`
`With recent advances in wireless communications and low-power electronics, accurate position location may now be accomplished
`by a number of techniques which involve commercial wireless services. Emerging position location systems, when used in
`conjunction with mobile communications services, will lead to enhanced public safety and revolutionary products and services.
`The fundamental technical challenges and business motivations behind wireless position location systems are described in this
`article, and promising techniques for solving the practical position location problem are treated.
`
`Position Location Using
`Wireless Communications on
`
`Highways of the Future
`
`T. S. Rappaport, J. H. Reed, and B. D. Woerner, Virginia Polytechnic Institute and State University
`
`afety is the primary motivation for vehicle position
`location. In the United States, the Federal Communi-
`cations Commission (FCC) has required landline telephone
`companies to provide 911 emergency service for many years,
`and in 1994 began investigating similar services for U.S. cellu-
`lar and personal communication service providers [1]. Basic
`911 service automatically forwards any 911 telephone calls to
`a public safety agency. Enhanced 911 (E-911) service improves
`emergency responsiveness by including the caller’s automatic
`number identification (ANT) and street address information
`so that the nearest public safety agency may respond and
`return calls to the emergency caller. Today, the 911 operator
`receives very little, if any, of this information from a wireless
`caller. In fact, [1] indicates that at least one of every five 911
`calls is originated by a cellular telephone user, and 25 percent
`of these users do not know their location when placing the
`call. In June 1996, the FCC adopted a new rulemaking order
`based on [1], which requires wireless service providers to sup-
`port a mobile telephone callback feature and cell-site location
`mechanism by mid-1997, with completion required by early
`I998. For wireless E-911 services, private branch exchanges
`(PBXs) which connect wireless users to the public switched
`telephone network (PSTN) will be required to indicate the
`wireless caller’s telephone number, the base station location,
`and an estimate of the location of the caller. With these new
`requirements, public safety answering points (PSAPS) will
`have enhanced position location information on each wireless
`emergency call, and will have the option of requiring even
`more detailed position information within five years [2].
`While safety is the main motivation for wireless position
`location, other promising applications include accident report-
`ing, navigational services, automated billing, fraud detection,
`roadside assistance, and cargo tracking. Position location sys-
`tems will provide new services and revenue sources for wire-
`less carriers, greater crime—fighting capabilities for law
`enforcement personnel, and new methods for tracking people
`and parcels. Position location services will not only provide
`new consumer options and products for wireless carriers, but
`also features that could differentiate services and markets
`(i.e., differentiation between PCS, cellular, specialized mobile
`radio, and paging). Location systems will also provide wireless
`carriers and vendors who use position location the ability to
`
`charge for services based on location, within a particular city,
`cell site, or specific location such as an office, home, or car.‘
`This will allow wireless service providers to control customer
`usage by offering cost incentives that match service plans to
`the wireless infrastructure and networking resources.
`In 1991, the U.S. Transportation Research Board and the
`National Research Council defined research needs and imple-
`mentation requirements for Intelligent Transportation System
`(ITS) communication standards [3]. In 1993, the U.S. Trans-
`portation Research Board focuscd on seven unique aspects of
`ITS communications, including vehicle monitoring, highway
`automation, and traffic management systems. Specific prob-
`lems targeted for research included candidate technologies for
`on-board vehicle location and position location from wireless
`base stations [4]. Meanwhile, commercial forces in the United
`States have created nearly 100 percent coverage of analog
`mobile phone system (AMPS) and paging services, as well as
`worldwide coverage of the global positioning system (GPS).
`In the remainder of this article, we provide an overview of
`existing position location systems, followed by a survey of fun-
`damental concepts in position location, a summary of
`advanced algorithms for position location, and a discussion of
`research and future issues for ITS position location for wire-
`less systems.
`
`OVERVIEW OF EXISTING
`
`POSITION LOCATION SYSTEMS
`
`A number of position location systems have evolved over
`the years that are useful for ITS applications. More
`recently these systems have become synergistic with wireless
`Communications. Already, large shipping and trucking compa-
`nies such as Highway Master and United Parcel Service have
`location capabilities which use existing cellular systems and
`GPS. Qualcomm‘s OrnniTRACS® system provides satellite-
`based fleet management. whereas Highway Master uses the
`terrestrial cellular system. Below we provide an overview of
`some of the popular commercial position location systems and
`the communication technologies with which these systems
`work.
`
`IEEE Communications Magazine ° October 19%
`
`Ol 63-6804/9(i/$05 .00 © I996 IEEE
`
`GOOGLE 1020
`Page 1
`
`
`
`GLOBAL PosITIoNING SYSTEM
`(GPS)
`
`GPS is the most popular radio
`
`navigation aide and has overtaken
`
`virtually all other forms of radio
`
`as a signpost system, since each base
`station transmits a beacon signal on
`its forward control channel [11]. As
`part of the forward control channel
`structure, an overhead message con-
`taining a station identification num-
`ber (SID) and a digital color code
`(DCC) is sent every 0.8 s. The SID
`identifies the market covered by the
`cellular system, whereas the DCC
`and forward control channel number may be used by an intel-
`igent receiver to determine location within a cell site. When
`receivers have a priori knowledge of the location and DCC
`assignment for each base station, a standard cellular system
`nay be used as a course position locator.
`
`GPS is the most popular radio nav-
`igation aide and has overtaken vir-
`tually all other forms of radio
`navigation because of its high accu-
`racy, worldwide availability, and
`low cost. For ITS applications, a
`GPS receiver is often coupled with
`a wireless communications device
`to relay location information to the PSTN or PSAP.
`The principle behind GPS is simple, although the imple-
`mentation of this time-of—arrival (TOA) system is quite com-
`plex [5—7]. GPS uses precise timing within a group of satellites
`and transmits a spread spectrum signal to earth on L-band
`(centered at 1575.42 MHz). An accurate clock at the receiver
`measures the time delay between the signals leaving the satel-
`lites and arriving at the receiver. This allows calculation of the
`exact distance from the observer to each satellite. If three
`satellites are visible to the receiver, triangulation can be used
`to find the observer’s location. In practice, a lower-accuracy
`clock is used by the observer, and signals from a fourth satel-
`lite are used to correct receiver clock errors. The time trav-
`eled by each signal describes a sphere about the satellite. A
`receiver’s position lies at the intersection of three spheres,
`providing coordinates in latitude, longitude, and altitude.
`Currently GPS receivers can be found in quantity for
`under $200,/unit with accuracy of approximately 100 in. More
`sophisticated units, including those used by the military or
`using differential GSP, provide accuracy within a few meters.
`Prices of GPS units are dropping rapidly as production levels
`and demand increase.
`Reducing the cost of GPS receivers is the key to the suc-
`cessful deployment of GPS for ITS applications. NAVSYS
`Corp. has developed a low-cost GPS sensor called TIDGETTM
`that takes a 10 ms “snapshot” of the raw GPS sampled data
`and transmits this information via cellular radio to a remote
`site where the information and the GPS receiver location are
`determined [8]. A map database may be incorporated into the
`processing scheme to allow the position of the GPS receiver
`to be determined with as few as three satellites in view. The
`TIDGET receiver can be purchased in large quantities [or
`about $50/unit since only a partial GPS receiver is needed.
`GPS/TIDGET accuracy is being tested as part of the Col-
`orado Mayday Project with positive early results [9].
`LORAN C
`
`navigation because of its high
`
`accuracy, world wide availability,
`and low cost.
`
`GLOBAL NAVIGATION SATELLITE SYSTEM
`
`The Global Navigation Satellite System (GLONASS), an ini-
`iative by the Russian government to provide a similar system
`to GPS, is in its final stage of development [12]. Although the
`system uses principles similar to GPS, its operation differs in
`several aspects. The synchronization period for GLONASS
`akes only 1/3 as long as GPS, typically under a minute. The
`integration of GLONASS and GPS receivers offers a synergis-
`ie combination to substantially reduce position errors [13].
`AUTOMATIC VEHICLE MONITORING
`
`Automatic vehicle monitoring (AVM) systems provide posi-
`tion location capabilities for handling large numbers of vehi-
`cles. Typical applications include fleet management, vehicle
`security, and emergency services. AVM systems have been
`available in the United States for a number of years, starting
`in 1968 as experimental systems, continuing in 1974 under
`temporary FCC rules, and in 1995 under permanent rules that
`recognize the new technologies and new ITS services provided
`by AVM [14]. In 1995, the FCC changed the name of these
`systems to “location and monitoring services (LMS).” In the
`United States, the primary band for LMS is the 902—928 MHZ
`industrial, scientific, and medical (ISM) band, although LMS
`is supported to a lesser extent in several bands below 512
`MHz. LMS systems are licensed systems with up to 300 W
`peak power for the forward link; however, they share the
`band with low-power unlicensed devices, such as cordless
`phones, wireless local area networks, and utility meter-reading
`systems. The band is also used by federal government radiolo-
`cation systems and amateur radio operators, so the prospect
`of interference between L.\/IS and other users of the spectrum
`is an issue in the deployment of LMS systems [15].
`CELLULAR GEOLOCATION
`
`Cellular geolocation uses principles described in the next sec-
`tion, and relies on the existing infrastructure of cellular base
`stations. Geolocation offers position estimates of mobiles as
`they transmit over standard cellular frequencies. This method
`was demonstrated by Raytheon E-Systems (Falls Church, Vir-
`ginia) in the Cellular Applied to IVHS Tracking and Location
`(CAPITAL) project in Northern Virginia [16, 17]. Other ven-
`dors such as KSl (Annandale, Virginia) and Associated Com-
`munications Corp. (Bala Cynwyd, Pennsylvania) are also
`working on this approach.
`Geolocation offers some advantages to GPS since it con-
`centrates cost at each base station and allows position location
`to be performed without the need of GPS at the mobile.
`Thus, standard cellular phones, including handheld portahles,
`may be tracked. Service providers may also use geolocation to
`accurately determine capacity needs for a particular region,
`and may adapt the network accordingly. This approach sup-
`
`IEEE Communications Magazine - October 1996
`
`Loran C, developed in the 19505 by the US. Department of
`Defense, operates in the low frequency (90—l 10 kHz) band
`and uses a pulsed hyperbolic system for triangulation. It has
`repeatable accuracy in the 19-90 in range and is accurate to
`about 100 m with 95 percent confidence and 97 percent avail-
`ability. Like GPS, its performance depends on local calibra-
`tion aud topography. The system offers localized coverage to
`the United States and selected countries [10]. GPS has
`replaced Loran C in most applications.
`SIGNPOST NAVIGATION
`
`Signpost Navigation employs a large number of simple radio
`transmitters to accurately determine position at a mobile.
`These transmitters are spaced along highways and typically
`serve as coded beacons, where the code designates the lati-
`tude and longitude of the signpost. The transmitter signal
`strength indicates the relative position of the receiver to the
`transmitter. This navigation aid works well for limited areas
`such as a small city. While not originally designed as such,
`today’s AMPS analog cellular radio system may actually serve
`
`GOOGLE 1020
`Page 2
`
`
`
`ports an F.-911 implementation
`that is compatible with any exist-
`ing mobile phone, and the posi-
`tion location information may be
`used simultaneously for vehicle
`traffic management,
`incident
`detection, and wireless network
`management.
`In the CAPITAL operational
`test, geoloeation equipment is
`located at selected cellular towers
`to collect phone usage statistics
`and to gcolocate phones on desig-
`nated roadways. In this applica-
`tion, it has been shown that traffic
`monitoring via cellular has several
`advantages over conventional ITS traffic monitoring tech-
`niques such as buried magnetic sensors or video cameras.
`These advantages include lower cost as compared to magnetic
`loop-based approaches, high reliability and low maintenance,
`and no disruption of road service for installation or repairs.
`The CAPITAL system geolocates the target mobile by
`monitoring (at base stations) the reverse voice channel or
`reverse control channel transmissions from the mobile user.
`Multiple base stations receive the mobile signal, and the tar-
`get position is determined by combining angle of arrival
`(AOA) estimates from each base station and time difference
`of arrival (TDOA) estimates between multiple base stations.
`AOA measurements at each base station are made using an
`adaptive array and a variation of the maximum likelihood
`techniques described in [18, 19] (discussed later). Signal time
`of arrival data are measured at each base station and time-
`stamped with a GPS time reference to determine TDOA posi-
`tion estimates. The impact of multipath is minimized by using
`highly directional adaptive antennas that offer spatial filtering
`[17, 20, 24]. However, it is still necessary to do additional pro-
`cessing to sort multipath components from direct components
`and to identify interfering components. Experimental results
`showed that position estimates were, for the most part, within
`100 m of the true location, and within the accuracy proposed
`for E-911cellular service [16, 17]. Positions are typically fixed
`in less than a second, which is faster than a typical GPS con-
`figuration. The technology also
`works for a variety of cellular
`standards such as AMPS, nar-
`rowband AMPS (N-AMPS),
`and US. digital cellular (USDC).
`Furthermore, cellular and per-
`sonal communication services
`(PCS) service providers are
`likely to use adaptive arrays in
`the future to increase system
`capacity. Thus, position location
`may become a natural by—prod-
`uct of future wireless systems.
`
`\“~--:§f3y3)
`c
`
`tion systems may be unilateral or
`multilateral. In a unilateral sys-
`tem, a mobile unit forms an esti-
`mate of its own position based on
`signals received from transmitters
`at known locations. The GPS and
`the three-point problem from sur-
`veying are classic examples of uni-
`lateral systems [7, 2l—23]. In a
`multilateral system, an estimate of
`the mobile location is based on a
`signal transmitted by the mobile
`and received at multiple fixed
`base stations. Most cellular geolo-
`cation proposals are multilateral,
`where the estimate of the mobile’s
`position is formed by the network, rather than by the mobile
`itself.
`Position location via wireless can be accomplished by two
`general methods: the AOA method and the time of arrival
`(TOA) method. AOA, also called direction of arrival (DOA),
`has been used widely in surveying, radar tracking, and vehicle
`navigation systems [22—24]. The location of the desired target
`in two dimensions can be found by the intersection of two
`lines of bearing (LOBs), each formed by a radial from a base
`station to the mobile target. A single measured angle forms a
`pair of LOBs and provides the target location. Instead of
`using the intersection of just two lines, many pairs of LOBs
`are used in practice, and highly directional antennas are
`required, making AOA difficult at the mobile. As shown in
`Fig. 1. AOA methods may use three base stations located at
`points (A, B, C), and two measured angles to deduce the loca-
`tion of the target at the point of intersection of two circles.
`This method, known as “resection” or triangulation, may be
`solved using trigonometry or analytic geometry, or through
`table lookup [22].
`For radio frequency (RF) signals, AOA is usually deter-
`mined at a base station by electronically steering the main
`lobe of an adaptive phased array antenna in the direction of
`the arriving mobile signal. Typically, two closely spaced anten-
`na arrays are used todither about the exact direction of peak
`incoming energy to provide a higher-resolution measurement
`of the AOA. The many adap-
`tive algorithms to accomplish
`this steering are discussed in
`the folowing section. AOA is
`applied to the problem of direc-
`tion finding (DF), where the
`target attempts to locate the
`direction of fixed sensors in
`order to obtain a position fix,
`often using high-resolution spa-
`tial analysis tcehniques that
`have been developed [24—27].
`The second primary method for
`determining position location is
`with TOA measurements [28].
`Since electromagnetic waves
`propagate at the constant speed
`of light (c = 3 X 108m/s), or
`approximately 1 ft/ns in a free
`space medium, the distance
`from the mobile target to the
`receiving base station is directly
`proportional to the propagation
`time. If the signal propagates in
`time t,- from the target transmit-
`
`F2—R1
`
`3_R1
`
`I Figure 1. The three-point problem, also known as
`triangulation. Three fired beacons (A, B, C) provide
`signals which allow the mobile to determine its loca-
`tion. The mobile must have an accurate method of
`measuring angles.
`
`POSITION LOCATION
`
`FUNDAMENTALS
`
`The primary function of a
`position location system is
`to locate the coordinates of a
`desired mobile user (called the
`target) with respect to a set of
`objects (base stations) with
`known positions. Position loca-
`
`IEEE Communications Magazine ' October 1996
`
`I Figure 2. 2-D hyperbolic position location. solution. Two
`hyperbolas are formed from TDOA measurements at
`three fixed receivers to provide an intersection point which
`locates the targrt source. S1, S2, and S3 represent the fired
`receiver locations, and Eq. (2) is used to determine the
`two hyperbolos.
`
`GOOGLE 1020
`Page 3
`
`
`
`ter to the ith fixed receiver, then the receiver lies at
`range R,-, where
`
`(1)
`Rt = Ch‘
`Therefore, if a free space signal arrives at a base
`station receiver 10 us after it is transmitted, the tar-
`get transmitter must lie on a sphere of radius 3000
`m from the base station. If TOA measurements are
`made at a second base station at a second location,
`the target position can be determined to lie on a
`circle since the intersection of two spheres is a cir-
`cle. The thrce-dimcnsional position of a transmitter
`is uniquely determined by the intersection of three
`spheres using TOA measurements from three base
`stations [28, 29].
`In general, direct TOA results in two problems.
`First, TOA requires that all transmitters and
`receivers in-the system have precisely synchronized
`clocks (e.g., just l us of timing error could result in
`a 300 in position location error). Second, the trans-
`mitting signal must be labeled with a timestamp in
`order for the receiver to discern the distance the
`signal has traveled. For this reason, TDOA mea-
`surements are a more practical means of position
`location for commercial systems [30].
`The idea behind TDOA is to determine the rela-
`tive position of the mobile transmitter by examining the dif-
`ference in time at which the signal arrives at multiple base
`station receivers, rather than the absolute arrival time. There-
`fore, each TDOA measurement determines that the transmit-
`ter must lie on a hyperboloid with a constant range difference
`between the two receivers. The equation of this hyperboloid is
`given by
`
`3 38.83-77.31
`
`:
`-77.29 -77.28 -77.27
`
`_L_
`|
`-77.26 -77.25
`
`I
`-77.2’!
`
`i
`-77.23
`
`-77.3
`
`-77.22
`
`ifi Figure 3. Geo/ucatioh 0/at cellular signal based on TDOA hyperbolas
`(gold) and lines of bearing (orange) from three base station sites in North-
`em Virginia. This figure illustrates repeated attempts on the same signal,
`with the black “x” indicating the position locations. (Compliments ofJoe
`Kennedy, Raytheon/E-Systems, Falls Church, Virginia).
`
`Rm‘:\i(Xg‘X)2+(K‘y)2+(Zi’Z)2
`—[<X, —x>? +<jY_, —y>2 +<z,- —z>2.
`
`(2,)
`
`where the coordinates (X,-, Y,-, Z,~) and (Xi, Y],-, Zj) represent the
`fixed receivers i andj, and make up the unknown coordinate
`of the target transmitter [31]. If the source and all receivers
`are coplanar, a two-dimensional source location can be esti-
`mated from the intersection of two or more independently
`generated hyperboloids generated from three or more TDOA
`measurements, as shown in Fig. 2. Three-dimensional source
`location estimates require at least four independent TDOA
`measurements.
`
`Unlike TOA measurements, the transmitted signal need
`not contain a timestamp, and TDOA measurements require
`only that the fixed location receivers have precisely synchro-
`nized clocks. This corresponds to the timing standards already
`provided at cellular base station sites, making TDOA more
`realistic than requiring each mobile unit to have an accurate
`clock. Atomic clocks, such as a Cesium time source, or a GPS
`receiver clock are typically used for timing at base stations.
`It is possible to combine TDOA and AOA techniques into
`hybrid systems. For example, the position location system
`developed by Raytlieon E-Systems for the CAPITAL project
`employs both techniques (Fig. 3) [17]. These two main classes
`of position location systems may be supplemented with dead-
`reckoning or inertial navigation techniques. Dead—reckoning
`can be particularly useful when buildings and terrain obscure
`linc—of-sight propagation between a transmitter and receiver.
`In this case wheel rotation sensors, which measure distance
`traveled, or inertial navigation systems using gyroscopes are
`used to update the location from the last previously known
`position until a new position fix can be obtained.
`
`The basic position location techniques described above
`work well if the signals are not corrupted by noise, multipath,
`or interference. In practical systems, however, position errors
`occur due to imperfections ir1 the channel [32]. VVhile both
`AOA and TDOA techniques require a minimum of two or
`three base stations to determine a unique position location,
`new cellular or PCS systems are often designed to ensure only
`one high signal-to-noise ratio (SNR) link between a transmit—
`ting mobile and a base station. This is because in a conven-
`tional cellular system, base station count (infrastructure cost)
`and interference between adjacent cells must be minimized
`when first deploying the system [1 l]. The ability of multiple
`base stations to hear the target mobile is paramount to the
`design of position location systems. This problem is referred
`to as hearability, and it is where the design philosophy of mul-
`tilateral position loeation systems diverges from that of wire-
`less communications systems. Hearability is more of an issue
`in rural cellular systems, where coverage issues dictate the sys-
`tem design rather than capacity demands, which leads to more
`base stations and redundant coverage in cities [34].
`Both AOA and TDOA techniques also rely on a direct
`line—of—sight path from the transmitting mobile to the base sta-
`tion receivers.
`Howcver, both urban and mountainous rural environments
`induce significant path blockage and multipath time disper-
`sion due to reflections from and diffraction around buildings
`and terrain. Multipath components may appear as a signal
`arriving from an entirely different direction, and can lead to
`catastrophic errors in an AOA system [33]. Although diffrac-
`tion around a building may have less severe consequences for
`the relative TOA of a signal in a TDOA system, multipath
`reflcetions from distant objects can lead to time distortions of
`several microseconds [11]. Because of their ability to resolve
`and reject multipath, wideband spread—spectrum systems and
`directional antennas will offer advantages for position location
`in a multipath environment [20].
`Real~world channel impairments require special processing
`techniques to improve the resistance of both ADA and
`TDOA methods to noise, multipath, and interference. It is
`often advantageous to use more than the minimum number of
`TDOA or AOA receivers for a unique solution. in order to
`
`[HEP Communications Magazine - October 1996
`
`GOOGLE 1020
`Page 4
`
`
`
`lncident plane
`wave
`
`I Figure 4. Illustration ofa plane wave incident on a linear equi-
`spaced array. The dotted lines represent the phase fronts of the
`incident wave.
`
`average out errors induced by the radio channel. While this
`affords improved performance by combining additional infor-
`mation, it is usually impossible to obtain a single consistent
`solution in this overdetermined case. As a result, processing
`algorithms must be capable of combining many noisy and
`inconsistent measurements. These algorithms are discussed in
`greater detail in the next section.
`
`ADVANCED ALGORITHMS FOR
`POSITION LOCATION
`AOA ALGORITHMS FOR POSITION LOCATION
`
`This section presents an overview of some of the more popu-
`lar methods for estimating the AOA of a signal impinging on
`an array of antenna elements. Because a vast body of litera-
`ture exists on these direction finding (DF) methods, the dis-
`cussion here is kept brief, with emphasis placed on those
`methods most applicable to the cellular/PCS radio environ-
`ment. More details on the algorithms discussed here may be
`found in the cited references. In particular, the overviews
`given in [24, 61] are excellent sources of background informa-
`tion on the problem of AOA estimation.
`In general, an angle of arrival estimate is made from a
`base station using a directional antenna such as a phased
`array of two or more antenna elements to measure the AOA
`of the incident signals (Fig. 4). In general. the sensor (e.g.,
`antenna element) spacing used in an AOA measurement is on
`the order of half the wavelength of the signal carrier frequen-
`cy. The relatively close spacing of the antenna elements allows
`the time delay seen by a signal as it propagates across the
`array to be modeled as a phase shift. This is referred to as the
`“narrowband model,” and is assumed to be appropriate in the
`development of most AOA estimation algorithms.
`The accuracy of the narrowband model is dependent on
`the signal bandwidth, the antenna element spacing, and the
`quality of the receiver hardware. The narrowband model is
`only accurate if the signals received at each antenna element
`are processed (filtered, downconverted, sampled, etc.) in an
`identical manner. This means that each channel of the receiv-
`er (RF front-end for each antenna element) must have nearly
`the same frequency response, be highly linear. and use the
`same oscillators for all mixing and sampling operations. This
`type of receiver is generally known as a coherent receiver. The
`receiver is a major contributor to the cost of an AOA estima-
`tion system, where the cost increases as the number of anten-
`na elements (and hence number of receiver channels)
`increases. Therefore, it is highly desirable to keep the number
`of antenna elements in the array to a minimum. The number
`of antenna elements needed in the array is strongly dependent
`on the signal environment and the specific AOA estimation
`algorithms employed. A critical assumption made for most DF
`
`IEEE Communications Magazine ° October 1996
`
`techniques is that the number of incident signals is strictly
`less than the number of antenna elements. As discussed later,
`this requirement can be relaxed if properties of the incident
`signal are exploited; if, for example, it contains a known
`training sequence, or the sequence can be estimated. It
`should be noted that implementation of adaptive beamform-
`ing requires the same type of coherent receiver. Therefore, if
`smart antennas (i.e., adaptive phased arrays) are deployed at
`the base station, AOA estimation can be incorporated with
`modest additional signal processing, and such antenna sys-
`tems show great promise for emerging high-capacity wireless
`systems [20]. The array must be carefully calibrated over all
`measured angles, as well as frequency and temperature. This
`is an expensive operation, in terms of the cost of both compu-
`tational storage and periodically performing the array calibra-
`tion.
`The most straightforward AOA estimation approach is
`phase interferometry. A phase interferometer directly measures
`the phase difference between the signals received on multiple
`pairs of antenna elements and converts this to an AOA esti-
`mate. This approach works quite well for high SNR but will
`fail for strong co-channel interference and/or multipath.
`Another conceptually simple approach is beamforming.
`This method can be viewed as measuring the output power of
`a beamformer while steering the main—beam of the array over
`the angular field of interest. This yields a tr11e spatial spec-
`trum, that is, an estimate of power distribution versus AOA.
`A diagram illustrating the concept of beamforming is shown
`in Fig. 5. The beamformer weights w,, control the spatial
`response of the beamformer. Capon’s method is closely relat-
`ed but has better angular resolution [35]. However, neither of
`these methods work well in coherent multipath.
`Methods that work well in multipath can be derived using
`the maximum likelihood (ML) framework [19, 24, 36]. Differ-
`ent ML algorithms are obtained by making different assump-
`tions about the incident signals. This leads to the so-called
`deterministic and stochastic ML methods. In multipath envi-
`ronments the MI. methods will estimate the AOA of each
`path. However, implementation of these methods requires a
`complex multidimensional search. The dimensionality of the
`search is equal to the total number of paths taken by all of the
`received signals. This search is further complicated by the fact
`that the total number of paths is not known a priori and must
`be estimated.
`Another class of methods that will work well in multipath
`can be derived by combining spatial smoothing with subspace-
`based algorithms. Examples of subspace methods include
`MUSIC [25] and ESPRIT [27, 37]. Normally these methods
`fail in multipath, but using a spatially smoothed covariance
`matrix in place of the conventional one allows them to oper-
`ate properly [38]. Spatial smoothing methods have been com-
`bined with propcrty-exploiting adaptive beamforming methods
`which estimate the spatial signature directly [38]. Estimating
`AOA from spatial signature vectors has several advantages
`over estimating AOA directly from the observed data. The
`principal advantage is that the search is reduced from one
`where the AOA of all paths must be estimated to one where
`only the paths contributing to the estimated spatial signature
`of the desired signal must be estimated. Another advantage is
`that more multipath components can be processed by a fixed
`array [33].
`A class of ML methods with very useful properties can be
`derived by assuming that the incident signals are known rather
`than unknown stochastic processes [39—41]. This allows
`exploitation of, for example, the training sequences that exist
`in most digital cellular standards. An interesting application of
`these methods to the problem of estimating the AOA of code-
`
`GOOGLE 1020
`Page 5
`
`
`
`Figure 5. Illustration of the beamforming
`structure for a phased array omteima.
`
`division multiple access (CDMA) sig-
`nals was recently proposed in [42]. All
`of the AOA estimation algorithms pro-
`posed so far assume that the number
`of antennas in the array exceeds the
`number of eo—channel signals. This is
`clearly not practical for CDMA where
`the number of co-channel signals is
`very large; therefore, none of the ADA
`estimation methods discussed above
`are applicable to CDMA. However, by
`assuming that the CDMA signal may
`be demodulated with low bit error rate
`(BER), an estimated waveform may be
`substituted for the known waveform.
`One implementation of this approach
`uses the despread soft decisions from
`each antenna together with the hard decisions made by the
`existing CDVIA demodulation process. Result