Transportation Research Record 1069 83

Timing Design of Traffic Signals

MAS AMI SAKIT A

ABSTRACT

A method of optimizing signal timing by the use of a linear programming tech-

nique is described. The linear program is based on the premise that the green
time given to each traffic movement in the signal cycle must be long enough to
handle the traffic coming into the intersection in one cycle. The linear pro-
gramming method gives minimum signal phase times, minimum cycle time, and
critical movements. The optimum signal phase times and cycle time can also be
obta i ned in a nondirect manner. The linear programming method may be used as an
off-line tool to analyze signal timing or it may be adopted as the basis for
developing a new local controller for critical intersections of centrally con-
trolled systems. When used as an off-line tool, the linear programming method
is very effective for "what if" analysis of traffic signal timing problems.

In recen t year s , renewed attentio n has been pa id to /J2 156

analyzing capacity and signal timing at signalized
in t e rse c tions . Two representative wor ks are those o f
Akcelik (1) and JHK and Associates (2). The former
emphasizes the analysis of signal timing and the
latter, capacity applications in the real world. In
both analyses , however , e r i tical movements play a
key role . The importance of c ri t ical movements in
the analysis o f i nte r section capacity and signal
timing may be due to the introduction of movement-
or iented signal controllers as alterna tives to
phase-oriented signal controllers.
The old-fashioned phase-oriented controllers are
very inflexible relative to signal phasi ng : a fixed
sequence of phases is repeated until t he traffic 113 ~7

engineer changes the signal phase sequence. Usually

not many overlapping movements of tra ffic can be
accommodated between phases. In contrast, the move-
ment-oriented traffic controllers are much more
flexible in the selection of phase sequence, and
they can handle overlapping traffic movements from
one signal phase to another.
To illustrate the point, Figure 1 is a diagram of 4
an eight-phase signal controller. Each box indicates 3 _}
s_
r 7
a phase that consists of a combination of noncon-
flicting traffic movements. As Figure 1 shows, many
phase sequences are possible with this type of local
controller. In addition, the overlapping of move-
ments between phases permits reduction of cycle time
by allowing add i tional phases without increasing the FIGURE 1 Typical phase sequence diagram for an eight-phase
total lost time in a cycle. (Lost time is time not controller.
efficiently used to handle traffic.)
The cycle-by-cycle local-actuation method is most
commonly used a s f o r signal timing and phas e sequenc- LINEAR PROGRAMMING METHOD
i ng f or isolated signals . Both pretimed and ac t uated
types are commonly used f o r cen tr ally controlled Consider a single lane in the approach to a signal-
signals . A linear prog ramming method for optimizing ized intersection. The traffic in that lane gets a
signa l timing a nd phase s eq uenc ing f o r pretimed loca l green signal f or a c ertain d urati on o f t ime in one
controllers is discussed here. This method may be cycle of signa.l. The i ncomi ng flow during the e ntire
used as an off-line tool for signal timing and c ycle l ength mus t be di scharged dur ing t he gr een
capacity analysis, similar to other methods devel- time i nter val. It is assume d that m t raffic move-
oped for the analysis of signalized intersections men ts ar e controlled at this i ntersection a nd t ha t
(~),or it may be used as a basis for on-line phase the signal controlle r hand les pretimed n..,phase s i n
seque nce t r i ggering a nd signal tim ing plan generation o ne cycle . Fur t he r, th e traffic movement is the ith
i nco rpora te d withi n a l ocal controller . The utility moveme nt, and this movement recei ves th e green signal
of the line ar programm ing method is demonstrated by ind icatio n i n a se t of signal phases iden tified as
applying it to a real-world problem. i . The minimum green t ime duration assigne d to t r af-
fic movement i is expressed as (_!)
Masami Sakita and Associates, 1259 El Camino Real
No. 121, Menlo Park, Calif. 94025. (1)
84 Transportation Research Record 1069

where
f,v
Gi green time assigned to traffic movement i, if'\•
.>
movement (green) 2 (green)
C cycle time, ·:·!
\"";
qi incoming traffic flow rate, and movement 5 ( green ) 6 (green)
;.~
si practical saturation flow rate for traffic
movement i.
¢1 1¢2 I ¢4

The practical saturation flow rate is the rate that

can be maintained throughout the signal cycle under (a) All Signal Phase Ti mes Are Longe r than
the prevailing conditions. the Inter,P,'ree n Ti.mes
Inequality 1 merely means that the total signal
phase times assigned to handle traffic movement i
are long enough to handle all the vehicles (of that
movement) corning into the intersection in one cycle
time. Traffic flow in a movement can be handled movement 1 (green) movement 2 (green ~~.
.;,..,
either by one signal phase or by multiple signal movement 5 (green) movement 6 (green ~.;J
phases. For example, movement i may be handled by ~·~

signal phases j and j + 1, whereas movement i + 1

may be handled by phases j + 1, j + 2, and j + 3. In
sll
Lhis t:a>ie both movements i an~ i + 1 are handled by
multiple phases, and phase j + 1 handles both i and
i + 1 movements. (b) One Signal Phase Time (Phase 2) ls Shorter
than the Intergreen Time
If Xj denotes the phase time for signal phase j
and Xi d e notes the sum of signal phase times that FIGURE 2 Example of phase sequence combinations.
h a ndle movement i,

Gi = Xi - Li = l Xj - Li (2) phase do not get green time simultaneously. Under

¥j£iprni such circumstances, Equation 3 is rewritten as

where <Pmi is a set of signal phases that handles x~ gj (3a)

traffic movement i and Li is lost time for traffic J

movement i. From Equations 1 and 2, signal timing constraints

The lost time corresponding to a phase change can be expressed as
consists of the starting delay ending lag and all-red
duration. It is closely related to the intergreen I Xj - Li ~ c qi/Si for i 1, 2, ... , rn (4)
time following the green time that carries the move- Vj£<t>mi
rnent--the amber duration (when no all-red duratio.n
is used) or the amber and all-red duration time. The The cycle time (Cl is expressed as
lost time is associated with the traffic movement,
and the intergreen t ime is associated with the signal (5)
phase. In reality, however, the intergreen time could
be considered as being defined by the movement it- The minimum cycle time that satisfies Constraints 4
self, because of the relationship between the move- can be obtained by solving a linear program the ob-
ment and the signal phases. In this paper the terms jective function of which is to minimize the cycle
"signal phase time" and "movement phase time" are time as expressed in Equation 5 subject to Con-
used. The signal ph~se time denotes the time duration straints 4. (The minimum cycle time is the cycle
consisting of the green time and the intergreen time time that is theoretically just sufficient for the
following the green time, or, in some cases, the traffic to pass through the intersection.)
green time itself. The movement phase time is the Constraints 4 can be simplified by introducing an
effective green time plus the lost time for that m x n matrix (P) expressing the relationship between
movement. signal phases and traffic movements, as follows:
If the signal phase time is longer than the in-
tergreen time, the phase time for signal phase · j, (6)
xj, is expressed as
The P-rnatrix indicates the movements handled by each
of the n phases of the signal cycle. The matrix con-
s is ts of elements that are either 0 or 1. In the
where gj is green-time durat i on of signal phase j matrix aij = l indicates that the ith movement gets
and lj is intergr ee n-t ime duration between signal a green indication in the jth phase, and aij = 0 in-
phases j and j + 1. dicates otherwise.
Sometimes a phase time is so short that the ending Then the linear program shown in Constraints 4
of the intergreen time for one traffic movement and Equation 5 can be rewritten as follows:
overlaps with the starting time of the intergreen
time that handles another traffic movement, This To find nonnegative x1 1 x 2 , ••• , Xn that minimize
situation occurs rather frequently with movement-
or iented local controllers. The phase sequence q,l, rn
ip2, and $4 in Figure l is used to illustrate this c l Xj (7)
very short phase time. Figure 2 shows two phase-time j=l
cases for illustrative purposes. In Case a, q,l,
ip2, and <1>4 have phase times longer than the inter- subject to
green times. In Case b, the phase time of <1>2 is
shorter than the intergreen time between ip2 and n
ip4. When the phase time is shorter than the inter- I (aij - qi/si) xj > Li for i 1, •••I m (8)
green time, the traffic movements in that short j=l
Saki ta 85

The dual problem is expressed as follows: linear program identical to the one described in
expressions 7 and 8, except that the right-hand side
To find nonnegative Ylr Y2r ... , Ym that maximize of Inequality 8 will be rLi instead of Li• In the
corresponding dual problem, expressions 9 and 10,
m rLi will also replace Li· Thus, the optimum timings
c (9) can be obtained either in two stages or in one. In
Li Yi
i=l the first stage of the two-stage approach, the mini-
mum signal phase times and cycle times are solved to
subject to obtain the r-value i and in the second stage, the
optimum signal times are obtained with the r-value.
m The one-stage approach is to solve the linear pro-
l: (aij - qi/Si) Yi < 1 for j 1, 2, ... , n (10) gram by using an approximate value for r, for
example, 2.0.
i=l

The unknown variables Yl• Y2r ••• , Ym of the dual

problem are dimensionless. The nonzero values of y 1 •s SIGNAL PHASE SELECTION AND OTHER CONSIDERATIONS
obtained as the solution indicate that the corre-
sponding movements are the critical ones that define The construction of a phase sequence chart is es-
the signal phase times and the cycle time. Those non- sential in the signal timing design process. The
zero Yl 's, however, do not reflect the movement chart is used identify signal phases and traffic
phase times of the critical movements. In fact, all movements and their relationships before construc-
nonzero y i's will have the same value as long as tion of linear programs. The chart is also used to
no other types of constraints are introduced into identify the phase sequence after the linear program
the linear program. has been run because the linear program will not
directly identify the phase sequence: It will select
the phases and critical movements that satisfy the
requirements. The chart should include all the pos-
OPTIMUM SIGNAL TIMING
sible signal phase sequences. For example, Figure 3
is a phase sequence chart of a typical two-way/two-
So far, the discussion has dealt with the minimum
way intersection, where right turns are allowed all
required green times and cycle time. The direct
the time and the left turns are handled by separate
mathematical programming approach to solve the opti-
green arrow signals. In the figure, phase sequences
mum timing of signals (least delay for all vehicles)
can be identified by tracing a sequence of phases,
is not simple because the objective function for the
following the arrows, and closing a loop that covers
problem cannot be expressed as a linear combination
of unknown variables (4). The approach used here is both north-south and east-west legs of the intersec-
tion. For example, $1 1 $4 1 $5 1 and $8 are a possible
an indirect one. The linear programming method has
phase sequence. Similarly, a possible phase sequence
been combined with the findings from earlier signal
timing studies. chart can be generated for such configurations as T-
The minimum cycle time in seconds, when all the intersections or five-legged intersections.
critical movements are known, is expressed as (~)

cm = L/(l - Y) (11) iS2

where L is the total lost time in a cycle (in sec-

onds) and Y is the sum of degrees of saturation for
all critical movements. The total lost time is not
the sum of all the intergreen times but is the sum
of lost times for all the critical movements. The
minimum cycle time (Equation 11) can easily be veri-
fied by the use of the linear programming method.
Webster and Cobbe (5) showed, by the use of com-
puter simu lation, that- the optimum cycle time (C0 )
that give s the least average delay to all vehicles
using the intersection is
iS3 iS7
C0 = (l.5L + 5)/(1 - Y) (12)

where C0 and L are in seconds.

Thus, from Equations 11 and 12 the optimum cycle
time can be expressed as a function of the minimum
cycle time and the lost time:

C0 = ((1.5 L + 5)/L] Cm= r Cm (13)

The CofCm ratio (r) is not sensitive to variation

of the L value: At L = 16 sec (which is considered to
_J1(L
be a long total lost time): At L = 16 sec (which is
~__)
,-7
-4
considered to be a short total lost time), r = 2.13.
In most cases, the r-value is in the range of 1.8 to
2.2. The findings by Webster and Cobbe (~l show that,
-
in a practical sense, the optimum cycle time is rea-
sonably well approximated by twice the minimum cycle
time.
The optimum signal phase times and cycle time can FIGURE 3 Possible phase sequences for a two-way/two-way
be obtained by solvi ng the primal probl e m of the intersection.
86 Transportation Research Record 1069

An important factor to consider in determining ¢2

signal timing is the requirement for noncritical
movements. Signal phase times must be determined to
satisfy the timing requirements not only for critical
movements but also for noncritical movements. TO
illustrate the problem, refer to the north-south
half of the phase sequence diagram presented in
Figure 3. If the selected phase sequence is <j>l,
<1>2, and <!>4, the critical movements that determine
the phase times are either l and 2 or 5 and 6. If l
and 2 are the critical movements, the linear program
with constraints consisting of only those require-
ments for the er i ti cal movements will select signal
phase times based on the timing requirements for
traffic movements 1, 2, and 5 or for movements 1, 2, ¢3
and 61 but the selected signal phase times will not
be based on the requirements for all four traffic
movements simultaneously. Thus, the selected signal
phase times may yield a situation where <1>1 and
<1>2 are both fully saturated in both directions but
<1>4 is saturated only in one direction (movement 2)
or <1>4 is saturated in both directions but both -s
<1>1 and <1>2 are saturated only in one direction (move-
ments 1 and 2). The requirement for signal-phase-
time and movement-time selection is that it be in
3
-
such a manner that the degrees of saturation for
noncritical movements (movements 5 and 6 in the
example) are equal when signal phase times are ad-
FIGURE 4 Phase sequence chart and
justable. This can be achieved by adding these re-
intersection layout for text example.
quirements to the constraints of the linear program.
The new constraints are expressed in any equality
form. For example, in the north-south direction of
TABLE I Traffic Characteristics Data for Example in Text
the two-way/two-way intersection, this requirement
for the minimum cycle time and phase times is ex-
Practical Lost Interfeen
pressed as Volume Saturation Flow Time 8 Time
Movement (veh/hr) Rate (veh/hr) (sec) (sec)
(x3 + x4 - 1 180 1,440 4 6 (2+4)
L4)/(~/s4) (14)
2 840 2,520 4 6 (2+4)
3 620 3,060 4 6 (2+4)
(x2 + x4 - 4 400 1,440 4 6 (2+4)
(15) 5 600 2,700 4 6 (2+4)
L4l/(q2/s2l 6 400 3,060 4 6 (2+4)

The same requirement for the optimum cycle time and ~Lo:ii:t tjme corresponding to a phase change.
Red+ amber,
phase times is expressed in the same manner as in
Equations 14 and 15, except that L1 1 L2, L3, and L4
are replaced by rL 1 , rL 2 , rL 3 , and rL 4 , respectively. to operate the east-west traffic. For this example,
Other signal timing considerations, such as pedes- the phase-movement relationship matrix P will be
trian crossing time or minimum green time require-
ments, can also be put in an inequality form as a

(
constraint for the linear program. These require-
ments, however, must be formulated as part of the ~ ~ ~ ~ ~)
0 0 0 0 1
constraints cf the linear program that seeks for the p l l 0 0 0 (16)
optimum timing of signals and should not be incor- 0 0 1 1 0
porated in the linear program seeking for the minimum 0 0 0 0 l
timing.
The resultant linear program is to find nonnega-
tive xi, ••• , x 5 that minimize

THE EXAMPLE
C = x1 + x 2 + ••• + x 5 (17)
To show how the linear programming technique is ap-
plied, it was used to solve the following example of subject to the constraints
a real-world problem. The example is a typical four-
leg intersection that has a total of six movements. 0.875x1 - O.l25x2 + 0.875x3 - 0.125x4 - 0.125x5 > 4.0
The major street is controlled with full separate -0.333x 1 + 0.667x 2 - 0.333x3 + 0.667x4
turning signals, and the minor street is controlled - 0.333x 5 > 4.0
by one phase for all movements. Data required to -0.202x1 - o.202x2 - 0.202x3 - 0.202x4
compute signal timing are given in Table 1 and Figure + 0.798x5 > 4.0
4 shows possible phasing schemes and the layout of 0.722x1 + 0.722x 2 - 0.278x3 - 0.278x4 - 0.278x5 > 4.0
the intersection. -0.222x 1 - 0.222x 2 + 0.778x 3 + 0.778x4
The traffic data (Table 1) indicate that at least - o.222x 5 ~ 4.0
two phases are required to handle traffic for the -0.13lx1 - 0.13lx2 - O.l3lx3 - 0.13lx4
north-south street, but only one phase is required + 0.869x 5 ~ 4.0 (18)
Saki ta 87

The linear program has two sets of solutions. The nondesired phases are zero. In this manner, it is
first set is x 1 9 sec, x2 = 6 sec, x4 = 13 sec, possible to conduct "what if" types of analyses by
x 5 = 12 sec, and C = 40 sec. In the second set, the varying constraints in the linear program.
solutions are x 2 = 15 sec, x 3 = 9 sec, x 4 = 4 sec,
x5 = 12 sec, and C = 40 sec. In the first set Phases
1, 2, 4, and 5 are selected; and in the second set CONCLUSIONS
Phases 2, 3, 4, and 5 are selected by the linear pro-
gram. In both cases the binding dual variables are The linear programming method can be used not only
y3, y4, and y5, i mplying that movements 3, 4, and 5 to optimize signal timings but also to find critical
are the critical moveme n ts. If any special phase se- movements. The linear programming method may be used
quence is desired, that requirement must be added to as an off-line tool to analyze intersection signal
the constraints. For example, adding timing and capacity, or it may be used as a oasis
for an on-line signal-timing-optimization and phase-
(19) sequence-tr igger ing mechanism that can be installed
within the local controllers.
to the constraints will force the linear program to
take the first set in the foregoing example.
In addition to the basic signal timing require- REFERENCES
ments, the timing requirement that the degrees of
saturation of noncritical movement phase times be 1. R. Akcelik. Traffic Signals: Capacity and Timing
equal whenever possible is added. In this example, Analysis. Research Report 123. Australian Road
this is expressed as Research Board, Nunawading, Victoria, Australia,
March 1981.
xl - 0.375x2 + x 3 - 0.375x4 = 2.50 (20) 2. Signalized Intersection Capacity Method. JHK and
Associates, San Francisco, Calif.: Traffic In-
The solution of the foregoing linear program with stitute, Northwestern University, Evanston,
this additional requirement is x 1 = 9 sec, x2 = 6 Ill. I May 198 2 o
sec, x 4 = 13 sec, x 5 = 12 sec, and C = 40.0 sec. (The 3. A.O. May. Description of Some Fixed-Time Signal
binding dual variables are naturally the same as Control Computer Programs. Institute of Trans-
those without the new constraints.) portation and Traffic Engineering, University of
The total lost time for a cycle is 12 sec. Thus, California, Berkeley, Sept. 1973.
the r-value is 1. 92. The optimum signal timings are 4. G.F. Newell. Lecture Notes for CE 263 Highway
x1 = 17 sec, x2 = 12 sec, x 4 = 25 sec, x5 = 23 sec, Traffic Control. University of California,
and C = 77 sec. Berkeley, Spring 1960.
In this specific case, the total lost time in a 5. F.V. Webster and B.M. Cobbe. Traffic Signals.
cycle is 12 sec for any phase combination. There- Technical Paper 56. U.K. Transport and Road Re-
fore, it could have been possible to directly solve search Laboratory, Crowthorne, Berkshire, Eng-
the linear program for the optimum timings as de- land, 1966.
scribed in the previous section. Other requirements,
such as m1n1mum green time and pedestrian green
times, are also possible constraints. If so required,
it is also possible to predefine the phase sequence Publication of this paper sponsored ~y Committee on
by putting in the constraints that phase times for Traffic Control Devices.