Reprint: AIAA-97-0222

35th Aerospace Sciences Meeting & Exhibit, Reno, NV, January 6-10, 1997

PROBABILITY-BASED COLLISION ALERTING LOGIC FOR

CLOSELY-SPACED PARALLEL APPROACH

Brenda D. Carpenter^*and James K. Kuchar⁺

Department of Aeronautics and Astronautics

Massachusetts Institute of Technology

Cambridge, MA 02139

Abstract

transgression, and when the controller is able to clear

the frequency and issue breakout instructions. To

provide adequate safety, the NTZ must be relatively

large and is 2,000 ft wide.

A prototype airborne collision alerting logic was

developed for aircraft on approach to closely-spaced

parallel runways. A novel design methodology was used

based on collision probabilities instead of traditional

spatial or temporal alerting criteria. With this technique,

an alert is issued when the probability of a collision

exceeds an acceptable threshold value. The logic was

based on a hazard level corresponding to the current

Precision Runway Monitor System of one accident in

every one thousand worst-case blunders. Probability

contours were constructed through Monte Carlo

simulation over a range of aircraft position, speed,

heading, and turn rate conditions. These contours were

stored in look-up tables that were accessed in real time

for evaluation during numerical simulation of

approaches. Three runway spacings were investigated:

3,400 ft, 2,500 ft, and 1,700 ft. The results show that

the unnecessary alert rate at 1,700 ft runway spacing

was double that at 3,400 ft runway spacing.

Additionally, the logic induced collisionsin two low-

closure-rate situations, suggesting that the "worst case"

blunder may not be a drastic heading change.

2000'

3400'

Fig. 1 Parallel Approach with

No Transgression Zone (NTZ)

Because of the potential for large time delays, PRM has
been estimated to be unable to resolve approximately 1

in 250 "worst-case blunders" (defined as a sudden 30º

heading change intrusion by one aircraft).⁶When

runways are spaced closer than 3,400 ft, PRM is unable

to provide this safety level without producing an

excessive number of false alarms.⁵Also, at close

runway spacings, a spatial NTZ cannot provide enough

lead time to prevent a collision. Position derivatives
such as closure rate or acceleration are required to detect

a blunder in time to successfully resolve it.

An airborne alerting system would have the benefit of

eliminating much of the time delay by warning the

flight crews directly. However, current airborne alerting

systems such as the Traffic Alert and Collision

Avoidance System (TCAS) would produce an

unacceptable false alarm rate at runway spacings closer

than approximately 3,000 ft.⁷This is because TCAS

has been designed for enroute and terminal area

maneuvering operations and is not optimized for flight

in close proximity to other aircraft. Thus, a specialized

airborne alerting logic is required if independent parallel

approaches are to be conducted in IMC to runways less

than 3,400 ft apart.

To meet this need, and to serve as a testbed for

examining research issues, a prototype airborne alerting

logic was developed for closely-spaced parallel approach.

The logic uses datalinked aircraft state information

(position, velocity, heading, turn rate) to determine if an

alert is needed. The primary design goal was to maintain

Introduction

Independent parallel approaches in Instrument

Meteorological Conditions (IMC) can currently be

conducted to runways as close as 3,400 ft apart with the

use of the Precision Runway Monitoring (PRM)

System.^1-6PRM uses a high-resolution ground radar to

provide a depiction of the approach situation to an air

traffic controller. The controller monitors the aircraft on

approach relativeto a No Transgression Zone (NTZ)

between the runways (Figure 1). When an aircraft strays

into the NTZ (termed a blunder), the controller issues

break-out instructions to prevent a collision.

A major limitation of PRM is the fact that there can be

significant time delays between when an aircraft begins

to transgress, when a controller observes that

*_{Graduate Research Assistant, Member AIAA}

+_{Assistant Professor, Member AIAA}

Copyright © 1997 by MIT. Published by the American Institute of
Aeronautics and Astronautics, Inc. with permission.

current PRM safety levels while keeping an acceptable

nuisance alarm rate. This paper describes the

development of the prototype alerting logic and its

subsequent evaluation in simulation studies.

Development of Alerting Logic

To simplify its development, the prototype alerting

logic focuses on the final approach segment when both

aircraft are established on the approach path, and

assumes that the aircraft are coaltitude until an alert is

generated. Runway stagger and the turn onto the

localizer are ignored. Also, the large-scale traffic

management of aircraft in the terminal area isnot

considered.

In contrast to current alerting systems (e.g., PRM or

TCAS), the logic described here does not use a standard

spatial or temporal (time to impact) alerting criterion.

Instead, the logic bases the alerting decision on the

estimated probability of a collision. Thus, alerts are

issued at a consistent level of safety rather than, for

example, at a consistent time before impact.

Because the logic is based on the probability of

collision, it was necessary to develop a dynamic model

of aircraft onapproach that included consideration of

uncertainties in sensor measurements and in the

intentions of the aircraft. A series of Monte Carlo

simulations were then used to estimate the probability

of a collision as a function of aircraft position, speed,

heading, and turn rate. Alerting thresholds were designed

to correspond with a specified probability of collision

and were stored in a series of look-up tables. In

operation, the logic compared measurements of aircraft

state against the look-up tables and issued an alert if the

estimated probability of collision was above the

threshold.

Dynamic Models

An analysis of the dynamics of approaches and blunders

was conducted to determine the importance of having

access to information about aircraft states such as

relative position, heading, and turn rate. Estimates of

these parameters are assumed to be available onboard the

aircraft through datalink such as Automatic Dependent

Surveillance - Broadcast (ADS-B) or through a

measurement filtering technique. It was determined that

knowledge of the relative position, speed, heading, and

turn rate (or bank angle) of parallel traffic greatly

enhances the ability to determine whether a situation is

hazardous. Higher derivatives such as rate of change of

bank angle are generally too noisy to produce reliable

projections into the future.

Because all measurements contain noise, the alerting

system must decide whether a given measurement

indicates a potential blunder or whether it results from

normal oscillations on the approach path. The alerting

system must therefore balance the probability of a

missed or late detectionagainst the probability of an

unnecessary alert. To view this tradeoff directly, a

probabilistic approach was taken.
Figure 2 shows the potential future flight paths of a

blundering intruder currently in a left turn. Based on an

estimate of aircraft heading and bank angle, the aircraft

is projected to fly a certain predicted trajectory.

However, because of uncertainties in the state estimate
and because the intruder may modify its maneuver, the

actual path followed by the intruder may be different

(shown as a shaded region). The size and shape of the

shaded region are determined from a probabilistic

dynamic model.

Fig. 2 Potential Intruder Trajectories

In the dynamic model, the intruding aircraft is

nominally projected to fly a constant-rate turn at

constant altitude based on its measured position,

heading, speed, and turn rate. To account for

uncertainties in this projection, random errors are

introduced into the starting conditions during Monte

Carlo simulation. In the dynamic model, the position

estimate of the aircraft is modeled as a zero-mean

normally-distributed random variable with standard

deviation of 35 ft (corresponding to Differential GPS

accuracy). Aircraft heading and bank angle are modeled

with standard deviations of 2.5º and 5º respectively. The

magnitudes of these uncertainties were determined

through analysis of aircraft state data during simulated

approaches in varying turbulence conditions.⁸

When an alert is issued by the prototype system, the

threatened aircraft is assumed to fly a specified avoidance

maneuver consisting of a climbing turn away from the

2

American Institute of Aeronautics and Astronautics

parallel traffic. Following a two-second response delay

from the alert, a 0.25 g pull-up maneuver and a 45º turn

at a 30º bank angle is performed.

Monte Carlo Simulation

The models of the intruding aircraft and of the threatened

aircraft's avoidance maneuver were used in a series of

Monte Carlo simulations to determine the probability

of a collision over a range of conditions. By

numerically extrapolating the positions of the two

aircraft on approach, it can be determined whether a

collision (defined as separation less than 500 ft) will

occur in a given situation. The probabilitythat a

collision will occur is then estimated by counting the

number of collisions that occur over a large number of

Monte Carlo simulations.
Monte Carlo simulations were performed at each of a

number of initial conditions specified by aircraft

position, speed, heading, and bank angle. To each initial

condition were added the random errors described in the

previous section, and the aircraft's trajectories were

numerically simulated to determine if a collision

occurred.

Table 1 shows the set of initial intruder conditions used

in the Monte Carlo simulations. Initial intruder

position was varied within a grid with dimensions of

4,400 ft laterally and extending 9,200 ft behind to 9,200

ft ahead of the threatened aircraft. Grid points were

placed every 400 ft in this region. Intruder speed was set

from 120 - 180 kts, in increments of 20 kt. Aircraft

heading was set from 40º away to 40º toward the

threatened aircraft in increments of 10º. Finally, intruder

bank angle was set from 20º away to 40º toward the

threatened aircraft in increments of 10º. The total

number of combinations of states in Table 1 results in

142,128 different conditions. At each condition, the

random errors described previously were introduced and

the aircraft were simulated to determine if a collision

occurred. 10,000 Monte Carlo runs were performed at

each condition.

The result of these Monte Carlo simulations is a

multidimensional map that describes the probability of

a collision as a function of intruder position, speed,

heading, and bank angle. The result applies only to the

specific avoidance maneuver performed by the threatened

aircraft: a different map is generated for different actions

by the threatened aircraft. This map is best visualized by

using probability contours. Figure 3 shows a schematic

of the probability contours for a single combination of

intruder speed, heading, and bank angle, and assuming

that the threatened aircraft performs the climbing turn

avoidance maneuver. A different set of contours was

constructed for each combination of intruder speeds,

headings, and bank angles in Table 1.

In Figure 3, the threatened aircraft is shown in black at

the origin of the coordinate system. The probability of a

collision, as determined from the Monte Carlo

simulations, is marked in the figure using contours.

Example intruders are shown in white at positions A,

B, and C. As a reference, if an intruder is located on the

dark curved line, it is projected to have a direct collision

with the threatened aircraft if it does not change its

speed or turn rate.

Alerting Threshold Definition

The alerting threshold was designed to correspond to a

constant probability of collision. This probability was

set at p = 0.001, which is the same order of magnitude

as the PRM system.⁶Thus, in Figure 3 the extent of

the alerting threshold is defined by the p = 0.001

contour. Because the shape of this contour varies

depending on the intruder's state, the alerting threshold's

spatial extent also varies.

Table 1Intruder State Initial Conditions

Lateral Position

Longitudinal Position

0 ft - 4,400 ft

400 ft

400 ft

20 kt

10º

10º

9,200 ft behind - 9,200 ft ahead

Speed

Heading

Bank Angle

120 kt - 180 kt

40º away - 40º toward threatened aircraft

20º away - 40º toward threatened aircraft

3

American Institute of Aeronautics and Astronautics

Fig. 3 Example Probability of Collision Contours (Schematic)

In the example situation shown in Figure 3, an intruder

in position Adoes not trigger an alert because it has to

greatly decrease its speed or increase its turn rate to

cause a collision. An aircraft in position Bis also

outside the threshold and does not trigger an alert:

although the intruder is projected to collide with the

threatened aircraft, time still remains with which to

delay the alert before the probability of a collision

reaches the threshold value of p = 0.001. An aircraft at

Cgenerates an alert because the probability of a

collision has risen above the threshold. Similar plots

and alerting thresholds were developed at each

combination of speed, heading, and bank angle from

Table 1.

Calculation of the probability of collision in real time

is not feasible for the types of processors currently

available on aircraft. Accordingly, it was necessary to

store the shape and size of the alerting threshold rather

than calculate it explicitly during operation. To

facilitate storage and processing, the alerting threshold

shapes were simplified so that they could be accessed

rapidly in real time.

Two criteria were used to define the shape and size of

the alerting thresholds. First, the intruder must be

projected to pass within 800 ft of the threatened aircraft.

The value of 800 ft corresponds approximately to the

longitudinal width of the p = 0.001 contours relative to

the curved direct collision line. Second, the intruder

must be within a certain range of the threatened aircraft

(corresponding to the lateral extent of the p = 0.001

contour). This range parameter is also a function of

airspeed, heading, and bank angle. If both tests are

passed, an alert is issued.

The alerting threshold parameters were codified in a

series of look-up tables for real-time access. In use, the

logic takes a set of state measurements and compares

them to the look-up table parameters to determine if an

alert should be issued. Parameter values were

interpolated between tables when the intruder states

varied between the conditions in Table 1. The alerting

algorithm was implemented in C code and used in the

MIT part-task Advanced Cockpit Simulator and in part-

task simulations at the NASA Ames Research

Center.^9,10

Evaluation of the Logic

The performance of the prototype alerting logic was

evaluated using different approach trajectories developed

from flight simulation tests at Rockwell-Collins.⁸

These included normal approaches and six categories of

blunder trajectories: a slowconstant-rate turn at a 5º

bank angle; heading-changes of 10º, 15º, 30º; and two

cases in which the intruder began a blunder but returned

to its approach path before crossing the threatened

aircraft's approach path. Separate trajectory data were

available for calm and turbulent conditions and at

airspeeds of 130, 145, and 160 kt. The same trajectories

were used at three runways spacings (1,700 ft, 2,500 ft,

and 3,400 ft) and over a series of initial longitudinal

spacings (within ±1.5 nmi) to cover a range of possible

4

American Institute of Aeronautics and Astronautics

encounter situations. A total of 42,822 simulations

using 39 different types of trajectories were performed

for the evaluations.

Procedure

In the evaluations, the threatened aircraft followed a

normal approach path while the intruder followed one of

the blunder or normal approach paths described

previously. The alerting logic was implemented and if

an alert was issued, the threatened aircraft performed the

specified climbing-turn avoidance maneuver. The

outcome of each approach was recorded, including (1)

whether an alert was generated, (2) whether a collision

occurred, and (3) whether an alert was deemed necessary.

Six categories were used to define the possible

outcomes, listed in Table 2. A collision was defined to

occur if separation at any point in the approach was less
than 500 ft. An alert was considered to be necessary if a

collision would have occurred without an alert. Thus,

for example, an alert in a situation in which separation

would have been 501 ft without the alert was

categorized as unnecessary. Even though such an alert

could be considered to be warranted, a specific definition

of unnecessary alert is required as a performance metric.

From Table 2, if an alert was not issued at any time

during a run it was classified as either a Correct

Rejection (if a collision did not occur) or as a Missed

Detection (if a collision did occur). If an alert was

issued, the outcome was placed in one of four

categories. An Unnecessary Alert was a case where the

intruder was not on a direct collision course, an alert

was issued anyway, and a collision was still avoided. If

a collision occurred because of the alert, it was classified

as an Induced Collision. A Correct Detection occurred

when a collision was averted because of an alert.

Finally, a Late Alert was a case in which an alert was

issued but was too late to prevent a collision.

Results

The results of the evaluations are compiled in Table 3

as the observed rate of occurrence of each of the six

possible outcomes at each of the three runway spacings.

It must be stressed that the observed rates are dependent

on the mix of the specific types of blunder scenarios

used in the evaluations and are notindicative of the

expected rates during a typical approach. Thus, the

values in Table 3 are useful as indicators of performance

but are not absolute measures.

In those scenarios in which the intruder followed a

normal approach trajectory and did not blunder, no alerts

were issued. Alerts were only issued in scenarios in
which a blunder occurred. Therefore, it appears that the

logic was able to distinguish between nominal approach

oscillations and blundersdown to 1,700 ft runway

spacing. However, a more complete study of aircraft on

approach is required to verify this finding.

As shown in Table 3, there were no Missed Detections

or Late Alerts at any runway spacing. This indicates

that alerts, when necessary, were issued early enough

that collisions could be avoided. A more complete

safety assessment would require modeling the
probability that the system fails to operate as designed

(e.g., because of loss of datalink between aircraft).

Table 3Logic Performance

(Rates based on 14,274 simulations at each runway spacing)

1700'

2500'

3400'

9.1x10^-1

9.4x10^-1

9.5x10^-1

0

0

0

4.7x10^-2

3.0x10^-2

2.3x10^-2

4.2x10^-4

4.2x10^-4

0

3.8x10^-2

3.2x10^-2

3.2x10^-2

0

0

0

5

American Institute of Aeronautics and Astronautics

These effects could result in Missed Detections or Late

Alerts but were not modeled in this evaluation.

Some situations at 1,700 ft and 2,500 ft runway

spacings resulted in Induced Collisions. That is, there

were some cases during a blunder in which an alert

causedthe collision. Had the system not alerted, there

would not have been a collision. These Induced

Collisions occurred only in the 5º bank angle and 10º

heading change blunder trajectories, suggesting that the

alerting logic is deficient when the closure rate between

aircraft is relatively low. This deficiency is likely due to

the approximations used to develop the alerting

thresholds, and it is believed that the logic can be

modified to reduce or eliminate the Induced Collision

rate.

No collisions occurred in any of the other blunder

scenarios, including the traditional "worst-case" 30º turn

blunder. Thus, if the 30º blunder is used as the criterion
for safety (as it is for PRM), then the prototype logic

performs well down to 1,700 ft runway separation. The

fact that some collisions did occur in the more benign

blunder cases points to the important fact that the

"worst case" scenario for a given system may not be the

trajectory that appears at first to be the most

threatening. A gradual closure-rate scenario may in fact

be more dangerous with some algorithms.

Some of the alerts, as shown in Table 3, were

categorized as Unnecessary. These Unnecessary Alerts
were cases in which acollision did not occur, but a

collision would also not have resulted had the alert not

been issued (i.e., the intruder was not on a collision

course). Unnecessary Alerts occurred in increasing

proportion as the runways were placed closer together.

At 1,700 ft runway spacing, the observed Unnecessary

Alert rate is more than double that at 3,400 ft.
Recall that as defined here, an Unnecessary Alert is a

case in which more than 500 ft separation would have

existed had no alert been issued. Figure 4 shows the

distribution of miss distances had no alerts been issued

(for the Unnecessary Alert cases only). In the figure, the

majority of Unnecessary Alerts were such that the miss

distance would have been less than 1,000 ft. At 2,500 ft

runway spacing, for example, 97% of all Unnecessary

Alert situations would have resulted in less than 1,000

ft separation had the alert not been issued. Thus,

although Unnecessary Alerts were relatively common,

many of them could be considered to be warranted

because miss distances of less than 1,000 ft would have

occurred.

Note that there is a significant change in the miss

distance distribution between 1,700 ft and the larger two

runway spacings. While 97% of the cases at 2,500 ft

runway spacing would have resulted in less than 1,000

ft miss distances, at 1,700 ft runway spacing this rate

drops to 75%. This indicates that the alerting logic

begins to have difficulty in rejecting Unnecessary Alerts

as runway spacing decreases below 2,500 ft.

1.0

0.8

0.6

0.4

0.2

0.0

<600

<700

<800

<900<1000<1100<1200<1300<1400<1500<1600<1700

Miss distance had no alert been issued (ft)

Fig. 4 Miss Distance Distribution During

Unnecessary Alerts

Again, it must be stressed that all Unnecessary Alerts

occurred during blunder scenarios and therefore they all

could be considered to be warranted. No alerts occurred

during normal approach scenarios at any runway

spacing.

Additional Considerations

The alerting logic presented here is generally effective in

the cases that were studied because the climbing

component of the avoidance maneuver provides vertical

separation. If the intruder also climbs, however,

additional collisions could result because this vertical

separation would be reduced or eliminated. The

incorporation of vertical state information (relative

altitude and vertical rate) will be necessary in an

operational system. These additional states will be

required both to improve safety in cases in which the

intruder climbs and also to aid in rejecting alerts when

the intruder is clearly above or below the threatened

aircraft's flight path.

An additional issue regards the availability of alternative

avoidance maneuvers such as level turns or straight-

ahead climbs. The ability to select one of several

avoidance strategies could enhance the performance of

the alerting system. For example, in a case in which the

intruder climbs during a blunder, it may be more
effective to command the threatened aircraft to perform a

turn at constant altitude (or even to continue the

6

American Institute of Aeronautics and Astronautics

approach) rather than climb. One issue that must be

examined, however, is the potential impact that

multiple maneuvers could make on pilot response.

Pilots may be able to react more quickly to an alert

when their response is known beforehand. When the

pilot must perform one of several alternative actions in

response to an alert (either through choice or by

command), response time may increase, thereby

reducing the intended benefit of providing multiple

maneuver options. A related issue is that the

consistency with which pilots can fly a prescribed

avoidance maneuver will impact system performance. In

the evaluations conducted here, the avoidance maneuver

was followed perfectly. In actuality, this will not be the

case and some scatter in pilot response will occur.

One final issue involves the ability of the pilot to

understand the underlying alert logic. Although a

probability-based alerting logic may be more effective

than a simpler spatial (NTZ) method, it may be more
difficult for a pilot to understand why alerts occur. As a

result, pilots may distrust the system, reducing its

effectiveness.

type of blunder occurs. Alternatively, as assumed here,

an alert could be considered necessary only if it is

required to avoid a collision.

Several collisions were induced by the alerting logic.

These collisions occurred in low-closure-rate situations,

and it is believed that the collision rate can be reduced

through modifications to the alerting logic. However,

the fact that the traditional "worst case" 30º blunder was

resolved more effectively than more benign blunders

indicates that the "worst case" may not always be

obvious.

Additional enhancements to the logic are required,

including incorporation of altitude-related alerting

criteria. Because the current logic uses a climbing-turn

avoidance maneuver, more collisions are expected in

situations in which the intruder is also climbing.

Altitude criteria are also required to reject alerts in cases

in which the intruder is clearly above or below the

threatened aircraft's flight path.

It may also be necessary for the logic to select one of

several avoidance maneuvers (e.g., climbing turn, level

turn, or straight climb).Each of these avoidance

maneuvers can be examined in terms of probability

contours similar to those used for this study. The

contours could then be used to determine which

maneuvers have the highest probability of success.

There are several interesting design issues forsystems

in whichmultiple avoidance maneuvers are available.

For example, a conservative system design could alert

when any one of several avoidance maneuvers becomes

unsafe due to an intruder. An alternative design could
delay alerting until only one of the avoidance options

remains. The former system allows the pilot more

latitude in determining the best action but will have

more unnecessary alerts than the latter system, in which

the pilot must accurately perform the single prescribed

avoidance maneuver.

The probabilistic approach taken in the development and

evaluation of the alerting logic can be applied to
evaluate other types of alerting systems. For example, a

spatial No Transgression Zone alerting system can be
evaluated in terms of the probability of collisionor

unnecessary alert rate. Thus, when alerting methods are

proposed for closely-spaced parallel approach, the

methodology used here can be applied in evaluating

their performance.

A complete safety analysis of a proposed system for

closely-spaced parallel approach will require estimates of

the likelihood and composition of blunders. This

information is not currently available, so the overall

impact of an airborne alerting system on safety cannot

Conclusions

A prototype airborne alerting logic for closely-spaced

parallel approach has been designed and evaluated. A

novel design methodology was applied in which the

probability of a collision was directly used to set

alerting thresholds. This approach contrasts with

conventional design methods in which alerting

thresholds are based on spatial or temporal (time to

impact) criteria.

The logic was evaluated through numerical simulation

of a variety of blunder and normal approach situations.

This allowed a wide range of encounters to be examined

but these encounters may not be representative of actual

operations. Statistical data on the frequency and

geometry of blunders is not currently available, so it is

difficult to assess the safety or unnecessary alert rate of

a proposed system.

For the encounters used in the evaluations, no

Unnecessary Alerts were generated during normal

approaches. This suggests that the logic performs well

in rejecting alerts due to nominal tracking oscillations.

The only Unnecessary Alerts that were observed

occurred in blunder situations in which the minimum

separation would generally have been less than 1,000 ft.

The Unnecessary Alert rate at 1,700 ft runway

separation was twice the rate at 3,400 ft runway

separation. However, the definition of "unnecessary" is

arbitrary and mustbe considered when evaluating the

results. An alert could be considered necessary if any

7

American Institute of Aeronautics and Astronautics

be determined. Second, it will be important to evaluate

the ability of a system to reject alerts during normal

approaches. Given that GPS-guided approaches are

presumed in concepts for closely-spaced parallel

approach, it will be necessary to incorporate data

regarding aircraft tracking performance using GPS (e.g.,

lateral deviation, heading, and bank angle variability).

Third, the impact of alerts on the overall traffic flow

must be examined. At this point, the alerting logic only

provides protection for the immediate collision hazard

but does not aid the pilot in returning into the approach

sequence.

Finally, the issues raised during this study are important

not only for closely-spaced parallel approaches, but

apply to alerting system design in general. As advanced

alerting systems are proposed forFree Flight or terrain

avoidance, for example, issues such as the tradeoff

between single or multiple response options will arise.

Thus, there is a need to further develop generic models

of alerting that can be applied to different systems.

Closely Spaced Parallel Runways", MITRE

Document MTR-94W0000056, July, 1994.

8_{Chamberlain, S., and S. Koczo, "Close Parallel}

Runway Operations (CPROPS) Prototype Alerting

System Definition and Evaluation", Rockwell-

Collins Draft Report Paper, August 14, 1995.

9_{Pritchett, A., Carpenter, B., Asari, K., Kuchar, J. K.,}

and R. J. Hansman, "Issues in Airborne Systems

for Closely-Spaced Parallel Runway Operations",

Proceedings of the 14th AIAA/IEEE Digital

Avionics Systems Conference, Cambridge, MA,

November, 1995.

10_{Carpenter, B., and J. K. Kuchar, "Prototype Alerting}

System Logic for Closely-Spaced Parallel

Approach", MIT Aeronautical Systems Laboratory

Report ASL-95-6, December 28, 1995.

Acknowledgment

This research was supported by the NASA Langley

Research Center through the MIT Lincoln Laboratory.

References

1_{Federal Aviation Administration, "Precision Runway}

Monitor Demonstration Report", Document

DOT/FAA/RD-91/5, February, 1991.

2_{Ebrahimi, Y. S., "Parallel Runway Requirement}

Analysis Study", NASA CR-191549, Vols. 1 & 2,

December 1993.

3_{Gladstone, C. X., "Overview of the Simultaneous}

Instrument Approach Model (SIAM) Version 1.0",

MITRE Document No. MTR-94W0000074, June,

1994.

4_{Massimini, S. V., "The Blunder Resolution}

Performance Model", MITRE Working Paper WP-

91W00147, September, 1991.

5_{Owen, M. R., "The Memphis Precision Runway}

Monitor Program Instrument Landing System

Final Approach Study", Lincoln Laboratory Report

ATC-194, Document DOT/FAA/NR-92/11, May

23, 1993.

6_{Shank, E. M., and K. M. Hollister, "A Statistical}

Risk Assessment Model for the Precision Runway

Monitor System", ATCA Conference Proceedings,

1992.

7_{Folmar, V. A., Szebrat, X. P., and N. E. Toma, "An}

Extension to the Analysis of Traffic Alert and

Collision Avoidance System (TCAS) Advisories

During Simultaneous Instrument Approaches to

8

American Institute of Aeronautics and Astronautics