Considerations and Challenges in Real Time Loating Systems Design Dr. Brian Gaffney DeaWave Ltd. Email: brian.gaffney@deawave.om Abstrat Real Time Loating Systems (RTLS) are a ombination of hardware and software that are used to ontinuously determine and provide the real time position of assets and resoures equipped with devies designed to operate with the system. There are many appliations alling for RTLS, partiularly now that it has beome affordable and mobile wireless devies have beome small and onvenient. This paper disusses some of the possible tehnologies and algorithms to be onsidered when designing these systems and highlights some of the important hallenges faed by this industry. I. INTRODUCTION Real Time Loating Systems (RTLS) are a ombination of hardware and software that are used to ontinuously determine and provide the real time position of assets and resoures equipped with devies designed to operate with the system. Future RTLS systems are envisioned to be based on low power eletroni tags used to trak and/or monitor assets, people or anything of value with very high auray and mobility. For example, a urrent appliation of RTLS is asset traking in hospitals, where valuable equipment an be instantly loated anywhere in the area overed by the network by a entral loation engine stored on a entral server. The market for RTLS systems is expeted to grow to 2.7 billion dollars in 2016 [1]. And new innovative RTLS and Loation Based Servies (LBS) tehnologies should even inrease this. However, urrent RTLS tags suffer from two major drawbaks. Wideband based tags provide auray but typially are energy detetor based systems with limited range, while relatively narrowband based tags do not provide the auray some appliations require. The seond major drawbak is that the tags an be large and power hungry. The goal of RTLS design is to eliminate this ompliated tradeoff between auray, power onsumption and range. This paper is arranged as follows. In setion II, some of the possible algorithms for RTLS are disussed, with details of the desired wireless signal properties, and the advantages and disadvantages disussed. Setion III gives a detailed desription of time of flight and time of arrival ultrawideband systems. A number of the major hallenges faed by RTLS systems are highlighted in setion IV and the onlusions are given in setion V. II. RTLS ALGORITHMS Fundamentally, RTLS algorithms an be divided into three lasses: algorithms whih aim to estimate the distane between the tag and the appliation point 1, algorithms whih use arrival angle at the appliation point and algorithms whih use a ombination of both. In addition, the entral loation engine may have a priori information on the environment, suh as a detailed floorplan with obstrutions or a hannel sounding database, whih allows the algorithms to produe a more aurate estimate of a tags loation. A. Estimating the distane between a tag and appliation point A popular tehnique in RTLS is to estimate the distane between a tag whih transmits a paket and a appliation point whih reeives this paket. Using multiple appliation points, the loation of the tag an be estimated from some form of trilateration (or higher order) algorithm [2]. However, there are multiple possible system level methods of estimating this distane. The first to be disussed is the reeived signal strength (RSS) tehnique. This uses an estimate of the energy of the reeived signal to estimate the distane that the signal has traveled. By assuming some path loss exponent n, the estimate of the distane an be alulated from the following relation ( ) ˆd ˆP RX = P TX 10n log 10 PL 0 (1) d 0 where ˆP RX is the estimate of the reeived power in deibels (dbs), P TX the transmitted power in dbs, d 0 is the referene distane whih has a path loss of PL 0 dbs and ˆd the estimate of the distane. This method has some advantages. In theory, the auray is often assumed to be independent of the bandwidth of the signal. Hene, popular preexisting networks an be utilised. For example, a WiFi network an be used with additional software and WiFi based tags to produe a RTLS system. This is a huge advantage due to popular deployment of WiFi and is popular in urrent RTLS produts [3]. However, using the expression in equation 1 to obtain ˆd has two major problems whih affets auray signifiantly. The first is that the path loss exponent n is unknown. This typially an range from 2-6 in WiFi hannels and without an extremely aurate estimate of this value, the estimate of the distane an 1 In this paper, the term tag refers to the transeiver whih is to be loated and appliation point refers to the transeiver whih is a known point of referene.
have a signifiant error omponent. The seond problem is the onstrutive and destrutive interferene present in almost all wireless hannels. With a relatively small bandwidth, this interferene an ause a large variation in the reeived power over the expeted value. Tehniques exist to ombat these two issues, but to date, do not produe a mean auray less than 5 meters. The seond possibility for estimating the distane between transmitting tag and reeiving appliation point is using an estimate of the time of flight. Given an estimate of the arrival time at the reeiver ˆt rx and information on the transmit time at the tag t tx, the estimate of the distane between transmitter and reeiver an be alulated as p g (g) 0.7 0.6 0.5 0.4 0.3 0.2 0.1 20 Mhz Signal 500 Mhz Signal ˆd = (ˆt rx t tx ) (2) where is the speed of light. This tehnique has the advantage that it is muh more robust to the multipath hannel than the RSS tehnique. By assuming that there is a diret path between transmitter and reeiver and that the loks in both transmitter and reeiver are synhronised (i.e. ˆt rx and t tx are from the same referene lok), the estimate ˆd an be extremely aurate. However, neither of these of these assumptions are trivial. Assuming a diret path between transmitter and reeiver arries some important system design hoies. Firstly, in a wireless hannel, obstrutions an ause this path to be extremely heavily attenuated. Steps must be taken in order to be able to resolve this attenuated path in the presene of noise. Seondly, the diret path should be as free of onstrutive and destrutive interferene as possible. Destrutive interferene an render an already attenuated path undetetable at the reeiver. Thirdly, and most importantly, to get a very aurate estimate of the time of arrival, the transmitted signal must have a very fast leading edge rise time. Of ourse, the faster the rise time, the larger the bandwidth. All of these issues justify the growing interest of ultrawideband (UWB) signals (greater than 500Mhz) for RTLS, where the signal is extremely robust to onstrutive and destrutive interferene as shown in figure 1 (assuming that an energy detetor based reeiver is not used) and where an UWB pulse has a very fast rise time. However, low power UWB reeivers are more diffiult to implement than onventional narrowband systems due to the high sampling rates required. The seond assumption on the synhronisation of loks is of onern no matter what the bandwidth due to lok differenes whih are aused by rystal effets. This assumption will be set aside until setion III where UWB RTLS will be disussed in detail. B. Using the Angle of Arrival (AOA) to estimate a transmitters loation The seond tehnique in RTLS uses the angle of arrival to estimate the orientation of the transmitter relative to the reeiver. By measuring the differene in arrival times of a signal on the elements of an antenna array, the diretion of 0 15 10 5 0 5 10 Channel Gain, g (db s) Fig. 1. Channel gain for ultrawideband and relatively narrowband hannels for a hannel with large delay spread ( 250ns). Channel gain is defined here as the resulting energy gain due to onstrutive and destrutive interferene. ( ) AOA = os t d antenna AOA t Antenna 1 d antenna Antenna 2 Fig. 2. An angle of arrival system using two antennas. By utilising the differene in the time of a arrival of a signal at the antenna pair ( t) along with the antenna spaing d antenna, the arrival diretion an be estimated. arrival an be estimated as in figure 2. This an be viewed as the reverse of beamforming. One of the main issues with AOA systems is the need for multiple antennas (and aompanying RF front ends) at the reeiver. For arrier frequenies less than 10GHz, where most wireless systems exist, the antenna size an be quite large. Multiple antennas inrease the form of the reeiver, leaving it somewhat unattrative. In addition, the angle of arrival of a narrowband signal is largely dependent on the dominant path whih may not be the diret path. This leads to an error in the arrival angle (whih ould be anywhere from π to π from the orret arrival angle) and an lead to a signifiant error in the loation estimate. Combinations of the disussed tehniques exist, but are outside the sope of this paper whih is intended as an introdution to the tehniques. Also, interested readers are direted to a MATLAB tool set alled the Sensor Network Loalization Explorer (SeNeLex) whih demonstrates the different methods [4].
ˆd 1 ˆd 2 AP 2 AP 1 AP 3 Fig. 3. Example of using 2D trilateration for loating a tag. By estimating the distane between the tag and eah of the three appliation points (TOF/RSS), the loation of the tag an be estimated by finding the intersetion of the three irles. III. UWB FOR PRECISION LOCATING As disussed in setion II, there are multiple system design onsiderations for RTLS. In this setion UWB systems ombined with two related, but different, approahes to preision loating of tags will be disussed in detail [5]. The first is ommonly known as Time of Flight (TOF) or Time of Arrival (TOA). In this system the time of flight is measured at three appliation points 2. Let ˆτ n equal the estimate of the time of flight between the tag and appliation point n, (x n, y n ) be the loation of appliation point n (in artesian oordinates), whih are known to the CLE and (x t, y t ) be the loation of the tag, whih is the unknown of interest. Estimating the loation of the tag is equivalent to finding the intersetion of three irles, as in figure 3. These irles are given by the following set of equations (ˆτ 1 ) 2 = (x 1 x t ) 2 + (y 1 y t ) 2 (3) (ˆτ 2 ) 2 = (x 2 x t ) 2 + (y 2 y t ) 2 (4) (ˆτ 3 ) 2 = (x 3 x t ) 2 + (y 3 y t ) 2. (5) This is alled trilateration. Trilateration is a method of determining the relative positions of objets using the geometry of triangles in a similar fashion to triangulation. Unlike triangulation, whih uses angle measurements (together with at least one known distane) to alulate the tags loation, trilateration uses the known loations of two or more appliation points, and the measured distane between the tag and eah appliation point. 2 Throughout this paper, we will assume two dimensional (2D) loating. This larifies the mathematis involved with the need for only three appliation points. However, an inrease to three dimensions only requires an additional appliation point. ˆd 3 However, this algorithm requires that all the tags and appliation points have the same referene lok. Considering rystal effets, this requirement is hard to ahieve. Ideally the tags are heap and therefore expensive rystals are not an option. Synhronising the loks of all devies is possible, but in an environment with a very high density of tags, this requirement redues the network effiieny (density of the tags) signifiantly. An alternative is time differene of arrival (TDOA). Consider the TOF/TOA based algorithm and the requirement that the loks are all synhronised. At a time t 0, the tag transmits a paket, whih is reeived at times t 1, t 2 and t 3 at appliation points one, two and three respetively. the times of flight are therefore τ 1 = t 1 t 0, τ 2 = t 2 t 0 and τ 3 = t 3 t 0. The CLE has knowledge of all these times, and they are all based on the same referene lok and an therefore use a trilateration algorithm to estimate the loation (x t, y t ). Multilateration, also known as hyperboli positioning, is the proess of loating an objet by aurately omputing the time differene of arrival (TDOA) of a signal emitted from the objet to three or more reeivers. In multilateration the arrival time of a transmitted paket is observed at three appliation points, eah with a known loation. These appliation points are assumed to have the same lok. The time differene of arrival (differene in time of flight) between AP one and two is written τ 12, whih is defined as τ 12 = τ 1 τ 2 = t 1 t 0 t 2 + t 0 = t 1 t 2 (6) Next, one of the three AP s is taken as the system origin. For example, we an assume AP three is loated at (0, 0). Using the relationships defined in equations 3 to 6, we an write τ 13 = 1 (x1 x t ) 2 + (y 1 y t ) 2 1 (x3 x t ) 2 + (y 3 y t ) 2 = 1 [ (x1 ] x t ) 2 + (y 1 y t ) 2 x 2 t + y2 t (7) τ 23 = 1 (x2 x t ) 2 + (y 2 y t ) 2 1 (x3 x t ) 2 + (y 3 y t ) 2 = 1 [ (x2 ] x t ) 2 + (y 2 y t ) 2 x 2 t + yt 2 (8) Equations 7 and 8 define the hyperbola whose intersetion gives the estimated tag loation (x t, y t ). An example of this is shown in figure 4. However, the auray of both of these algorithms depends
AP 2 AP 1 AP 3 Fig. 4. Example of using 2D multilateration for loating a tag. Appliation point two is taken as the system origin and two hyperbola traed using the time differene of arrival values τ 12 and τ 32. The intersetion point of these two is the loation of the tag. on the auray of the arrival time estimates. Errors an our due to noise, harsh NLOS hannels or interferene from other tags. To ahieve sub-meter auray, UWB pulses are neessary. Their wide bandwidth allows almost all the multipath omponents to be resolved, whih allows the diret path to be resolved aurately and with a very fine time resolution. However, this potential auray requires the UWB signal to be reeived oherently, whih require high sampling rates. This results in the signifiant hallenge to UWB RTLS designers to design low power systems whih an harness the full potential of UWB. IV. CHALLENGES FOR RTLS There are numerous hallenges for RTLS. In this setion, a small subset of these will be introdued and disussed. A. Number of tags per network ell Firstly, the number of tags per network ell is an important fator. Depending on the appliation (asset traking in a warehouse or people traking in a hospital) the number of tags per network ell will vary. For some appliations, the large number of tags needed will be hallenging to the RTLS systems. The two main fators when onsidering tag density is the duration of the transmitted paket and the rate of transmission (blink rate). The number of pakets reeivable, N tags, per blink rate period is alulated from N tags = eb R T P (9) where e is the effiieny of the network protool (for example 0.184 for the aloha protool), B R is the blink rate in seonds and T P is the length of the paket in seonds. By reduing the length of the paket, the number of tags an be inreased. However, eah paket must ontain a ertain amount of information (for example, a tag identifiation number), so the length of the paket is ditated by the data rate of the system. Hene, high data rates are desirable. By inreasing either e or B R in equation 9, the number of tags an also be inreased. These two values an atually be onsidered to be losely related. If we assume some network protool whih allows a ertain onfidene in reeiving pakets (i.e., a protool whih results in a ertain perentage of ollision free pakets) would need to be employed in ombination with a blink rate whih allows the loss of pakets. As an example, if a network protool results in the loss of five perent of the pakets through ollisions, over two blink intervals only 0.25 perent of the pakets have not been reeived in either interval. If the blink interval is short enough (whih is ditated by the required traking speed), this loss maybe aeptable and results in a large inrease in the overall number of tags while retaining the traking speed. B. Long battery life An important hallenge for RTLS is designing a system whih is of low enough power to ensure that the battery lasts long enough for the required appliation. Ideally, a tag should be small enough to tag an asset unobtrusively and with a battery life of the order of years so that it does not need to be replaed. This hallenges the system design to deliver on the potential performane, while using as little power and silion area as possible. C. Channel environment However, the most signifiant hallenge to any RTLS devie (be it narrowband or UWB) is the hannel environment. Most environments result in a signifiantly degraded reeived power and impulse response as the distane between transmitter and reeiver inreases. This limits the possible range of RTLS and their performane. Inreasing the signal bandwidth an redue the hannel effets somewhat (or, at least, make them onsistent), but onrete walls and metalli objets/strutures still degrade the auray in the estimate of the loation. This presents signifiant hallenges to both system level and software design. In addition, the antenna adds distortion to the signal waveform, whih introdues a hallenge for antenna designers. V. CONCLUSIONS This paper disusses some of the important system design onsiderations and hallenges in real time loating systems. The two main methods (distane between transmitter and reeiver and angle of arrival) used in estimating the loation of a transmitting tag were introdued and disussed. This is followed by a detailed disussion of time of arrival based algorithms and their use in estimating the distane between transmitter and reeiver. Comparison of the different methods shows that eah have their assoiated advantages. RSS shemes using narrowband transeivers (WiFi) are a good solution where the infrastruture is pre-existing and battery life is not a major onern. However, it was learly shown that ultrawideband signals are essential to ahieve the sub meter auray required by many appliations. Finally, some of the major hallenges to RTLS system design were highlighted and their impat on the system design desribed.
REFERENCES [1] P. Harrop, G. Holland and R. Das, Real Time Loating Systems 2008-2018, IDtehEX report, www.idtehex.om. [2] N. Bulusu, D. Estrin, L. Girod and J. Heidemann, Salable Coordination for wireless sensor networks: Self-Configuring Loalization Systems, In Proeedings of the Sixth International Symposium on Communiation Theory and Appliations (ISCTA 2001), July 2001. [3] H. L. Moen and T. Jelle, The Potential for Loation-Based Servies with Wi-Fi RFID Tags in Citywide Wireless Networks, 4th International Symposium on Wireless Communiation Systems (ISWCS), Ot. 2007, pp.148-152. [4] The Sensor Network Loalization Explorer, http://www.ee.osu.edu/ ashj/loalization/ [5] Part 15.4: Wireless Medium Aess Control (MAC) and Physial Layer Speifiations for Low-Rate Wireless Personal Area Networks (WPANS). Amendment 1: Add Alternate PHYs. IEEE Computer Soiety, Annex D.