PH I LI PS TECHNICAL REVIEW-----------Volume43, No.11/12,December1987-----------

CARIN, a car information and navigation system

M. L. G. Thoone

Whatever we think of the car, it is now a part of modern life. A t the very least it is useful forgetting from A to B. In fact, however, this is not always so easy. We have to be able to readthe road map and we need to know where we are. Planning the most economical route canalso present problems, particularly if we are unexpectedly held up during the journey. A teamat the Philips Project Centre at Geldrop, in the Netherlands, are just rounding off their workon prototypes of an electronic vehicle-guidance system called CARIN (an acronym for Car In-formation and Navigation System). During thejourney CARIN speaks through a speech syn-thesizer to tell the driver which road to take at crossings and road junctions. Before the jour-ney the system also plans the best route, finding a way round traffic delays. In the design ofCARIN the vast storage capacity of the CD-I (Compact Disc Interactive) was of inestimablevalue for recording the digitized map data.

Introduction

In a car it is the passenger sitting next to the driverwho usually has to navigate: plan the route, read themap, relate actual position to the map, and give direc-tions to the driver. The passenger performs thesetasks with varying degrees of success, and we all knowthat any inadequacies of performance can lead to dis-harmony. Perhaps more serious than this is that mis-takes in planning and following a route can mean thatgreater distances are travelled than is necessary. Forprofessional road users, such as police, fire brigades,delivery and freight services, ambulances and taxis,

Ir M. L. G. Thoone, formerly with Philips Research Laboratories,Eindhoven, is now with the Philips Consumer Electronics Division.He is the leader of a combined team (from the Research Labora-tories and the Consumer Electronics Division) working on theCARIN project. Various members of this team made valuable con-tributions to this article.

economic route planning and good vehicle guidancecan give substantial savings. This is certainly so if theguidance helps drivers to avoid traffic delays by anearly change of route. All this can become reality ifthe 'navigator"s tasks are taken over by an electroniccomputer-controlled system. Electronic navigationand information systems are well established in mar-ine and aviation applications. Car and electronicsmanufacturers are now trying to develop systems thatare suitable fot automotive applications.Almost all of the car navigation and information

systems now known to be operational have the dis-advantage that they do not have a suitable mediumfor storing topographical information. Some of thesystems also require substantial investments in 'infra-structure', such as beacons for radio range finding.

318 M. L. G. THOONE Philips Tech. Rev. 43, No. 11/12

Nor are most of the systems very user-friendly, sincethe driver has to enter the starting position anddesired destination as a map grid reference [11.

At the Geldrop Project Centre, which is part ofPhilips Research Laboratories, a car navigation andinformation system that does not have these disadvan-tages is now being developed in cooperation with thePhilips Consumer Electronics Division [21. The system,called CARIN (Car Information and NavigationSystem), plans the best route, guides the driver withthe aid of a speech synthesizer, periodically deter-mines the position of the vehicle, selects an alternativeroute if there are coded digital radio signals reportingtraffic obstructions, and can also give tourist informa-tion; see fig. J. The special feature of CARIN is that ituses a highly efficientmedium for storing digital data,the Compact Disc. The version used for CARIN is theCD-I (Compact Disc Interactive), specially developedfor storing audio, video and computer data for con-sumer applications. This new medium is also ideally

less than that of a magnetic-tape cassette; about asecond as compared with about a minute. (The accesstime is longer, however, than that of a semiconductormemory; the problems connected with the non-negli-gible access time of the Compact Disc will be dealtwith later in this article.)

The units that are carried in the car are:• an on-board computer for controlling and proces-sing the information, containing a semiconductormain memory with a capacity of 1 Mbyte (8 Mbit);• a Compact Disc player with amplifier and loud-speakers;• sensors for determining the position of the vehicle(if wheel sensors are fitted for an antiloek brakingsystem (ABS), it may be possible to use them for someaspects of position determination);• an electronic speech synthesizer;• a simple keyboard; and• a small flat liquid-crystal display (LCD); for show-ing a simplified map of the next intersection.

Fig. I. The interior of a car fitted with the CARIN vehicular guidance system. The car was de-signed by the British firm l.A.D. (Industrial Automotive Design), of Worthing, West Sussex.The prototype is the result of ideas for 'the car of the not too distant future'.

suited for the storage of topographical information.The amount of digital data that can be stored on oneCompact Disc is vast: 4800 Mbit. The storage capac-ity of the largest semiconductor memory of the dy-namic-RAM type (Random-Access Memory) as yetavailable is far less: 1 Mbit. The storage capacity ofone Compact Disc corresponds to 150000 typed A4pages, or - and this is important for navigation sys-tems - to the detailed topographical information fora medium-sized country. The access time is also much

These units complete the basic CARIN system. Otherunits can also be added:• a touch-screen graphic display for interactive con-sultations of map data for a larger area;• a special car radio that receives messages about traf-fic delays, road hazards, etc. and inputs them to thesystem; and• an interface with sensors present in the car for mon-itoring the status of oil temperature and pressure,vehicle lighting, and fuel.

Philips Tech. Rev. 43, No. 11/12

The driver can only consult the map data on thetouch screen when the car is stationary. The keyboardis used for entering the starting position and destina-tion, and any special requirements. The driver doesnot have to enter a map reference, but can simplyspecify the names of countries, towns and streets,since lists of such names are stored with their mapreferences on the Compact Disc. The computer callsthe required application programs from the disc, cal-culates an efficient route from the starting position tothe destination, and ensures that the information re-quired for following this route is read from the discand stored in the main memory. Then.the player canbe used for playing music again. During the journeythe driver is guided by spoken directions from thespeech synthesizer, such as 'take first turning left','fork right', 'take first exit at roundabout', and 'wehave arrived'. The information on the small fiat dis-play gives extra support. In CARIN the emphasis ison auditive guidance for the driver, for reasons oftraffic safety.The principle of the vehicle-locating procedure is as

follows. Sensors that measure the number of revolu-tions of the non-driven wheels provide informationabout the distance travelled and changes in the direc-tion of the vehicle. An electronic compass, whichmeasures the direction of the vehicle in relation to theEarth's magnetic field, gives information about theheading. The computer uses both kinds of informa-tion to periodically calculate the location and size ofan area in which the car is located with a certain prob-ability. This area is compared with the topographicalinformation in the main memory. Since the car shouldbe on a road whose data is stored in the memory, thiscomparison determines the part of the road where thecar is located. This procedure is continuously re-peated. The positioning error with this method isabout 20 m. In later generations of CARIN we alsointend to make use of a satellite navigation systemcalled GPS (Global Positioning System), due to be-come operational in the nineties [31.

The European Broadcasting Union has recentlyreached agreement on RDS (Radio Data System), asystem that can be used to transmit digital data at arate of 1200 bits/s in a free frequency band within theFM broadcast signal. RDS will be introduced in manyWest European countries within the not too distantfuture [41. In future versions of CARIN, if the systemreceives an RDS message about traffic delays or roadobstructions, it will calculate the time to reach theaffected position. If the received message says that theobstruction will have disappeared within that time,the route will remain unchanged. If the obstructionwill still be there, CARIN will then calculate a route

CARIN 319

that bypasses the obstacle and gets the car to its des-tination in the shortest possible time.

In the rest of this article we shall first consider themethod of digitizing map data. Then we shalllook atthe navigation problems, and see how vehicle locationcan be improved with the aid of the stored map data.Next we shall consider some aspects of the storage ofdigital information on a Compact Disc, with specialattention to the partitioning and optimum arrange-ment of the topographical information. Finally, someaspects of the data management will be considered.

Digitizing the map data

In modern cartography some of the topographicalinformation is now processed digitally. Two methodsare used, the raster scanning method and the vectormethod. In the raster scanning method a map is div-ided into picture elements (pixels), say 0.1 mm by0.1 mm. The colour of each pixel is represented by adigital code. To digitize the map of Amsterdam (cov-ering 14 km by 12 km), for example, on a scale of1/15000, a storage capacity of 375Mbit is required forthe raster scanning method [21. In the vector method,as we use it, the axes of the roads are approximated bystraight-line segments, which each represent a'vector'. This method requires far less storage capac-ity. Digitized in this way the map of Amsterdam oc-cupies 1 Mbit, based on the assumptions made in thefirst article of note [2]. If other information is addedto the purely topographical data, such as the names ofstreets and general information about filling stationsand restaurants, etc., this will require a further 1Mbitof memory. As we noted earlier, the Compact Dischas a capacity of 4800 Mbit, so that more than 1000maps like the map of Amsterdam can be stored onone disc.Fig. 2 illustrates the digitization procedure we use in

our vector method. The road network is translated[I) W. Zimdahl, Guidelines and some developments for a new

modular driver information system, Proc. IEEE VehicularTechno). Conf., Pittsburgh, PA, 1984, pp. 178-182;H. Friedl, Guidance of vehicle traffic flows - ALl, RadioElektron. Schau 51, 266-269, 1975;D. J. Jeffery, Options for the provision of improved driver in-formation systems: the role of micro-electronics and inforrna-tion technology, Proc. lEE Colloquium on Vehicular RouteGuidance, London 1985, pp. 1-3.

(2) M. 1. G. Thoone and R. M. A. M. Breukers, Application ofthe Compact Disc in car information and navigation systems,SAE Int. Congo and Exposition, Detroit, Mich., 1984, Tech.Paper 840156; .M. 1.G. Thoone, 1.M. H. E. Driessen, C. A. C. M. Herrnusand K. van der Valk, The car information and navigation sys-tem CARIN and the use of Compact Disc Interactive (CD-I),SAE Int. Congo and Exposition, Detroit, Mich., 1987, Tech.Paper 870139;

(3) T. A. StanselI, Jr, Civil GPS from a future perspective, Proc.IEEE 71, 1187-1192, 1983.

(4) Main characteristics of a unified European VHF Radio DataSystem, Doe. Gt RI 290, Eur. Broadcast. Union (EBU), 1982.

320 M. L. G. THOONE Philips Tech. Rev. 43, No. 11/12

into a graphic structure of points connected bystraight-line segments. The position of each point isgiven as the coordinates of a map reference, i.e. as(x,y)-coordinates in a rectangular system. (In a next-generation system these could be geodetic coordi-nates.) There are two kinds of points, called nodesand intermediate points. A node can be:• a point where at least three segments meet (1);• an intersection with the edge of the map or withsome other artificial boundary (2);• an intersection with an administrative boundary (3);• the end of a cul-de-sac (4); or.' a point where one or more items of data for a streetchange, for example its name or the type of road sur-face (5).All other points are called intermediate points. Theyare found at a bend in a road or they are used to ap-proximate the curvature of a road by a number of seg-ments, like the points 11 and 12 in fig. 2. The succes-sion of points in a sequence that starts and ends at anode is called a 'chain'. In the figure, 1, 12, 11 and 5form a chain, and so do 1 and 4, 1 and 3, and 2 and 5.The graphic structure formed by the various chains

can be regarded as the' skeleton' of the digitized map.'Attributes' can then be added to the skeleton, such as:• street names;• classes of road: motorways, trunk roads, secondaryroads and limited-access roads;• directions of one-way streets.At the start of a journey (and perhaps during the

journey as well) some of the map data on the Com-pact Disc has to 'be read and stored in the main mem-ory. It is therefore important to partition the data and

MB

'" I'" I //. //3' /I ,II 1>\I I '"I I '

I "-I .

I ",~/ ~/I/IIIII4 I

'ol

Fig.2. Vector method for digitizing information from the map. Theaxes of roads are approximated by line segments that connect pointswhose position is fixedby the coordinates of map references. 1 to 5nodes. At node 5 the condition of the road surface changes. 11, 12intermediate points. MB map boundary. CB local-governmentboundary. The segment 1-4 is a cul-de-sac. Points 1,12.11 and 5form a 'chain'.

arrange it on the Compact Disc in such way as to giveshort access times. The CARIN project group havemade extensive studies of the problem of the optimummethod of data storage, since agreement on properstandards here is of considerable importance toPhilips. The theories of the management of topo-graphical information use the topological concepts ofthe O-cell, the I-cell and the 2-cell [51. The numbers 0,I and 2 represent the 'dimension' of a cell. In our case,O-cells correspond to nodes, l-cells to chains and2-cells to areas within joined-up (concatenated) chains;seefig. 3. We shall deal with this in more detail at theend of the article.

6

Fig.3. Some concepts from topology[61. 1 to 6 O-cells(nodes). a toh l-cells (chains). A to C 2-cells (areas inside concatenated chains).

Vehicle location

General equations

The electronic navigation system of CARIN has toobtain a 'fix' for the vehicle at regular intervals, sayevery five metres, by dead reckoning. This amounts todetermining the centre of the Vehicle Location Prob-ability Area (VLPA), the area in which the vehicle ismost likely to be travelling, with a probability of, letus say; 95070. The coordinates of this centre are x« andYn for the n-th dead-reckoned fix. (The y-axis pointstowards geographic north.) The navigation systemtakes the coordinates Xn and Yn and calculates newcoordinates Xn+l and Yn+l from the equations

Xn+l = Xn + On sin ~ (f/Jn+l + f/Jn) (la)

and

Yn+l = Yn + On cos~ (f/Jn+l + f/Jn), (lb)

where On represents the distance travelled between thenth and the (n + I)th fixes, and f/Jn and f/Jn+l representthe headings of the vehicle with respect to geographicnorth at the nth and the (n + I)th fixes.

Philips Tech. Rev. 43, No. 11/12

The periodic dead-reckoning of a fix requires peri-odic measurement of the heading and the distancetravelled. As noted earlier, the car has sensors forthis: an electronic compass and two wheel sensors or'odometers'. These measure the number of revolu-tions of non-driven wheels about their axes. We shallnow take a closer look at these sensors and at the pro-cessing of the signals they produce.

The sensors

Both the electronic compass and the wheel sensorscontain transducers that convert a magnetic flux intoa voltage. The transducers are made by Philips Valvo,and their type number is KMZ10. They are based on athe magnetoresistance effect, an effect in which the re-sistance of certain materials depends on the angle be-tween the direction of the current and that of the mag-netization [61. The effect is strongest in ferromagneticmaterial. The transducer is therefore made from alarge number of strips of a nickel-iron alloy. Thestrips are connected in four groups to form a Wheat-stone bridge; see fig. 4a.When the external magnetic field is zero, the pre-

ferred direction of magnetization of the strips infig. 4a will be along the longitudinal axis. This corre- bsponds to the x-axis. The angle between the directionof the current and the magnetization, and hence thevalue of the resistance, changes when a magnetic fieldis applied in the y-direction. The change in resistanceas a function of this field-strength H; is approximatelyparabolic, as shown by the dashed curve in fig. 4b. IfH; = 0 for such a curve, the output signalof thetransducer for a small change in H; is very weak. Thisproblem is solved by applying a structure of conduct-ing material to the strips so that the angle between thedirection of the current and the magnetization atH; = 0 is equal to 450

; see fig. 4c. The relation estab-lished in this way between the strength of the externalfield and the change in resistance is approximatelylinear in a fairly wide range, as shown by the continu-ous curve in fig. 4b. The maximum sensitivity of sucha transducer is about 0.1 mVper Alm at a supply volt-age of 5 V.The electronic compass may contain three such

transducers for the magnetic field. They measure thecomponents of HE, the strength of the terrestrialmagnetic field, in three orthogonal directions. For themoment only the components HE sin<pnand HE COS<Pnare of interest. (In future we shall use the verticalcomponent to determine the gradient of the road.) HEis about 15Alm. Measurements of such a weak mag-netic field are very difficult because of the slow varia-tion of the output signal, due to temperature changes.An answer to this problem is to convert the output

CARIN 321

y l" 1f:J r~ ..x-- ......-..a....i

0.5

OL_--L_--l_--L_--~~~~

-1 0 1- Hy/Ho

c

Fig.4. a) The interior of a KMZIO magnetic transducer (a PhilipsValva product); three of these can be used in the electronic com-pass. The transducer measures magnetic fields via the magneto-resistance effect. The transducer is composed of ferromagneticstrips, connected to form a Wheatstone bridge (shown schemat-ically in blue). b) The relative change in the resistance of a strip in(a) as a function of the ratio of the magnetic field-strength in they-direction to a reference field-strength Ho [6]. The dashed curve re-lates to an untreated strip. The continuous curve, which is a straightline over much of its range, is applicable if a structure of conductivematerial is applied to the strip. c) A strip treated in this way. Thegrey areas correspond to conductive material. In the regions in be-tween these areas the angle between the current J and the magneti-zation is 45° for H; = 0.

signalof the transducer into a 'square-wave' signal.This is done by periodically changing the direction ofmagnetization in the transducer strips by energizing acoil around the transducer with alternate positive andnegative pulses of current. Synchronous demodula-tion of the rectangular voltage then converts it into adirect voltage, with none of the low-frequency com-ponents that interfered with the original signal.

[6] S. Lefschetz, Introduetion to topology, Princeton Univ. Press,Princeton, N.l., 1949.

[6] W. J. van Gestel, F. W. Garter and K. E. Kuijk, Read-out of amagnetic tape by the magnetoresistance effect, Philips Tech.Rev. 37,42-50,1977.

322 M. L.G. THOONE Philips Tech. Rev. 43, No.ll/12

Transducers of the same type are used for meas-uring the number of revolutions of the non-drivenwheels. Fig. 5 shows how two transducers are mountedin relation to a strip bonded to the inside edge of awheel rim. The strip is permanently magnetized sothat the magnetic field is radial and alternately in-wards and outwards, and so that 26 periods of themagnetization, each of length p, correspond to a rota-

Fig. 5. Principle of a wheel sensor. ST permanently magnetizedstrip. p length of the period of the radial magnetization. wRperipheral velocity of the strip; w angular velocity, R radius. Triand Tr2 magnetic transducers; see fig.4. The output signal isapproximately sinusoidal with a period T = p/wR. The phase dif-ference between the output signals from the two transducers is T/4and is positive or negative, depending on the direction of rotationof the wheel. El circuit that converts the signals into square waves.E2 circuit that derives pulse trains of period T/4 from the twosquare waves. D circuit that decides whether a positive or negativevalue should be assigned to the pulses. Cl counter that supplies abinary number ZL (ZR), which is a measure of the angular displace-ment of the left-hand (right-hand) wheel. (The circuits are digitalexcept for El-)

tion of the wheel through 3600• The spacing of the

transducers is p/4. The figure also shows how thetransducer signal is electronically processed. Whenthe wheel is turning, the transducers each supply anapproximately sinusoidal signal, with a 900 phasedifference between the two signals. The direction ofrotation of the wheel determines which of the two sig-nals leads in phase. The sinusoidal signals are ampli-fied and converted into square-wave voltages. Theseare transformed into a pulse train in such a way thatone wheel revolution corresponds to 4 X 26 = 104pulses. At the same time the decision is made as towhether each pulse should be positive or negative.Finally, a binary number ZL is produced, which is ameasure of the angular rotation of the left-hand wheelwith respect to the previous dead-reckoned fix. In thesame way a binary number ZR is produced, which is ameasure of the angular rotation of the right-handwheel.

Values for the quantities Jn and fIJn defined in equa-tions (la) and (lb) are found from the measured re-sults from the sensors. The distance travelled Jn andthe change in heading f/J': - f/J':-.lare determined withthe aid of the wheel sensors by using the equations:

(ZL + zR)nRJn =

104(2)

and

w W (ZL - zR)nRfIJn - f/Jn-l= 52S '

where R is the wheel radius and S the track width. The

(3)

heading measured by the electronic compass is de-noted by f/J~. A weighted mean of the measured valuesfor the heading is given by:

Af/J': + Bf/J~fIJn = A + B (4)

where A and B are weighting factors. Now we shalldiscuss a relatively simple way of obtaining a valuefor fIJn that is virtually free from interference.

Combining the measured values for the change inheading

We have seen how the signal from the electroniccompass and the difference signal from the wheel sen-sors can both provide a measure for the heading ofthe vehicle. The compass signal supplies an absolutevalue of the heading relative to magnetic north; thewheel signal supplies a relative value with respect tothe initial value of the heading. The signals must thusbe made comparable. Both signals are also subject tointerference because of the way in which the signalshave been obtained. The difference signal from thewheel sensors has a slow drift due to differences intyre pressure and wear, and so on. The compass signalcontains very fast fluctuations due to the magneticfields of other vehicles, steel bridges, structural metalin buildings, and tramlines. The compass signal alsohas a constant error due to the magnetic field of thesteel body of the car; see jig. 6.Fig. 7 shows how the two signals for the heading are

corrected and the interfering signals removed. In thecomputing unit C2 the successive numbers of pulses ZL

and ZR from the left-hand and right-hand wheels areconverted into a signal f/Jw, which still contains an un-wanted low-frequency component. In the computingunit C3 the effect of the car's own magnetic field iseliminated from the signals for HE cosf/Jand HE sine.This is done by calculating the equation of a circle ofbest fit in the diagram for HE cosf/J as a function ofHE sin é (see fig. 6) for a large number of measuredresults. Next, C3 calculates the signal f/Jefor the head-

Philips Tech.Rev.43,No.llfI2

ing. The values of </Jwand </Jcare each multiplied by aconstant factor to make them comparable with eachother in magnitude. The signal </Jw is then passedthrough a highpass filter and the signal (/Jcis passedthrough a lowpass filter. Both filters have the samecut-off frequency jà, The interfering components arealmost completely removed in the filters.

The special feature of our method is the continuousadjustment of the cut-off frequency Ic of the filters bythe control unit R. The interfering components in </Jwand </Jcare measured in R during a certain period oftime by subtracting the mean value from the signals.

i

Q

I•... /

". >1,'/;

. I

b

Fig. 6. a) The component HE coscp of the Earth's magnetic field HEas a function of the component HE sincp; cp heading of the vehicle.The graph is a recording, made during a journey, of measurementsmade with two magnetic transducers in the electronic compass.Theoretically the measured points should lie on a circle with centre O.The best-fitting circle, however, has M as its centre, because of theeffect of the magnetic field of the vehicle. The other deviations aredue to external magnetic fields from passing vehicles, steel bridges,electrical cables and so on. b) Photograph of such a recording, com-bined with the interior of an electronic compass with two trans-ducers (see also fig. 4).

CARIN 323

+

,--I

cpW--------_------

Fig.7. Processing the signals ZL and ZR from the wheel sensors andHE sin cp and HE coscp from the electronic compass. C2, C3, C4computing units. FU filter unit. In C2 the signals ZL and ZR are con-verted into a signal cpW for the heading. In Ca the compass signalsare converted into a signal cpc comparable with cpw. HP high passfilter. LP lowpass filter. Ic cut-off frequency of the two filters. Rcontrol unit that continuously controls the magnitude of Ic, bymeasuring the r.m.s. value of the interfering components in cpwandcpc. After addition, the resultant signal cp is periodically convertedinto values Xn and Yn for the coordinates of position, with aid of thesum signal from the wheel sensors.

The root mean square (r.m.s.) value of each of the in-terfering signals is then determined. If the r.m.s. valueof the interference in (/Jw is large, Ic is increased; if thisr.m.s. value is large for </Jc, th en je is reduced. Finallythe two filtered signals are added together. Setting thevalue of je also weights </Jwand </Jcas in equation (4).The computing unit C4 calculates coordinates of pos-ition Xn and Yn from samples of the signals and fromthe sum signalof ZL and ZR, which is a measure of thedistance travelled On. This calculation makes use ofequations (la) and (lb). The corrections described arenot executed with samples of analog signals, but arederived from calculations with binary numbers, sothat analog/digital converters are used where re-quired. The calculations are made in a special com-puter for navigation.

Improving the dead-reckoned fix from map data

Errors in the fix arrived at by dead reckoning maybe divided into systematic and random errors. Themagnitude of systematic errors is essentially predic-table, and they can be eliminated from the results ofthe measurements by calibration. Random errors can-not be eliminated by calibration.

324 M. L. G. THOONE Philips Tech. Rev. 43, No. 11/12

Once the system has been installed in a car, severalcalibrations are necessary. First the car is drivenaround a closed loop for the compass to be calibrated.The effect of the car's magnetic field is then eliminatedand the transducers are matched to each other. Thecomputer does this by fitting an imaginary circle (a re-gression circle) to a diagram resembling the one infig.6. Then the car is driven along a straight line, sothat the wheel sensors can be calibrated by setting thedifference signal equal to zero and comparing the sumsignal with the distance actually travelled. Such cali-brations mayalso be necessary after new tyres orsnow chains have been fitted.

Calibration is also carried out during a journey.The sum signalof the wheel sensors is then comparedwith the length of map segments and the electroniccompass is calibrated as described above. The remain-ing errors, the ones that cannot be eliminated by cali-bration, have to be treated as random errors.

The magnitude of random errors can be estimated,since their standard deviation can be calculated with afair probability from repeated measurements. Thestandard deviations of the random errors in the meas-urements of heading and distance can be obtainedfrom a comparison of a large number of dead-reck-oned fixes with the actual position of the vehicle. Themagnitude of the random errors is expressed by thedimensions of the VLPA, the Vehicle Location Prob-ability Area. In theory this area is bounded by an el-lipse whose dimensions are given by rules for thepropagation of random errors.

The size of the VLPA increases continuously dur-ing the journey. The increase is abruptly interruptedwhen a position correction is made, where a positioncorrection is a correction of the dead-reckoned fix, as-sociated with a reduction in the uncertainty of posi-tion. A position correction can be made when thecomputer has recognized the pattern of a number ofsuccessive fixes by comparing them with topograph-ical information in the main memory. The procedureused, which will now be described, requires a consid-erable amount of computation. To keep this withinreasonable bounds, the VLPA ellipse is approximatedby its smallest enclosing rectangle.

Fig. 8 shows how the comparison of map data anddead-reckoned fix is made. Whenever a dead reckon-ing is made, the computer finds out which segments lieinside the VLPA rectangle. A record of these seg-ments is kept in the form of 'trees' of possible routes,as illustrated in fig. 8b. The lower-case letters in thetrees refer to segments whose direction does not reallycorrespond to the heading of the vehicle. If such asegment lies outside the VLPA in the next dead reck-oning, it is removed from the tree. A position correc-

2 A-bIC

3 AIC-eID-f

, A

4 AI

/C"o EI IF G

5 AI

/C"o EI IF GI IH H

a b

Fig.8. Comparison of map data with dead-reckoned fix. a) Thecomputer finds out which map segments lie inside a 'vehicle loca-tion probability area' (VLPA). b) 'Trees' of possible routes result-ing from this comparison. Segments whose direction does not cor-respond to the heading of the vehicle are shown in lower-case let-ters. The size of the VLPA steadily increases because of the in-creasing uncertainty of position. The tree at VLPA 5 shows thatthere are two possible routes. (In reality the successive VLPAs over-lap each other, since their centres are typically spaced at a distanceof 5 m.)

tion (an update) is made if the computer can decidewith a good degree of certainty which of the possibleroutes has been followed.

Fig. 9 shows schematically the position correctionmade if the computer has recognized the pattern ofthe centres of successive VLPAs as that of the seg-ments A, C, D, F and H in fig. 8. (For clarity fewerVLPAs are shown in fig.8 than in fig.9.) The dis-tances between the segments on the map and the linesof best fit (shown red in the figure) through thesecentres give two components of a correction vector.After the position correction the VLPA is muchsmaller and grows again until the next position correc-tion is made.

In addition to position corrections there is also a'map-matching procedure'. This is a procedure inwhich, at every dead-reckoned fix, the centre of theVLPA is projected on to a segment inside it. If thereare two or more segments inside the VLPA, the choiceof the segment is then determined by factors such aswhether the segment is on the planned route or not, orthe correspondence between the direction of the seg-ment and the heading. Map matching can be used, for

Philips Tech. Rev. 43, No. 11/12

Fig. 9. A position correction after the computer has recognized theroad pattern. This pattern is compared with the routes A-C-D-F-Hand A-C-E-G-H on the map, see fig. 8, and the second route is re-jected. Best-fitting straight lines (red dashes) through the consecu-tive centres of VLPAs yield two vectors. The resultant gives the dis-placement of the centre of the last VLPA, and the positional uncer-tainty immediately becomes smaller. It grows again as the distancetravelled increases.

example, to display the correct segments, nodes andintermediate points on the LCD screen. The positionresulting from map matching is always at the centre ofthe display, which therefore always corresponds to apart of a road.

The Compact Disc as a carrier of digital map data

As mentioned earlier, we decided to use the Com-pact Disc Interactive, or CD-I, as the storage mediumfor the CARIN system. An international standard forCD-I is being prepared. The use of a standardizedmethod of storage for the information has the advan-tage that the data carriers for different navigation sys-tems are interchangeable and that savings can bemade in research and development costs. The CD-Istorage medium was developed for consumer-elec-tronics applications by the licensors of the CompactDisc Digital Audio system. A CD-I can store audioand video signals of different quality levels, combinedwith computer data. The proposed CD-I standardexpresses detailed agreements relating to the storageof computer data, coding and error-correction sys-

CARIN 325

tems, the operating system and the connection ofperipherals. This means that discs and systems fromdifferent manufacturers can be compatible.

In addition to digital map data, which will occupymost of the disc, a CD-I for geographical applicationswill shortly contain applications software. Examplesof this are parts of the navigation software, the route-planning program and the driver-guidance program.Programs can also be included that will make theCD-I with its player in the home into a kind of inter-active atlas. The disc can also be used for teachinggeography, if the topographical information is aug-mented by photographs, drawings and even soundand - with a few restrictions - moving video pic-tures. Depending on the options available with thehardware, it will soon be possible to plan a route athome, study it on a video display and take a printedcopy.

In the development of a geographical CD-I, use ismade of a topographical database. The rules to beobserved in such a database are also laid down. Thedatabase will be regularly updated by publishers of(electronic) geographical data, and it can also be usedfor other applications, e.g. for town and regionalplanning, for keeping records of traffic accidents andfor processing census data. The part of the databaseintended for use by CARIN is given the correct struc-ture by a conversion program. The file thus obtainedis combined with application software and possiblywith other data. The result is then formatted to theCD-I standard. In simple terms, this means that thedata is divided into blocks of 2 kilobytes (16 384 bits),each with a 'header', with synchronization bits andwith bits for later error correction. After coding andmodulation (EFM: eight-to-fourteen modulation) thedata is in the correct form for writing to the disc [71.

The blocks are converted into sectors, and the addressis included in the header.

The result of these operations is that the successivesectors form a sequence of 'channel bits'. Each' l' inthe sequence of channel bits is a land/pit or pit/landtransition in the spiral track of pits on the final disc [81.

The pattern of pits and lands is transferred to a photo-sensitive layer on a rotating disc with the aid of alaser. Development of this layer produces a 'master',and the moulds for manufacturing the discs are madefrom the master in a number of intermediate steps.As noted earlier, the storage capacity of a Compact

Disc is 4800 Mbit. The main memory of the compu-

[7] J. P. J. Heemskerk and K. A. Schouhamer Immink, CompactDisc: system aspects and modulation, Philips Tech. Rev. 40,157-164, 1982.

[8] M. G. Carasso, J. B. H. Peek and J. P. Sinjou, The CompactDisc Digital Audio System, Philips Tech. Rev. 40, 151-155,1982.

326 M. L. G. THOONE Philips Tech. Rev. 43, No. 11/12

ter, on the other hand, has a capacity of only about8 Mbit. While this is still a very respectable capacityfor a semiconductor memory, it is small comparedwith that of a Compact Disc. The time required forreading all the information from a disc is about anhour. The player used in the CARIN system is anordinary Compact Disc player for music playback incars. Moving the optical pick-up in the player fromthe smallest radius to the largest radius of the usefularea on the disc takes one or two seconds. It will takeless time in the later generations of players. It is notpossible to determine beforehand where exactly thepick-up will arrive after a movement. Since it is notpossible to read 'backwards' when the pick-up has tobe moved to a particular address on the disc, the datareading must start well before the address is reached.This means that extra information has to be read be-fore the correct address is found, and this takes time.(This extra information is not transferred to the mainmemory, of course.)

As explained, the access time is not negligible, andso the data required for navigation cannot be readfrom the Compact Disc at the exact moment when it isrequired. At the start of a journey, therefore, all thetopographical information required for navigation isas far as possible retrieved, read and stored in themain memory. Since this must be done fairly quickly,it makes sense to store the data on the Compact Discin the most advantageous way.

The topographical information is divided into 'par-cels'. The information in each parcel is stored as a'block' (or 'bucket') of data on the Compact Disc andis distributed over an integer number of successivesectors. Also stored on the Compact Disc is a list ofaddresses, or 'directory'. This includes the address ofthe first sector where a block is stored, the number ofsectors used for the block, and the coordinates of thelines that define the corresponding parcel on the map.A request for map data from the computer generallymeans that map data is called for a rectangular area;see fig. JO. The directory is then consulted to find outwhich blocks contain information about the requestedarea. The information can then be retrieved fromtheir addresses.

The time required for these operations can be shortif two requirements are satisfied. The map must bedivided into convenient parcels, and the correspond-ing blocks should be stored on the Compact Disc in asequence that does not require too much movement ofthe pick-up. The first requirement means that all theblocks should contain about the same amount ofdata, which must not be so large that the blocks re-quested for a particular area take up too much spacein the main memory. Nor, on the other hand, should

Fig. 10. Partitioning of the map into 'parcels' - the white and greyrectangles. When the computer requests map data from the rec-tangle shown red, a list of addresses (the directory) supplies thelocation of the sectors on the Compact Disc where the nine blockswith information about the required nine parcels are stored.

there be so many blocks that too much space is re-quired for the directory or that too many movementsof the pick-up are necessary. The second requirementmeans that parcels adjacent on the map should corre-spond to blocks close to each other on the disc. Theseproblems will be discussed at greater length in the nextsection.

Partitioning and arranging the map data

A map should be partitioned into parcels in such away that the data content in each corresponding blockon the disc is about the same. Parcels will therefore besmaller in towns than in rural areas. Several partition-ing algorithms are known from the literature. Thealgorithm we have used is known as the 'region quadtree' [91. In this algorithm the map is continuouslysubdivided into four rectangles until the informationcontent per parcel is smaller than a preset value; seefig. l l a. We have adapted the algorithm so that, afterthe repeated partitioning, some pairs of adjacent par-cels are joined together again to make the data con-tent of the sum no more than slightly larger than thepreset value; see fig. 11b.The problem of arranging the data blocks resulting

from the map partitioning on the disc is equivalent to

[9] H. Samet, The quadtree and related hierarchical data struc-tures, Comput. Surv. 16, 187-260, 1984.

[10] G. Peano, Sur une courbe, qui remplit toute une aire plane,Math. Annalen 36, 157-160, 1890;E. A. Patrick, D. R. Anderson and F. K. Bechtel, Mappingmultidimensional space to one dimension for computer outputdisplay, IEEE Trans. C-17, 949-953, 1968.

Philips Tech. Rev. 43, No. 11/12

I HlI I

b

IIII

-----

I --HI II I

II II I

:II

II

Fig.n. Map partitioning with the 'region quad tree' algorithrnl'".a) Repeated partitioning into four rectangles until the data contentin each block that corresponds to a parcel is smaller than a presetvalue. b) Some of the pairs of adjacent parcels are then joined sothat the block has a data content never more than slightly largerthan the preset value. The parcels formed from these mergers areshown grey.

translating a two-dimensional structure into a one-dimensional structure. To solve this problem we used'space-filling curves' whose construction was de-scribed in 1890 by Giuseppe Peano [101. A space-fillingcurve passes through each point in a space once only.In our case we are working in a plane, and Peano'sconstruction is based on a square network of points;the result is called a Peano curve. Points close to each

CARIN 327

Q n N

b

5

Fig. 12. Generation of a Peano curvellPl. a) The two possible seedcurves for four points of a rectangular network. b) A PeanoNvcurve, generated from the seed curve on the right. The curvepasses once through all 64 + 2 points of the network. c) The PeanoNscurve that connects the parcels of fig. lie and starts at S. In thecase of a parcel formed from a summation, the centre of the firstoriginal parcel is taken; the other parcel is then ignored.

other in sequence along the curve are also generallyclose together in the plane. The converse is also usual-ly true. A Peano curve is obtained by starting from a'seed curve', see fig. 12a. The two curves drawnthrough four points of the network are the only pos-sible seed curves of these dimensions - apart fromrotations or reflections. A Peano curve with the left-hand seed curve is sometimes referred to as a Hilbert

328 M. L. G. THOONE Philips Tech. Rev. 43, No. 11/12

curve, and a Peano curve with the right-hand seedcurve is known as a Peano-N curve. Fig. I2b showshow a Peano-N curve is generated. Beginning with thefour points at the lower left, blocks of 16, 64, 256points and so on are connected up.

The Peano curve is particularly suitable for arrang-ing the parcels resulting from region quad tree parti-tioning. Fig. I2c shows how a Peano-N curve is gen-erated for the map partitioning in fig. 11b. The curveconnects the centre-points of the different parcels onthe map, with S as the starting point. First of all an at-tempt is made to describe an 'N' that contains the par-cels whose area is a quarter of that of the map; theseare the first-degree parcels. When such a parcel hasbeen partitioned, an 'N' is first described through thecentres of the parcels of the second degree, and so on.When two parcels of a particular degree have beenjoined together the process continues as if they hadnot been joined together. The centre of the first orig-inal parcel is taken and that of the other one is nowignored. The process is applied consecutively to par-cels of increasing degree, i.e. with a smaller and smal-ler area.It can be seen from fig. I2c that the small parcels in

the 'town area' to the right of the centre are closetogether on the Peano curve. Arranging the corre-sponding blocks on the Compact Disc in the samesequence as the points on the Peano curve thereforehas the intended result that the retrieval of informa-tion about adjacent areas requires little or no move-ment of the pick-up. The algorithm we finally chosefor arranging the blocks works in a slightly differentway, but the result is the same - a relatively shorttime for retrieving them.

Aspects of data management

In the previous section wesaw how the data blocks thatcorrespond to the parcels arestored sequentially on a Com-pact Disc. For efficient man-agement of the informationread from the Compact Disc itis most important that the in-formation is efficiently struc-tured inside each block. As wesaw at the beginning of this ar-ticle, the topographical infor-mation is built up from the co-ordinates of nodes and inter-mediate points. A sequence ofpoints starting and ending at anode, together with the inter-

mediate points on the roadconnecting the nodes, forma chain. The data management works with the con-cepts of a O-cell, which is a node, a I-cell, which is achain, and a 2-cell, which is the area inside a numberof connected chains. While navigating it is sometimesnecessary to retrieve information from the main mem-ory about the chains that surround a 2-ce11.This couldbe necessary if the car is no longer on a road on themap, e.g. on a rough track, in a car park, or on privateland. The information about the chains correspon-ding to a 2-cell on the right or left of the I-cell wherethe car was last travelling is then requested by thecomputer. As soon as the car leaves the area of the2-cell, the computer can usually determine the l-cellwhere the car is located. In the pattern-recognition

6

key information pointer to adjacent I -cell

I xxxx c

2 xxxx b

3 xxxx h

4 xxxx h

b

start node end node

key information pointer thread- pointer thread- 2-cell left 2-cell rightpointer pointer

a ~ I c - -4.. b A D

l2..F-- xxxx 2 ~ _3 ~..Á!Y A B

c xxxx I ..W 4 .......<.,.g D A

d xxxx 2 'Jfa y "'W"'\ B D

e xxxx 6 d V5 Cf)« B D

f xxxx 3 b / 5 g C B

g xxxx 4 \ h JI 5 e D C

h xxxx 4 W 3 f C A

Fig.13. a) O-cells, I-cells and 2-cells; see the caption to fig. 3. b) Schematic representation of thelist of O-cells forming part of the directory on the Compact Disc. c) List of I-cells. A separate listfor 2-cells is unnecessary, because the thread-pointers in the fourth and sixth columns pointunambiguously to the I-cells that form the boundary of a 2-cell to the left or right of a I -cell. Thethread-pointer at the start or at the end node is the I-cell calculated clockwise from the I-cell thatcontains the node. For I-cell b the figure shows the method of finding the I-cells of the 2-cell onthe right (B, shown red) and the 2-cell on the left (A, shown blue). For more details, see text.

Philips Tech. Rev. 43, No. 11/12

procedure described above it mayalso be necessary torequest the chains for a 2-cell.

Let us now consider the structuring of the informa-tion inside a block. The addresses in the memorywhere the data for 0-, 1- and 2-cells are stored arelinked by means of 'lists' . Fig. 13 shows two such lists,which relate to the parcel shown schematically infig. 3. (For simplicity, this parcel does not include anyroads that cross its boundaries.) The lists use ref-erence addresses called 'pointers' and 'forwarding ad-dresses' called 'thread-pointers', which link the vari-ous memory locations. Fig. 13b gives a list of theO-cellsinside the parcel. The first column contains thedesignation or 'key' of the cell, the second the infor-mation in the form of coordinates, etc., and the thirdthe 'pointer' to a l-cell that has the O-cell as the startor end node. (We shall not consider here which l-cellis most suitable for this.) Fig.I3c gives a list of thel-cells in the parcel. The columns contain from left toright the key of the l-cell, the corresponding informa-tion, the pointer and thread-pointer for the startnode, the same for the end node, and the keys of the2-cells located to the left and right of the l-cell,The thread-pointers in the list given in fig. I3c can

be used to find out which l-cells enclose a 2-celllo-cated to the left or right of a l-cell. The thread-pointerindicates a l-cell which, as seen from the node, isclockwise from the l-cell to which the row in the list

CARIN 329

relates. The l-cells that surround a 2-cell on the rightof a l-cell in the list are found by first taking thethread-pointer of the start node of the l-cell and thenalways taking the thread-pointer that belongs to theopposite node of the l-cell indicated by the thread-pointer. The same procedure is applied to the 2-celllocated on the left of the l-cell, but starting with thethread-pointer that belongs to the end node of thel-cell, This is illustrated in fig. I3a and c for the 2-celllocated on the right (B, shown in red), and the 2-cellon the left (A, shown in blue) of l-cell b.

The structure of the data on the Compact Disc andin the CARIN main memory is built up from manysuch carefully designed links. It has enabled us to in-crease the speed of data retrieval, yet without intro-ducing redundancy.

Summary. The CARIN (CAR Information and Navigation) systemplans the best route and guides the driver with spoken directionsfrom a speech synthesizer during the journey. The system does thisby making periodic fixes by dead reckoning, with the aid of wheelsensors and an electronic compass. The dead-reckoned fix is con-tinuously updated by comparing it with map data read from a CD-I(Compact Disc Interactive), which can store 4800 Mbit of digitaldata. Methods have been developed for conveniently partitioningthe map data into 'parcels' and arranging the corresponding datablocks in the most efficient sequence on the Compact Disc. An im-portant data-management aspect is the efficient way in which the in-formation is structured in each block: the minimum amounts of in-formation, which correspond to the nodes and 'chains' on the map,are linked via pointer addresses.