8501 Mo-Pac Blvd.
P.O. Box 201088
Austin, TX 78720-1088
RCN 298-018-01-00	(512)454-4797
DCN 91-298-018-01-02
Prepared for:
Certification Division
Environmental Protection Agency
2565 Plymouth Road
Ann Arbor, Michigan 48105
Prepared by:
Timothy H. DeFries
Rob Klausmeier
Radian Corporation
P.O. Box 201088
Austin, TX 78720-1088
18 July 1991

3.1	Two Possible Bases for a Certification Test
Procedure		3-1
3.2	Studies Needed to Support Certification Test Procedure
Development		3-4
3.3	Selection of Driving Parameters to be Measured 		3-5
3.3.1	Parameters that Affect Vehicle Driving
Characteristics 		3-5
3.3.2	Important Driving Parameters to Measure		3-7
4.1	Method #1 - Instrument Private Vehicles to Obtain
Continuous Data		4-1
4.2	Method #2 - Instrument Vehicles to Obtain Summarized
Data		4-6
4.3. Method #3 - Perform Chase Car Surveys		4-8
4.4	Method #4 - Use Vehicle Diaries 	4-11
4.5	Method #5 - Make External Observations With Stationary
Observers 	4-12
4.6	Comparison of Different Approaches and Recommendations
for the Approach to be Used in the Driving Pattern
Study 	4-13
5.1	Elimination of Bias 		5-1
5.1.1	Instrumented Private Vehicles 		5-1
5.1.2	Chase Car Surveys 		5-3
5.2	Definition of Most Appropriate Survey Site 		5-4

6.1	Summary Statistics 		6-1
6.2	Signal Processing Techniques 		6-1

Evaluations have been made of various measurement techniques to collect
data on the driving patterns of light-duty vehicles in the population. This data would be
used to evaluate the current certification test procedure, and revise, update, or replace it
if necessary. To better evaluate the techniques, consideration has been given to the
broader issues of the development of a certification test procedure: selection of
measurement parameters, selection of measurement techniques, measurement of driving
parameters, analysis of driving pattern data, and certification test procedure
A new certification test procedure could be made in various formats, but
all candidate procedures would probably specify dynamometer speed versus time. A new
procedure could be representative of either population driving patterns or population
emissions patterns. A clear recommendation on which approach to take is difficult; each
has strong advantages and disadvantages. The following compares the attributes of
certification test procedures which are based on either driving patterns or emissions
	Vehicles tested using a driving pattern based certification test
procedure would spend dynamometer time for various driving
features in proportion to the occurrence of the same driving features
in the vehicle population. This would produce emissions
representative of the vehicle population. Emissions characterization
would be good for today's emission control technology, fuel
composition, and emission types and also for future technologies,
fuels, and emission types. However, because the certification test
procedure would be representative of driving patterns, this type of
certification test procedure would be inefficient. That is, most
dynamometer time would be spent on driving features that produce
very low emissions; a small fraction of the dynamometer time would
be spent on high emitting features.
	Vehicles tested using an emissions based certification test procedure
would spend dynamometer time for various driving features in
proportion to the mass emissions of the same driving features in the

vehicle population. Because this new certification test procedure
would be representative of emissions patterns, the procedure would
be efficient. That is, all dynamometer time would be spent
measuring the emissions of driving features in proportion to the
importance of the driving feature emissions to the total emissions of
the vehicle population. However, as vehicle manufacturers improve
vehicle emission control technology in response to the new
certification test procedure, as new fuels and fuel compositions are
introduced into the marketplace, and as different pollutants become
of concern (for example, perhaps formaldehyde), the new
certification test procedure will become less relevant for measuring
the emissions of new vehicles. At some point, a need for a new
certification test procedure will become apparent. Still, an
emissions pattern based procedure may be better than a driving
pattern based procedure for some time.
Both the driving pattern based and the emissions pattern based
certification test procedures will require driving pattern data for their development. In
addition, an emissions based procedure would require the relationship between emissions
and driving. However, Radian suggests that even if the emissions based procedure is
desired, the emissions versus driving study would best be performed separately from the
driving pattern study. This separate approach benefits the driving pattern study in
several ways:
	Since fewer parameters need to be collected for the pure driving
pattern study, simpler instrumentation of private cars is required.
	Fewer parameters means that early, as well as late model, cars can
be used in the study to get a picture of driving pattern over the
useful lifetime of cars.
	Simpler instrumentation means that more cars can be instrumented
for the same funds and therefore more driving pattern data will be
	Simpler instrumentation means that fewer refusals to participate will
be encountered because the equipment installation is simpler, faster,
less intrusive, and less inconvenient for the owner.

	Fewer refusals to participate means that the results of instrumented
vehicles would potentially be less biased and representative sampling
efforts to overcome bias would need to be implemented less often.
	Requiring fewer parameters means that the chase car technique will
produce data which is more similar to the data from the
instrumented private vehicle technique.
Radian recommends that the driving patterns be measured with a program
that contains two data collection studies:
	Private vehicles instrumented for clock time (seconds), vehicle
speed, ignition on/off, and manifold vacuum or manifold absolute
pressure (MAP). In addition, information about the vehicle should
be recorded: Vehicle Identification Number (VIN), power/weight
ratio, and transmission type.
	Instrumented chase car with differential speed ranging device for
clock time (seconds), vehicle speed, and ignition on/off. In
addition, information about the vehicle should be recorded: license
number, VIN, power/weight ratio, and transmission type.
Because the instrumented private car technique and the chase car
technique are complementary, Radian recommends that both techniques be used so that
the advantages of each are present in the program. For both studies considerable
attention and effort needs to be used to attempt to minimize the effects of vehicle and
driver sampling biases.
The instrumented private car technique provides continuous detailed
information about vehicle speed wherever the vehicle is driven; however, potential biases
exist in the collection of data because not all owners will want their vehicles
instrumented and drivers may drive differently if they know the vehicle is instrumented.
It seems that the most attractive procurement location for instrumented private cars is at

centralized inspection/maintenance (I/M) lanes.
The chase car technique provides detailed speed information for more
vehicles at a lower cost per vehicle than can be obtained from instrumented vehicles.
However, since it is not practical to chase a single car for a complete trip, chase cars are
more effectively used to monitor behavior over selected links in the road system. Since
the driver of a vehicle being chased will probably not be aware that he is being followed,
the driving behavior will probably not be biased. However, to avoid other kinds of bias,
the chase car technique must also use appropriate sampling techniques for the selection
of links to monitor and specific vehicles to chase. Urban Transportation Systems
Planning can be used to help make link selections, but all possible driving paths are not
covered by these models.
Radian also recommends that at each survey site, both techniques should
collect data at the same time. This will eliminate the possibility of differences between
the two study results caused by city or date. The following guidelines can be used to
chose the survey sites:
	The site should be a non-attainment area but should not be
dependent on whether it is an ozone or CO non-attainment area.
	The site should be at low altitude, be relatively flat, and have low
precipitation during the survey period.
	The typical traffic should be neither congested nor light.
	A site using centralized I/M and having a good Urban
Transportation Planning System is highly desired to reduce bias in
the instrumented private car study and in the chase car study,
	A site which has an air quality staff which is willing and able to
assist with various aspects of the studies is desirable.

The Clean Air Act Amendments of 1990 require EPA to evaluate the
Federal Test Procedure (FTP) and revise it as necessary to ensure that vehicles are
tested under circumstances which reflect actual current driving conditions. These
conditions include fuel, temperature, acceleration, and altitude. EPA currently is making
appropriate changes in the procedure to address concerns over the representativeness of
the fuel, temperature, and altitude provisions. However, the current driving cycle used in
the FTP has not been evaluated to determine if it is representative of vehicles in actual
EPA's Certification Division (CD) of the Office of Mobile Source Air
Pollution Control has been given the lead responsibility to analyze and modify, if
necessary, the FTP. CD issued a work assignment to Radian Corporation to collect
information on in-use driving patterns and define the most appropriate method to
proceed with the task of modifying the FTP.
Characterizing in-use driving conditions is a complex problem that must
consider several interrelated factors. Of key importance is how driving pattern data
should be collected. The data collection method must be straightforward and be
representative of the population.
The choice of measurement techniques (to) characterize driving
patterns in real world fleets depends on the data that must be collected, how the data
will be analyzed, and the desired certification test procedure to be developed. This
report will cover the five stages which are used to develop a certification test procedure:
	select driving parameters;
	select measurement techniques;
	measure driving parameters;

	analyze driving pattern data; and
	develop certification test procedure.
Each of these stages of development will be discussed in the following sections so that
the needs of the measurement program can be put in perspective. Evaluation of
candidate driving pattern measurement techniques will be presented in greater detail.

The ultimate goal of the driving pattern test program will be to evaluate
the current certification test procedure, the FTP. This data will be used to determine
whether the FTP is adequate, needs to be updated, or needs to be replaced. The current
test procedure specifies the speed of the vehicle versus time, and it is anticipated that
any new procedure which may replace it would use a similar approach.
3.1	Two Possible Bases for A Certification Test Procedure
A new certification test procedure can be developed using one of two
bases. The cycles can be representative of today's:
	driving patterns; or
	emissions patterns.
Each basis has advantages and disadvantages as discussed below.
A driving cycle which is based purely on driving patterns would be
representative of the driving pattern of the vehicle population. Emission measurements
taken while vehicles are driven in traffic indicates that such a cycle would spend most
time testing driving features which produce low emissions and a small amount of time
testing high emission driving features. During certain episodes, for example, during cold
start operation or during heavy acceleration, HC and CO emissions are high. However,
these high emission events occur during a small fraction of the total driving time. This
means that a driving cycle which is representative of vehicle population driving patterns
will be inefficient at measuring the performance of the vehicle emission control systems
during high emissions episodes, even though it is accurately measuring the emissions
performance of the vehicle in a cycle which is representative of real world driving.

Alternatively, a certification test procedure which is based on the emissions
pattern of the vehicle population will reflect the range of driving in the real world but
will be weighted by the response of the vehicle emissions to those driving features.
Because of this, the emissions based driving cycle would contain more high emission
driving features and fewer low emission driving features. Thus, the procedure would be
efficient at measuring the high emissions behavior of vehicles.
An example will help clarify the important distinction between a driving
pattern based certification test procedure and an emissions pattern based certification
test procedure. Suppose that analysis of the driving pattern data and a separate study of
the relationship between emissions and driving indicated that the emissions rate of
hydrocarbon could be explained by just two kinds of driving modes: Type A and Type B.
Suppose that it were found that vehicles spend 95% of their driving time in Type A
driving but generate only 20% of the total fleet hydrocarbon emissions in this mode.
Suppose that vehicles spend only 5% of their driving time in Type B driving but generate
80% of the total fleet hydrocarbons in this mode.
The driving based certification test procedure based on this hypothetical
information would be a cycle with 95% of the time in the A mode and 5% of the time in
the B mode. The emissions based certification test procedure would be a cycle with 20%
of the time in the A mode and 80% of the time in the B mode. This second procedure
emphasizes the time spent on high emissions features.
A driving cycle which is based purely on driving patterns will need to be
updated only when the driving patterns of the vehicle population change. Changes may
occur relatively slowly over perhaps twenty years. A driving pattern based test procedure
will also continue to representatively measure the emissions of vehicles even though the

vehicle emission control technology and fuel types may change dramatically in the next
several years.
A driving cycle based on emissions patterns depends on the emissions
behavior of current vehicles, which in turn depends on the type of fuel and emission
control technology used when the current vehicles were being certified. Thus, for
example, a new driving procedure developed in the next few years would be based on the
emission characteristics of 1992 vehicles. These vehicles are, of course, creatures of the
current federal test procedure, since they have been designed to pass that procedure. As
vehicle manufacturers continue to develop more advanced control technologies and as
new types of fuels become available in the market place, a driving cycle based on
emission behavior of 1992 vehicles may become a poor measure of the emission behavior
of the latest vehicles. Consequently, the new driving cycle may become obsolete in a
short time. A continually revised certification test procedure in response to changes in
vehicle technology and fuels will mean that the rules for manufacturing a new vehicle
would be continually changing. Manufacturers would find such procedural changes
difficult to quickly respond to, given the long lead times needed to integrate a new
control system with a future model year vehicle. Also, the EPA would need to update
the certification test procedure on a regular basis. In addition, there is no guarantee that
this circular process would converge toward lower vehicle emissions and a simpler
certification test procedure.
Another disadvantage of the emissions based test procedure is that the
desired weighting of the high and low emitting driving features may not be the same for
HC, CO, and NO, emissions. Road emissions data indicates that HC and CO emissions
increase greatly during cold start operations and heavy accelerations, but the response of
NOx emissions to driving may be different.

The choice of whether to use a driving pattern based or an emissions
pattern based test procedure is difficult because which will be more effective depends to
a large extent on the speed and degree of changes which may occur in emission control
systems, fuel composition, and emissions of concern. If changes are slow and small, and
if the new procedure can be implemented rapidly, then the emissions based procedure
will be more effective for a long time by concentrating on the high emitting driving
features of today's vehicles and vehicle population. If changes are fast and large, then
the driving pattern based procedure will remain effective for a long time.
2>2	Studies Needed to Support Certification Test Procedure Development
The data needed to develop a certification test procedure depends on
which of the bases discussed above is desired. If a new driving cycle is to be based
purely on driving patterns of the population, then only the driving pattern of the vehicle
population needs to be measured. However, if a new cycle is to be based on the
emissions pattern of the vehicle population, then another study must be performed to
measure the emissions response of 1992 vehicles to driving. This study could be either a
dynamometer or a road study. It could be conducted at the same time that the driving
pattern measurement of the vehicle population is being measured. The emissions
response of the vehicles would be used to weight the occurrence of corresponding driving
features in the new procedure.
Thus, it is clear, that regardless of whether the new test procedure will be
based on driving patterns alone or the emissions patterns of the vehicle population,
characterization of the driving patterns is needed. Therefore, the study on driving
patterns can proceed before a decision is made on how to develop the certification test

Selection of Driving Parameters to be Measured
Since both approaches to develop the certification test procedure need
measurement of the population driving patterns, parameters related to driving behavior
must be measured; parameters related strictly to emissions behavior do not need to be
collected. In this section, the parameters that affect vehicle driving characteristics will be
examined, and then those parameters which are the minimum required to define the
driving patterns will be identified.
3.3.1	Parameters that Affect Vehicle Driving Characteristics
The parameters that affect vehicle driving characteristics can be broken
down into four groups:
	Driving pattern parameters;
	Vehicle environment parameters;
	Vehicle characteristics; and
	Engine operating parameters.
Table 3-1 shows a list of the parameters in these four categories.
Driving pattern parameters include those which are determined by the
operator of the vehicle and the environment in which the vehicle finds itself. The most
important driving pattern parameter is the speed of the vehicle as a function of time. If
speed is known on a frequent basis, for example second-by-second, then the trip length
and acceleration of the vehicle can be determined. The engine-on period is also an
important driving parameter, since this helps determine trip length and periods of cold
and warm start operation.

Table 3-1
Driving Pattern
Speed (t)
Engine On/Off (t)
Vehicle Environment
A/C Status
Fuel Characteristics
Vehicle Characteristics
Drive Train Characteristics
Engine Operation
Engine Load (t)
Engine RPM (t)

Vehicle and environmental parameters also may affect vehicle operation.
Altitude, for example, may affect the emissions of the vehicle and also the way a vehicle
is driven, since internal combustion engines produce less power at high altitudes. The
demands on the performance of a vehicle will be affected differently in flat and hilly
terrains. The performance of vehicles with small engines can be greatly affected by the
operation of air conditioning. Finally, the fuel characteristics used in vehicles can affect
driving patterns - especially if low octane fuel is used and the driver backs off on the
throttle when engine knock is encountered.
Driving characteristics are also affected by vehicle design. The
characteristics of the transmission and other drive train components will affect the speed
of the engine with respect to the vehicle. Finally, the power/weight ratio and the weight
of the vehicle may have an influence on how the vehicle is driven.
While measures of engine operation are not directly relevant to vehicle
driving patterns, they can be used to estimate the vehicle emissions by knowing the
response of engine emissions in different modes of operation. Two parameters which
can be used to approximately define engine operation are manifold vacuum (or manifold
absolute pressure) and engine speed (rpm). In addition to engine operation, manifold
vacuum can provide some indication of terrain, cargo, and air conditioning loads,
although it cannot distinguish among them. Comparison of engine speed with vehicle
speed can be used to determine which transmission gear is being used at any given time.
3.3.2	Important Driving Parameters to Measure
From the list of parameters reviewed in Section 3.3.1, a subset should be
chosen to be used in the driving pattern study. These parameters should be ones which
can be used to define a new certification test procedure or ones which could be used to
understand population driving patterns. Based on our review of the literature, interviews

with experts, our judgment, and discussions which occurred at the June 6, 1991 meeting
in Atlanta, we recommend that the parameters given in Table 3-2 be used to determine
the driving patterns of the vehicle population.
The identity of each vehicle should be documented. This would include
the vehicle identification number (VIN), the vehicle power/weight ratio, and whether the
vehicle has a manual or automatic transmission. The VIN could be derived from the
vehicle license number, and the power/weight ratio and transmission type could be
derived from the VIN. Specific second-by-second driving data should include vehicle
speed and engine on/off period.
We feel that this relatively short list of data to be obtained in the study will
result in a study which has more vehicles being monitored and will result in less potential
bias for data obtained from instrumented vehicles.
The above parameters are the minimum ones required to define vehicle
driving patterns and can be effectively used to evaluate the current FTP or develop a
new certification procedure. Manifold absolute pressure and engine speed are two
additional, but optional, parameters that would be useful to define engine operation and
to understand how drivers and drive trains achieve the speeds that make up driving
patterns. Also, through the use of an engine map and emissions model such as
VEHSIME, these two parameters can be used to better estimate vehicle emissions if
engine maps for the specific vehicle drive train configurations are available. The
collection of these two optional parameters is attractive for future analysis of the data if
the incremental cost of collecting them is not too great.

Table 3-2
Parameters to be Recorded in Driving Pattern Study
Vehicle Identification Number (VIN)
Transmission: Automatic or Manual
Vehicle Speed (sec)
Ignition On/Off (sec)
Engine Speed (rpm)
Manifold Absolute Pressure

The following approaches have been identified to collect data for the
purpose of evaluating and revising the FTP:
	Method #1-Instrument private vehicles to obtain continuous data;
	Method #2~Instrument private vehicles to obtain summarized data;
	Method #3--Perform chase car surveys of individual vehicles;
	Method #4--Use vehicle diaries to collect information about driving;
	Method #5--Make external observations with stationary observers.
Each of the above methods is described below. Each technique has advantages and
disadvantages. In each discussion, measures which could be used to mitigate
disadvantages are summarized.
4.1	Method #1--Instrument Private Vehicles To Obtain Continuous Data
A popular approach toward characterizing in-use driving is to instrument
private vehicles to obtain continuous data on parameters important to characterization of
driving cycles. Instrumentation is available to collect at least the following parameters
on a second-by-second basis:
	Engine speed;
	Vehicle speed;
	Clock time;
	Demand air/fuel ratio;
	Actual air/fuel ratio;

	Air flow;
	Manifold absolute pressure;
	Throttle position; and
	Coolant temperature.
The Motor Vehicle Manufacturers Association (MVMA) and the
Association of Imported Automobiles (AIA) have proposed to provide at least 20 sets of
instruments to record the above data. Initially, it is proposed that a pilot study be
performed whereby at least three vehicles would be instrumented to measure the above
parameters. Based on the analysis of the pilot program data, recommendations will be
made on parameters, frequency of data collection, and equipment needed for an in-depth
study of driving cycles.
Based on the specific goals of the driving pattern measurement study, the
list of parameters which need to be monitored may be reduced. A minimum list would
contain clock time, vehicle speed, and manifold absolute pressure, as discussed in the
previous section.
Instrumentation of loaner vehicles is a related approach that could be used.
The use of loaner vehicles introduces a bias in the data caused by people who do not
want to give up their vehicle or who would drive a leaner differently from their own

The primary advantage of instrumenting vehicles and recording continuous
data is that a complete history of the operation for the particular vehicle is obtained.
Unlike the situation with external observations or chase cars, the data recorded by
instruments on the target vehicJe precisely measures the parameters for that vehicle. As
a result, this method accurately monitors cold-, warm-, and hot-start operation, as well as
trip beginnings and ends. In addition, this method provides accurate data on engine
load. With other data collection methods, engine load can be inferred from acceleration
and information on the vehicle, but it cannot be accurately measured.
Other advantages of this approach are listed below:
	Provides high-frequency information on parameters. Information
can be provided on a second-by-second basis or even more frequent
	This method is able to detect small variations in speed and load
which may have significant impact on emissions. For example, the
classic 'foot-pumper" that is constantly changing throttle while
maintaining a relatively constant speed could be detected by this
	Has capabilities of maintaining average (summarized) as well as
instantaneous data.
	Data are recorded in machine-readable form, and therefore no key
punching or data entry is needed.
	Provides an accurate measure of the distance travelled by the
Data analysis technique can be changed without the need to collect
additional data. When several key engine parameters are
monitored on a continuous basis, versus being logged into bins, the
data analysis technique can be revised as more knowledge is gained
about driving cycles and emissions.

Another possible advantage of this method is that air/fuel ratio also can be
continuously recorded. This may simplify data analysis by allowing the analysis to
concentrate on open-loop rather than closed-loop operation. Based on discussions,
manufacturers apparently attempt to maintain closed-loop operation through most of the
FTP. During closed-loop operation, emissions are minimized. However, open-loop or
closed-loop operation may not be an appropriate measure of driving pattern, since the
mode of operation depends on vehicle technology.
The primary disadvantage of instrumenting vehicles is the
representativeness of the data collected. As vehicle procurement programs have
discovered, a majority of vehicle owners do not positively respond to procurement
requests. Rejection rates often exceed 70 percent. This raises concerns over the
representativeness of the vehicle owners that do respond. Vehicles in EPA's emission
factor program typically show much less tampering than the overall vehicle population,
which implies that these people tend to be "do-gooders." Accordingly, the same people
may be easier on the vehicles than the average vehicle owner or the more aggressive
vehicle owner. Note that the issue is fleet emissions, which often are determined by a
small percentage of the overall vehicle population. Therefore, if motorists with
aggressive driving behavior are not included in the study, the results may fail to indicate
a significant cause of excess emissions due to off-FTP driving conditions. To reduce the
impact of participation refusals, considerable effort would need to be made to sample
representatively and replace refusals with the same type of driver but who is willing to
Another related issue concerns the representativeness of the vehicles that
are procured for the program. The many-parameter type of monitoring discussed above
would best be performed on late-model vehicles, where most of the parameters can be
pulled off the on-board computer. Owners of late-model vehicles may have significantly

different driving behaviors than owners of older vehicles. It is likely they may be easier
on their vehicles and, accordingly, may encounter less off-FTP operation. To avoid
measurements on only late-model vehicles, we favor the measurement of just a few
parameters (speed, manifold vacuum) which can be monitored easily on any technology
vehicle. Another advantage of the few-parameter approach is that the data collection
equipment can be installed quickly. The relationship between emissions and driving can
more appropriately and be better measured in a study separate from this driving pattern
The above concerns with instrumented vehicles most likely apply more to
driving on moderate or uncongested roads rather than to congested roads. On congested
roads and in parking lots, driving patterns more likely are influenced by the driving
behavior of nearby motorists and less by individual preference.
Other disadvantages of obtaining continuous data from instrumented
vehicles are listed below:
	The estimated cost to procure and instrument vehicles is high;
	Large quantities of data are generated by this method, which makes
data analysis complex;
	This method requires a high-capacity storage medium and/or
frequent off-loading of data; and
	Data recovery requires personnel involved in the study.
Note that there are alternatives to the traditional use of an onboard data acquisition
device with hard disk drive. Datalinks by radio or mobile telephone to a central
computer are also possibilities that may be attractive.

Method #2--Instrument Vehicles to Obtain Summarized Data
This method is similar to Method #1, except that it involves recording
summarized data rather than continuous data. In this case, key observables such as
acceleration rates, speed, load, coolant temperature, etc., are logged into pre-defined
bins. For example, a data logger may create histograms of different acceleration rates.
The histograms are continuously updated as the vehicle is being driven, but the memory
requirements of the data recording system usually do not grow.
Data loggers are currently available to continuously update histograms
describing different parameters. Some of these data loggers can maintain both
continuous records of data as well as histograms. Data loggers can be programmed to
maintain histograms describing the overall driving and continuous records of certain
episodes, such as times the vehicle is operating open-loop. The cost for a data logger
that only maintains histograms is much less than one that maintains both continuous and
histogram data.
Method #2 has many of the same advantages as Method # 1. Most
notably, Method #2 accurately represents the vehicle being studied; is able to monitor
all vehicle operation including cold starts, hot starts, and trip end; provides good
information on engine load if manifold pressure and throttle position are recorded;
provides good information on percent of time in off-FTP operation if air/fuel is being
monitored; and records data in machine-readable form (i.e., no key punching is
Method #2 has some inherent advantages over Method #1. The most
significant advantage is that data analysis is greatly simplified, because the data are
essentially processed on board the vehicle. In addition, data does not need to be

retrieved as frequently as Method #1. In fact, it is conceivable that data recovery can be
done by the vehicle owner. In addition, as mentioned earlier, histogram data loggers
generally are less expensive than continuous data loggers.
Method #2 shares many of the same disadvantages with Method #1. In
particular, concerns are raised over the representativeness of the owners and/or the
vehicles being procured for testing. Some of these concerns are partially alleviated if the
equipment can be installed quickly and is not too obtrusive. However, it is still likely
that certain segments of the population will refuse to have their vehicles instrumented.
Appropriate selection of study participants can be used to minimize the bias produced.
In addition to the problems shared with Method # 1 over vehicle
procurement, Method #2 has some additional disadvantages. The most significant
problem with recording summarized data versus continuous data is that there is no
chance to reanalyze the data. This means that correct observables must be chosen
before the study. Therefore, the key parameters affecting emissions must be identified
before the instruments can be installed on the vehicles.
An example will illustrate potential problems with this approach. Suppose
that engine load is recorded onto histograms and the analysis of the data indicates that
full-throttle engine operation is important. Suppose, however, that the duration of full-
throttle accelerations are not recorded. Thus it would be difficult to describe how long
full-throttle operation must be monitored during the revised emission test procedure.
Requiring that full-throttle acceleration be monitored for an excessively long period of
time creates problems for the vehicle manufacturers in control system design, because
methods must be developed to prevent catalysts from overheating and/or excessive
engine wear during full-throttle operation. However, if it is determined that these
episodes generally are very short, EPA can devise a test procedure involving full-throttle

operation that does not greatly limit design options or increase emission control system
cost for the automobile manufacturers.
Identifying cases of the classic "foot-pumper" would be another example of
the difficulties associated with the using histogram data loggers. Unless the data logger
is programmed in advance to recognize patterns of frequent on and off throttle
operation, these cases will not be identified with this approach. The data logger might
pick-up a foot-pumper from the vehicle speed data. However, usually bin widths would
be set too wide to detect such small changes in speed. This may be a significant problem
if these types of driving are important contributors to off-FTP emissions.
4.3	Method #3--Perform Chase Car Surveys
As opposed to instrumenting vehicles, driving characteristics can be
determined by using chase cars to follow motorists over pre-determined paths. The
chase car is instrumented so that it can keep continuous traces of speed versus time of
the vehicle being followed. Transportation models typically are used to estimate the
traffic intensity between zones in the city. A variety of chase car routes are chosen so
that most driving conditions are encountered, The chase car tends to follow only one
vehicle on the route, changing to other vehicles only if that particular vehicle exits the
Sierra Research is in the process of developing significant enhancements to
the chase car technique to help overcome some of its shortcomings. In particular, they
are funding the development of an infrared laser device that will accurately measure the
differential between the speeds of the chase car and the vehicle being followed. This
eliminates the need to exactly mimic the speed of the target vehicle. Other
enhancements include using multi-channel recorders to allow the occupants of the chase
vehicle to record parameters describing the type of vehicle being chased, road conditions,
and other factors that will be useful in analyzing the data.

The primary advantage of using chase cars is that motorists are
participating in a study without being aware of it. Consequently, the driving of all types
of drivers, not just a subset of the drivers that might be willing to participate in a study,
can be characterized. In addition, chase cars are able to follow all types of vehicles, so
any interaction between vehicle type and driving pattern can be identified.
Chase cars also are able to track the driving characteristics of a larger
number of vehicles, although day-to-day variations in individual vehicle characteristics
are not possible to determine.
Other significant advantages of using chase cars are listed below:
	Provides instantaneous speeds on a second-by-second basis;
	Is able to monitor most accelerations on the target vehicle;
	Time to begin data collection is short once the chase car has been
	Data recovery is simpler than in instrumented vehicles, because it is
concentrated on one vehicle; and
	Like instrumented vehicles with continuous recording, the data can
be analyzed in several ways, and the data analysis technique can be
changed as new information is generated about important
The infrared speed tracking system being developed by Sierra Research
will eliminate one of the major disadvantages of chase cars, that is, the inability to

accurately determine the speeds and accelerations of the target vehicle. However, other
disadvantages remain.
The most significant disadvantage of chase cars is that they are unable to
monitor manifold pressure, and throttle position. Consequently, engine load and off-FTP
operation cannot be directly determined. Engine load can be inferred based on the type
of vehicle and the observed acceleration, but there is uncertainty on the actual load. For
example, a manual transmission car, operating under full load in a high gear, could
experience the same acceleration as the same vehicle under part load in a lower gear.
Certainly the emission characteristics would be different under these two conditions.
Another significant disadvantage of chase cars is that they are probably are
unable to detect cycling of vehicle speed due to driving characteristics such a pumping of
the accelerator. As previously mentioned, these types of driving behavior may affect
Other disadvantages of chase cars are listed below:
	The chase car is unable to determine soak times, being able to
follow the vehicle only when it is moving.
 The chase car does not provide an accurate indication of vehicle
operation in the cold-start or warm-start mode.
	The chase car generates large quantities of data that may be
difficult to process, particularly if the chase vehicle has to switch
target vehicles several times during a trip.
	Selection of the chase car routes may result in certain routes being
neglected that may have a disproportionate impact on emissions.
For example, a road system that has a lot of stop-and-go driving due
to stop signs and/or traffic could generate more emissions than an
equal length system more conducive to steady-state operation. If
the former road system is not included in the survey, then this type
of driving pattern may be under-represented.

There may be an unintentional bias in selecting the target vehicles,
which may influence study results. This concern can be partially
alleviated by well-defined methods of selecting the vehicles and/or
data analysis techniques to normalize the data based upon expected
vehicle population.
4.4	Method #4--Use Vehicle Diaries
Vehicle diaries have been used in the past to generate statistics about
average trip length, miles travelled per day, number of trips performed per day, when
trips occur, and lengths of soak periods. Currently, several surveys of this type are in
progress to collect more data on trip characteristics.
In addition to providing the statistics discussed above (average trip length,
miles travelled per day, number of trips per day, soak times, etc.), diaries can provide
information on average speeds. Other advantages of diaries are listed below:
Does not require installation of data logging equipment.
Permits different data analysis techniques.
Does not require high capacity storage medium.
Data recovery does not require personnel at the vehicle.
The primary disadvantage of diaries is that they do not provide specific
data on maximum speeds, accelerations, load, and other parameters important in the
characterization of driving patterns. Other disadvantages are listed below:

 This method generates large quantities of data that must be
keypunched or manually entered into a computer system.
	Procurement of participants may introduce bias into the sample.
	Drivers may neglect to accurately record data on all trips.
4.5	Method #5--Make External Observations With Stationary Observers
Stationary observers can record data on vehicle driving patterns. Sierra
Research is currently using this method to characterize trip ends and trip beginnings.
Observers are being positioned in residential neighborhoods to record data on driving
characteristics immediately following engine start-up. Observers also are being
positioned in parking lots and other likely trip end locations to characterize vehicle
driving at the end of the trip. Data collected in this manner can complement chase car
and other data, and help characterize the complete spectrum of driving.
External observations also can be used to characterize driving behavior
from points along a trip length. For example, observers with radar guns or with the
infrared laser ranging device discussed previously under chase cars, could characterize
speeds and accelerations as vehicles enter on-ramps.
Other methods of performing external observations include using
helicopters or airplanes to make aerial observations. However, these methods probably
are not feasible to characterize speeds and accelerations, although they could help
characterize vehicle travel in the transportation system.

The primary advantage of external observations is that they provide data
on several different types of vehicles and drivers. As with the case of chase car surveys,
this method is not constrained by possible procurement problems. All vehicles that are
visible from the street can be surveyed.
The primary disadvantage of external observations is that they are limited
to characterizing vehicle operation within the line of sight of the observer. Another
significant disadvantage is the inability to monitor engine loads, particularly during key
modes, such as cold starts.
Other disadvantages of external observations are listed below:
	This method requires information on trip origins and trip ends, so
that observers can be placed in representative sites;
	Positioning observers in residential neighborhoods triggers suspicion
by residents and can possibly lead to police calls; and
	This method cannot be used to characterize vehicle travel
throughout the whole trip.
4.6	Comparison of Different Approaches and Recommendations for the
Approach to be Used in the Driving Pattern Study
Table 4-1 summarizes positive and negative features of each of the five
methods to collect data for purposes of evaluating and revising the FTP. The features
have been broken down into three general areas:

Table 4-1
Evaluation of Measurement Techniques
Instrumented DataLogger on Driving Chase Stationary
Private Car Private Car Diary	Car Observer
Driver does not know he is being monitored
Representative of vehicle population
Representative of drivers
Nonintrusive installation of data logging equipment
Representative of all types of trip segments
Follows each car all the time - moving or not
Contains high frequency information
Measures average speeds
Measures instantaneous speeds
Measures accelerations
Measures number of cold starts
Measures during cold starts
Measures number of warm starts
Measures during warm starts
Measures distance travelled
Contains load information
Permits various data analysis techniques
Correct observables not needed at beginning of study
Generates small quantity of data
Can measure drive train operation
Low procurement cost
Low installation cost per car
High capacity storage medium not required
Data recovery does not require personnel at vehicle
Does not require data keypunching
P = Measureable if pre-processor is programmed appropriately

	Sources of bias
	The technical features of the method? (For example, frequency of
data collection, parameters monitored, and quantity of data.)
	Cost of the technique (For example, installation cost, vehicle
procurement cost, data recovery costs).
A review of Table 4-1 indicates that no single technique is without
significant drawbacks. Instrumenting private vehicles has the advantage of being able to
collect detailed data on different parameters important in driving pattern
characterization, but there are significant concerns over the bias of the sample. Chase
car surveys can collect much driving information without the knowledge of the target
vehicle but still must deal with appropriate sampling techniques and are unable to record
some of the significant parameters in driving pattern characterization, such as engine
load. Chase cars are also not able to record data on vehicle operation during the cold
start modes, hot start modes, and trip ends. Diary surveys provide good information on
trip characteristics, but no information on acceleration, load, high frequency variables,
and other key parameters important to vehicle emissions. External observations of trip
beginnings and trip ends may provide information on an unbiased sample, but the type of
data that can be collected is very limited.
No single approach appears capable of addressing all the needs of a driving
pattern survey. The complete picture of driving behavior can be seen using data
obtained from a combination study using three techniques. Radian recommends that:
	Private cars should be instrumented to collect continuous data on
vehicle speed and engine load parameters;
	Chase cars should be used along representatively selected segments
to characterize speed versus time history, including accelerations, of
randomly selected vehicles; and

 Existing diary data should be analyzed to better characterize soak
times, trip lengths, and possibly average speeds. (A
recommendation is not being made to collect new diary data as part
of this driving pattern study.)

For the implementation of the instrumented private car and chase car
studies, it is important to consider the elimination of bias and criteria for selection of
survey site.
There exists considerable potential for the collection of biased data in both
the instrumented car survey and the chase car survey. Considerable effort needs to be
spent trying to minimize the bias. The reason for this is at the conclusion of each of the
studies, the results from the studies will be compared and combined to obtain the overall
picture of driving patterns. If a considerable degree of bias is present in either or both
of the studies, then the comparison will show that they do not agree. At that point it will
be necessary to resolve the differences either by throwing out data or by altering data so
that the characteristics of the driving patterns obtained from the two studies are
In the discussions below, the instrumented private car and chase car studies
are considered for sources of bias and techniques which may be used to try to eliminate
For the instrumented private car study, bias may be introduced by two
types of driver behavior: 1) some drivers will refuse to participate in the study; and 2)
knowledge that their car is instrumented will affect their driving.
Elimination of Bias
Instrumented Private Vehicles

People who are willing to let their private cars be instrumented may be a
subset of the driving public that has a significantly different driving behavior pattern
than the entire driving population. For example, it could be expected that young drivers
driving sporty vehicles would have a higher refusal rate than older drivers driving sedans.
Demonstration that the owners of instrumented vehicles have the same demographic
description as the population as a whole is not sufficient for guaranteeing that the driving
behavior of the drivers of instrumented vehicles is the same as the driving behavior of
the entire population.
Several statistical and operational techniques can be used to try to
eliminate this bias. Either random sampling or stratified random sampling techniques
can be used.
Stratified random sampling involves breaking the drivers into demographic
descriptions which are perceived to be related to their driving behavior. Such
demographic descriptors as driver's age, sex, marital status, and vehicle descriptors as
power/weight ratio, sedan versus sport car, and type of transmission can be used to
describe the different strata which need to be sampled. The strata would then be filled
with the number of participants which is proportional to the entire population.
A random sampling technique may be most easily implemented by
sampling on a centralized inspection and maintenance lane. With this technique, the
driver of every hundredth car having his car inspected would be asked to participate in
the test program. The I/M lanes to be sampled would be randomly chosen across the
city so that different socio-economic strata could be covered. A strategy would be
needed to deal with bias caused by drivers who refuse to participate. Demographic data
of those who refuse may be used to identify other drivers from the same stratum. It may
also be possible to use the description of the vehicle as a surrogate descriptor of the
driver. For example, if a young driver who is driving a red sports car refuses to

participate in the program, then another young driver who is driving a red sports car
could be requested to participate even though he is not the hundredth person getting his
car inspected.
It is possible that the knowledge that a vehicle is instrumented will affect
the way the driver drives. One way to reduce the potential for this type of bias is to tell
the participant that the study involves something other than driving patterns or to not tell
the driver the complete story of the study. For example, the driver may be told that the
study involves the characterization of the vehicle emissions. It may be advantageous to
provide a dummy sensor which is placed in the tailpipe to confirm the driver's belief
about what is being measured by the instrument on the vehicle. There is a possibility of
legal complications in deliberately misleading the driver of the vehicle.
It may also be advantageous to use a follow-up interview after the data has
been collected from each vehicle to determine if the driver has changed his driving
habits during the study, if he went out of town on a long trip, or if he had mechanical
problems with his car.
5.1.2	Chase Car Surveys
Biases can be encountered in the chase car driving pattern measurement
technique. The vehicles which are difficult to follow will tend to be under-represented in
the data that is obtained using a chase car. Thus, a vehicle which drives fast and
changes lanes quickly, weaving in and out of traffic, may be under-represented simply
because the chase car has difficulty following the vehicle.
An urban transportation planning system (UTPS) model will be used to
determine where private vehicles should be chased. Apparently, these models describe
primarily interzonal trips. Specifically, they count the trips between city zones.

However, the precise path which vehicles take to move from one zone to the other is not
part of the UTPS. In addition, trips within a zone, so-called intrazonal trips, are not
described at all by the UTPS models. Accordingly, random sampling techniques will
need to be developed to choose the interzonal trip paths and intrazonal trip paths which
will be monitored.
In the chase car technique, as it is proposed, target vehicles will be
monitored until they turn off the trip path. At that point, the next target vehicle needs
to randomly be chosen.
5.2	Definition of Most Appropriate Survey Site
Desirable characteristics of the survey sites should be considered when
choosing the city which will be used for the measurement of driving patterns. These are
discussed below.
It is appropriate to use an emissions non-attainment area for the study.
However, the question arises whether ozone non-attainment areas, CO non-attainment
areas, or areas in which both are not attained should be considered. We do not see any
reason why either type of non-attainment area which is surveyed would affect the driving
pattern behavior of the vehicle.
From a practical point of view, the survey site which has centralized I/M
would make the procurement of instrumented vehicles easier. For the chase car study, a
survey site with a good urban transportation planning system model is required.
The terrain and altitude of the survey site will have an affect on the way
the vehicles are driven. Since most vehicles are at low altitude, it would seem
appropriate that low altitude survey sites are most reasonable. Since the presence of

many steep hills can affect vehicle driving patterns, and most large cities are relatively
flat, it is appropriate to chose a city which is relatively flat.
The presence of precipitation on the road most likely affects driving
patterns, so cities which are expected to have a large amount of precipitation during the
period of the study, should be avoided.
The characteristics of the traffic and road systems of the survey site will
greatly affect the results of this study. Driving characteristics of a city which has
congested traffic will be significantly different from a city which has relatively open
traffic. Trips in large, spread out cities will tend to be longer than those in small,
compact cities. For example, downtown Boston would have different driving patterns
than Los Angeles.
A list of the non-attainment areas within the continental United States that
have centralized I/M programs is shown in Table 5-1. The table contains information
about the I/M program, the altitude, and the precipitation in these 25 areas. This
information along with other information specific to the collection of data by the chase
car technique can be considered to select candidate cities for the measurement of vehicle
driving patterns.
All 25 areas use centralized I/M programs; however, only 20 use
centralized I/M exclusively. For example, in New Jersey vehicles may be inspected at
the centralized state station, or for a higher fee they may be inspected at decentralized
stations. While most vehicles go to the state stations, it is possible that the vehicles that
are inspected at the decentralized stations are driven differently than those that are
inspected at the centralized stations. Consequently, the five areas that have a
combination of centralized and decentralized I/M (New Jersey, Washington DC,
Jacksonville FL, Miami FL, Tampa FL) are less attractive than those that have only

TABLE. 5-1
Attributes of Non-Attainment Areas with Centralized l/M Programs

Model Years ol
l/M Program Type
Average Per Cent
Average Monthly
Average Number

Light Duty Vehides

ot Days
ot Days

Tested for Emissions

With > 0.01 Inch

With > 11nch Snow


{C - Centralized)

(D - Decentralized)



Connecticut (e.g. Hartford)
Frequent snow
Delaware (e.g. Wilmington)
New Jersey
C+ D
Partly decentralized l/M

BaMmore MO
00 122

Washington suburbs MO
001 21
Washington DC
C+ 0
00 1 21
Partly decentralized I'M
Jacksonville FL
Partly decentralized l/M
Miami FL
Parity decentralized l/M
Tampa FL
Partly decentralized l/M
Memphis TN

Nashville TN
00 1 1 1
Ctevoland OH
Frequent predp, frequent snow
Louisville KY

Louisville suburbs IN
001 21
Chicago suburbs IN
Frequent snow
Chicago IL
Frequent snow
EStLouts IL
0 1122
Milwaukee Wl
Frequent snow
Minneapolis MN
Frequent snow
Phoenix AZ
0000 0

Tucson AZ
Higher altitude
71 +
Frequent predp
Portend OR
71 +
Frequent predp
Seattle WA
63 >
Frequent predp
Spokane WA
Higher altitude, frequent predp.
frequent snow

centralized inspection. More detailed information about the specific I/M programs used
in these five areas might allow some of these areas to remain under consideration.
All of the sites except Tucson AZ and Spokane WA have altitudes below
1300 feet. The higher altitudes of these two cities may cause different driving behavior
because vehicles will not respond to driver demands as do the majority of vehicles which
are operated at lower altitudes. Also, the terrain of the 25 areas needs to be considered
in the final selection of survey sites.
Precipitation can produce differences in measured driving patterns by
allowing wheel spin and by inducing drivers to drive more carefully than they would on
dry pavement. On the other hand, since wet or snowy roads are part of the environment
in which most vehicles are occasionally driven, it is reasonable to measure driving
behavior under wet conditions during part of the study. However, because the data in
this study will likely be collected in autumn and winter, but the results must be relevant
to year-round driving, an excessive number of wet days should be avoided.
Data collection is expected to occur sometime between October 1991 and
February 1992. The average precipitation for the months of October through February
at each of the 25 areas is given in Table 5-1 in terms of the average percentage of days
that have measureable precipitation, the average monthly precipitation in equivalent
inches of water, and the monthly average number of days that snowfall is greater than
one inch.
Consideration of the percentage of days with precipitation shows that the
18 areas that remain on the list can be broken into three groups: 12-22%, 30-36%, and
43-63%. Those areas with 43-63% of the days which are wet (Cleveland OH, Medford
OR, Portland OR, Spokane WA, Seattle WA) should be thrown out.

The amount of precipitation is also shown in the table. Large amounts of
rainfall in comparison with small amounts are not expected to be a problem because
roads will dTain and dry; however, days with snow can be a problem, especially in cold
climates, because roads may remain slippery for days after the snowfall. Accordingly, the
table also shows the average number of days with snowfall by month. In general, days
with snowfall are low in October and November, but are high in December, January, and
February. Thus any of the remaining 14 sites could be used if measurements are
completed by the end of November; however, if the measurements cannot be completed
before December, then the sites with greater than 9 snow days (Connecticut, Chicago
suburbs IN, Chicago IL, Milwaukee WI, Spokane WA, Minneapolis MN) are less
Nine areas remain that are characterized by strictly centralized I/M
programs, lower altitudes, less than 36% of the days with precipitation, and less than 7
days snowfall for the five month test period:
Delaware (e.g. Wilmington)
Baltimore MD
Washington suburbs MD
Memphis TN
Nashville TN
Louisville KY
Louisville suburbs IN
E.St.Louis IL
Phoenix AZ
Any of these areas appear to be acceptable for an instrumented private car
study. Considerations will be imposed by the chase car study requirements and
especially by the need for the existence of an Urban Transportation Planning System.

An additional, although perhaps minor, consideration to be made for either
type of study might be the presence of a nearby large city which could attract vehicles
from the area of interest and thus increase the amount of interstate driving that would
otherwise not be present. Areas with nearby large cities include the suburbs:
Washington suburbs MD, Louisville suburbs IN, and E.St.Louis IL. Wilmington DE and
Baltimore MD might be expected to be large enough to have driving patterns like cities
rather than suburbs even though other large cities are nearby. This would leave in the
Wilmington DE
Baltimore MD
Memphis TN
Nashville TN
Louisville KY
Phoenix AZ
Finally, to assist in this study, a city with a local air quality staff which is
available and willing to assist is desirable. Several areas of assistance could be used to
improve the efficiency of the survey program, and as a result reduce the cost of the
measurements. Because they are familiar with the area in which the study will be made,
local air quality staff can help coordinate activities with the local I/M authorities, help
procure private vehicles for instrumentation, tabulate meteorological data during the
measurement program, assist with data handling at the I/M lanes, and help get
information from the Urban Tansportation Planning System.
The cost of an instrumented private vehicle measurement program depends
on many factors: the driving parameters which are to be recorded, the cost of
instrumentation, the number of vehicles to be surveyed, the number of cities to be
surveyed, the participation rate of drivers who are solicited, the time required to

instrument each car, the amount of time each car will be monitored, losses incurred from
instrument breakdown, claims by drivers that their cars were damaged by the instrument,
and the extent to which local air quality staff can assist the measurement program. We
have estimated to instrument from 50 to 250 private vehicles at each of two cities
including data analysis and reporting could cost between $315,000 and $600,000
depending on the makeup of the survey. In most cases, the largest expenses in the
different scenarios are the labor and travel costs associated with staff installing and
retrieving vehicle data logging equipment.

The analysis of the data obtained in the two driving pattern studies will be
used to evaluate the current FTP and potentially be used to develop a new certification
test procedure. The analysis of the driving pattern data can be made with a variety of
techniques. The two discussed here are summary statistics and signal processing
6.1	Summary Statistics
Summary statistics can be calculated for the data obtained in each of the
surveys. This would include the distribution of specific statistics as well as average
values. The distribution is important because extreme values of a given statistic may
describe the high emission behavior which the FTP currently does not now describe.
Distributions of statistics for the survey data can be compared with the statistics for the
FTP to get an idea of where the FTP fails to represent the driving of the vehicle
The statistics for the same driving mode between the chase car and the
instrumented private car studies should be similar. If they are not similar, this is an
indication that either of the studies may have bias, in which case, appropriate resolution
of the data would need to be made. This would mean that either the data which is
considered unusual should be discarded or data should be adjusted so that the results of
the two studies will be consistent.
62	Signal Processing Techniques
All vehicles are driven differently, and yet many similarities exist in the way
different vehicles are driven. For example, a vehicle with a large power to weight ratio

may be driven differently from one with a small power to weight ratio; a young driver
may drive differently than a retired person; and vehicles may be driven differently in
congested traffic and in open traffic. On the other hand, all vehicles stop at stoplights,
accelerate up to a speed limit, and then stop at the next stoplight; they are usually not
driven in the early morning hours and therefore cool down overnight; and during cruises
a small amount of speed variation is present.
A more descriptive and more detailed analysis of the driving pattern data
can be used to describe these differences and similarities present in the driving pattern
data by using signal processing techniques. These techniques use advanced, but standard
mathematical tools applied with the guidance of engineering experience to find and
describe the speed versus time profiles of vehicle driving patterns in terms of driving
pattern descriptors. Thus, the driving patterns of the entire study would be characterized
by several descriptors and a distribution of values for each descriptor. The goal of the
signal processing analysis would be to seek a set of descriptors which along with
appropriate values of each descriptor could completely describe any trip. Thus, a value
chosen for each descriptor would be sufficient to define a specific driving cycle which is
consistent with the driving patterns measured in the field.
Because signal processing may not be familiar to the reader, it is described
briefly below. A small list of books which may introduce the interested to signal
processing is included in the references.
Signal processing is the study of sequences of numbers. The driving speed
history for an individual vehicle is an example of a signal. Here the numbers represent
speed, and the sequence represents equally spaced samples in time. A sequence often
also represents sampling in space, for example, in meteorology a pressure field is often
represented as a 4-dimensional signal in space and time. The signal processing field has
developed tools and techniques for manipulating signals, to better understand the

sampling process, noise in the measurements, to extract useful parameters from signals,
and to display results.
Signal processing has been applied to many technical areas, such as speech,
biomedical engineering, acoustics, radar, sonar, seismology, telecommunications,
vibration, and process control. Because of the deep technical interest in these
applications, signal processing is a mature field with a solid mathematical foundation,
theoretical and applied journals, textbooks, and commercial software and hardware
products. The signal processing aspects of vehicle driving pattern analysis are simple and
straightforward compared to some of these other applications listed above, and quite a
bit could be gained just from applying the basic tools of the field to the driving pattern
To demonstrate one method of applying signal processing analysis to the
driving pattern problem, some parallels can be drawn between speech processing and
driving pattern analysis. To use speech recognition, a voice waveform is captured as a
temporal sequence of sound pressure level samples, signal processing procedures convert
the waveform signal to phonemes, words, and eventually to the meaning of the utterance,
and then some action is based on this meaning.
Speech recognition is usually accomplished by dividing the low level signal
into distinct sections in time, representing the separate phonemes. Then each phoneme
is analyzed, and the sequence of phonemes are appended together to form words.
This analysis is very difficult, and the most successful recognition systems
use a hybrid approach. Detailed models of the speaking processing and the physics of
the vocal chords/mouth cavity are used to help in the dividing into separate phonemes.
More basic and general techniques, such as Fourier analysis, Markov processes, adaptive
filtering, and artificial neural networks are used to analyze each phoneme.

Just as a speech signal naturally consists of a sequence of phonemes, a
driving history consists of a sequence of trip segments, where each segment starts and
stops with the vehicle at rest. Different kinds of trip segments (between city blocks, a
stretch on an interstate, etc.) correspond to the different phonemes.
So a natural representation for a driving pattern might be as a sequence of
segments, with a few parameters (max speed, duration, speed fluctuations, accelerations,
etc.) required to describe each one. The signal processing analysis for driving patterns
should be much simpler than for speech, but a similar approach should work well.
The model of the speech system is a slowly varying linear system, with just
a few parameters required to represent the differences between speakers. In a similar
manner, a driving pattern model should be a simple system, requiring few parameters.
Speech synthesis (e.g. automatic reading from text) is an inversion of the
analysis task above. The analogous driving pattern task would be generation of typical
random driving patterns.
An example of the process used in speech recognition and synthesis is the
telephone communications application now being used by AT&T. For essentially all
long distance communications, the voice signal is sampled, compressed using digital
algorithms and special high-speed dedicated signal processing hardware, and then
transmitted. At the receiving end, the compressed digital signal is decoded (again using
special hardware) and the analog signal is generated and sent to the telephone receiver.
The compression achieved is a factor of 10 to 50, and yet the telephone users can easily
recognize the familiar voices of acquaintences.
A variation on this is secure communications, where the compressed digital
signals are scrambled before transmission and unscrambled at the receiving end.

Because of the short time between the end of the field testing and the due
date of the report, it would be helpful to develop signal processing data analysis
procedures using an existing set of data while the new driving pattern data is being
collected. This could be performed on the OCS Columbus, Ohio data which contains
vehicle speed and temperature data on many vehicle/driver combinations or on the
MVMA instrumented car pilot data which will have vehicle speed and several engine
operation parameters on a few vehicles. Then, when the data in the driving pattern field
studies becomes available in November 1991, the data analysis techniques could be
applied to it in a reasonably short period of time. It is expected that the driving
descriptors for the OCS data set and the new data set would be the same; the
distribution of values for each descriptor will be different.

The final development of the new certification test procedure or updating
of the current FTP will depend on the basis of the new procedure and the format which
the new procedure might have. (As discussed in Section 3.1, the certification test
procedure may be focused on driving patterns or emission patterns. The test procedure
could be developed in a straightforward manner as a direct consequence of the signal
processing data analysis technique.) In the discussion below, options for the revised test
procedure are discussed.
Several different certification test procedure formats can be envisioned for
an updated or new procedure. These might include procedures which are based either
on driving patterns or emission patterns. These formats include:
	Current FIT;
	Revised FTP (Bag 4);
	New deterministic cycle;
	Multiple deterministic cycles; and
	Multiple stochastic cycles.
After analysis of the driving pattern data, it may be found that the current
FTP is a good representation of today's driving and therefore does not need to be
revised. However, the current FTP is believed not to cover the full range of
accelerations which are present in the vehicle population. Since in today's technology
vehicles, heavy accelerations are known to produce emissions, the FTP may be
underestimating the emission potential of today's new vehicles. One approach to correct
this situation is to add a Bag 4 to the end of the current FTP. This fourth bag would
contain the higher accelerations and other emission features which need to be added to
the FTP to make it more representative of actual emission patterns. The driving
features which would be included in Bag 4 can not include all of the driving pattern

features which would make the FTP representative of driving patterns in the population.
If this were done, most of the driving features in Bag 4 would not contribute significantly
to emissions produced by the FTP cycle. Since the features which should be included in
Bag 4 are based on emissions, it is imperative that the relationship between emissions
and driving be known. This would have to be determined in a separate study or be
based on existing information.
If the signal processing approach is used to analyze the data, and if the
driving pattern descriptors are evaluated for their contributions to emissions, then the
current FTP can be evaluated to determine the types of driving features that need to be
added to make it more representative of emissions pattern of the vehicle population.
Another approach is to replace the current FTP with a new deterministic
cycle. A most likely deterministic cycle could be produced by using a Monte Carlo
simulation of the most likely values for each of the descriptors developed in the signal
processing analysis of the data. If this new cycle is based on emission patterns, then the
most likely cycle will emphasize the high emission driving features of the vehicle
population. However, if this cycle is based on driving patterns, then the most likely cycle
will not include very much high emission feature content since the test procedure would
be representative of population driving.
An alternative approach is to use multiple deterministic cycles. Such a
battery of cycles could be used to test the vehicles near the extremes of driving behavior
which were measured in the driving pattern study.
Rather than deterministic cycles, which are fixed, multiple stochastic
(randomly generated) cycles could be used to certify new vehicles. Such cycles could be
generated using a Monte Carlo simulation based on the distribution of values and the
descriptors developed from the signal processing analysis of the driving pattern data.

Earlier, this technical note made an anology between driving patterns and
speech patterns. Signal processing techniques have successfully been used to recognize
speech patterns and to perform the reverse, that is, to synthesize speech from printed
text. For the signal processing analysis of driving patterns, the reverse of determining
the structure of the underlying random and non-random features of driving patterns is
the synthesis of driving cycles which are consistent with the structure of the driving
pattern data collected in the field study.
A general example of the driving pattern analysis and the stochastic driving
cycle synthesis processes might help clarify this discussion. The goal of the signal
processing analysis would be to find an algorithm to break each driving pattern down
into a set of functions (descriptors) which is common to all driving and a set of values
(e.g. coefficients) for each function that describes the random differences present in
driving caused by the many random influences that cause vehicles to be driven
differently. At least several functions can be expected in the real set of functions that
describe real-world driving patterns. But for the purposes of the example, suppose there
are two, A and B. Also, suppose that each of these descriptor functions has only one
value, a and b, that is required for each function to make it specific so that a driving
pattern can be described. The result of the signal processing analysis might be that each
vehicle speed versus time driving pattern, v(t), could be described by:
v(t) = A(a) + B(b)
In this simple example, the algorithm is simply the sum of the two
functions, A and B, when appropriate values of a and b are used. Since the functions
are independent of vehicle and driver, it is the a and b which describe the differences in
driving behavior. For the driving measured in the driving pattern study, distributions of
a and b will be observed. A stochastic cycle could be generated by randomly selecting
values of a and b from the distributions and then applying the algorithm discovered in

values of a and b from the distributions and then applying the algorithm discovered in
the signal processing analysis, in this example, A(a) + B(b).
Randomly generated cycles would have a variety of driving features; the
severity of each feature would be randomly selected from the range of severities found in
the driving pattern study. For a new vehicle to pass certification, it would have to pass a
certain fraction of a certain number of randomly generated driving cycles. A computer
program which generates the cycles would be made available to anyone for testing
purposes. The fraction of stochastic cycles which must be passed and what determines a
pass would have to be decided using statistical and probablistic techniques.
The advantage of using the stochastic approach to develop a driving
pattern based test procedure is that the approach would work regardless of the emission
control technology, the fuels being used by the vehicles, or the emissions of concern. In
addition, the stochastic approach would be effective because the procedure would
encourage manufacturers to develop emission control systems which could pass most
randomly generated cycles. EPA's testing of vehicJes to be certified would be inefficient,
but the development of an emission control system would not require the testing of
numerous stochastic cycles. In the development of a vehicle emission control system the
manufacturer might choose to test the vehicle for only those features where emissions
were high. Such a certification test procedure would be good as long as the driving
pattern data used to develop the driving pattern descriptors and their distribution of
values are good.

Signal Processing Bibliography
J. S. Lim and A. V. Oppenheim, Advanced Topics in Signal Processing. Prentice-Hall,
Inc., Englewood Cliffs, New Jersey, 1988.
L. Ljung, System Identification: Theory for the User. Prentice-Hall, Englewood Cliffs,
New Jersey, 1987.
S. L. Marple, Jr., Digital Spectral Analysis. Prentice-Hall, Inc., Englewood Cliffs, New
Jersey, 1987.
J. M. Mendel, Lessons in Digital Estimation Theory. Prentice-Hall, Inc., Englewood
Cliffs, New Jersey, 1987.
A. V. Oppenheim, Applications of Digital Signal Processing. Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey, 1978.
A.	V. Oppenheim and R. W. Schafer, Digital Signal Processing. Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey, 1975.
B.	Widrow and S. D. Stearns, Adaptive Signal Processing. Prentice-Hall, Inc., Englewood
Cliffs, New Jersey, 1985.