Detection with the Help of an OS CFAR Processor in CDMA Networks in the Presence of Multipath Interference*

Transcription

1 БЪЛГАРСКА АКАДЕМИЯ НА НАУКИТЕ. BULGARIAN ACADEMY OF SCIENCES КИБЕРНЕТИКА И ИНФОРМАЦИОННИ ТЕХНОЛОГИИ Том 4, 2 CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 4, No 2 София. 24. Sofia Detetion with the Help of an OS CFAR Proessor in CDMA Networks in the Presene of Multipath Interferene* Christo Kabakhiev, Vladimir Kyovtorov, Ivan Garvanov Institute of Information Tehnologies, 1113 Sofia s: kabakhiev@iit.bas.bg kabakhiev@yahoo.om vladimir_ak@yahoo.om igarvanov@iit.bas.bg igarvanov@yahoo.om Abstrat: In this paper we disuss the problem of seondary appliation of CDMA wireless ommuniation networks for low flying target detetion. It is redued to a PN signal detetion in multipath interferene by using a passive orrelation reeiver with an OS CFAR proessor. We perform noise level estimation in both windows, applying the order statisti (OS) approah. The Minimum of Average deision threshold is used as an OS estimation riterion of effiieny. The parameters of the orrelation reeiver with an OS CFAR are obtained by Monte Carlo simulation. The results may be applied for target detetion in multistati radars using the existing ommuniation networks. Keywords: passive orrelation reeivers, OS CFAR proessor, multipath interferene. 1. Introdution The ontemporary ommuniation networks reveal interesting seondary appliations. One of these appliations is the flying target detetion [1, 2]. Our aim is to use the existing CDMA wireless network for low flying target detetion and traking. For this purpose small passive radar reeivers ould be added to the base stations. It is well known that several small radars are more effiient than one, whih overs the same * The work reported is supported by IIT- 159/24, MPS Ltd. Grant RDR and Bulgarian NF SR Grant No TH-135/23. We wish to express our kind gratitude to Asso. Prof. Vladimir Baronkin. 11

2 area [3]. Considering the harateristis of the pilot signal in CDMA networks, the high energy, ontinuous ode sequene with big base, we hoose it as a signal in our pseudo-radar network [4, 5]. We suppose the target goes through only one ell, and only the pilot signal is reeived by the base station. The multipath interferene is a forward sattered pilot signal from different adjaent objets and bakground. Aording to Turin s investigations [9-11], the multipath propagation is typial for the spread spetrum ommuniations. In these systems, the RAKE reeiving approah is well known where the signals have the same information [4]. In our ase this approah is inappliable beause we estimate the target range. The task is resolved as detetion of PN signals (CDMA IS 95 pilot signal), in multipath interferene [1, 2], applied in wireless ommuniation systems. We suppose that the target signal is a pulse train from forward sattered signals in the presene of multipath interferene on the input of a oherent passive reeiver (radar reeiver). We assume that at the input of the orrelation reeiver the target eho from ommuniation signal flutuates aording to Swerling II ase model, and the multipath interferene is with a Poisson distribution of probability in the flow and Rayleigh distribution of their amplitudes. The Turin s model for multipath propagation is used [9-11]. The signals phase distribution is not onsidered. We work with the pilot signal and a orrelation reeiver, aording to CDMA IS-95 standard [4]. The signal has good auto-orrelation and ambiguity funtions, whih is important for radar appliations. The pilot signal is easily detetable, beause it has no data modulation [4]. Moreover, this signal is emitted ontinually and does not appear asually, whih makes it different from the paging or traffi hannels. Suh approah an be used with other omplex signals, whih are used in other ommuniation systems (GPS, CDMA 2, WCDMA [12-14]). Deteting in the presene of multipath with a fixed threshold laims high interferene to noise ratio (INR). Using CFAR ould improve this ratio, as it is done with the detetion in the pulse jamming presene. The multipath an ause an interferene, whih an be interpreted as a jamming proess. The detetion in the jamming environment with CFAR is desribed in [15-19]. The proposed by Himonas [18] and further developed by B e h a r et al. [15] API CFAR proessor with ensoring in both test and referene window needs minimum ADT (average deision threshold) in the ambiene of intensive jamming with Poisson distribution and Rayleigh amplitude distribution. Behar elaborates this problem in [15] with a more omplex and more general statistial desription model than the model used in [18]. Furthermore, we investigate the problem onerning the synthesizing of CFAR proessor with a ommuniation signal (IS-95 pilot signal) in multipath interferene. Our hypothesis was to use Behar s proposed API CFAR detetor for target signal detetion in multipath presene. After a number of simulations of the ensoring algorithms behavior of this detetor, it was found that the ensoring doesn t work properly. The reason is the lak of learly defined differene between the white gaussian noise and the multipath effet performane, and the height dynami of the average order statisti. For this reason we have to use an order statisti approah (OS) for ** The glossary used is found on page

3 noise level estimation in both windows [2, 21]. The Minimum of Average deision threshold is the riterion of this estimations effiieny [2]. The best interferene appraisal is made with the help of the Monte-Carlo simulation results. The CFAR effiieny is evaluated by the detetion and the false alarm probability for different values of the interferene-to-noise ratio (INR), and for different probability of appearane of multipath interferene. All results are obtained in MATLAB environment by Monte-Carlo simulation. The results ahieved in this paper may be suessfully applied for moving target detetion by using existing ommuniation networks. 2. Detetion in CDMA networks in the presene of multipath interferene problem formulation The ontemporary CDMA ommuniation networks are used for data transfer via air interfae for mobile subsribers [4]. Generally, the networks onsist of base stations (BS), eah BS overs an area (ell). The ells have different sizes depending on the environment (pio ell <1 m, and miro ell <1 km). The air interfae onsists of a pilot signal, a synhronization signal, paging and traffi hannels. The pilot signal is the same for the whole network and it is used for phase initialization of demodulation (supports the system oherene) [4]. The synhronization hannel is demodulated by all mobiles and ontains important system information onveyed by the synh hannel message, whih is broadast repeatedly. The paging hannels are used to alert the mobile for inoming alls, to onvey hannel assignments and to transmit system overhead information. The Traffi hannels arry the digital voie or data to the mobile user. In the overing area of the ommuniation networks all moving and fixed objets reflet these signals. Our aim is to use the existing CDMA wireless network for seondary appliation radar detetion of low flying targets. For this purpose, BS ould be added with passive radar reeivers. As a result, we ould obtain a pseudo radar oherent network from small radars (Fig.1). Detetion range of small radar Diretion of flight Fig. 1. Small radar net [3] In this ase the target ould ross the pseudo-network from these small radars, during the flying over the wireless ommuniation network. Our reeivers ould be 13

4 obtained in the global synhronization time from GSP or CDMA network, be phased synhronized with the CDMA network, and managed through own or entral ontrol system. We suppose the target goes through only one ell and only the pilot signal is reeived from the base station. From all other signals used in CDMA network, we hose the pilot signal as a signal used in our radar. It is the most powerful signal, with ontinued ode sequene with big period of repetition and uninterrupted in time. The refleted signal from the target onsists of many independent refleting elements, and many plains in the mirowave range have suh distributions [22]. We suppose relatively fast target flutuations, so that the target ross setion is independent from pulse to pulse in one sanning beam. Suh distribution is known as Swelnig II model with apriori known parameters. A speifi interferene in the CDMA ommuniation networks is the multipath, aused by the spread spetrum harater of the signals. We use the wellknown Turin s model for multipath propagation in urban areas with apriori known parameters. It desribes a flow with a Rayleigh distribution and Poisson probability of appearane of amplitude. Time delay is assumed up to the flow of approximately 2 s or 3 km. These interferenes together with the bakground interferene overed all small ells (miro and pio). As a differene from the bakground interferene, whih oneals the target signal, the multipath propagation forms many not real targets and worsens detetion and estimation. The problem of radar detetion in the CDMA ell is to redue the detetion with unknown oordinates and veloity of the pilot signal in presene of multipath interferene with apriori known parameters of the signal and interferene. We use the approah for detetion and estimation of the moving targets in the surveillane radar together with some of the well-known methods for bakground suppressing (for example Moving Target Detetion MTD, Adaptive Moving Target Detetion AMTD or Spae Time Adaptive proessing STAP) [6, 7, 8]. In this ase the target detetion with unknown oordinates and veloity is redued to the multihannel range target detetion in a fixed veloity (azimuth) hannel [6, 7, 8]. The use of CFAR approah enables the keeping of false alarm rate in all the range distane. As a result, the target detetion in one ommuniation ell ould be transformed as a target CFAR detetion in the moving window on the range in all hannels of veloity (in any hannel azimuth). In our paper we do not investigate moving targets. Then the target detetion is redued to pilot signal CFAR detetion in the moving window on the range (in any hannel azimuth), in the presene of multipath interferene. We use a base-band model of a oherent passive reeiver (a orrelator and CFAR detetor, Fig.2). From the HF blok Correlation reeiver CFAR detetor 1 Fig. 2. Simple blok diagram of reeiver 14

5 3. Signal and environment model In this paper we study the signal and environment models similar to the ones proposed in [9-11]: (1) w( t) a s( t t ) a s( tt ) n( t), d k 1 where: a s(t t d ) is the refleted signal in every range element of the signal matrix (a ell); a has an amplitude flutuating independently aording to Rayleigh distribution law (Swerling II target model); s(t) is the PN ommuniation pilot signal; t d is the time delay of the diret signal; a s( tt ) is Turin s multipath model, where a k 1 k k k is the amplitude flutuating aording to Rayleigh distribution law, t k is the delay time with Poisson probability of appearane; n(t) is the additive white gaussian noise. We do not onsider the uniform random distribution of phases in the multipath model. The emitted signal is ontinuous, but the reeived signal an be onsidered as pulse signal after the use of a orrelator with fixed length. The reeived signal is sampled in range by using M+1 resolution ells resulting in a matrix with M+1 rows on range and L olumns on azimuth. Eah olumn of the data matrix onsists of the values of the signal obtained for L pulse intervals in one range resolution ell. This matrix is used for signal proessing in the orrelator. The output matrix is used by the CFAR proessor. Тhe first M/2 and the last M/2 rows of the data matrix are used as a referene window in order to estimate the noise-plus-interferene level in the test resolution ell of the radar. In this ase the samples of the referene ells result in a matrix X of the size ML. The test ell or the radar target image inludes the elements of the M/2+1 row of the data matrix and is a vetor Z of length L. In the presene of a desired signal from a target, the elements of the test resolution ell are independent random variables with distribution law (1). The elements of the referene window are independent random variables, whih an be defined as (1), setting a s(t t d ) = Signal model In our ase, the refleted signal is a PN ode ommuniation pilot signal. The IS-95A standard is based on spread spetrum signals (SSS) with diret spreading (DS) [4]. It is established that these signals ould possibly be used as radar signals, beause they are synthesized with a low ross orrelation funtion in order to minimize the influene between partiular hannels [4, 5]. We fous our attention to the Pilot signal, whih is uninterrupted and onsists of simple signals. It is used primarily as oherent phases referene for demodulation of the other hannels. For this reason, IS-95 requires that the hip timing and arrier phase of eah downlink hannel be in very lose agreement. The pilot signal inludes onstant logial, it is modulated at Walsh hip rate of Mps by the -th row of the 6464 Hadamart matrix, whih is the Walsh sequene onsisting of 64 zeros-thus, in effet, it is not modulated at all. The two distint short PN odes I and Q are maximal length sequenes generated by 15-stage shift registers k k 15

6 and lengthened by the insertion of one hip per period in a speifi loation in the PN sequene Thus, these PN odes have periods equal to the normal sequene length of = plus one hip, or hips [4]. The PN-ode spreading is followed by lassi QPSK modulation of the radio frequeny arrier. This signal is perfet for seondary surveillane as it is more powerful than the others. Beause it is a synhrosignal, the standard demands aquisition in signal-to-noise ratio of 15 db. In this paper Swerling II model is used for desribing the signal flutuations in the pulse train oming diretly from the target Multipath model Three types of ommuniation hannel models exist empirial, deterministi, and a ombination of both. We use the suggested by Turin ommuniation hannel model [9-11]. It is desribed as pulse train with Rayleigh amplitude distribution, Poisson probability of appearane and uniform random distribution of phases. Despite of the fat that the model has been developed in 1975, it is still valid and used [23]. We perform the Turin s model simulation as a pulse train with Poisson probability of appearane in one ell of the signal matrix, similarly to that in [15-17, 19, 21]. In aordane with Turin s model, we hoose (98) width for the signal matrix and probability of appearane P a =.2 at the input of the CFAR. 4. A passive reeiver for target detetion in the presene of multipath interferene 4.1. A orrelation reeiver We use the baseband aquisition diagram for the pilot signal of CDMA-IS 95 A [4]. It onsists of a orrelator and a threshold detetor with a fixed threshold (Fig.3). In our ase we use an OS CFAR proessor for target detetion in the presene of multipath interferene. The quadrature omponents at the baseband filter input are [4]: E (2.1) r ( t) r ( t) 2 os( t) C( t) n ( t) I lowpass I I T E C ( t)os( ) I T (2.2) r ( t) r ( t) 2 sin( t) Q lowpass E C ( t)os( ) Q T E C ( t)sin( ) n ( t) Q I T, E C ( t) n Q T ( t) E C ( t)sin( ) n ( t), I Q T Q 16

7 where C I (t) and C Q (t) are PN sequenes in the I and Q hannels respetively, n(t) is the additive Gaussian white noise (AWGN), n I (t) and n Q (t) are the orresponding noise in both hannels (statistially independent white gaussian noise [4]). 2 os( t ) r(t) LP CI(t τ) dii(t) r I(t) diq(t) CQ(t τ) CQ(t τ) + + X(t) mt () dt m1 Z1 () 2 Zm > < LP rq(t) dqq(t) dii(t) - Y(t) mt () dt m1 Z2 () 2 2 sin( t ) CI(t τ) Quadrature detetor Fig. 3. Filter diagram The probability density funtion Z m (Fig.3) at the output of the orrelator is given with [4]: (3) p Z m 1 1 ( 2 ( ) e 2 2 othrewise, ) I ( 2 ) 2,, where T N 2 E, λ is a nonentrality parameter, 2 4NR ( ) and N = T/T N, N is the number of pulses (hips) used in the integration, Т is the observation interval, Т is the duration of one hip, E is the hip energy OS CFAR proessor Our hypothesis was that we an use the proposed by B e h a r a t a l. [15] API CFAR detetor for signal detetion in the presene of multipath interferene. After performing a number of simulations, we established that the ensoring algorithm does not work effetively. Due to the lak of learly defined differene between the white Gaussian noise and the multipath effet performane in the test and referene windows, it stops the using of the ensoring algorithm in the ordered statistis. Additionally, the signal after the orrelation reeiver is very dynami. Differently from pulse radars, for whih it is onsidered that there is no signal in the referene window, in our ase, due to the fat that the model works with ontinuous signals, the statistis in both, the test and referene windows, have similar strutures, inluding white noise, signal, and multipath interferene. The differene is that the statistis in the test window ontains 17

8 a pulse train refleted from the target. Therefore we use the OS approah for noise level estimation in both windows. The minimum of the average deision threshold is used as a riterion of effetiveness of these estimations [2]. The rank-ordered parameters, giving the best OS estimation, are hosen by using Monte Carlo simulation. The effetive estimations are equivalent to those elements of the ordered statistis in both windows, where the minimum SNR ours for P d =.5 and fixed P fa. In order to optimize these estimations, we hange dependently and independently the rank-ordered parameters (from (3/4)L to (1/8)L), as it is done in [2]. The results revealed that the minimum ADT is ahieved for the rank-ordered parameter ombination k 1 =(1/8)L и k=(3/16)r. For the other ombinations, the ADT is higher approximately with 5-6 db. We use in our work a new modifiation of the two-dimensional OS CFAR proessor, whih is very effetive in the presene of multipath interferene Analysis of two-dimensional OS CFAR proessor The algorithm onsists of the following two stages. Stage 1. The elements of the referene window x ( x, x,..., x ), R=ML and the 1 2 R test resolution ell z ( z, z,..., z ) are rank-ordered aording to inreasing magnitude: 1 2 L (4) x x... x... x and i N z z z. 1 2 j L z Stage 2. The main idea of an OS CFAR proedure is to selet one main value 1 x, k 1, 2,..., R and z 1, k 1, 2,..., L k k from the sequene in (4) and to use it as 1 1 estimates V and q for the average noise power and the average signal power of the observed referene and test window: (5) V x and 1 k q z. The rank-ordered parameters k and k 1 of the OS CFAR proedure are hosen in suh a way that the average deision threshold of the OS CFAR proessor to be with a minimum value. The target is then deteted aording to the following algorithm: 1 k 1 (6) H H 1: : ( q ) 1, : ( q ), q T V, a q T V, a where H 1 is the hypothesis that the test resolution ells ontain the ehoes from the target and H is the hypothesis that the test resolution ells ontain white noise, signal, and multipath interferene. The onstant T a is a sale fator, whih is determined in order to maintain a given false alarm probability onstant. The probability of target detetion is determined as: 18

9 (7) P P( q T V H ) P V dv P ( q / H ) dq, d a 1 where P V V is the probability density funtion (PDF) of the noise level estimate in the referene window and P q q H 1 is the onditional PDF of the test window, under hypothesis H 1. The probability of false alarm is determined by substituting s =, that is: (8) P P( q T V H ) P V dv P ( q / H ) dq, fa a V where P q q H is the onditional PDF of the test window, under hypothesis H. On the output on the orelator, we have not the analytial equation for the probability density funtion (PDF): of the noise level estimate in the referene and the test window, under hypothesis H 1 and H. Then we use Monte-Carlo simulation approah for estimation of probability performane of OS CFAR proessor. 5. MATLAB Simulation model of the passive reeiver in the presene of multipath and additive white gaussian noise (desribing the MATLAB model) In our investigation we use a Monte Carlo simulation approah. In this ase, it is important to use a orret model, beause all statistial results, whih desribe the behavior of our system, depend on the model orretness. Due to this reason, we use some adjustments to ontrol the orretness of the separate parts of the model. In that way we an hek the orretness of the omplete model (we perform the superposition adjaent of the system). The model that we onsider, onsists of a simulation model of the input signals and interferene, a orrelator, CFAR and resolving system. The blok diagram of our model is shown in Fig. 4. The blok diagram shows the bloks for formation of the pilot signal, the multipath signal, the forward sattered signal, white gaussian noise, the orrelator and the CFAR detetor. In the present work the PN pilot signal waveform is reeived from the MATLAB 6.5R13, Simulink CDMA referene blokset [24]. We use the model of base-band gaussian noise generator from [4]. It is based on Riian deomposition (9)-(1), whih is easy for modeling. For this reason it is widely used in the omputer simulations (Fig. 5) [4]. As it is said in theory [4], the two hannel statistial independene is ahieved by two independent uniformly random generators: x( t) x ( t)os( t) x ( t)sin( t), s x ( t) R( t)os( ( t)), (9) x ( t) R( t)sin( ( t)); x ( t) (1) s ( t) artg. x ( t) s R( t) x ( t) x 2 2 s V ( t), T V a q T V a q 1 19

10 Fig. 4. Simulation model of the passive reeiver in the presene of multipath Independent uniform random generator {U 1n } ln -2σ 2 {ζ sn } Independent uniform random generator {U 2n } 2π {θ n } os sin {ζ n } Fig. 5. Diagram of the generation of two independent Gaussian random variables from two independent uniform random variables The MATLAB simulation model of the multipath with a diret signal issue is shown in Fig.6. The model onsists of: probability of Poisson appearane ontroller (1), two synhronized tape delay lines (2), and a diret hannel to the target. The omputer model of the orrelator is made aording to the theory desribed in [4] and 3.2. The model of OS CFAR is aording to

11 Fig. 6. The MATLAB simulation model of the multipath propagation (blok diagram) 6. Simulation results 6.1. Model verifiation We use the mathematial model of the orrelator in [4], implemented in SUMILINK of MATLAB environment. We have obtained: simulation results of P d depending on P fa, and SNR of the pilot signal in white gaussian noise presene, and a numerial analysis in MATLAB, and P d versus P fa and SNR of the pilot signal is performed [4]. To make a verifiation of our simulation model we have ompared both results. The hosen P fa is not appropriate for radioloation appliations, beause of its low order. We have hosen these quantities in order to be ompared with the theoretial results in the theory of ommuniation appliations. Fig. 7 shows the results obtained from the simulation analysis of P d versus P fa in the E /N = 15 db for different lengths of the signal used (a part of the pilot signal) N = 32, 64, 96 hips. The simulation results are obtained with Monte-Carlo simulation (1 runs). N denotes the number of hips used in the referene sequene. From the obvious oinidene between the two results we an draw the onlusion that the model of the orrelator is orret. The more the referene hips inrease, the more the probability of detetion rises (Fig.7). In Figs. 7 and 8 the orrelator probability harateristis are shown. These results are obtained with the help of numerial and simulation analysis for a signal with different length. 111

12 Fig. 7. Upper bound on aquisition detetion probability versus false alarm probability for the ase of E /N = 15 db [7] In Fig.7 P d versus SNR is shown with onstant probability of false alarm P fa =.1. We an see from the figure that in the SNR= 15 db the probability of detetion is P d =.5.9 aording to the length of the signal used N = 32, 64, 96. The distintions in the Average Deision Trheshold (ADT) for different lengths of the filter are 3 5 db. The probability of false alarm P fa I versus the detetion threshold, with E /N = 15 db with different lengths of the used signal is shown in Fig. 8. We an see that to maintain the same probability of false alarm P fa =.1, for different lengths of the referene signal in the orrelator the threshold hanges drastially from 3 to 11. From the slope of the urve it follows that the longer the length of the used signal is, the higher threshold is needed for onstant P fa. These results are onfirmed in the next hapter too. As total onlusion onerning Fig.8 and Fig.9, it is obvious that there is very good oinidene between the numerial and the simulation result, whih onfirms again the regularity of the results obtained. A simulation analysis of the multipath generator is performed. At the Fig. 11 are presented a graphial results of the probability of appearane P a in one detetion ell for a different period of observation T e =1Δ, 15Δ, 2Δ, 25Δ, 3Δ versus the number of arrived rays (Δ is the signal s range resolution). This number is up to 3 aording to the Turin s experiment. Graphial results of the theoretial Poisson distribution (dashed line), and simulation results from the blok generating the multipath (solid lines) are shown. We estimate the fitting between the two results (theoretial and simulated) using the riterion for Cramer-Von Mises. 112

13 Fig. 8. Probability of false alarm versus sale fator Fig. 9. Detetion probability of the orrelation reeiver 2 * 2 (11) P ( x) P ( x) dp( x), 2 where is the error oming from the theoretial P(x) and experimental drawn * P ( x). Fig.1 shows the urve fitting error versus the time of observation T e. As we know [24] a oinidene over 5% is a bad fit, we an onlude that the model has the right probability of appearane on the ase of T e =1Δ or P a =.2 and a fixed number of oming rays 3 (Figs. 1 and 11). T e is the duration of observation window, or the length of a row from the signal matrix in our ase. Therefore, in our investigations we use the probability of appearane P a =.2 with an observation time T e =1Δ. In that 8 113

14 way, with the urve fitting method we limit our model. This restrition doesn t mean that in the investigations other probability of appearanes in one ell, or times of observations (see Turin s model) are allowed [9-11]. Fig. 1. Comparison of the goodness between the theoretial urve and the experimental urve aording the Cramer-Von Mises test Fig. 11. Rays time of appearane 114

15 6.2. Simulation results The statistial harateristis of the output of a passive reeiver in the presene of multipath interferene are ahieved by Monte-Carlo simulation, for example performing 25 runs for alulation of the detetion probability and 1 4 runs for alulation of the false alarm probability for P fa = In our work we onsider that the average signal energy oming from the multipath propagation (the interferene to noise ratio) is onstant for all simulation runs and its value is INR=1 db. The passive reeiver is modeled as a sequene of a orrelator for a PN sequene and an OS CFAR proessor. The length of the referene orrelation sequene is fixed to N=96. Fig.12 shows the signal reeived at the output of the orrelator, as a result of mixing of the target refleted signal and the multipath interferene. The investigation is performed for a fixed matrix size of the referene window 88. In order to ahieve a referene window with this size, the matrix size at the input of the orrelator should be (N+8)8. This matrix ontains target refleted signal, multipath interferene and white Gaussian noise. Therefore, they overlap in this part of the matrix, and their average energies are summed in the orrelator. The following harateristis are obtained by Monte-Carlo simulation: statistial average ordered statistis (OS) for the referene and the test windows; average deision threshold; detetion and false alarm probability. Fig. 13 shows the statistial average OS of the mixture of white gaussian noise and ompressed signal (at the output of the orrelator) in the referene window. Fig. 12. The effet of mulptipath propagation on the orrelation reeiver s output This piture larifies the influene of the signal level over the estimation. The OS has a highly expressed exponential harater with great dynamis. It is obvious that the average level of the OS hanges muh when the SNR is inreased over 1 db. Fig.14 shows a mixture of the white gaussian noise, the signal and the multipath interferene at the output of the orrelator. This omponent has a different SNR=1, 2 db and a probability of appearane of the multipath omponents P a =.1 and.2. The referene window is of the size 88. The typial differene between the two domains is observed when multipath propagation is present in the referene window and there is no target signal in the averaged OS. If the mixture ontains the target 115

16 Fig. 13. Average order statisti in the presene of signal and noise for the referene window refleted signal, the multipath interferene and the white Gaussian noise, then the statistial average OS aquires the typial exponential mode, and the inreasing of the SNR ratio leads to the disappearane of the existing differene between the two domains. It is due to the inreased average energy resulting from the overlapping of the signals in the orrelator. The absene of learly defined border between the output noise level and the multipath interferene, and the high signal dynamis, makes the proposed by H i m o n a s [18] ensoring algorithm ineffetive in this situation. Fig. 14. Average order statisti in the presene of signa, noise, and multipath interferene for the referene window Fig.15 presents the hanges of the average deision threshold (ADT) in the OS CFAR proessor versus the hanges of the rang-ordered parameters. The results are obtained for P fa =.1, INR = 1 db, P a =.2. The ADT of an optimal detetor of a PN signal in the presene of white gaussian noise with fixed threshold and the ADT of the 116

17 additional multipath interferene are presented with solid lines. Both lines are obtained for fixed thresholds. It an be observed that for an appropriate ombination of the two rang-ordered parameters, in the test and referene windows, the average deision threshold beomes smaller than the fixed threshold. The minimum ADT is obtained for the ombination of the rank-ordered parameters k 1 =(1/8)L and k =(3/16)R. In this ase the improvement is about 1 db. Fig. 15. Comparison between OS CFAR proessor and fixed threshold detetor in the presene of multipath interferene and an optimal detetor Fig. 16. Detetion probability of OS CFAR in the presene of multipath interferene with different probability of false alarm and rank-order parameters 117

18 The OS CFAR probability of detetion versus the SNR is presented on Fig.16. This figure is similar to Fig.12. The results are reeived for the input parameters P fa =.1 и.1; P a =.2; INR=1 db; k 1 = (1/8)L and (1/4)L; k = (3/16)R and (1/4)R. The detetion probability is ahieved by Monte-Carlo simulation and the numerial solution of equation (3) in the presene of white noise. A detetor with fixed threshold demands a higher SNR for a determined value of the P d in the presene of multipath interferene. The 1 db improvement, ahieved for the OS CFAR proessor with parameters k 1 = (1/8)L and k = (3/16)R in omparison with the detetor with fixed threshold, is invariant to the hange of the P d. 7. Conlusion The task for seondary appliation in the CDMA wireless ommuniation networks is presented in this paper detetion of low flying target in the presene of multipath interferene. The task is onsidered only for one ell or one pseudo-oherent radar reeiver. We use the approah for detetion and estimation of the moving targets in surveillane radar together with the use of some of the well-known methods for bakground suppressing. In this ase the target detetion with unknown oordinates and veloity is redued to multi hannel range target detetion at fixed veloity (azimuth) hannel. The use of CFAR approah enables the keeping of false alarm rate in the whole range distane. The PN sequene, used in the CDMA ommuniation networks, is hosen as a signal model. Swerling II target signal flutuation is used as a pulse train model. The multipath propagation onsidered is a flow with a Poisson probability of appearane and Rayleigh amplitude distribution. As a result when the target does not move, the target detetion in one ommuniation ell ould be transformed to a target OS CFAR detetion in the moving window in the range (in any hannel azimuth) with apriori known parameters for signal and interferene. The riterion for effiieny of OS estimations ensures minimum of the average deision threshold [2]. The use of the ensoring algorithm [15] in both windows is not effiient beause of lak of learly defined differene between the white Gaussian noise and the multipath interferene in the test and referene windows, as well as the high dynamis of the average OS. For these reasons, we use the OS interferene evaluation. The rank-ordered parameters, for whih the best evaluation of the interferene is ahieved in the test and referene windows, are alulated by Monte-Carlo simulation. The minimum ADT is obtained for a ombination of rank-ordered parameters k 1 =(1/8)L and k = (3/16)R (for a referene window of size LR). 1 db improvement for the OS CFAR proessor is ahieved for P d =.1, ompared to the detetor with fixed threshold, and 5 db for P d = The results an be applied for target detetion in multistati radars using the existing ommuniation networks. 118

19 R e f e r e n e s 1. C h e r n i a k o v, M., M. K u b i k. Seondary appliations of wireless tehnology (SAWT). In: European Conferene on Wireless Tehnology, Paris, C h e r n i a k o v, M., D. N e z l i n, K. K u b i k. Air target detetion via bistati radar based on LEOS ommuniation signals.iee Pro. Radar Sonar navigation, Vol 149, February 22, No B a k e r, C. J., A. L. H u m e. Netted Radar Sensing. IEEE AESS Systems Magazine, February L e e, J., L. M i l l e r. CDMA Systems Engineering Handbook. Arteh House, K y o v t o r o v, V. A. Analysis of the possibilities for of seondary appliations of CDMA ellular systems. ICEST, B a r t o n, D. Modern Radar System Analysis. Arteh House, H a y k i n, S. Adaptive Radar Detetion and Estimation. John Wiley, N i t z b e r g, R. Adaptive Signal Proessing for Radar. Arteh House, T u r i n, G. L. e t a l. A statistial model of urban multipath propagation. IEEE Trans. Vehiul.Tehnol., Feb T u r i n, G. L. e t a l. Simulation of urban vehiule-monitoring systems. IEEE Trans. Vehiul. Tehnol., Feb S u z u k i, H. A. Statistial Model for Urban Radio Propagation. IEEE Transations on Communiations, Vol. Com-25, July 1977, No L a z a r o v, A., C. M i n h e v. ISAR Tehnique with omplementary phase ode modulated signals. In: PLANS 24 Conferene, Monterey, CA, April 26-29, L a z a r o v, A., C. M i n h e v. Three dimensional image reonstrution proedure over Barer s ode modulated ISAR signals. In: Pro. of IEEE Radar Conferene 23, Huntsville, Alabama, USA, 5-8 May L a z a r o v, A. Reurrent Kalman proedure for ISAR image reonstrution from Barer s phase ode modulated trajetory signals. Cybernetis and Information Tehnologies, Vol 2, 22, No 2, B e h a r, V., C. K a b a k h i e v, L. D o u k o v s k a. An adaptive CFAR PI proessor for radar target detetion in pulse jamming. VLSI, SP-26, 2, G a r v a n o v, I., C. K a b a k h i e v. Sensitivity of API CFAR detetors towards hange of input parameters of pulse jamming. In: Pro. of the International Radar Symposium IRS 24, Warszawa, Poland, 24, G a r v a n o v, I., C. K a b a k h i e v. Sensitivity of CFAR proessors toward the hange of input distribution of pulse jamming. In: Pro. of IEEE, International Conferene on Radar Radar 23, Adelaide, Australia, 23, H i m o n a s, S. CFAR Integration proessors in randomly arriving impulse interferene. IEEE Trans., Vol. AES-3, 3, July, 1994, No 3, G a r v a n o v, I., V. B e h a r, C. K a b a k h i e v. CFAR proessors in pulse jamming. In: Conferene Numerial Methods and Appliations 22, NMA 22. Preprint: Letures Notes and Computer Siene, LNCS 2542, 23, R o h l i n g, H. Radar CFAR thresholding in lutter and multiple target situation. IEEE Trans., Vol. AES-19, July, 1983, No 4, A k i m o v, P., F. E v s t r a t o v, S. Z a h a r o v. Radio signal detetion. Mosow, Radio and Communiation, 1989, (in Russian). 22. S k o l n i k, M. I. Radioloation Manual. Vol. 1. Mosow, Sovetskoe RaChannel. IEEE Transations on Communiations, Vol. 39, Ot.1991, No R o h l i n g, H. 25 years researh in range CFAR tehniques.in: Pro. IRS-23, Germany, 23, G a r v a n o v, I. Methods and algorithms for supporting onstant frequeny of false alarm under onditions of haos-pulse disturbanes. Ph. D. Thesis, Sofia, IIT BAS, 23 (in Bulgarian). 119

20 G l o s s a r y ADT Average deision Threshold PDF Probability Density Funtion API Adaptive Post-detetion Integration P d Detetion Probability AWGN Additive Gaussian Noise P fa Probability of False Alarm CDMA Code Division Multiply Aess QPSK Quadra Phase Shift Key CFAR Constant False Alarm Rate SNR Signal to Noise Ratio INR Interferene to Noise Ratio M I Multipath Interferene OS Order Statisti OS CFAR детектиране на сигнали в CDMA мрежи в присъствието на многолъчево разпространение Христо Кабакчиев, Владимир Кьовоторов, Иван Гарванов Институт по информационни технологии, 1113 София s: kabakhiev@iit.bas.bg kabakhiev@yahoo.om vladimir_ak@yahoo.om igarvanov@iit.bas.bg igarvanov@yahoo.om (Р е з ю м е) В статията се разглежда задачата за вторично приложение на CDMA комуникационни мрежи откриване на ниско летящи обекти. Тази задача е редуцирана до детектиране на псевдослучаен (PN) сигнал с използването на пасивен корелационен приемник с OS CFAR детектор в смущения, породени от многолъчевост. Като се използва подхода на подредените статистики (OS), се оценява шумовото ниво в двата прозореца на CFAR процесора (тестов и обучаващ). Като критерий за ефективност на оценката се използва минимумът на средния праг на откриване (Average Deision Threshold). Параметрите на корелационния приемник с OS CFAR детектор са получени с Монте-Карло симулация. Резултатите могат да бъдат приложени за откриване на цели в многопозиционни радари или мрежи от радари, използващи съществуващи комуникационни мрежи. 12