- Research Article
- Open Access
Multifrequency and Multistatic Inverse Synthetic Aperture Radar, with Application to FM Passive Radar
© Guillaume Ginolhac et al. 2010
- Received: 27 April 2009
- Accepted: 20 October 2009
- Published: 9 December 2009
This paper deals with the imaging of a moving target using a multifrequency and multistatic radar consisting in one receiver and several narrowband transmitters. Considering two hypotheses about the studied target, we derive two multistatic inverse synthetic aperture radar processors: the first one, which models the target as a set of isotropic points, performs a coherent sum of bistatic images; the second one, which models the target as a set of nonisotropic points, performs an incoherent sum of bistatic images. Numerical simulations are done, which demonstrate the efficiency of the second processor. We also apply both processors to a multistatic passive radar scenario for which the transmitters are FM stations located in a realistic configuration. We study the system performance in terms of resolution and sidelobe levels as a function of the number of transmitters and of the integration time. Both processors are applied to similar complex targets for which the scattered fields are simulated by a numerical electromagnetic code. The resulting multistatic radar images show interesting characteristics that might be used by classification algorithms in future work.
- Synthetic Aperture Radar
- Isotropic Point
- Scattered Field
- Mean Absolute Error
Passive radars use one or several transmitters of opportunity to detect, locate, and classify targets. This technique is cost effective because users only need to install receivers. Nevertheless, the use of civil communication signals, which are often narrowband signals, makes the detection and the classification difficult. The main limitations of passive radars are due to the center frequency and the bandwidth of these signals. For instance, for FM and TV signals, the center frequency varies from 100 MHz (for FM) to 800 MHz (for some TV stations in France) with bandwidths ranging from 75 kHz (for FM) to 1 MHz (for TV stations). For comparison, the center frequency and the bandwidth of a monostatic active radar such as the RAMSES ONERA SAR system  in the X-band are approximately 10 GHz and 1.2 GHz. For reaching equivalent performances (resolution and sidelobe levels) as an active system, a passive radar has to illuminate the target a long time and to consider several narrowband transmitters (FM or TV) in order to compensate for the low values of the center frequency and the bandwidth. In the following, such a radar system will be called a multifrequency and multistatic radar.
Passive bistatic radar systems consisting in one transmitter and one receiver located at different places and using FM  and TV signals  have demonstrated the feasibility of passive radars. Detection, localization, and tracking algorithms have been developed and validated using real data. However, classification is better performed using multiple transmitters or receivers. Therefore, different approaches have recently been developed. An approach to Automatic Target Recognition (ATR) consists in comparing the modeled Radar Cross Section (RCS) of different targets with those measured by the multistatic radar system [4–7]. Another approach, which is the one this paper focuses on, relies on the development of classification algorithms based on a target image obtained from the multistatic radar measurements.
In previous works [8, 9], we studied the resolution attainable by a multifrequency and multistatic radar using analytical and numerical approaches. Results showed that multifrequency and multistatic radars could achieve sufficient resolution using several narrowband transmitters and a long integration time. In this paper, we propose imaging algorithms for multifrequency and multistatic radars, which could be useful as a preprocessing for classification methods. Most of imaging algorithms are based on the Synthetic Aperture Radar (SAR) concept . The main purpose of imaging algorithms is to achieve enough resolution to make the classification step easy. For instance, the final images corresponding to two similar targets must exhibit enough differences so that it is possible to make the difference. For narrowband radar, it is well known that data from transmitter-receiver pairs distributed all around the target allow to have an image [11, 12] with such properties. These particular systems are similar to tomographic ones. Recently, Wu and Munson  simulated the imaging of a moving airplane using a multistatic radar composed of several TV transmitters and a single receiver. The imaging algorithm in  relies on the assumption that the target is composed of isotropic points and uses 2D interpolation on the frequency domain to achieve the processing. However, the resulting images are subject to degradation because the target has to be illuminated over a long period (in this case, the target is often nonstationary). Several methods have been proposed for solving this problem: time-frequency-based methods [14–16], deconvolution methods , or optimization techniques .
This paper proposes new Multistatic Inverse SAR (MISAR) processors where SAR processors were showed to be particular cases of detection algorithms [19, 20]. Two MISAR processors are developed, depending on the hypotheses made about the target: in the first case the target is assumed to be equivalent to a set of isotropic points, and in the second case the target is assumed to be equivalent to a set of nonisotropic points. We demonstrate that the first processor is equivalent to the one proposed in  while the second one is slightly different.
Next, we apply both MISAR processors to a realistic multistatic FM passive radar configuration for which the FM transmitter locations are chosen according to those of the FM transmitters in France . A study of image resolution based on methods presented in [8, 9] is performed as a function of the number of transmitters and of the integration time. Both MISAR processors are tested on two similar complex targets presented in . This paper does not consider the estimation of the trajectory. This problem is addressed with an extended Kalman Filter or a particle filter in [2, 3, 7, 23–26].
Section 2 describes the hypotheses that are made about the target as well as the notations that will be used in this paper. Section 3 develops both MISAR processors and illustrates their potential via a numerical simulation. Section 4 details the geometric configuration of our multistatic passive FM radar scenario and presents the resulting image.
We first present the notation needed to derive the desired detection algorithms. We next present two conjectures: the scattered field of target is either a set of isotropic points or a set of nonisotropic points. The corresponding detection problems are finally presented.
2.1. Context and Notation
2.2. Target Hypotheses
In this paper, we consider narrowband transmitters. As a consequence, the scattering function of the general target, which is denoted by , can be assumed to be independent of the transmitted frequency. Moreover, assuming that the target can be described as a set of points,
the isotropic case for which the scattering coefficient neither depends on the transmitter nor depends on the receiver location; in this case, the target can be considered to be a set of isotropic points;
the nonisotropic case for which the scattering coefficient depends on the transmitter and receiver locations (indexed by ); in this case, the coefficients of each transmitter are different and the target is considered as a set of nonisotropic points.
In the following, we denote by the scattered field of the isotropic point located at pixel and by the scattered field seen by transmitter of the nonisotropic point located at pixel .
2.3. Detection Problem for the Isotropic Case
We consider the following detection problem for the pixel at ; the received signal may be composed either of noise (hypothesis ) or of noise plus the reflected target signal (hypothesis ):
where is the signal received from transmitter , is complex zero mean Gaussian noise with known variance , and is the noiseless signal reflected by a normalized isotropic point, originating from the transmitter . The expression for is
where is the time delay along the transmitter -target-receiver path. The scattering coefficient is unknown.
The initial detection problem (2) may be rewritten as
where is unknown.
2.4. Detection Problem for the Nonisotropic Case
We consider the following detection problem for the pixel at , where the received signal may be composed either of noise (hypothesis ) or of noise plus the reflected target signals (hypothesis ):
where the definitions of , , and are the same as those of Section 2.3. The scattering coefficients are unknown.
For the nonisotropic case, is defined as follows:
The initial detection problem (6) may be rewritten as
where are unknown.
Both the isotropic (5) and the nonisotropic (8) detection problems described above are addressed in Sections 3.1 and 3.2. Moreover, simulations of the two resulting MISAR processors are provided and analyzed in Section 3.3.
3.1. MISAR Processor for Isotropic Case
The Generalized Likelihood Ratio Test for the detection problem (5) leads to
where is the probability density function of vector under the hypothesis . The threshold is usually adjusted to fulfill a given probability of false alarm. In Gaussian context, the above test may be rewritten as
Estimation of consists in determining
which leads to the least-squares solution
where denotes the conjugate transpose operation. Substituting into (10),
We propose to compose the target image by plotting the test value of (13) for every position:
Let denote the bistatic complex image created using only the transmitter . It is to be noted that (14) makes a coherent sum of every bistatic complex image which is identical to classic multistatic algorithms . We call this algorithm Multistatic Coherent ISAR (MCISAR) processing.
The computation cost of the bistatic complex image can prove to be heave. Nevertheless, an approximation such as the bistatic PFA proposed in  could be used with a view to better computational efficiency.
3.2. MISAR Processor for Nonisotropic Case
The Generalized Likelihood Ratio Test for the detection problem (8) leads to
In Gaussian context, the above test may be rewritten as
Estimation of consists in determining
which leads to the least-squares solution
Substituting into (16),
We propose to compose target image by plotting the test value of (19) for every position:
Equation (20) is an incoherent sum of every bistatic complex image. This result is consistent with the fact that the scattering coefficients are assumed to be different for each transmitter. Equation (20) thus differs from MCISAR (14) and classic multistatic algorithms that usually perform a coherent sum of every bistatic complex image. Such a result emphasizes our interest in modeling the received signal more accurately. We call this algorithm Multistatic Incoherent ISAR (MIISAR) processing.
3.3. Qualitative Comparison of MCISAR and MIISAR
Center frequency (MHz)
Scattering coefficient of isotropic point
Scattering coefficient of nonisotropic point (modulus, phase)
( 23.8, 54.45)
(0.66, 61. )
(2.19, 85.5 )
(0.22, 144.1 )
( 38.4, 40.2)
(0.73, 8.6 )
In the next section a quantitative study of both MISAR processors is performed. Resolution and sidelobe levels are studied in a realistic ISAR context and the MCISAR and MIISAR processors are applied to complex targets.
The first subsection describes the FM passive radar configuration and the considered targets. The second subsection is dedicated to the study of the impact of the system geometry on the image resolution for the MIISAR. Finally, in the third subsection, the MCISAR and MIISAR processors are applied to complex targets and the results are analyzed.
4.1. Simulation Context
The center frequencies of each of the narrowband transmitters are all different and are separated by at least kHz. We assume that the center frequency for transmitter is equal to MHz. This choice was made so as to have the total frequency coverage in the FM band and therefore to obtain the best image resolution in the chosen geometric configuration.
4.2. Study of the Impact of the Geometric Configuration on Image Resolutions
In this subsection, we focus on the final image resolution and sidelobe levels that could be obtained by the MIISAR processor in the configuration of Figure 5. The MIISAR processor is chosen because its result corresponds to the envelope of the MCISAR one. Therefore, the impact of parameters, such as the integration time and the number of transmitters, on resolution and sidelobe levels is obviously the same for both processors.
Our resolution study is similar to the one reported in [8, 9]. First, let the target be an isotropic point located at coordinate . The result for this target and for a given configuration is obtained by the MIISAR processor. An ellipse is obtained from this image by computing the image level at the maximum minus 3 dB. Several parameters, defined in Figure 10, are determined from this ellipse to measure the quality of the image resolution: the area of the ellipse in , and the maximal and minimal resolution in (the large and small axes of the ellipse).
To study sidelobe levels, the following parameter, denoted the Mean Absolute Error (MAE), is used:
where and are the number of pixels of the image and is a reference image, which is assumed of high quality in terms of resolution and sidelobe levels.
Now, we describe all cases tested as a function of the and from the geometric configuration of Figure 5.
SAL = 4 km and transmitters.
SAL = 8 km and transmitters.
SAL = 16 km and transmitters.
SAL = 32 km and transmitters.
SAL = 32 km and transmitters, with a frequency spacing of 200 kHz.
SAL km and transmitters, with a frequency spacing of 800 kHz.
Results of resolution parameters and MAE for Cases 1–6.
Number of the case
Number of transmitters
Frequency Spacing (kHz)
Resolution Area ( )
Maximal resolution ( )
Minimal resolution ( )
It is first to be noted that relatively good image resolution is obtained with a small SAL when all transmitters are considered. This might mislead us into thinking that a small SAL is sufficient to obtain usable images of complex targets. Unfortunately, the sidelobe levels are high for these cases, as indicated by the MAE value. The MAE decreases as the SAL increases, which is reasonable. In particular, the result for an SAL of 32 km and 103 transmitters seems promising in terms of resolution and sidelobe levels. However, we are interested in reducing the number of transmitters to reach a less complex radar system; also, it is unlikely that so many transmitters could be used in practice.
4.3. Simulation Results
In this subsection, we apply the MCISAR and MIISAR processors within the simulation context of Section 4.1. In this discussion, copolarization refers to polarization and cross polarization refers to polarization.
These results show that both processors applied to copolarization and cross-polarization data might help developing classification rules. Moreover, the processors appear to be robust with respect to violations of the hypotheses they were derived under.
The first processor, called MCISAR, is based on the isotropic point hypothesis. It is analogous to classical processors as it coherently sums bistatic images.
The second processor, called MIISAR, assumes that the scattered field of the target is different for each narrowband transmitter (realistic hypothesis in our context). This processor incoherently sums bistatic images.
Numerical simulations have been done in order to qualitatively illustrate the potential of the MIISAR processor in cases where the isotropic point hypothesis is not respected.
The last section applied the two considered processors to two similar complex targets in a realistic multistatic FM passive radar scenario. A study of the resolution and sidelobe levels has shown that it is not necessary to take every transmitters into account to get an optimal result. However, using transmitters that provide with wide frequency spacing and a large SAL is needed to achieve decent results.
Copolarization results from the MCISAR processor allowed us to determine some simple parameters of both the considered targets, such as the length and the width of the cylinder; however, images of both targets were similar and the wings and the cone were not seen. The MIISAR processor is proved to be more performant for discriminating the two targets because the images provided in copolarization are different. Moreover, the cross-polarization images obtained by the MIISAR processor exhibited many differences. According to those results, the proposed imaging algorithms may be used as first steps of a classification process using multifrequency and multistatic passive radar.
Future work will deal with more complex target models and the development of associated imaging algorithms based on the monostatic SAR processors presented in [19, 20] where subspace approaches are used to integrate complex target model in monostatic SAR processing.
- Cantalloube H, Dubois-Fernandez P: Airborne x-band sar imaging with 10 cm resolution: technical challenge and preliminary results. Proceedings of the 5th European Conference on Synthetic Aperture Radar (EUSAR '04), May 2004, Ulm, GermanyGoogle Scholar
- Howland PE, Maksimiuk D, Reitsma G: FM radio based bistatic radar. IEE Proceedings: Radar, Sonar and Navigation 2005, 152(3):107-115. 10.1049/ip-rsn:20045077Google Scholar
- Howland PE: Target tracking using television-based bistatic radar. IEE Proceedings: Radar, Sonar and Navigation 1999, 146(3):166-174. 10.1049/ip-rsn:19990322Google Scholar
- Ehrman L, Lanterman AD: Automated target recognition using passive radar and coordinated flight models. In Automatic Target Recognition XIII, April 2003, Orlando, Fla, USA, Proceedings of SPIE Edited by: Sadjadi FA. 5094: 196-207.View ArticleGoogle Scholar
- Ehrman LM, Lanterman AD: A robust algorithm for automatic target recognition using passive radar. Proceedings of the 36th IEEE Annual Southeastern Symposium on System Theory, March 2004, Atlanta, Ga, USA 102-106.Google Scholar
- Ehrman LM, Lanterman AD: A robust algorithm for automated target recognition using precomputed radar cross sections. In Automatic Target Recognition XIV, April 2004, Orlando, Fla, USA, Proceedings of SPIE Edited by: Sadjadi FA. 5426: 197-208.View ArticleGoogle Scholar
- Herman SM: A particle filtering approach to joint passive radar tracking and target classification, Ph.D. thesis. University of Illinois, Urbana, Ill, USA; 2002.Google Scholar
- Maussang F, Daout F, Ginolhac G, Schmitt F: Multistatic scenarios for a GPS SAR system. Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS '08), July 2008, Boston, Mass, USA 3: 270-273.Google Scholar
- Daout F, Schmitt F, Ginolhac G: Performance analysis of a bistatic/multistatic passive sar system based on a GPS signal. submitted to IEEE Transactions on Geoscience and Remote SensingGoogle Scholar
- Soumekh M: Synthetic Aperture Radar Signal Processing. Wiley-Interscience, New York, NY, USA; 1999.MATHGoogle Scholar
- Mensa DL, Heidbreder GR: Bistatic synthetic-aperture radar imaging of rotating objects. IEEE Transactions on Aerospace and Electronic Systems 1982, 18(4):423-431.View ArticleGoogle Scholar
- Walker JL: Range-doppler imaging of rotating objects. IEEE Transactions on Aerospace and Electronic Systems 1980, 16(1):23-52.View ArticleGoogle Scholar
- Wu Y, Munson DC Jr.: Multistatic synthetic aperture imaging of aircraft using reflected television signals. In Algorithms for Synthetic Aperture Radar Imagery VIII, April 2001, Orlando, Fla, USA, Proceedings of SPIE Edited by: Zelnio EG. 4382: 1-12.View ArticleGoogle Scholar
- Wu Y, Munson DC Jr.: Wide-angle ISAR passive imaging using Smoothed Pseudo Wigner-Ville distribution. Proceedings of the IEEE National Radar Conference, May 2001, Atlanta, Ga, USA 363-368.Google Scholar
- Wu Y, Munson DC Jr.: Multistatic passive radar imaging using the Smoothed Pseudo Wigner-Ville distribution. Proceedings of IEEE International Conference on Image Processing (ICIP '01), October 2001, Thessaloniki, Greece 3: 604-607.Google Scholar
- Lanterman AD, Munson DC Jr., Wu Y: Wide-angle radar imaging using time-frequency distributions. IEE Proceedings: Radar, Sonar and Navigation 2003, 150(4):203-211. 10.1049/ip-rsn:20030712Google Scholar
- Lanterman AD, Munson DC Jr.: Deconvolution techniques for passive radar imaging. In Algorithms for Synthetic Aperture Radar Imagery IX, April 2002, Orlando, Fla, USA, Proceedings of SPIE Edited by: Zelnio EG. 4727: 166-177.View ArticleGoogle Scholar
- Cetin M, Lanterman AD: Region-enhanced passive radar imaging. IEE Proceedings: Radar, Sonar and Navigation 2005, 152(3):185-194. 10.1049/ip-rsn:20045019Google Scholar
- Durand R, Ginolhac G, Thirion L, Forster P: New SAR processor based on matched subspace detectors. IEEE Transactions on Aerospace and Electronic Systems 2009, 45(1):221-236.View ArticleGoogle Scholar
- Durand R, Ginolhac G, Thirion L, Forster P: SAR processor based on a CFAR signal or interference subspace detector matched to man made target detection in a forest. Proceedings of IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP '07), April 2007, Honolulu, Hawaii, USA 2: 293-296.Google Scholar
- Agence National des Fréquences : Emplacement des émetteurs fm en france. http://www.cartoradio.fr/netenmap.php?cmd=zoomfull
- Daout F, Schmitt F, Ginolhac G: Computation of bistatic ser with nec2 in a context of passive isar system. Proceedings of IEEE Geoscience and Remote Sensing Symposium (IGARSS '05), July 2005, Seoul, South KoreaGoogle Scholar
- Petsios MN, Alivizatos EG, Uzunoglu NK: Manoeuvring target tracking using multiple bistatic range and range-rate measurements. Signal Processing 2007, 87(4):665-686. 10.1016/j.sigpro.2006.07.003View ArticleMATHGoogle Scholar
- Forster P, Lesturgie M: Passive radar for target tracking and classification. Proceedings of the International Radar Conference, October 2004, Toulouse, FranceGoogle Scholar
- Tobias M, Lanterman AD: Probability hypothesis density-based multitarget tracking with bistatic range and Doppler observations. IEE Proceedings: Radar, Sonar and Navigation 2005, 152(3):195-205. 10.1049/ip-rsn:20045031Google Scholar
- Tobias M, Lanterman AD: Techniques for birth-particle placement in the probability hypothesis density particle filter applied to passive radar. IET Radar, Sonar and Navigation 2008, 2(5):351-365. 10.1049/iet-rsn:20070051View ArticleGoogle Scholar
- Rigling BD, Moses RL: Polar format algorithm for bistatic SAR. IEEE Transactions on Aerospace and Electronic Systems 2004, 40(4):1147-1159. 10.1109/TAES.2004.1386870View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.