The design of a real-time

engineering flight simulator for the

rapid prototyping of avionics systems
and flight control systems
by D. J. Allerton

This paper describes the design of a real-time flight Fighter Aircraft programme at British Aerospace, the
simulator, which is based on a modular architecture of PCs Boeing-777 programme and the continuing development
coupled by Ethernet. The simulator is required to provide a of flight control laws for Airbus aircraft by Aerospatiale.
rapid prototyping environment to support the design and As a result of the advances in PC technology and the
evaluation of avionics systems and flight control systems. supporting chip sets for commercial applications, there
Methods are described to ensure real-time implementation has been an increasing emphasis on the use of ’commer-
of the equations of motions using a standard PC and the cial aff the-shelf’ systems (COTS) in the procurement of
provision of real-time graphics to simulate aircraft displays training equipment. Up to the last 5 years, the intensive
using SVGA. The paper includes analysis of the perform- processing, which is necessitated by real-time simulation,
ance of the flight modelling methods and the simulation of could only be achieved by high performance mini-
aircraft displays. The paper concludes that high fidelity computers and custom hardware. However, with the
flight models and aircraft displays can be implemented increases in the processing capacity and memory
using standard PC platforms and Ethernet cards, if capacity af FCs, combined with a concomitant reduction
attention is given to the design of the simulator architec- in the cost of processors, PCs can now be used for the
ture. The resultant simulator provides a rapid prototyping real-time computation required in flight simulation.
environment, standardising on the packet format and low- This paper focuses on three aspects of engineering
level packet protocols of Ethernet, simulation:
Keywords: Flight simulation; rapid prototyping; avionics o real-time modelling methods in the simulation of
system design. vehicle dynamics - to ensure that the model has
sufficient rigour and realism to replicate aircraft and
engine dynamics accurately, while sustaining the itera-
1. Introduction tion rates which are essential in real-time simulation
9 real-time computer graphics to simulate aircraft
Flight simulators have had a major impact on airline displays - to replicate the functionality of specific
safety during the last 20 years. As a result of the aircraft displays and to meet the real-time require-
improvements in simulator fidelity, which has only been ments for display update rates
possible with the advances in computer technology, the o a distributed approach to the organisation of simu-
flight simulator is now accepted as an essential element lator modules - to partition the overall processing
of pilot training (International Standards, 1996). For load and to ensure the data throughput of the
many major airlines, specific pilot training programmes simulator (Valentino, 1994).
are conducted completely in a flight simulator rather
than an aircraft. This paper describes a COTS approach to the design of
In recent years, simulators have also been used in a an engineering flight simulator based on PCs, SVGA
different role; that is, in the development and evalua- graphics and Ethernet network cards. The paper
tion of aircraft systems; these simulators are termed establishes the requirements for real-time engineering
’engineering flight simulators’ rather than training flight simulation and shows that standard PCs can
simulators. These simulators allow engineering designs provide sufficient fidelity and performance, provided that
to be integrated into the flight deck of a civil aircraft or care is given to the design, organisation and implementa-
the cockpit of a military aircraft and to be evaluated in a tion of the simulator architecture. The paper focuses on
realistic manner that can avoid potentially dangerous or real-time modelling techniques and the provision of
expensive flight trials. Indeed, engineering simulation computer graphics to provide a rapid prototyping
has been a major component of several recent aircraft environment to develop and evaluate avionics systems
development programmes including the European and flight control systems. This approach to engineering
simulation includes the acquisition of simulator data
during simulator-based trials and the storage and
College of Aeronautics, Cranfield University, Bedford- presentation of flight data, both during trials and in
shire MK43 OAL. UK. off-line analysis.

51

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

Fig 2 Aircraft dynamic model

2. Simulator organisation simulator modules, it is straightforward to change the

The organisation of the modules of a typical flight
partitioning between hardware and software modules
without incurring extra programming effort to reorgan-
simulator is shown in Fig I . In early flight simulators, ise the modules.
the majority of these functions were computed by a
single high-performance processor (Allen, 1993). More
recently, as the performance of microprocessors has 3. Flight modelling
improved, and with the availability of low-cost local area
network cards, the partitioning of simulator functions Flight models are based on Newtonian mechanics: the
has become distributed, in the form of an array of aircraft position in space is determined by numerical
processors, each performing a number of the functions integration of the aircraft velocities which, in turn, are
outlined in Fix 1. 1&dquo;~~~Lr~h~ie~~, this partitioning still derived from numerical integration of the accelerations
remains critical, as the overall throughput of data must on the body which can be computed from the forces
be s-Jflicien£ to avoid any disce-rnible lags the to the body. The principle to the
simulator. Consequently, if large amounts are trans- determination of aircraft attitude in space, which derives
ferred for every frame of the simulator iteration, it is from a double integration of angular accelerations on the
necessary to provide effective communication links airframe, which are readily computed from the moments
between simulator modules ~~.~tz et al, 1989). applied to the body (role and Staples, 1986).
The modular approach developed in this paper allows This six degree-of-freedom model (three linear degrees
the partitioning to be achieved at either a processor level, of freedom and three angular degrees of freedom) is
where the interface between modules is an Ethernet outlined in Fig 2. The aircraft forces derive from
packet transferred between modules, or, alternatively, at aerodynamic lift (from the wings, tail and fuselage),
a software level, where data is transferred between propulsion forces from the aircraft engines and propel-
modules by means of parameter passing mechanism. lers, side forces and drag caused by the airframe,
By careful construction and organisation of the propellers and engine intake systems. The moments

52

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

Fig 3 Equations of motion

arise from the application of these forces as a set of requiring extrapolation and interpolation at run-time,
moments about the centre of gravity and secondly, from incurring further increase in the processing load
the aerodynamic moments caused by the airframe and * forces, moments, accelerations, velocities and position
by the inceptors used to control aircraft attitude. need to be derived in the most appropriate axis
The computation of the forces and moments are more system - inevitably, this requires transformations
straightforward in specific axes systems. For example, between axes systems
propulsion forces are more readily computed in aircraft * the transformations and computations of forces and
body axes, whereas aerodynamic data is commonly moments involve trigonometric functions, typically in
derived from forces and moments measured in stability the form of series expansions. The accuracy required
axes. Similarly, the aircraft position and attitude seen by in these computations, which is determined by the
the pilot is best represented in ’world co-ordinates’ requirements for fidelity, also affects the number of
defined by the axes system of the simulator visual system. instructions executed per frame.
A detailed definition of the reference frames used to
define aircraft motion is given in Baarspul (1990).
The simulation of the equations of motion of any 4~a . ~~r~~~~~t~~~~~ issues
vehicle can be decomposed into three stages: Most applications in simulation require the precise
o
computation of the forces and moments modelling of system characteristics. However, real-time
simulation extends these requirements to the repeated
@ transformation of forces, accelerations and velocities
solution of the system equations within the frame time.
between axes systems
In practice, this requirement imposes a minimum update
a integration of accelerations and velocities to derive rate of 50 frames per second, or 20 ms per frame. This
position and attitude. requirement is exacting in two senses. Firstly, the com-
This simplified outline of the equations of motion for an plexity and fidelity of the modelled equations must be
aircraft is shown in Fig 3. These equations are solved acceptable. In other words, although a simplified model
repeatedly, with a sufficiently fast iteration rate, so that a may reduce the amount of processing and thus increase
pilot cannot discern delays between an input and its the update rate, the equations must be solved with
effect, in comparison with the response of an actual sufficient rigour that all aspects of the modelling are
aircraft. Typically, the solution of these equations is acceptable to a pilot. Secondly, the update rate must be
required at rates in excess of 50 Hz, giving 20 ms per maintained at all times and under all conditions. This
iteration or frame. In practice, part of this frame time is second requirement is particularly apposite in a dis-
dedicated to communications (data transfers) with other tributed system, where delays in one sub-system are, in
systems and secondly, a margin is required to ensure that effect, passed on to other systems. Both of these
the frame time is never exceeded. In order to maintain requirements are paramount in the acceptance phase of
this overall processing rate, there are a number of a simulator, where it is to demonstrate that

implications for the real-time solution of the equations both the fidelity and latencies of the simulator fulfil the
of motion: specific requirements for a reai-time simulator.
A number of methods are used to ensure that the real-
o a processor has a ~r~i~e ~i~c4~ speed and thereifore,
there is a finite limit to the number of instructions performance be art all times:
which can be executed per frame. In addition, the o timing individual modules, particularly for ’worst-
number of instructions executed per frame may vary case’ conditions to estimate the delay in executing each
with time. Nevertheless, there is a requirement to procedure. For example, extra code may be executed
maintain the minimum frame update rate at all times, during stalling if a high-incidence model is invoked
including ’worst-case’ situations under these conditions
e aerodynamic data is likely to be provided in tabular * estimation of delays from inspection of the code,
form, derived from flight tests, wind-tunnel data or based on timing information from processor data
predictions (Smetana, 1984) - this data may be sheets
inconsistent and may be a function of two or three . selection of an optimising compiler
variables (not necessarily defined at regular intervals), . selection of a faster processor

53

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

Fig 4 Aerodynamic data Cmrx

s time-out code which abandons non-critical code if the The use of cubic spline methods avoids the introduction
frame time is exceeded. of discontinuities in the data. Interpolation is used
between tabulated data points and extrapolation is used
Data is acquired by the flight model from two sources:
to obtain points outside the range of generated data
analogue and digital data from inceptors and commands {Sinnett et al. 1989). The majority of aerodynamic data
issued by the instructor station. Analogue data is
tables are functions of two or more variables (e.g. mach
sampled autonomously by programming a 12-bit number, flap position, angle of attack) and multi-
analogue-to-digital (A/D) converter to interrupt the dimensional interpolation is used on these forms of
processor on the completion of every A/D conversion, at tabulated data.
the rate of 32 conversions per frame (32 channels) or
1600 interrupts per second. On completion of each
conversion, the processor is interrupted, the converted
data is copied to a shared buffer and the AID is reset to ~e Display
sample the next channel. There are two advantages to Computer generated graphics offers several advan-
this approach. Firstly, the processor never waits for
tages in the rapid prototyping of aircraft systems:
completion of an A,~’I3 conversion and secondly, data is
simply read from a shared buffer when it is required ~ it is
straightforward to change (i.e. to re-program) the
during the frame. The data is guaranteed to have been displays
sampled during the previous frame, and on average 8 EFIS (Electronic Flight Instrument Systems) aircraft
within 10 ms at a 50 Hz frame rate. This latency can be displays are commonplace on modern flight decks
reduced by increasing the sampling rate, but there is still @
integration of a display is reduced to a software task,
a context switch in responding to each of the 1600 minimising the disruption to the simulator configura-
interrupts per second. If the response time for each tion.
interrupt is 100 ~ts, say, then 3200 ps (over 3 ms) is ’lost’ However, there are also several problems associated with
per frame in handling the interrupts from the A/D card. the provision of aircraft displays in a PC-based rapid
A similar approach is adopted to the receipt of environment:
Ethernet packets from other processors. Each packet is prototyping
copied to a shared buffer on interrupt. As only one 8 the SVGA graphics architecture of the PC is not
packet is received per frame, the overhead of Ethernet strictly designed to support real-time graphics
transfers is negligible. However, there is an overhead in * the display resolution is limited to 640:480 or 1024:768

constructing Ethernet packets for transmission and in pixels

* the rendering rate of an SVGA card limits the amount
copying received packets and in extracting data from
these packets. the graphical drawing per frame and hence, the
Aerodynamic data is typically provided in the form of complexity of computer generated displays.
tabular data as a function of several variables, probably
This last point is a major factor in the use of PC graphics
measured at specific points of a flight test programme or
for rapid prototyping. The standard PC SVGA -sol-
wind tunnel experiment. Figure 4 shows the pitching
ution of 640:48Q pixels requires that 15 million pixels are
moment derivative C~,, as a function of angle of attack
re-drawn per second. In practice, even with direct writing
for a military jet, based on the following 14 specific
to the SVGA frame store, it is difficult to sustain
values, obtained from flight-test data: [0.0, 0.0], [7.0,0.0], rendering rates better than 2 million pixels per second,
[9.0, 0.002], [10.0, 0.0035], [11.0, 0.008], [12.0, 0.013], without the use of special-purpose hardware.
[12.2, 0.0135], [13.0, 0.005], [13.4, 0.0], [14.0. -0.004], ,
Despite these limitations, there are specific aspects of
[15.0, -0.0075], [16.0, -0.009], [18.0, -0.01], [20.0,
, 1 1 y
aircraft displays which can be exploited to reduce the
-0.01]. rendering requirements to a manageable level:
The resultant graph is generated from a cubic spline fit
to the data given above at intervals of 0.2° for an angle of * a relatively small proportion of an aircraft display
attack from 0 to 20°. changes every frame
Data fitting is performed prior to the computation of * the size and shape of display components are fixed
the aerodynamic data tables used in the flight model. (invariant)

54

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

Fig 5 Boeing 747-200 SVGA display

o typical aircraft displays contain instrument pointers 6. ~~~~-t~~~ displays,

and symbols, i.e. vectors and characters
The animation of an aircraft display, which includes
ø the maximum number of colours in an aircraft display
is typically of the order I ~. moving pointers, rolling digits and rotating compass
cards, etc., is achieved by erasing objects and re-drawing
them in a. new position at sufficiently fast rates that the
It is thereforepossible to optimise graphic operations for
aircraft displays subject to the limitations outlined
operation of erasing and re-drawing cannot be discerned
above. The resultant library of graphical primitives is
by the human eye. In practice, the displays are updated at
50 Hz, in line with the overall iteration rate of the flight
restricted to the drawing primitives needed for aircraft model. This implies that, within every 20 ms frame, every
displays, and omits the more conventional graphics graphic object that alters during the frame must be
functions of a general-purpose graphics library. The re-drawn. A typical aircraft display, in this case a Boeing
library routines are optimised to exploit the features 747-200 aircraft with Pratt and Whitney JT9D engines, is
inherent in the SVGA graphics architecture, which are shown in
matched to specific real-time display requirements.
Fig 5.
There are two requirements in simulating aircraft
Unfortunately, there is a wide range of SVGA displays in real-time:
graphics cards and display resolutions for SVGA. How-
ever, the VESA standard (1994) overcomes the problems o to re-draw objects at the fastest possible rates, i.e. to
of software portability for SVGA cards and the display optimise the vector drawing speed and the character
formats. Despite this common standard, the applications drawing speed
software must still contain the code to draw lines and a to minimise the graphical changes to the display -
characters and to set the colour mapping between pixels ideally, the updating process is limited to the graphics
and the colour palette chip. The rendering of lines and
objects which have altered in the previous frame.
characters is achieved by writing specific pixels in the
SVGA RAM. The speed of vector (straight line) drawing is achieved by
The standard adopted for the flight simulator means of Breshenham’s line drawing algorithm (1965).
described in this paper is a display resolution of This well known algorithm is based on determining the
640:480 pixels with 8 bits per pixel providing 256 pixels closet to a straight line between the start point
possible colours. The VESA graphics interface manages and the end point and avoids the use of any multiplica-
the control of the frame store registers to ensure the tion or division operations. A line-drawing algorithm
correct pixel addressing and page selection (640:480 was written in assembler language to optimise the
requires over five 64K pages of video memory). A real- drawing rate. Writing each pixel to the SVGA video
time graphics library was developed to support line RAM involves a single byte access to external memory
drawing, character drawing, colour selection and ceneral per pixel and is dependent on the memory access speed
display management. of the video RAM. Figure 6 shows the line drawing rate

55

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

Fig 6 Pixel drawing rate

Fig 7 Distribution of vector lengths

for a 33 MHz 486 PC. The pixel drawing rate varies from rendering rate. However, gains in graphics speed are also
6.9 gs per pixel to 0.63 gs per pixel according to the line dependent on the interface between the processor and
length and orientation owing to the variation in the the graphics card and the access speed of the memory on
number of arithmetic operations for each line. The the SVGA graphics card.
distribution of vector lengths for a typical aircraft Several of the aircraft displays contain both vectors
display is shown in Fig 7. This clearly the characters, for the compass of
majority of vectors are less than 30 pixels in length and the Horizontal Situation Indicator and the Radio
that a worst case figure of 1.5 gs per pixel can be Magnetic Indicator. Although characters can be drawn
assumed. as a set of straight line segments, it is also possible to
In addition, specific aircraft instruments contain large store character set fonts and to extract the appropriate
areas of in-filled colour, for example, the attitude character from the font set as an array of pixels, as
indicator. If the complete region occupied by the attitude shown in Fig 8.
indicator illustrated in Fig 5 is re-drawn, then approxi-
mately 10000 pixels need to be re-drawn per frame, con- 7. Colour plane mapping
suming 15 5 ms of the 20 ms frame. Of course, one solution
is to use a faster PC, where the gain in clock speed is (to a In the 256 colour mode, each pixel corresponds to a
large degree) proportional to the improvement in pixel single byte of SVGA RAM. As each byte is accessed by

56

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

Fig 8 Character fonts

the display, it is transformed from an 8-bit byte to a This approach provides eight distinct planes and a
video signal containing red, green and blue components maximum of 256 colour options for overlapping planes.
by the colour palette chip. The mapping of the colour For example, a display pointer could be drawn in plane
palette register allows each of the 256 possible bytes to be 6 with the remainder of the display drawn in the other
specified in terms of three 6-bit colour components, planes. To move the pointer, it is erased but, for each
giving 21’ possible colours. For example, the following pixel of the pointer, this only requires bit 6 to be cleared,
components produce the colours defined below: leaving the pixels ’under’ the pointer unaffected. More-
over, if the pointer overlays a region of red, then the
mapping of the white pointer over a red object can be
defined as white, to give the effect of the pointer lying on
top of the red plane or alternatively, it can be set to red,
to give the effect of the pointer passing under the red
plane.
Although this arrangement appears to be complicated,
the SVGA RAM is organised as a conventional memory,
allowing logical operations in addition to read and write
operations. For example, pixels can be written in an
individual plane by a logical OR operation without
Although the mapping of the colour palette can be
changed at run-time, in practice it is programmed with a altering the pixel values in other planes and similarly, a
set of colour mappings, prior to the generation of any pixel can be erased by a logical AND operation, without
graphics. For the application of aircraft displays, it is affecting pixels in other planes.
possible to consider each bit of each pixel byte to Consider the example of the Radio Magnetic
Indicator (RMI), shown in Fig 5. The rotating compass
represent a colour plane. For example, bit 2 could
card could be drawn in plane 0, the fixed objects could be
represent plane 2, which maps to red, say:
drawn in plane 1, the green pointer drawn in plane 2 and
the yellow pointer in plane 3.
One further benefit of this approach is that one plane
can deliberately occlude another plane, simply by

arranging the colour palette mapping. Consider the

case where the colour plane mapping is used to hide

Similarly, bit 5 could define plane 5 which maps to ~r~e~a4 parts of a display, for example the rounded bezel of an
say EFIS display, shown in Fig 9.
Region (2) represents pixels in both blue and grey
planes. The mapping allows the blue region to be written
to a rectangular boundary (for example, to simplify any
clipping operations) but to appear to be grey, i.e. under
the grey region ( ~ ).
Another example where this feature can be exploited is
A combination of these two bytes is shown below. This the provision of sliding scales, which are commonly used
byte, which represents the value 3610, can map to a third in EFIS displays, as shown in Fig 10(a), which shows the
colour, purple say, such that a pixel drawn in planes 2 and Primary Flight Display (PFD) of a Boeing 747-400
5, maps to purple. aircraft. The digits of the expanded airspeed indicator,
shown in Fig 10(b), are occluded by the window, as a
result of the colour palette mapping function, avoiding
any computation of the hidden portion of the digits.
This technique can also simplify clipping operations. For

57

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

example, the pitch lines and the regions of blue and
brown of the attitude indicator are clipped to a rec-
tangular region, partially obscured by the surrounding
instrument bezel. The alternative approach, clipping to a
circular region would involve considerably more pro-
cessing.
Handley and Allerton (1994) have shown that a
considerable proportion of the frame time can be
expended on the in-filling operations required to render
the attitude indicator. Although the number of graphical
operations can be reduced to an acceptable level, the
algorithm is further simplified by means of the colour
mapping, as shown in Fig 1 l. Only half of the attitude
’ball’ is drawn. The background colour is set to blue. The
brown region is overlaid on the blue region, mapping to Fig 9 Colour plane organisation
brown. The white pitch lines can occur in eight regions,
as follows: memory at the packet level. Ethernet offers several
advantages for real-time simulation:
* it is relatively low-cost
* the software interface to a packet driver is straight-
forward
e data is transferred at 10M bits per second (approx-
imately 1M byte per second)
a data integrity is excellent
e cards operating at 100M bits per second are also
available.
One disadvantage with the use of Ethernet in flight
simulation is that an Ethernet card can detect, but
cannot resolve traffic conflicts. In other words, if two
nodes transmit at e~aetly the same time, the data on the
The operations are reduced to: bus will be corrupted and has to be re-transmitted,
@ erasing and re-drawing the white lines in a single plane requiring a robust protocol to ensure bus integrity and
@ re-drawing the blue region in a single plane. throughput. However, for simulation applications, it is
possible to synchronise the data transfers such that only
one PC transmits at a time. Ethernet also provides a

8. Ethernet ’promiscuous’ receive mode so that a node can detect all

packets transferred over the bus. It is straightforward to
Ethernet cards for the PC provide a thin-wire co-axial program the bus transfers, so that packets are trans-
cable interface at a data rate of 10M bits per second. mitted in an agreed sequence, avoiding the need for
These cards also support DMA transfers to or from PC higher level protocols.

Fig 10(a) Boeing 747-400 primary flight display; (b) airspeed indicator

58

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

Fig 11I Colour plane organisation for the attitude indicator

In practice, although the overall data rate is 1(~~ bits off-line, the packets are obtained as sequential disk
per second, time is also required to form transmitted blocks.
packets and to extract received packets. For a 33 MHz This approach also provides the basis of a real-time
PC, an overall data rate of 2.~I~I bits per second was flight data recorder for flight test applications. Ethernet
achieved for back-to-back transmissions. At 2M bits packets are buffered in XMS memory as they are
per second, the data rate is approximately 200K bytes received. As the XMS memory fills, packets are copied
per second or 4K bytes per frame. Limiting the transfers to disk, as a background task. However, applications
to 3 ms of the 20 ms frame time, reduces the throughput software can access the current packet, packets stored in
to 600 bytes per frame. This ensures that packets are XMS or packets saved to disk, in a totally transparent
received at the start of each frame. However, data that is manner. In effect, the limit on data storage is the disk
not time critical can also be sent during the remainder of capacity rather than the XMS memory capacity.
the frame and transferred as a background task with
minimal disruption to the real-time performance.
The equations of motion are solved at 50 Hz and 9. Results
aerodynamic data, engine data and navigation data is
transmitted by the flight model computer as a single The flight models have been validated in a number of
512 byte Ethernet packet every frame. This data includes ways:
accelerations, velocities, position, attitude, engine fuel o standard test inputs were applied to the flight model,
flows, propeller blade angles, glide-slope deviation, etc., for example, step inputs and doublets and compared
in an agreed format. At this rate, 25K bytes of data is
with published aircraft data
acquired per second or 1.5M bytes per minute. For o the simulator flight models have been evaluated by an
example, a 24M byte system can store up to 16 min of experienced test pilot
continuous flight data, without data compression and o data was obtained from flight tests and applied to the
without discarding any data.
simulator in order to compare the simulator response
This data acquisition time is achieved by means of the
with equivalent aircraft inputs.
Extended Memory System (XMS) provided under the
DOS operating system for the PC. A small library was Figure 12 illustrates the short period phugoid response
written to read or write 1/2K byte blocks to XMS of a Boeing 747-200 the manual application of 18° of
to
memory. Each Ethernet packet is copied directly to elevator for 2s. The acceleration ~, pitch
XMS memory during the Ethernet interrupt response. attitude, angle of attack and elevator input are plotted
Data for specific frames is then acquired from XMS (at a 50 Hz sampling rate) over a period of 12 s.
memory. This method ensures that data packets are One further advantage of this modular approach is
stored as a background task, without disrupting the that simulator modules can be re-used off-line to
instructor station functions but, more importantly, no undertake analysis of the flight model and to investigate
data packets are lost. Subsequently, XMS data packets the various modes of motion. By careful organisation of
are written directly to disk and can be recovered as the simulator modules, it is possible to use actual
512 byte blocks as the equivalent of ’raw’ Ethernet simulator modules for c~~ line analysis.
packets. For teaching, this approach allows identical The resultant timing for typical instrument display
data analysis software to be used on-line or off-line; on- updates for the Boeing 747-200. illustrated in Fig 5, is
line, the packets are obtained as sequential XMS blc~~~s; shown in Fig 13, which also indicates the amount of

59

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

Fig 12 Short period phugoid response Boeing 747-200

Fig 13 Overall timing

processing required for the aerodynamic model and the these conditions requires the maximum number of
engine model. These results are ’worst-case’ figures, vectors and characters to be re-drawn per frame.
where fast roll and pitch rates were applied to the From this graph, it is clear that a single 66 MHz
simulator to exercise the primary flight displays over a 486 PC is capable of updating the flight model for a
period of 10 s. The amount of graphical updating under Boeing 747-200= an engine model for four Pratt and

60

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

Whitney JT9D turbofan engines and 25 flight instru- It has been shown that a standard PC with SVGA
ments within 14 ms, a margin of 6 ms of the
leaving graphics can provide the real-time environment neces-
20 ms frame. Typically, the instrument display occupies sary for the flight modelling and simulation of aircraft
3-5 ms of the frame, with occasional excursions up to displays. The flight models provide sufficient rigour in
10 ms, for example, where both compass cards of the terms of the fidelity of the model dynamics and, by
RMI and HSI are updated. The full six degrees-of- providing an iteration rate in excess of 50 Hz, this
freedom flight model requires less than I ms of the frame approach has been used in pilot-in-the-loop studies
and the engine model takes approximately 0.2 ms. The where the minimisation of system latency is an essential
remainder of the frame, shown by the overall timing, requirement.
includes the Ethernet transfers to the visual system and The use of Ethernet provides a straightforward
the instructor station, solution of the navigation mechanism for system partitioning and ensures data
equations and management of the 1/0 functions. These throughput. It also offers an extensible route to optimise
functions amount to approximately 3 ms of the frame processor loading for a wide range of applications and,
from inspection of Fig 13. These values, which were in addition, affords a modular approach to the con-
obtained by a pilot flying relatively high rate manoeuvres figuration of real-time systems.
in pitch and roll, correspond to timing figures taken from The organisation of graphics planes, which is possible
off-line performance analysis of individual simulator by exploiting the SVGA architecture of the PC, lends
modules. itself to the real-time simulation of aircraft displays. The
These results also show that a 66 MHz 486 PC with an functionality of the full range of aircraft displays and
SVGA display can display a full range of aircraft instruments has been demonstrated but, more impor-
instruments well within the frame update rate of 20 ms. tantly, it is possible to achieve display update rates in
A library of aircraft instrument display primitives has excess of 50 Hz with low-performance 486 PCs.
been developed covering the range of primary flying This modular approach to the design of a flight
instruments. This approach allows a systems designer to simulator allows flight models, engine models, flight
select primary flight displays, engine instruments and control laws and aircraft displays to be configured in a
navigation displays from a portfolio of aircraft displays straightforward manner and used to develop and
appropriate to a specific task. This library is optimised evaluate avionics systems. The provision of rapid proto-
for real-time aircraft displays, as follows. typing in a low-cost flight simulator, offers opportunities
to undertake research in flight control systems design,
@ the displays are limited to eight separate colour
avionics and human factors, simply by combining a set
planes which are organised to form any selection of of simulator modules in an arbitrary configuration of
256 colours - in practice, the full range of display
PCs. In addition, simulator modules can be combined
functionality and colour is supported with applications, independent of the language or
e the
graphics is limited to 2-D vector generation, platform, by reading and writing Ethernet packets. The
character generation and a restricted set of in-filled
only requirement of the applications software is com-
polygon operations to provide the complete set of pliance with the Ethernet packet formats and the
graphical operations required for instrument displays, provision of a low-level Ethernet driver.
including EFIS displays With the availability of high performance work-
a the minimum content of the display is re-drawn
per stations and PCs in recent years, engineering simulation
frame, reducing the number of graphical operations, has become an essential element of systems design in the
in order to sustain a real-time display.
development of aircraft systems. The approach outlined
One of the advantages of this modular approach to in this paper extends this approach to standard PC
research flight simulation (Green, 1994) is the benefits of platorms, in particular, to provide a realistic design
rapid prototyping. To illustrate the efficiency of this environment for teaching and research in avionics
approach to system design, a postgraduate student in the systems and flight control systems.
College of Aerodynamics, who participated in the
Garteur Flight Mechanics Programme (Gautrey, 1995),
implemented a flight model based on a variant of the
Airbus A300-600R in one day, coded a C* flight control 11= References
law for this aircraft in 2 days and integrated an auto-
Allen, L. 1993. ’Evolution of flight simulation’, AIAA
throttle -in 2 h. The student was able to acquire Sight test Simulation and
data from pilot-in-the-loop studies within 3 days of
Conference on Flight Technologies,
AIAA-93-3545, August.
obtaining the data package for the aircraft. Baarspul, M. 1990. ’A Review of Flight Simulation
Techniques’. Progress in Aerospace Sciences, 22, 1-20.
Bresenham, J. E. 1965. ’Algorithm for computer control
of a digital plotter’, IBM Systems Journal, 4(1),
10. Conclusions
25-30.
The provision of a real-time flight simulator, which Gautrey, J. 1995. ’Robust flight control design in a
can be easily configured and programmed for a range of computational control aircraft engineering environ-
applications, is essential in programmes involving the ment’, 5th Garteur Flight Mechanics Action Group
design and evaluation of avionics systems. A modular FM(AG08), Toulouse, November.
approach to the organisation of the simulator, provides Green, B. 1994. ’ADA autocoding and object-based
the level of flexibility that is needed to undertake studies software design - an application for real-time flight
ranging from the design of aircraft displays and flight controls simulation’, AIAA Conference on Flight
control systems to human factors. Simulation Technologies, AIAA-94-3408-CP.

61

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016

Handley, S. J., and Allerton, D. J. 1994. ’The design of an Sinnett, M. K., Steck, J. E., Selberg, B. P., and Oetting,
EFIS attitude indicator for a flight simulator’, 19th R. B. 1989. ’An alternate approach to table look-up
ICAS Congress, AIAA, Anaheim, pp. 282-290, routines for real-time digital flight simulation’, AIAA
September. Conference on Flight Simulation Technologies, AIAA-
International Standards for the Qualification of Airplane 89-3310, p. 336.
Flight Simulators.1996. The Royal Aeronautical Smetana, F. 1984. Computer Assisted Analysis of Aircraft
Society, July. Performance, Stability and Control, McGraw-Hill,
Katz, A., Allen, D. M., and Dickson, J. S. 1989. New York.
’Synchronisation and time tagging in distributed Valentino, G. J. 1994. ’Shared-memory networks
real-time simulation’, AIAA Conference on Flight (SMNs): a key enabling technology for distributed
Simulation Technologies, AIAA-89-3300, p. 259. real-time
systems’, AIAA Conference on Flight Simula-
Rolfe, J. M., and Staples, K. J. 1986. Flight Simulation, tion Technologies, AIAA-94-3427-CP.
Cambridge University Press, Cambridge, UK. VESA BIOS Extension (VBE) Core Functions Standard.
1994. Version 2.0, Video Electronics Standards
Association, November.

Transaction Papers
Papers are invited on all aspects of measurement and control - research, development,
application and education - and from any discipline.
The technology covered by the Transactions is naturally inter- or cross-disciplinary and for this
reason the field is very wide. The Institute has technical groups currently working in the following
subject areas:

Measurement and Instrumentation Technology concerned with design and application of instruments
for the measurement of physical and chemical variables, Also with signals, signal processing and
display of information.
Systems and Control Technology concerned with the theory and practice of control systems;
modelling and analysis of systems; identification of systems; application of control; digital control
systems ir~~ludin~ on-line computation and software systems; and applications of programmable
logic controllers.
Social and Biological Systems concerned with the application of measurement and control techniques
to areas outside the more ’traditional’ areas of the Institute’s coverage; eg, to management and
economic systems, systems engineering, measurement and identification of biological systems,
aids to medical diagnostics, etc.
Educational Activities concerned with the advancement and communication of knowledge of
education and training in the art of measurement and control.
Standards Policy concerned ~.~ith standards and codes of practice c~~rerin~ measuring instruments and
control equipment.
Robotics
§~~~;.U~i’P..Yv&dquo;&dquo; *3’.gLP4’.~I~~.E be welcomed on Safety and information £4av&48~~Y.Ja’iX~lB,
Guide Notes for intending contributors to the Transactions are available from:
Transactions of the Institute of Measurement and Control, Editcar~al Administration
Office, Queens University, Department of Mechanical and Manufacturing Engineering,
Ashby Building, Stranmillis Road, Belfast BT9 5AH (Tel: 01232 274112; Fax: 01232
661729; e-mail: [email protected])

62

Downloaded from tim.sagepub.com at PENNSYLVANIA STATE UNIV on September 13, 2016