TS-Diagram. 
4.1 Background about Time Space Diagram 
Time space diagram (TS-Diagram) is a popular visualization tool for traffic signal 
evaluation, and has been used widely for signal operations (Signal Timing Manual, 
2008). With the time and distance as the X-Y axes, vehicle trajectories and signal status 
are commonly shown in a TS-Diagram. With the graphical illustration of the queues and 
delays, the progression quality can be evaluated intuitively. Figure 4.1 plots one 
example of TS-Diagram, with trajectory of one vehicle and signal status at two 
intersections. Delay of the subject vehicle due to traffic signals can be clearly observed. 
If similar delays occur for the majority of the vehicles, a poor progression quality will 
be indicated. 
 
Figure 4.1 Example of TS-Diagram  
[Source: Traffic Signal Timing Manual, 2008] 
 
 
24 
 
 Because of such convenience, TS-Diagram has been utilized by several 
commercial software as an output of signal optimization. For instance, TRANSYT-7F 
plots TS-Diagram with the green bands, and Synchro provides options to plot simulated 
trajectories in its TS-Diagram (Wallace et al, 1998; Husch & Albeck, 2006). A 
screenshot of Synchro TS-Diagram is illustrated in Figure 4.2, with blue and red lines 
as the trajectories of southbound and northbound traffic, respectively. 
 
Figure 4.2 Screenshot of Synchro TS-Diagram 
 In the practice, the Synchro TS-Diagram is being used widely, to design or 
update signal timing parameters in the signal retiming process. An accurate Synchro TS-
Diagram can properly reflect the field traffic condition, and help engineers choose 
appropriate timing parameters. On the contrary, inaccurate Synchro TS-Diagram, could 
mislead the engineers, degrade the timing performance, and increase the fine-tuning 
efforts. More importantly, since the Synchro TS-Diagram only uses manually collected 
data during a limited period, it cannot reflect the changing traffic conditions over time. 
 On the other hand, many intersections in the US are now using actuated signals, 
whereas few agencies have archived or analyzed the detector and signal status data to 
help signal operation (Liu et al., 2008). Recently, increasing attentions have been paid 
 
 
25 
 
on using detector and signal data for performance evaluation and optimization of the 
traffic signal system. However, to the best of our knowledge, few studies have been 
done to generate the TS-Diagram for evaluation of signal coordination. This research 
attempts to fill in this gap. 
 In this research, we propose a practical procedure to generate TS-Diagram using 
the high-resolution event-based traffic data collected from the field. With data collected 
and archived automatically, the TS-diagram can be generated periodically without 
significant labor cost. This is particularly beneficial for the practitioners to evaluate 
traffic signal performance after the retiming process, or to fine-tune the system over 
time. The proposed procedure is demonstrated with field examples and validations were 
performed using detector and probe vehicle trajectories data. Then, an experiment is 
demonstrated to illustrate how signal parameter changes could be made by intuitively 
evaluating the generated TS-Diagram.  
4.2 Constructing Link TS-Diagram 
The proposed procedure is to first generate TS-Diagrams for each link individually. 
Then, the link TS-Diagrams are combined as the arterial TS-Diagram. Note that, for two 
opposite through directions, there are two links between two adjacent intersections. The 
TS-Diagrams of these two links are overlapped, similarly to the Synchro TS-Diagram. 
Then the TS-Diagrams of consecutive links are stacked up sequentially, to form the 
arterial TS-Diagram. 
 In light of the cyclic nature of the traffic signal system (Day et al., 2010), the 
TS-Diagram is constructed cyclically using data aggregated from multiple cycles. This 
indicates no variations of traffic patterns over different cycles in the plot. The reason is 
that traffic signal parameters are generally governed by dominant cyclical patterns while 
variations from cycle to cycle are usually limited with TOD operation. Moreover, the 
aggregation can help reduce the data noise from the detector, and make it easy to 
evaluate the average performance of a timing plan. Two steps are involved to generate a 
link TS-Diagram, described as follows. 
Step 1. Aggregating high-resolution event-based data 
 
 
26 
 
In this step, both detector and signal data are aggregated. The detector event data is 
aggregated as the average cyclic flow profile (CFP), i.e. the interval volume over time 
within a cycle. Note that, the CFP has been used in TRANSYT-7F and SCOOT system 
(Wallace et al., 1998; Robertson & Bretherton, 1991), as well as some other researches 
for offset optimization (Abbas et al., 2001;Gettman et al., 2007; Day & Bullock, 2011). 
Here, we calculate the CFP of the through traffic at the link entrance, based on the 
advance detector data. To do so, the time instances of vehicle arrivals at the link 
entrance are estimated first. Depending on the queue impacts on the detector data, two 
cases need to be considered. 
Case A: Queue does not propagate to the advance detectors  
In this case, all vehicles would travel from the entrance to detector location in free flow 
speed. Hence, the arrival time at entrance are estimated by simply shifting the arrival 
time at the detector location with free flow travel time, as expressed by formula (4.1). 
ݐ௘ሺ݇ሻ ൌ ݐௗሺ݇ሻ െ ݈ௗݒ௙   ሺ4.1ሻ 
Where: 
ݐ௘ሺ݇ሻ is arrival time of k th detector actuation event at link entrance, 
ݐௗሺ݇ሻ is start time of k th detector actuation event at detector location and is measured 
directly,  
݈ௗ is distance from detector location to the link entrance, and 
ݒ௙ is free flow speed.  
Case B: Queue propagates to the advance detectors 
In this case, vehicles may not all travel freely from the entrance to detector location, as 
some vehicles are delayed by the queue. Hence, the shifting in formula (4.1) is not 
applicable for all arrivals. Such cases can be identified with the queue over detector 
(QOD) phenomena from the detector data (Liu et al, 2009; Wu et al., 2010). This is 
illustrated in Figure 4.3. After the red starts, if a long queue exists, the detector will see 
vehicles standing at the detector location for unusually long time. The end time of this 
particular actuation event is when the vehicle on the detector starts to move after green 
 
 
27 
 
starts, followed by queuing vehicles passing the detector at a saturated departure rate. In 
Figure 4.3, Point A indicates the time when queue propagates to the detector location. 
Point B indicates the time when the queuing vehicle on the detector starts to move, and 
Point C indicates the time when the last queuing vehicle passes the detector. The 
arrivals between points A and C are delayed by the queue, and need further adjustment 
to estimate their arrival times at the entrance. The corresponding detector status is also 
shown in Figure 4.3. 
 To estimate the arrival time of the delayed vehicles, we adopt the algorithm in 
(Liu et al., 2009). Essentially, the algorithm attempts to identify three points A, B, and C 
based on the occupancy times and the time gaps from the vehicle-detector actuation 
event data. The points A and B can be identified as the start and the end of a detector 
event with long duration time. Point C, the end of queue departure, could be identified 
as the end of consecutive saturated departure events after green start, followed by a 
large gap. For more details of identifying point A, B, and C, we refer to (Liu et al., 
2009). After the points A, B, C are identified, the arrivals are assumed as uniform 
arrivals between Point A and C, as expressed in the formula (4.2). 
ݐ௘ሺ݇ሻ ൌ
ۖە
۔
ۖۓݐௗሺ݇஺ሻ ൅ ሺ݇ െ ݇஺ሻ݇஼ െ ݇஺ ሾݐௗሺ݇஼ሻ െ ݐௗሺ݇஺ሻሿ  െ
݈ௗ
ݒ௙ , ݂݋ݎ ݇ א ሾ݇஺, ݇஼ሿ
ݐௗሺ݇ሻ െ ݈ௗݒ௙ , ܱݐ݄݁ݎݓ݅ݏ݁
     ሺ4.2ሻ 
Where: 
݇஺ is index of detector event starts at Point A, and 
݇஼ is index of detector event at Point C. 
  
 
 
28 
 
 
Figure 4.3 Illustration of queuing at intersection with QOD 
 Based on the estimated arrival times at the entrance, the interval volume can be 
then calculated using formula (4.3). 
ݍሺ݆ሻ ൌ ෍ ૤ሼ݆∆ݐ ൑ ݐ௘ሺ݇ሻ ൏ ሺ݆ ൅ 1ሻ∆ݐሽ, ݂݋ݎ ݆ ൌ 1,2 … ܶ.   ሺ4.3ሻ
ே
௞ୀଵ
 
Where: 
ݍሺ݆ሻ is volume of j th interval over investigation period, 
∆ݐ is the length of interval, 
݆ is the index of interval over investigation period, 
ܰ is the total number of detector events, 
ܶ is the total length of investigation period, normalized by ∆ݐ, and 
૤ሼ. ሽ is the indicator function, with 1 if condition within bracket is satisfied, and 0 
otherwise: 
Then, the average CFP can be obtained by aggregating interval volume over cycles 
using formula (4.4). One example of the CFP with 5 sec interval is shown Figure 4.4. 
ݍොሺ݅ሻ ൌ ∑ ݍሺ݆ሻெ௢ௗሺ௝,஼ሻୀ௜
௖ܰ
, ݂݋ݎ ݅ ൌ 1, … , ܥ.    ሺ4.4ሻ 
 
 
29 
 
Where: 
ݍොሺ݅ሻ is flow of i th interval over time within cycle, as the CFP, 
௖ܰ is total number of cycles during investigation period,  
ܥ is for cycle length, which is the cycle length divided by ∆ݐ, and 
Mod (,) is the modulo operator. 
 
Figure 4.4 Example of CFP 
 Based on signal event data, the average green start and end of the associated 
phase are calculated by formula (4.5). For simplicity, yellow signal is treated as part of 
red signal. 
ۖە
۔
ۖۓ ݐҧ௚ௌ௧ ൌ
∑ ܯ݋݀ሺݐ௚ௌ௧ሺ݇ሻ, ܥሻே೎௞ୀ଴
௖ܰ
ݐҧ௚ாௗ ൌ
∑ ܯ݋݀ሺݐ௚ாௗሺ݇ሻ, ܥሻே೎௞ୀ଴
௖ܰ
     ሺ4.5ሻ 
Where: 
ݐҧ௚ௌ௧ is the average green start time over time of cycle, 
ݐ௚ௌ௧ሺ݇ሻ is the time of k th green start, 
ݐҧ௚ாௗ is the average green end time over time of cycle,  
ݐ௚ாௗሺ݇ሻ is time of k th green end, and 
all variables including ݐҧ௚ௌ௧, ݐ௚ௌ௧ሺ݇ሻ, ݐҧ௚ாௗ, ݐ௚ாௗሺ݇ሻ are normalized by ∆t. 
5 30 60 90 120 150 180
0
200
400
600
800
1000
1200
1400
1600
1800
Time Within Cycle (sec)
5 
Se
c 
In
te
rv
al
 V
ol
um
e 
(v
eh
/h
ou
r)
 
 
30 
 
Step 2. Generating Virtual Trajectories 
The CFP, average green start and end will be used to generate the link TS-Diagram. 
Before plotting, based on the CFP, a sequence of time instances at the link entrance will 
be generated, simulating virtual vehicles entering the link. This is a little tricky, due to 
the non-integer flow rate during an interval. For instance, with a 5 sec interval length, if 
a 1800 vehicle per hour (vph) flow rate is observed for the 1st interval, and 800 vph for 
the 2nd interval, , the problem is then to simulate 2.5 vehicles during 1st interval 
window, and 1.25 vehicles for the 2nd interval. One approach is to rounding up the 
numbers of vehicles, but it will introduce substantial rounding up errors. Instead, we 
simulate 250 vehicles equally spaced in the 1st interval, and 125 vehicles in the 2nd 
interval. Then, 100th, 200th, and 300th vehicles over the 10 sec horizon will be selected 
as the virtual arrivals to generate the trajectories. This approach can be viewed as 
generating fraction of vehicle based on flow rate. Then fractions are accumulated as 
whole vehicles. 
 At each time instance, a vehicle is simulated traversing from the entrance to stop 
bar, using the Newell's simplified car following model (Newell, 2002). In Newell's 
model, the follower's trajectory is simply a temporal-spatial translation of the leader in 
car-following mode. The calculation is expressed in formula (4.6).  
ݔ௜ሺݐ ൅ ߬ሻ ൌ ܯ݅݊ሼݔ௜ିଵሺݐሻ െ ݀, ݔ௜ሺݐሻ ൅ ߬ݒ௙ሽ   ሺ4.6ሻ 
Where:  
ݔ௜ିଵሺݐሻ, ݔ௜ሺݐሻ are the trajectories of the leader and follower respectively, 
߬ is the time displacement, and 
݀ is the space displacement. 
 To account for the signal impact, a dummy vehicle based on red signal is 
generated as the first leading vehicle within the cycle. The procedure is repeated cycle 
by cycle to expand the plot. The illustration of the dummy vehicle is shown in Figure 
4.5. 
 
 
31 
 
 
Figure 4.5 Illustration of car following and dummy vehicle. 
4.3 Validation of TS-Diagram 
Two cases are demonstrated, using data collected on Trunk Highway (TH) 55 and TH 
13 in Minnesota. Both arterials are operated with vehicle-actuated coordinated control. 
Here, the parameters of the car following model are not calibrated but using typical 
values. In details, the post speed limit was used as the free flow speed, around 50 mph. 
30 ft and 1.6 sec were used for the space and time displacement, approximately leading 
to jam density of 175 vehicle per mile (vpm) and capacity of 1800 vph. 
4.3.1 Case 1: TH 55 
Five intersections on TH 55 deployed with the SMART-Signal system are highlighted, 
with data collected on Nov 17th, 2008, from 7:15 AM-8:45 AM, when the morning 
peak timing plan was running. The TS-Diagram are shown in Figure 4.6, with arterial 
layout. In addition to the advance detectors, there were also temporary detectors at the 
link entrances for research purpose. The eastbound (EB) and westbound (WB) 
trajectories are plotted with blue and red lines, respectively. The associated signal status 
for each direction was plotted facing the corresponding trajectories. As can be seen 
clearly, although the link TS-Diagrams were constructed independently, the traffic 
patterns were consistent across intersections, reflecting the consistent progression 
impact from the signal. This can also be viewed as a preliminary validation of traffic 
patterns in the TS-Diagram. 
 
 
32 
 
 
Figure 4.6 Layout and TS-Diagram for TH 55 
 Using data from the entrance detectors, a comparison was conducted, regarding 
the estimated and observed CFPs at the entrances. For each link, the total volumes from 
advance detectors and entrance detectors were balanced to reduce the impacts from the 
sinks and sources. The result is shown in Figure 4.7. As we can see, the estimated CFPs 
match well with the observed CFPs, indicating good estimation by the proposed method. 
 
Int. 1 
Int. 2 
Int. 3 
Int. 4 
Int. 5 
Time 
Space 
EB Traffic
WB Traffic 
EB Signal 
WB Signal 
Consistent traffic 
pattern across 
intersection 
 
 
33 
 
 
 
Figure 4.7 Estimated CFPs VS Observed CFPs for EB(Up) and WB (Down) traffic 
4.3.2 Case 2: TH 13 
In this case, seven intersections on TH13 with the SMART-Signal system were selected. 
The data was collected during afternoon peak period, from 4:30 to 6:00 PM, on Jul 23, 
2013, with GPS data of nine field rides through the selected intersections. The TS-
Diagram is shown in Figure 4.8, with the field ride trajectories in black. Note that, the 
field rides were actually made during several cycles, but were plotted within a single 
cycle based on individual departure time within each cycle at both ends of the arterial. 
 As indicated in the figure, the EB trajectories match well with the generated TS-
Diagram, while some disagreements were found, indicated by letter A and B. This is 
expected as the TS-Diagram only reflects the average conditions, and cannot reflect the 
variations of the green start due to side street actuations and pedestrian calls. Similar 
disagreement was found for WB trajectories at Int. 6, denoted by C. However, 
significant disagreements were observed for WB trajectories approaching Int. 5, 
denoted by D. This was mainly due to the occurrence of WB oversaturation and residual 
queues at Int. 5. Due to the residual queues, the repetitive patterns were interrupted, 
1  60 120 180
500
1000
1500
2000
 
 
1  60 120 180 1  60 120 180 1  60 120 180
V
ol
um
e 
(v
eh
/h
ou
r) Observed
Estimated
Link 1-2 Link 2-3 Link 3-4 Link 4-5
1  60 120 180
500
1000
1500
2000
Time Within Cycle (sec)
 
 
1  60 120 180 1  60 120 180 1  60 120 180
V
ol
um
e 
(v
eh
/h
ou
r) Observed
Estimated
Link 2-1 Link 3-2 Link 4-3 Link 5-4
 
 
34 
 
resulting in amplified variations over cycles. This is evident from the significantly 
different trajectories traversing the intersection, while the vehicle entered the link at 
similar times within cycle. This indicates a limitation of the proposed TS-Diagram for 
the oversaturated intersections. 
   
Figure 4.8 TS-Diagrams for selected intersections on TH 13 
4.4 Fine-Tuning Using Ts-Diagram with Offset and Lead-Lag Sequence 
To illustrate the potential application of the TS-Diagram, an experiment was carried out 
to adjust the offsets and main street phase sequence, i.e. the lead-lag sequence, 
indicating whether the main street left turn phase is leading or lagging the through phase, 
for a eight-phase dual-ring NEMA controller. For the offset, we use green start of first 
coordinated phase as the reference point. Regarding the lead-lag sequence, there are 
EB rides
WB rides
A
B
D
C
Int. 1 
Int. 2 
Int. 3 
Int. 4 
Int. 5 
Int. 6 
Int. 7 
 
 
35 
 
four possible configurations for an eight-phase dual-ring NEMA controller, as shown in 
Figure 4.9. The configuration can be identified easily from the shape of overlapping 
green signals in the TS-Diagram. 
 
Figure 4.9 Offset reference point and combinations of lead lag sequence 
 So far, the TS-Diagrams are generated based on average data, reflecting the 
average traffic conditions. Many agencies, however, may concern about traffic signal 
performance for congested traffic conditions. To address such concerns, we also 
construct the TS-Diagram for relatively congested conditions, which will be the base of 
the analysis hereafter. To do so, when aggregating the raw data, the cycle volumes are 
calculated and ranked. Then, cycles with high cycle volume is kept and the rest will be 
filtered out. Here, we kept data with the highest 60-100 percentiles cycle volumes for 
each intersection, and constructed the TS-Diagram for the congested conditions. In 
practice, the criteria can be user specified, while our choice of 60-100 percentiles is to 
reflect relatively congested yet not extreme cases.  
 Both of the TS-Diagrams are shown in Figure 4.10, with Int.1 to Int. 5 of TH13 
highlighted for analysis. The data was collected on four consecutive weekdays, 
10/28/2013 to 10/31/2013, during 7:15 to 8:45 AM, for the evaluation of morning peak 
timing plan. Note that, while main traffic patterns remained similar, the queues were 
 
 
36 
 
clearly longer in the congested case than the average case, indicated by letter A and B. 
 
  
Figure 4.10 TS-Diagrams for selected intersections on TH 13 with average (left) 
and relatively congested conditions (right) 
 For a two-way coordinated arterial, trade-off exists for traffic in the opposite 
directions. With TS-Diagram, established method e.g. MAXBAND or combination 
method (Little, 1966; Gartner & Little, 1973; Day & Bullock, 2011) can be incorporated 
easily. could be incorporated for optimization. Here, instead of using a specific 
optimization method, we performed a quick, visual based evaluation of the TS-Diagram 
Int. 1 
Int. 2 
Int. 3 
Int. 4 
Int. 5 
Average Relatively Congested
B 
A
D
E
Lag-Lead 
F
Lead-Lead 
C
EB
EB Green 
WB Green 
 
 
37 
 
for the changes. Indeed, this would involve engineering judgment, and may lead to sub-
optimal solution. For the sake of illustration, we will the visualization approach with 
this analysis, leaving systematic optimization for future research. 
 First, the queue length and delays were identified from the TS-Diagram. In 
Figure 4.10, the EB cases are indicated by letter A and B, with the WB cases by C and 
D. The offsets were first fine-tuned for the direction with larger volume. The opposite 
direction was optimized next. Here, the EB was favored with a higher volume. The 
queues indicated by A and B, in the congested case, could be reduced by slightly 
shifting offsets to the left. The tradeoff exists with degraded WB performance at Int. 3 
and Int. 4. This, however, should be minimal with the lower volumes. The decision was 
made to shift the offset at Int. 4 to the left for 4 sec, and 4 sec further to the left for 
offset at Int. 5, with a total 8 sec to the left at Int. 5. The 4 sec was selected arbitrarily 
because we want implement small adjustments only. 
 For the WB cases, to reduce the queue labeled with C, either the offset at Int. 3 
needs to be shifted to the left, or offsets at Int. 4 and Int. 5 to the right. If offset at Int. 3 
were shifted to the left, the end of EB platoon at Int.3 would be cut off and delayed by 
the red. If offsets at Int.4 and Int.5 were shifted to the right, the EB queues and delays 
would increase at Int. 4. Considering queue labeled with D, the end of the WB platoon 
was close to the green end, and would be delayed if offset were shifted to the left. 
Considering the overall benefit, no offset changes were justified regarding the WB cases. 
 Next, lead-lag sequences were evaluated. Here, we search for configuration of 
the lead-lag sequences that provides the max green window for the platoons in both 
directions. One example of a proper lead-lag sequence is indicated by E. The 
overlapping shape of platoon arrivals (EB first), indicated by the blue and red 
parallelograms was consistent with the overlapping shape of the green windows (EB 
first) for the two directions. This indicates the current setting of Lag (phase 1)-Lead 
(phase 5) was appropriate at Int.2. The intersections were evaluated one by one with 
similar logic. One potential change was identified, indicated by F. In this case, the end 
of EB platoon was cut off by the red signal, and could be better served by changing the 
Lead-Lead to Lead-Lag. The change was then granted. In addition, to accommodate the 
 
 
38 
 
lead-lag change, the offset in Int. 3 was also shifted 7 sec to the left, so that there would 
not be unused EB green time after the end of EB platoon. 
 The offset and lead-lag changes are summarized in left part of Figure 4.11. For 
better illustration, the shifts of individual phases are indicated by the arrows. The 
changes were implemented from 11/05/2013 to 11/07/2013. Due to a connection issue, 
the data for 11/06/2013 was not available. The TS-Diagram generated using data 
collected on 11/05/2013 and 11/07/2013 is shown on the right part of Figure 4.11. 
 Slight changes of traffic patterns were found from Figure 4.11. For EB traffic, 
the shockwave at Int. 2, indicated by A, was slightly reduced. The EB arrivals near the 
start of red at Int.3 was better accommodated, indicated by B. No obvious change was 
found for the EB arrivals at Int.4, indicated by C. This is intriguing given the relatively 
large shift of EB phase at Int. 3. If observed carefully, the density of the lines for the 
arrivals has changed slightly as circled by the blue parallelograms, and both arrivals 
showed slight inconsistency with departures at Int. 3. The potential interpretation is as 
follows: in the before case, the drivers decelerated with anticipation of the queue 
formation, and the queuing profile at Int. 4 were based on the delayed arrival time, and 
were underestimated. For WB traffic, degraded performance was found at Int.1, Int.3 
and Int.4. Some improvements were found at Int. 2, labeled by E, likely because main 
street arrivals near the green start and arrivals from the site street at Int. 3 were better 
coordinated. 
  
 
 
39 
 
 
  
Figure 4.11 Before-After comparisons of TS-Diagrams with relatively congested 
condition 
 For further validation, the queue length and delay were estimated by the 
SMART-Signal system (Liu & Ma, 2009; Liu et al, 2009; Wu et al, 2010), and the 
results are summarized in Figure 4.12 and Table 4.1. As indicated in Figure 4.12, 
average maximum queues were reduced for most of EB phases, while slightly increased 
for the WB phases. This is consistent with the TS-Diagram, as the proposed changes 
favored EB over WB traffic.  
14 s 
7s 
4s 
4s 
8s 
Int. Offset 7s to 
the left & Lead-
Lead to Lead-Lag 
Int. Offset 
4s to the left 
D
Int. 1 
Int. 2 
Int. 3 
Int. 4 
Int. 5 
CB
CE
C
D
CA
Int. Offset 
4s to the left  
Before After 
EB
  
 
w
w
s
Figure 4.1
As in
ith 4.6 sec
ith the high
ec/cycle. 
Table 
E
E
W
W
Cycle
2 Before-A
dicated in T
/veh, while 
er volume 
4.1 Before-A
Average 
B Vol. (vph
B Delay (se
B Vol. (vph
B Delay (se
 Delay (Veh
fter compa
able 4.1, th
for WB tra
of EB traffic
fter comp
B
) 
c) 
) 
c) 
*Sec) 4
40 
rison of ma
Signal Syst
e average d
ffic, it incre
, the total d
arison of de
efore Aft
1314 13
32.4 27
1003 100
65.3 68
607.6 444
ximum que
em 
elay for EB
ased by 4%
elay was re
lay from SM
er Abs. C
10 -
.8 -4
7 4
 2
3.8 -16
ue length fr
 travel was r
, with 2.7 s
duced by 4%
ART-Sign
hange Pe
4 
.6 
 
.7 
3.8 
om SMAR
educed by 
ec/veh. Ove
, i.e., 163.8
al System
r. Change
0% 
-14% 
0% 
4% 
-4% 
 
 
T-
14%, 
rall, 
 veh 
 
 
41 
 
CHAPTER 5 GREEN SPLITS EVALUATION AND FINE-TUNING 
This chapter will introduce the performance evaluation and fine-tuning module for the 
green splits. 
5.1 Measure of Effectiveness for Green Splits 
Assigning green splits proportionally to traffic demands is a simple while effective rule 
to determine green splits. To evaluate the performance of green splits, classical 
measures of effectiveness (MOE), volume/capacity ratio (v/c ratio), is commonly used 
when designing signal timing parameters. However, in the conventional practice, the v/c 
ratios are only calculated during signal retiming process with manual collected data, and 
are not monitored continuously. Although the practice is limited, there are some 
researches trying to calculate v/c ratios using the detector and signal data for evaluation 
of the green split and cycle length (Balke & Herrick, 2004; Balke et al., 2005; Smaglig 
et al., 2007; Day et al., 2010; Sunkari et al., 2011). 
 The green splits fine-tuning module proposed here can be viewed as an 
extension from (Smaglig et al., 2007; Day et al., 2010). The main difference is, instead 
of using detector volume directly, we proposed a new MOE, the Utilized Green Time 
(UGT), extended from the queue service time (QST) for green split evaluation. The 
purpose is to address the issue that volume measured at stop bar detectors could be 
inaccurate, since the stop bar detectors are mainly for vehicle presence detection, 
especially when long or sequential detectors are used (Lawrence et al., 2006). 
5.1.1 v/c ratio 
 The v/c ratio indicates saturation level of a specified signal phase (Highway 
Capacity Manual, 2000), and can be calculated cycle-by-cycle, as: 
௜ܺሺ݆ሻ ൌ ௜ܸሺ݆ሻݏ௜ݔ௜ሺ݆ሻ       ሺ5.1ሻ 
Where: 
௜ܺሺ݆ሻ is the v/c ratio of phase i, for j th cycle, 
 
 
42 
 
௜ܸሺ݆ሻ is the flow rate of phase i, for j th cycle, 
ݏ௜ is saturation flow rate of phase i, 
ݔ௜ሺ݆ሻ ൌ ௚೔ሺ௝ሻ஼   is the green split, and, 
݃௜ሺ݆ሻ is the green time. 
When long or sequential detectors are used, vehicle counts could be inaccurate. 
This is because consecutive arrivals may only cause a single detector actuation event 
with long duration length. One example from a sequentially connected detector at stop 
bar is shown in Figure 5.1, with example of detector actuation events. In such cases, 
occupancy-based measures will be more desirable. 
 
Figure 5.1 Example of traffic events from long detector at stop bar 
5.1.2 Queue Service Time (QST) 
Another important MOE is the Queue Service Time, defined as the length of green time 
 
 
43 
 
used to clear the queue (Highway Capacity Manual, 2000; Balke & Herrick, 2004). 
Balke & Herrick (2004) proposes to use the time elapsed from the green start to the start 
of first time gap identified from stop bar detector data, as the QST. This is based on the 
assumption that queuing vehicles will constantly place calls on stop bar detector until 
queue is cleared, which may not be the case if the detector is not long enough. 
Alternatively, we propose following formula for QST calculation. The key is to identify 
the end of queue departures, by identifying the first large gap from detector events. Here, 
a threshold gap value of 2.5 sec is used. An illustration of QST is shown in Figure 5.2. 
ܳܵ ௜ܶሺ݆ሻ ൌ ݐொ,௜ሺ݆ሻ െ ݐீ௥,௜ ሺ݆ሻ     ሺ5.2ሻ 
Where: 
ܳܵ ௜ܶሺ݆ሻ is the queue service time for phase i, for j th cycle. 
ݐொ,௜ሺ݆ሻ is the end of queue departure, and is identified as the start of the first gap larger 
than 2.5 sec after green start, and 
ݐீ௥,௜ሺ݆ሻ is the start of green. 
 
Figure 5.2 Illustration of QST 
5.1.3 Utilized Green Time and Slack Green 
While QST represents green time to clear the queue, it does not account for green time 
used by the vehicles that are not queuing. Based on the QST, the following formula is 
proposed to calculate the Utilized Green Time (UGT), as the total green time needed to 
 
 
44 
 
serve the demand within a cycle. In the formula, the first part is the QST, while the 
second part represents the minimum green time needed to serve the rest of arrivals. 
ܷܩ ௜ܶሺ݆ሻ ൌ ܳܵ ௜ܶሺ݆ሻ ൅ ݄௦,௜ ൈ ݊௜ሺ݆ሻ   ሺ5.3ሻ  
Where: 
ܷܩ ௜ܶሺ݆ሻ is utilized green time at phase i, for jth cycle, 
݄௦,௜ is the saturation headway for phase i (2 sec in our study), and  
݊௜ሺ݆ሻ is number of random arrivals within green time after queue is cleared, at phase i, 
for j th cycle. 
 An example of calculation of QST and UGT is shown in Figure 5.3. As 
indicated in the figure, the green start is indicated as ݐ஺, and the end of queue departure 
is indicated as ݐ஻, followed by a relatively large gap with four additional arrivals. Hence, 
the QST is ሺݐ஻ െ ݐ஺ሻ, while the UGT is ሺݐ஻ െ ݐ஺ ൅ 4 ൈ 2 ݏ݁ܿሻ. 
 
 
Figure 5.3 Example of QST and UGT calculation 
 Then, the slack green, ݃ௌ௟,௜, unutilized to serve the traffic, can be calculated as: 
Detector Occupation Event 
Signal Status 
Queue Departure Random Arrival 
tB 
Time 
tA
 
 
45 
 
݃ௌ௟,௜ ൌ ݃௜ െ ܷܩ ௜ܶ   ሺ5.4ሻ 
 Based on the slack green, the phase failure can be identified for cycles without 
any slack green, meaning that the green time may not be sufficient to serve all the 
demands within one cycle. 
5.2 Green splits rebalancing based on UGT 
For vehicle-actuated signals, traffic engineers are typically concerned with phase 
failures for non-coordinated phases. Therefore, the recommended fine-tuning rule is to 
maintain acceptable slack green time for each phase to reduce the potential phase 
failures (Day et al., 2011). In addition, since the progression quality of coordinated 
phases is another major concern , the green splits rebalancing is restricted to non-
coordinated phases only. 
 In addition, since NEMA controller operates the two rings concurrently, in order 
to compare the phase utilization across different phases, the critical phases need to be 
identified first. The critical phases are phases with higher green utilization, hence, 
smaller slack green, than the concurrent phases on the other ring. General speaking, 
these are relatively more congested phases. For a typical eight phase NEMA dual ring 
controller, following rules can be used to identify the critical phases based on the slack 
green, excluding the coordinated phases.  
 For phase 1, 5, if ݃௦௟,ଵ ൏ ݃௦௟,ହ, phase 1 is critical phase. Otherwise, phase 5 is 
critical phase. 
 For phase 3, 4, 7, 8, if ݃௦௟,ଷ ൅ ݃௦௟,ସ ൏ ݃௦௟,଻ ൅ ݃௦௟,଼, then phase 3 and 4 are critical 
phases. Otherwise, phase 7 and phase 8 are critical phases. 
 Once the critical phases are identified, the splits can be first rebalanced across 
the critical phases. Since more slack greens exist at non-critical phases, further 
balancing across non-critical phases can be easily done to accommodate the changes at 
critical phases. 
5.3 Split Fine-tuning Example 
To demonstrate the split fine-tuning, an example is illustrated based on one of the 
 
 
46 
 
intersections on TH13. Figure 5.4 shows the layout of the selected intersection, at 
Nicollet Ave & Highway 13, Burnsville. The ring-and-barrier diagram is also shown in 
the figure. Six phases are used with a cycle length of 130 seconds, with phase 2 and 6 as 
coordinated phases, and the rest phases, phase 1, 3, 4, 5, non-coordinated. The numbers 
on the green bars in the ring-and-barrier diagram show the max green time, plus yellow 
and red clearance time. The detector and signal data used were collected during 1:30 pm 
to 3:00 pm, from 7/08/2013 to 7/12/2013, Mon to Fri.  
 
Figure 5.4 Intersection layout and ring-and-barrier diagram of the signal timing 
 In Figure 5.5, cycle-by-cycle green time with UGTs was plotted for the non-
coordinated phases. The X-axis denotes the cycle index, while the Y-axis shows the 
length of time duration. The purple lines indicate the separation between two 
consecutive days. From the figure, varying green time was clearly shown at phase 1, 3, 
4. Due to the fixed force off setting at coordinated phase 6, phase 5 was essentially 
operated as fixed-time. For phase 4 and phase 5, relatively high proportion of green 
have been utilized, with small slack green left, indicated by letter A. For phase 1 and 
phase 3, on the contrary, relatively large slack green was observed, indicated by letter B. 
This indicated opportunities to rebalance the green splits among the four. 
 
 
 
47 
 
 
 
Figure 5.5 Green Time VS Utilized Green Time 
 After the aggregation of 5 days data, average statistics were summarized in the 
table in Figure 5.6. The information was further tabulated within the ring-and-barrier 
diagrams, for more intuitive interpretation. In the ring-and-barrier diagram, the green 
bar indicates the average green time, while the blue bar indicates the average UGT. For 
brevity, the yellow and red clearance signals were not plotted. The values of the UGT 
and green time were also denoted nearby the diagram, and were stated in the format as 
"Phase ID-UGT/Average Green, (Phase Failure rate)". For instance, "P3-9/15 (19%)", 
represents for "Phase 3 with 9 sec UGT, 15 sec Average Green, and a 19% phase failure 
rate". 
 Then, the critical phases were identified. The slack green at phase 5 was 2 sec, 
which was shorter than slack green at phase 1; hence, the phase 5 was the critical phase. 
Since no phase 7 and 8 were used, phase 4 and phase 3 were critical phases. Among the 
critical phases 5, 4 and 3, it would be feasible to reduce max green time for phase 3 
slightly, with increase at phase 4 and 5, particularly when considering the relatively high 
phase failure rate with 57% and 36% at phase 5 and 4 respectively.  
P4
Mon  Tue  Wed  Thu  Fri 
P1  P3 
P5 
 
Cycle Index 
Time Length (s) 
Green time 
Utilized Green 
Green time (Moving Average) 
Utilized Green (Moving Average) 
A 
B 
 
 
48 
 
Table of average values for the MOEs 
Phase 
ID 
Max Green (s) Average Green (s) UGT 
(s) 
Slack 
Green (s) 
Phase Failure 
Rate2 
1 17 15 8 7 3% 
51 10 8.0 6 2 57% 
4 18.5 19 14 5 36% 
3 18.5 15 9 6 19% 
1: underline indicates this is a critical phase 
2: phase failure rate is the proportion of cycles with phase failures 
 
 
 
 
Figure 5.6 Ring-and-barrier diagrams with UGTs 
  
Phase 3, UGT: 9 Sec, Average Green: 
15 sec, Phase Failure Rate: 19%.
 
 
49 
 
CHAPTER 6 CONCLUDING REMARKS AND FUTURE 
RESEARCHES 
6.1 Concluding Remarks 
Due to the increasing needs to improve traffic management, performance measure for 
signalized networks has become an emerging focus in the U.S (Liu et al., 2008). 
However, despite of the availability of detectors, few agencies has archived or analyzed 
detector and signal data at signalized intersections to help signal operations. In this 
study, by using data collected from existing traffic signal systems, a systematic tool is 
successfully developed for performance visualization and fine-tuning of the arterial 
traffic signal system.  
In details: 
 To address the technical issue of data collection, DCUs were developed as a cost 
effective and vendor independent solution, to collect high-resolution event based 
traffic data.  
 For the evaluation and fine-tuning of offsets, we proposed a practical procedure to 
construct the TS-Diagram, to visualize progression qualities of signalized arterials. 
Based on the data collected automatically, the TS-Diagram can be constructed 
periodically, without significant labor cost. The field examples were demonstrated 
for the proposed TS-Diagrams, and reasonable agreements were found for validation 
using detector data and vehicle trajectories. A field experiment was also carried out 
to illustrate the application of TS-Diagram for adjusting offsets and lead-lag 
sequence. By intuitively evaluating the TS-Diagram, changes were recommended, 
and a 4% reduction of total delays was achieved. 
 For the evaluation and fine-tuning of green splits, an adjusted MOE, the UGT, was 
proposed, extended from queue service time. The ring-and-barrier diagram was 
adopted to tabulate the information for performance visualization and evaluation. 
Field examples were also illustrated to demonstrate the implementation potentials. 
There are two potential applications of the signal operations in current practice that the 
tool can help:  
 
 
50 
 
(1) It can be used to validate traffic patterns and fine-tune signal parameters during 
signal retiming process, thus to reduce associated labor costs. 
(2) It can be used to evaluate the signal performance periodically, and help engineers to 
identify adjustments opportunities for improvements, thus to maintain efficient 
signal operations over time. 
6.2 Limitations and Future Researches 
Limitations of the proposed approach are discussed in this section. For TS-Diagram, 
because traffic patterns at oversaturated intersections are not represented accurately, 
applications of the proposed approach to oversaturated cases should be avoided. Also, 
the proposed approach tends to underestimate queues and delays due to the mechanism 
that only data on the subject link is considered. In cases with long queues, it may be 
necessary to consider the upstream intersection jointly, to better evaluate the traffic 
conditions. 
Our next step is trying to improve the accuracy of TS-Diagram by incorporating 
data from adjacent intersections. Other direction is to develop a more systematical 
approach to facilitate automated optimization of offsets, lead-lag sequence, and the 
green splits, using the event based traffic data. Moreover, performance evaluation and 
fine-tuning of cycle length and time-of-day transition schedules should also be pursued 
in the future.  
  
 
 
51 
 
BIBLIOGRAPHY 
1. Abbas, M., Bullock, D. M., Head., L. (2001) Real-Time Offset Transitioning 
Algorithm for Coordinating Traffic Signals. Transportation Research Record 1748, 
26–39. 
2. Akcelik, R. (1994). Estimation of green times and cycle time for vehicle-actuated 
signals. Transportation Research Record, (1457). 
3. Ahmed , A. & J. Brian (2008). An Intersection Traffic Data Collection Device 
Utilizing Logging Capabilities of Traffic Controllers and Current Traffic Sensors. 
NIATT Report Number N08-13. 
4. Balke, K., & C. Herrick (2004). Potential Measures of Assessing Signal Timing 
Performance Using Existing Technologies. TTI Report No. 0-4422-1. 
5. Balke, K., H. Charara & R. Parker. Development of a Traffic Signal Performance 
Measurement System (TSPMS) (2005). TTI Report No. 0-4422-2. 
6. Birst, S & A. Smadi (2006). Use of Hardware-in-the Loop Traffic Simulation in a 
Virtual Environment," Technical Presentation at 2006 ASCE AATT. 
7. Bullock, D., & A. Catarella (1998). “A Real-Time Simulation Environment for 
Evaluating Traffic Signal Systems”. Paper presented at the 77th Annual TRB 
Meeting, Washington D.C. 
8. Cowan, R.J. Useful Headway Models. Transportation Research, Vol.9, No. 6, 1975, 
pp. 371-375. 
9. Day, C. M., D. M. Bullock, & J. R. Sturdevant (2009). Cycle-Length Performance 
Measures: Revisiting and Extending Fundamentals. Transportation Research 
Record, 2128: 48-57. 
10. Day, C. M., J. R. Sturdevant and D. M. Bullock (2010a). Outcome-Oriented 
Performance Measures for Management of Signalized Arterial Capacity. 
Transportation Research Record, 2192: 24–36. 
11. Day, C. M., R. Haseman, H. Premachandra, T. M. Brennan, Jr., J. S. Wasson, J. R. 
Sturdevant, and D. M. Bullock (2010b). Evaluation of Arterial Signal Coordination. 
Transportation Research Record, 2192: 37–49. 
12. Day, C. M., & D. M. Bullock (2011). Computational Efficiency of Alternative 
Algorithms for Arterial Offset Optimization. Transportation Research Record, 
 
 
52 
 
2259:37-47. 
13. Duncan, G. (2011). Signalized Intersection Performance Measurement. In ITE 2010 
Technical Conference. ITE. Accessible online: 
http://www.ite.org/meetings/2010TC/Session%2018%20Gary%20Duncan.pdf.Acce
ssed on June 1st, 2013. 
14. Econolite Group, Inc. (2012). ASC3 controller specification. Available online at: 
http://www.econolite.com/assets/pdf/controller_asc3_specification.pdf. Accessed on 
Aug 3rd, 2012. 
15. Econolite Group, Inc. (2013). CENTRACS System Datasheet. Available online 
at:http://www.econolite.com/assets/pdf/systems-centracs-data-sheet.pdf. Accessed 
on Oct 3rd, 2013. 
16. Gartner, N., & J. D. C. Little (1975). Generalized Combination Method for Area 
Traffic Control. Transportation Research Record, 531: 58–69. 
17. Gartner, N. H., S. F. Assmann, F. Lasaga, and D. L. Hou. MULTIBAND: A Variable-
Bandwidth Arterial Progression Scheme. Transportation Research Record, 1287: 
212–222, 1990. 
18. Gettman, D., Shelby, S. G., Head, L., Bullock, D. M., & Soyke, N. (2007). Data-
driven algorithms for real-time adaptive tuning of offsets in coordinated traffic 
signal systems. Transportation Research Record: Journal of the Transportation 
Research Board, 2035(1), 1-9. 
19. Gordon, R.L. & W. Tighe (2005). Traffic Control Systems Handbook. Federal 
Highway Administration, Report No. FHWA-HOP-06-006. 
20. Gordon, R.L. (2010) Traffic Signal Retiming Practices in the United States, National 
Cooperative Highway Research Program (NCHRP) Synthesis 409. 
21. Husch, D., & Albeck, J. (2006). Synchro Studio 7 user guide. Trafficware Ltd., 
Sugar Land, TX.  
22. James Bonneson, Srinivasa Sunkari, & Michael Pratt (2009). Traffic Signal 
Operation Handbook. TxDOT. Report No. FHWA/TX-09/0-5629-P1. 
23. Koonce, P., L. Rodegerdts, K. Lee, S. Quayle, S. Beaird, C. Braud, J. Bonneson, P. 
Tarnoff, & T. Urbanik (2008). Traffic Signal Timing Manual. FHWA. Report No. 
FHWA-HOP-08-024. 
 
 
53 
 
24. Lawrence, A. K., M. K. Mills, & D. R. P. Gibson (2006). Traffic Detector 
Handbook: Third Edition—Volume I. FHWA, Report No. FHWA-HRT-06-108. 
25. Little, J. D. C., M. D. Kelson, & N. H. Gartner (1981). “MAXBAND: A Program 
for Setting Signals on Arteries and Triangular Networks,” Transportation Research 
Record, 795:40–46. 
26. Liu, H. X., Ma, W., Hu, H., Wu, X., & Yu, G. (2008). SMART-SIGNAL: Systematic 
Monitoring of Arterial Road Traffic Signals. In Intelligent Transportation Systems, 
2008. ITSC 2008. 11th International IEEE Conference on (pp. 1061-1066). IEEE. 
27. Liu, H. & W. Ma (2009). A virtual vehicle probe model for time-dependent travel 
time estimation on signalized arterials. Transportation Research Part C, 17 (1): 11-
26.  
28. Liu, H., X. Wu, W. Ma, & H. Hu (2009). Real-time queue length estimation for 
congested  signalized intersections. Transportation Research Part C, 17 (4): 412–
427. 
29. Liu, H. X., J. Zheng, H. Hu, and J. Sun, Research Implementation of the SMART 
SIGNAL System on Trunk Highway (TH) 13. MnDOT Research Report, 
MN/RC2013-06. Accessible online: 
http://ntl.bts.gov/lib/47000/47100/47104/MnDOT2013-06_1_.pdf. Accessed on 
March 15, 2013. 
30. Lin, F.B (1982a). Estimation of Average Phase Durations for Full-Actuated 
Signals.In Transportation Research Record 881, TRB, National Research Council, 
Washington, D.C., pp. 65–72. 
31. Lin, F. B (1982b). Predictive Models of Traffic-Actuated Cycle Splits. 
Transportation Research, Vol. 16B, No. 5, pp. 361–372. 
32. Morgan, J.T. & J.D.C. Little (1964). Synchronizing Traffic Signals for Maximal 
Bandwidth. Operations Research, Vol. 12, pp. 896-912. 
33. National Electrical Manufacturers Association. (2003). NEMA Standards 
Publication TS 2. National Electrical Manufacturers Association. Rosslyn, VA. 
34. National Transportation Operations Coalition. (2012). 2012 National Traffic Signal 
Report Card- Technical Report. National Transportation Operations Coalition. 
Washington, D.C. 
 
 
54 
 
35. Newell, G. F. (2002). A simplified car-following theory: a lower order model. 
Transportation Research Part B, 36(3), 195-205. 
36. Petty, K., J. Kwon & A. Skabardonis (2005). A-PeMS: An Arterial Performance 
Measurement System. 2006 Annual Meeting Workshop. Washington, D.C. 
37. Robertson, D. I. (1969). TRANSYT: a traffic network study tool. 
38. Robertson, D. I., & Bretherton, R. D. (1991). Optimizing networks of traffic signals 
in real time-the SCOOT method. IEEE Transactions on Vehicular Technology, 40(1), 
11-15. 
39. Skabardonis, A. (1996). “Determination of timings in signal systems with traffic-
actuated controllers.” Transportation Research Record. 1554, Transportation 
Research Board, Washington, D.C., 18–26. 
40. Smaglik, E. J., A. Sharma & D. M (2007).  Bullock . Event Based Data Collection 
for Generating Actuated Controller Performance Measures. Transportation Research 
Record, 2035: 97–106. 
41. Smaglik, E. J., D. M. Bullock, D. Gettman, C. M. Day, & H. Premachandra (2011). 
Comparison of Alternative Real-Time Performance Measures for Measuring Signal 
Phase Utilization and Identifying Oversaturation. Transportation Research Record 
2259: 123–131. 
42. Sunkari, S. R., H. A. Charara, & P. Songchitruksa (2011). A Portable Toolbox to 
Monitor and Evaluate Signal Operations. TTI Report No. 0-6177-1. 
43. TRB (2000). Highway Capacity Manual. National Research Council, Washington, 
D.C. 
44. Wallace, C. E., Courage, K. G., Hadi, M. A., & Gan, A. C. (1998). TRANSYT-7F 
user’s guide. Transportation Research Center, University of Florida, Gainesville, 
Florida. 
45. Webster, F. V. (1958). Traffic Signal Settings. TRRL Report 39. U.K. Transport and 
Road Research Laboratory, London.  
46. Wu, X., H. Liu, & D. Gettman (2010). Identification of Oversaturated Intersections 
Using High-Resolution Traffic Signal Data. Transportation Research Part C, 18 (4): 
626–638. 
47. Wu, X. and Liu, H. (2011) A Shockwave Profile Model for Congested Urban 
 
 
55 
 
Arterials, Transportation Research Part B, 45(10), 1768-1786.7. 
48. Yin, Y., Li, M., & Skabardonis, A. (2007). Offline Offset Refiner for Coordinated 
Actuated Signal Control Systems. Journal of transportation engineering, 133(7), 
423-432. 
49. Zhang, L., & Yin, Y. (2008). Robust synchronization of actuated signals on arterials. 
Transportation Research Record, 2080(1), 111-119. 
50. Zhang, L., Yin, Y., & Lou, Y. (2010). Robust signal timing for arterials under day-to-
day demand variations. Transportation Research Record, 2192(1), 156-166.