Joint Wireless Positioning and Emitter Identification in DVB-T Single Frequency Networks

Transcription

1 Joint Wirele Poitioning and Emitter Identification in Single Frequency Network Liang Chen, Lie-Liang Yang, Jun Yan and Ruizhi Chen Abtract Digital televiion (DTV) ignal ha been recognized a a promiing ignal for navigation and poitioning. However, due to the ingle frequency network (SFN) tranmiion within the European tandard digital video broadcating terretrial () ytem, emitter confuion problem occur in navigation and poitioning, reulting in that a receiver i unable to know from which emitter a received ignal come. In thi paper, we conider the wirele poitioning with emitter confuion problem in SFN network. A joint wirele poitioning and emitter identification algorithm i propoed, which i baed on the expectation maximization (EM) method. The propoed algorithm i teted in a cenario, where ignal are received from to 5 emitter. Simulation reult how that, relying on more than emitter ued in the tet for D poitioning, the EM aited poitioning algorithm i feaible to achieve accurate poitioning reult in the exitence of the emitter confuion problem. Our tudie how that the performance achieved by the propoed algorithm approache the Cramér-Rao bound (CRB). Furthermore, the propoed algorithm i effective to identify the DTV emitter, and the poitioning performance i robut to the emitter identification error. Additionally, our methodology i general, and can be employed for time of arrival (TOA) baed poitioning in any SFN. Index Term, ingle frequency network (SFN), expectation maximization (EM), Cramér-Rao Lower bound (CRLB), wirele poitioning I. INTRODUCTION Terretrial televiion ignal have been deigned for both indoor and outdoor reception. Recently, wirele poitioning uing digital televiion (DTV) ignal ha attracted a growing interet after the DTV ytem have been put into operation for maive uer. It ha been recognized that the DTV ignal have a range of advantage for poitioning, when compared to the Global Navigation Satellite Sytem (GNSS) []. Baed on the DTV ignal, wirele poitioning benefit from a higher tranmiion power [], larger ignal tranmiion bandwidth [], le Doppler effect and ionophere diturbance [], and lower carrier frequency, hence reulting in better diffraction performance and better receiving quality for urban and L. Chen i with State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sening, Wuhan Univerity, Wuhan 7, China and alo with Collaborative Innovation Center of Geopatial Technology, Wuhan 79, China ( l.chen@whu.edu.cn). L.-L. Yang i with the faculty of Electronic and Computer Science, Univerity of Southampton, Univerity Road, Southampton, SO7 BJ, United Kingdom, lly@ec.oton.ac.uk. J. Yan i with Nanjing Univerity of Pot and Telecommunication, Nanjing, China, yanj@njupt.edu.cn. R. Chen i with the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sening, Wuhan Univerity, Wuhan 7, China and alo with Collaborative Innovation Center of Geopatial Technology, Wuhan 79, China ( ruizhi.chen@whu.edu.cn) Manucript received December, 6; revied April, 7; accepted April, 7.; (Correponding author: Liang Chen.) indoor propagation [5]. Furthermore, it i cot effective, due to the reue of exiting DTV facilitie [6]. The principle and recent progre in reearch about the DTV ignal aited navigation and poitioning have been overviewed in detail in [7], [8]. Among all the terretrial DTV tandard, the European tandard [] ha been employed in mot countrie worldwide. In the baed ytem, multiple emitter are deployed to cover a region by tranmitting the ame ignal on the ame frequency band, which i hence called a the ingle frequency network (SFN) []. The SFN i efficient in term of the ue of frequency reource. However, in poitioning, SFN generate the emitter confuion problem that a poitioning receiver i unable to know which DTV tation it receive ignal from. The emitter confuion problem ha been addreed in ome literature [9], [], [], []. Specifically, the watermarking technique have been propoed for the emitter identification in DTV ytem, which include the ytem baed on the Advanced Televiion Sytem Committee (ATSC) tandard [9] and that on the tandard []. However, applying the watermarking technique ha to change the tructure of the tranmitted DTV ignal, which lead to extra cot for updating the whole DTV ytem. By contrat, the author of [] have propoed a revere poitioning technique, while the author in [] have propoed to ue mobile velocity information, in order to avoid the emitter confuion problem. However, thee technique are only uitable for the poitioning receiver in moving, but not uitable for the tatic poitioning receiver. In thi paper, we propoe a novel approach, namely the expectation maximization aited poitioning (EMP) algorithm, which jointly etimate a uer poition and identifie the DTV emitter. Our EMP algorithm iteratively identifie the DTV emitter and etimate the poition of the receiver, with the aid of the peudo-range meaured by the receiver baed on it received ignal from the different DTV tation. In comparion with the exiting approache [9], [], our propoed method doe not require to modify the tructure of the exiting DTV ignal. Hence, it i uitable for both tatic and mobile receiver to carry out their poitioning. Additionally, we hould note that, although our method i derived and analyzed in the context of the ytem, the methodology i in fact very general, which may be employed in any SFN baed ytem for the time of arrrival (TOA) baed poitioning. The ret of the paper i organized a follow. Section II preent the ytem model and formulate the problem of DTV-baed poitioning without knowledge about emitter la-

2 bel. In Section III, we detail the olution to the problem. Section IV how the teting reult and provide correponding dicuion. Finally, in Section V, the concluion from tudie are ummarized. GPS Satellite GPS ignal Tranmitter Tranmitter RF II. SYSTEM MODEL AND PROBLEM FORMULATION A. TOA Etimation with ignal The tandard ue the OFDM (orthogonal frequency diviion multiplexing) modulation to achieve robut tranmiion in multipath cenario. In ytem, OFDM ignal are pecified by three parameter, namely, the number of ubcarrier or the FFT (fat Fourier tranform) ize, the length of cyclic prefix (CP) and the ampling period. The option (or mode) for etting up thee parameter are provided in the ETSI tandard []. In ytem, ignal are continuouly tranmitted and every OFDM ymbol i tranmitted within a fixed duration of time. The pilot ubcarrier in each OFDM ymbol are given by a known Peudo-Random Binary Sequence (PRBS) with booted power. From the perpective of wirele poitioning, the good autocorrelation property of the PRBS with booted power and the continuou tranmiion are helpful to accurately acquire and track the ignal for TOA etimation. Moreover, in ytem, multiple emitter are uggeted to be coordinated to the GPS time and imultaneouly tranmit the ame ignal in the ame frequency band, forming the o-called SFN tranmiion. Therefore, multiple emitter upported by network ynchronization make the timing-baed etimation required for wirele poitioning available. In our previou work [7], a oftware defined receiver ha been developed to etimate the TOA of ignal for poitioning. The reult obtained from the field tet campaign how that the TOA tracking on real ignal i capable of achieving good ranging accuracy, with the etimation error within - meter depending on the practical ignal to noie ratio (SNR). B. Sytem Model We conider a uer that can imultaneouly receive ignal from M tation located within an area covered by a SFN network. The eat-north (X-Y) coordinate of the uer location i expreed a x = [x r y r ] T, and the coordinate of the ith tation i expreed a c i = [x i y i ] T, {i,,..., M}. In network, the ditance between a tation and a uer can be meaured by the TOA etimation [7]. Let d i denote the ditance between the uer at x and the ith tation at c i, which can be expreed a d i = hi (x) = c i x, i =,,..., M () For the ake of implicity, let u conider only the line-of-ight (LOS) tranmiion environment. In thi cae, the ditance meaurement i only corrupted by the ytem meaurement noie expreed a n i for the ith tation, which can be modeled a an independent and identically ditributed (i.i.d.) white Gauian noie N(, σi ) with zero mean and a variance of σi. Conequently, when M tation are conidered, Ditribution Network Tranmitter Tranmitter RF receiver Fig.. Illutation of poitioning confuion in SFN. the vector collecting the meaurement of the ditance from the M tation to x can be written a z = h(x) + n () where z = [z, z,, z M ] T, h(x) = [h (x), h (x),, h M (x)] T and n = [n, n,, n M ] T. A n i N(, σi ), we have n N (, Σ), where Σ = diag{σ, σ,, σm } i a diagonal matrix. In network, the tation coordinate are fixed and can be aumed to be known to the poitioning receiver by aving them in it memory. However, due to the property of the SFN, the label of a tation i not necearily tranmitted. Conequently, even though a poitioning receiver know the coordinate of the tation invoked, it may be unable to identify from which tation the ignal are received, generating the emitter confuion problem. For example, in Fig., when given the M ditnce from the M tation to a poitioning receiver, there may poibly exit M! ditribution yielding thee ditance, when we conider all the poible combination of the ditance from the tation. Therefore, one of the important tak in baed poitioning i to identify the tation, in order for a receiver to know from which tation it receive ignal and, ultimately, to calculate accurately it own location. Baed on the above dicuion, our poitioning problem i formulated a follow. C. Problem Formulation Aume that the receiver know a et of M urrounding emitter expreed a {Tx}, from which the ignal are received. Define the et of permutation generated by {Tx} a S {,, L }, where L i equal to the factorial of the total M emitter in {Tx}, i.e., L = M!. For a permutation j = [ j, j,, M j ], j {,, L}, a number i {,, M} i tranformed to a number i j, i.e. j (i) = i j, where i j {,, M}. Accordingly, the peudo range meaurement z i between the imaginary tation at c i j and the receiver i z i = c i j x + n i () A we know, within the et, there i only one permutation, ay j, which correpond to the actually received peudo-

3 range z. Thu, the poitioning problem in the SFN conit of the ub-problem of finding the right permutation of j, and the ub-problem of etimating x baed on the peudo range meaurement z. In thi paper, we propoe a method, which jointly identifie the DTV emitter and etimate the receiver location by maximizing the likelihood of the oberved range meaurement. Furthermore, an EMP algorithm i deigned in order to find the olution, a formulated a follow. III. EXPECTATION MAXIMIZATION ASSISTED POSITIONING Baed on the oberved peudo-range vector z, a hown in (), the maximum likelihood etimation (MLE) can be applied, which etimate the receiver poition x according to the optimization of where p(z x) = ˆx = arg max p(z x) () x (π) M/ Σ / e (z h(x)) T Σ (z h(x)) Taking the logarithm on both ide of (5) and neglecting the additive contant, it i explicit that the maximization of (5) i equivalent to the optimization of ˆx = arg min x { (z h(x)) T Σ (z h(x) However, in our problem, the label of the emitter are unknown, making the receiver unable to ditinguih the ignal received from the different emitter. In thi cae, even the receiver ha the knowledge of the ditance from the M urrounding emitter, it i till unable to calculate it poition directly, when conidering that there are in total L poible permutation for a et of the meaurement z. In order to olve the problem, we introduce the permutation variable a the miing data in the MLE. Then, the logarithm of p(z x) can be rewritten a [], } (5) (6) log p(z x) = log p(z, x) log p( z, x) (7) Upon taking the expectation on both ide with repect to the ditribution log p( z, x old ), where x old i the current etimation of the poition, we obtain where E old i the expectation of with repect to the ditribution p( z, x old ). Note that, in (8) the left-hand ide i the ame a that in (7), a it doe not depend on. According to the Gibb inequality, the lat term of E old (log p( z, x)) on the right ide of (8) i maximized at x = x old. Therefore, if we chooe x to improve E old (log p(z, x)) beyond E old (log p(z, x old )), log p(z x) will be imilarly improved beyond log p(z x old ). In other word, the term of E old (log p(z, x)), viewed a the expected complete-data loglikelihood [], can be ued a the cot function for maximization of log p(z x), which for implicity can be denoted a Q(x, x old ) = p( z, x old ) log p(z, x) (9) According to [], with the aid of the miing data, the EM algorithm i capable of converging to a local maximum of the log-likelihood function Q(x, x old ). Let u factorize the complete-data likelihood a p(z, x) = p(z, x)p( x) () where p( x) act a the a-priori knowledge about the true poition of the tation. However, due to the lack of thi knowledge, we aume the uniform a-priori of. Meanwhile, we can aume that the election of the poible permutation i independent of x, i.e. p( x) = p(). Then, Q(x, x old ) in (9) can be further factorized a Q(x, x old ) = p( z, x old ) log p(z, x) + p( z, x old ) log p() = Q (x, x old ) + Q () where by definition, Q (x, x old ) = p( z, xold ) log p(z, x), and Q = p( z, xold ) log p(). Explicitly, Q i a contant, a it i not depended on x. Conequently, the maximization of Q(x, x old ) with repect to x i equivalent to maximizing Q (x, x old ). A. Derivation of Algorithm The EMP algorithm i divided into an E-tep and an M-tep, which are detailed below. ) E-tep: Given the current etimate of the poition x old, the conditional ditribution of p( z, x old ) can be obtained by Baye theorem a: p( z, x old ) p(z, x old )p( x old ) () In (), when we aume the uniform ditribution of, then log p(z x)= p( z, x old ) log p(z, x) we have p( x old ) = p(). According to the meaurement p( z, x old ) log p( z, model x) of (), when given a permutation, p(z, x old ) obey a Gauian ditribution with the mean of h (x old ) and covariance = E old (log p(z, x)) E old (log p( z, x)) matrix Σ. Furthermore, let u define γ old = p( z, x old ), (8) which repreent the weight coefficient of the permutation of obtained from the current etimate x old and the meaurement z. Thu, for the E-tep, we have Q (x, x old ) = where it can be hown that γ old = ) (x L N old ; h (x old ), Σ γ old log p(z, x) () ()

4 ) M-tep: Given the weight coefficient γ old derived from the E-tep, the maximization tep (M-tep) i to find x, o that x = arg max x Q (x, x old ) (5) Upon ubtituting (5) into () and then into (5), taking the logarithm a well a neglecting the additive contant, it i explicit that the maximization of (5) i equivalent to the optimization problem of ˆx = arg min x { ( z h (x) ) T (Σ/γ old) ) ( ) } z h (x) (6) Let u define e (x) = z h (x), e(x) = [e T (x),, e T L (x)] T, and a covariance matrix {( ), ( ) W = diag Σ/γ old, Σ/γ old L }. Then, the optimization problem of (6) i equivalent to minimize the objective function of f(x) = e(x) T We(x) (7) The minimum value of f(x) occur at the point with the f(x) gradient being zero, i.e. x =. Since the function (7) include both the independent variable and the parameter, the cloed form olution to the gradient i not traightforward to derive. Alternatively, by directly approximating the nonlinear h (x) by a Jacobi matrix, i.e. H (x) = h(x) = c x x T c, x an iterative weighted leat quare method (IWLS) [5] can be implemented to olve the optimization problem. The IWLS method i numerically robut, which can eaily make ue of almot any kind of meaurement for the purpoe of poitioning in mobile location etting [5]. A diadvantage of the IWLS method i that it require an initial gue of the poition. Poor etting of the initial poition may caue the algorithm to converge to a local maximum value of the likelihood, reulting in a location divergent from the real poition, which correpond to the global maximum value of the likelihood. B. Initial Value Setting For the lack of the a-priori information about the poition of x, the initial etimation of poition x can be taken a the average of all the etimated poition from different permutation, i.e., ˆx = E(ˆx ), where the expectation i in term of all the poible permutation of the emitter. Given that there are in total L permutation, we have x = L ˆx. C. Weight Threhold In order to accelerate the computing, a weight threhold γ threhold can be introduced to decide whether a permutation j i to be conidered for further computing. Specifically, after an E-tep, when the weight γ old j, a een in (), atifie < γ threhold, we may aume that the permutation j ha a very mall probability to match the current etimate x old and the meaurement z. Therefore, j can be deleted from the candidate permutation lit for further conideration. γ old j TABLE I TESTING SCENARIOS Emitter cite Longitude Latitude Altitude Tx power (W) Grande Etoile.868 N 5.67 E 58 k - k La Mille.785 N 5.9 E 6 6 Super Rouviere.55 N E 6 6 Promegue.7 N 5.8 E Mont De Mareille.589 N E 6 D. Algorithm Decription In ummary, our EMP algorithm can be tated a follow, which jointly identifie the SFN emitter and etimate the poition of the receiver. Algorithm EMP algorithm Initialization: the initial mobile poitioning ˆx i et baed on Section III-B, et the topping error tolerance threhold to δ, and et the maximum number of iteration to N max ; for k =,, N max, ) compute the weight γ old j of the permutation according to (); ) delete the permutation having mall weight, according to what tated in Section III-C; ) update the etimate ˆx k according to (6); ) If ˆx k ˆx k > δ and k N max, et k = k + and return Step ). Otherwie, top and return the etimate to the mobile poition. end for A. Tet Scenario IV. TESTS RESULTS The cenario ued to tet the EMP algorithm i elected in the city of Mareille in Southern France. According to the public information releaed by the Coneil Supérieur de L audioviuel (CSA) [6], there are five emitter covering the SFN in the area. Table I lit the information of all the five emitter. From Table I, we can know that the emitter of the Grande Etoile i the main tranmitter in the area. The ignal are tranmitted in term of the Effective Radiated Power (ERP) in the range of KW to KW for different channel. The tranmitter in Promegue i the econd mot powerful one in the area with an ERP of.5 KW. There are three emitter in the uburban of the city, whoe tranmiion power i relatively mall and in the range from 6 W to 6 W. In [7], a TOA etimation approach ha been propoed, which i capable of tracking weak ignal for the purpoe of ranging, even in the cae that the received ignal are not trong enough to meet the requirement for the quai error free (QEF) demodulation. Thi i becaue the TOA etimation method in [7] exploit the pilot of ignal for tracking, which are tranmitted with booted power in comparion to the ignal for payload data. A an example, Fig. how one naphot of the channel acquiition in one of the field tet provided by the oftware defined receiver developed in [7]. In thi tet, the receiver i located in a parking place

5 DTV emitter True EMP Etoil h North (m) 6 initial poition 6 ample Promegue Rouviere Mille 5 5 Eat(m) Fig.. Snaphot of the channel acquiition from one field tet in Mareille, where the receiver i located in a parking place near the Notre Dame de la Garde and the acquired ignal are tranmitted in the frequency band of Channel. near the Notre Dame de la Garde, having the two dimenional (D) coordinate of [.88 N, E] and the altitude of 9 meter, according to a GPS receiver. Additionally, the central carrier frequency of the acquired ignal i MHz, which i the central frequency of Channel. B. One Realization of the Tet To evaluate the performance of our EMP algorithm, we aume that there i a receiver, which i located within the line of ight (LOS) region of the four SFN emitter noted in Table I and Fig.. The meaurement noie i aumed to be the white Gauian noie ditributed with zero mean and a variance of σ n = m. Thi aumption i in conitent with the field tet reult in high ignal-noie ratio (SNR) region, a hown in [7]. In Fig., we how a one-tep realization of the EMP algorithm, when the initial poition i randomly ditributed within the conidered area. By contrat, in Fig., we depict the D urface of the likelihood evaluated from (5). In our evaluation for Fig., all the poible permutation in S for a pecific grid x i are conidered. For the lack of the knowledge, equal a-priori probability of p( = j ) = /L i aumed to compute the likelihood. From Fig. and Fig., it i oberved that in high SNR region, poitioning i very reliable with mall evaluation error, meaning that all the emitter can be correctly identified. Comparatively, when the SNR decreae, ome of the emitter might not be correctly identified, a hown in Fig. 5, where the noie variance i m. However, a hown in Fig. 5, the poition of the receiver etimated by the EMP algorithm till converge to the actual poition of the receiver, even though more iteration are required, and when ome of the emitter are not correctly identified by the EMP algorithm. C. Statitical Performance Fig. 6 depict the root mean quare error (RMSE) performance of the EMP algorithm, when different number of Fig.. One realization of the EMP algorithm experiencing mall meaurement noie, where the variance of meaurement noie i m. In the tet, the poitioning error i within.5 m and the poitioning error i computed a the ditance between the uer true poition and the poition etimated by the EMP algorithm. There are two number marked with each emitter. The upper (black) number i the true label of the emitter, while the lower (red) number i the label of the emitter etimated by the EMP algorithm. In thi tet, all the label of the emitter are correctly identified x 5 North (m) 5 5 Eat(m) 5 Fig.. D urface of the value related to the likelihood evaluated by (5). In order to have a good view of the global minimum in the D urface, the coordinate of z axi increae from the top to the bottom. emitter are deployed. The reult were obtained from the average of Monte Carlo realization. In our imulation, the poition of the receiver i randomly generated within the area of interet. In order to compare the performance of the EMP algorithm, the Cramér-Rao lower bound (CRLB) [7], [8] wa computed and alo hown in Fig. 6. We hould note that, the CRLB wa computed under the ideal knowledge that the receiver know which ignal i received from which emitter. Therefore, the CRLB i an over-optimitic etimation for the problem conidered in thi contribution. From the reult of Fig. 6, we oberved that when only emitter are ued for poitioning, the poitioning error i much higher than the CRLB. By contrat, when there are four or five emitter available for poitioning, the poitioning performance

6 North (m) 8 6 DTV emitter True EMP Promegue Etoil initial poition Rouviere Mille 5 5 Eat(m) x RMSE (m) EMP emitter CRLB emitter EMP emitter CRLB emitter EMP 5 emitter CRLB 5 emitter Standard Diviation of Meaurement Noie (m) Fig. 5. One realization of the EM aited poitioning under large meaurement noie, where the variance of meaurement noie i m. In thi tet, the poitioning error i. m and the two emitter of Etoil and Mille are not correctly identified. i cloe to the CRLB. A hown in Fig. 6, when the number of emitter increae, the performance of poitioning improve. In contrat to the ignificant performance improvement when the number of emitter i increaed from three to four, there i only marginal improvement, when the number of emitter i increaed from four to five. Therefore, our propoed EMP algorithm i a highly effective poitioning algorithm even when the receiver experience the emitter confuion problem. Furthermore, conidering the complexity-performance trade-off, in practice, it i deirable for the EMP algorithm to work with four emitter, a evidenced by the reult hown in Fig. 6. In Fig. 7, we how the error rate of the emitter identification, when three, four or five emitter are employed by the EMP algorithm. Explicitly, when the tet ue emitter, the error rate of the emitter identification i the larget among the three tet cenario. Thi i becaue, in thi cae, there are not enough peudo-range provided for the EMP algorithm. By contrat, when four or five emitter are ued in the tet, the error rate of emitter identification decreae ignificantly. However, it i intereting to find that the error rate of emitter identification in the cae of 5 emitter i lightly larger than that in the cae of emitter. In fact, thi i becaue, when more emitter are ued in the tet, the emitter confuion problem become everer, reulting in the increaed error rate of emitter identification. Neverthele, a hown in Fig. 6, the lightly larger error rate of the emitter identification ha little effect on the final poitioning accuracy, which i inferred by the EMP algorithm. Thi i due to the fact that the objective function of the EMP algorithm i to minimize the poitioning error, while the label of the emitter are the hidden variable to be identified. Therefore, from the reult of Fig. 6 and Fig. 7, we are implied that the EMP algorithm i robutne to the error of emitter identification. Furthermore, the reult of Fig. 6 and Fig. 7 ugget that it i highly efficient for the EMP algorithm to work with four emitter, epecially when we alo take the Fig. 6. Poitioning performance of the EMP algorithm experiencing Gauian ditribution meaurement noie Emitter Identification Error Rate (%) emitter emitter 5 emitter Standard Diviation of Meaurement Noie (m) Fig. 7. Error rate of emitter identification. implementation complexity into account. V. CONCLUSIONS Thi paper ha tudied the wirele poitioning in DTV network. Due to the SFN property in DTV network, the receiver cannot identify the pecific emitter where a received ignal come. Therefore, an EMP algorithm ha been propoed to jointly etimate the receiver poition and identify the label of the emitter. To evaluate the algorithm, a tet cenario in the city of Mareille ha been conidered, where the poition of five emitter are et according to the official information publihed by the Authority in France. The performance of the EMP algorithm ha been tudied by conidering different iue. Our tudie and imulation reult how that, in the cae of or 5 emitter, the EMP algorithm i highly effective to achieve good poitioning performance, even in the exitence of the emitter confuion problem. When or 5 emitter are available, the tatitical performance how that the EMP algorithm i capable of achieving the performance approaching

7 the CRLB that i only achievable with full emitter information. Furthermore, with or 5 emitter, the EMP algorithm i alo effective to identify the DTV emitter. Additionally, our imulation tet how that the EMP algorithm i robut to the tolerance for the error of emitter identification. Note that, the methodology propoed in thi work i general, which may be applied in the other SFN for TOA baed poitioning. ACKNOWLEDGMENT Thi work wa partially upported by Academy of Finland under the Grant No. 5. Liang Chen would like to thank SIGNAV Laboratory in École Nationale de L Aviation Civile (ENAC), France, for providing the aitance on the field tet in Southern France. REFERENCES [] E. D. Kaplan, Ed., Undertanding GPS: Principle and Application. Norwood: Artech Houe, 996. [] M. Rabinowitz and J. Spilker, A new poitioning ytem uing televiion ynchronization ignal, IEEE Tranaction on Broadcating, vol. 5, no., pp. 5 6, March 5. [] Framing tructure, channel coding and modulation for digital terretrial televiion, ETSI Standard Std. ETSI EN 7 V..., January. [] J. T. Ong, H. Yan, S. V. Rao, and G. Shanmugam, Indoor DTV reception: meaurement technique, IEEE Tranaction on Broadcating, vol. 5, pp. 9 99,. [5] M. Rabinowitz and J. Spilker, Poitioning uing the ATSC digital televiion ignal, U.S.A. Patent WO: /57 6 A, [6] L. Chen, L.-L. Yang, and R. Chen, Time delay tracking for poitioning in DTV network, in Proc. Ubiquitou Poitioning, Indoor Navigation, and Location Baed Service (UPINLBS),, Oct., pp.. [7] L. Chen, O. Julien, P. Thevenon, D. Serant, A. G. Pe?na, and H. Kuuniemi, TOA etimation for poitioning with ignal in outdoor tatic tet, IEEE Tranaction on Broadcating, vol. 6, no., pp , December 5. [8] L. Chen, P. Thevenon, G. Seco-Granado, O. Julien, and H. Kuuniemi, Analyi on the TOA tracking with ignal for poitioning, IEEE Tranaction on Broadcating, vol. 6, no., pp , December 6. [9] X. Wang, Y. Wu, and J. Y. Chouinard, A new poition location ytem uing DTV tranmitter identification watermark ignal, EURASIP Journal on Applied Signal Proceing, vol. 6, pp., 6. [] J. Yang, X. Wang, M. Rahman, S. I. Park, H.-M. Kim, and Y. Wu, A new poitioning ytem uing tranmitter ignature waveform in ingle frequency network, IEEE Tranaction on Broadcating, vol. 58, no., pp. 7 59,. [] P. Thevenon, S band air interface for navigation ytem: a focu on OFDM ignal, Ph.D. diertation, Univerite de Touloue,. [] J. Yan and L. Wu, Bae tation identification in ingle-frequency network poitioning ytem under mixed line-of-ight/non-line-of-ight condition, IET Communication, vol. 6, no. 8, pp , December. [] A. B. Gelman, J. S. Carlin, H. S. Stern, and D. B. Rubin, Bayeian Data Analyi, nd ed. Chapman & Hall,. [] A. Dempter, N. Laird, and D. Rubin, Maximum likelihood from incomplete data via the EM algorithm, Journal of the Royal Statitical Society, Serie B, vol. 9, no., p.?8, 977. [5] N. Sirola, Cloed-form algorithm in mobile poitioning: Myth and miconception, in Proceeding of 7th Workhop on Poitioning Navigation and Communication (WPNC),, Bremen, Germany, Oct., pp. 8. [6] tvignaud.pagepero-orange.fr/tv/tnt.pdf. [7] S. M. Kay, Fundamental of Statitical Signal Proceing: Etimation Theory. Prentice Hall, 99. [8] K. W. Cheung, H. C. Sho, W. K. Ma, and Y. T. Chan, Leat quare algorithm for time-of-arrival-baed mobile location, IEEE Tranaction on Signal Proceing, vol. 5, no., pp. 8, April.