Research Article  Open Access
Radar Coincidence Imaging under Grid Mismatch
Abstract
Radar coincidence imaging is an instantaneous imaging technique which does not depend on the relative motion between targets and radars. Highresolution, finequality images can be obtained using a single pulse either for stationary targets or for complexly maneuvering ones. There are two imagereconstruction algorithms used for radar coincidence imaging, that is, the correlation method and the parameterized method. In comparison with the former, the parameterized method can achieve much higher resolution but is seriously sensitive to grid mismatch. In the presence of grid mismatch, neither of the two algorithms can obtain recognizable highresolution images. The above problem largely limits the applicability of radar coincidence imaging in actual imaging scenes where grid mismatch generally exists. This paper proposes a joint correlationparameterization algorithm, which uses the correlation method to estimate the gridmismatch error and then iteratively modifies the results of the parameterized method. The proposed algorithm can achieve high resolution with fine imagery quality under the grid mismatch. Examples are provided to illustrate the improvement of the proposed method.
1. Introduction
Radar coincidence imaging is developed as the extension of classical coincidence imaging in microwave radar systems [1]. Classical coincidence imaging, which is realized in thermal optical systems, is a method to nonlocally image an object with high resolution by producing sharply fluctuating signals in the imaging plane [1, 2]. Similarly, the essential principle of radar coincidence imaging is to produce timespace independent signals in the detecting area. Then radar signals at different positions have mutually independent waveforms. Illuminated by such signals, targetscattering centers within a radar beam will also reflect echoes of independent waveforms associated with their respective positions. Therefore, the echo component of each scattering center can be extracted from the receiving signal and then correlated to their respective positions. Consequently, scattering centers within a beam can be resolved and the spatial distribution of the target scattering centers can be obtained. Obviously, it is quite different from the radar imaging techniques based on the rangeDoppler (RD) principle, where scattering centers are resolved based on the analysis of timedelay and Doppler frequency [3].
Because of the different imaging principles, the radar coincidence imaging technique has two advantages over most of the RD imaging methods. Firstly, radar coincidence imaging does not depend on the aspectangle integration or the Doppler gradient to achieve high azimuth resolution. Thus, it does not require relative motion between radars and targets and can obtain images of the targets which remain stationary with respect to radars. Furthermore, radar coincidence imaging can achieve highresolution using a single pulse. The extremely short imaging time, which is shorter than a pulse width, considerably decreases the impact of the noncooperative motion to imagery quality. Therefore, radar coincidence imaging can obtain highresolution, finequality images either for stationary targets or for the ones in complex maneuvers.
In radar coincidence imaging, the target area needs to be discretized to a grid and targetscattering centers are assumed to be located at the grid points. If scattering centers are located off the grid points, then the grid mismatch yields. There are two main imagereconstruction methods used in radar coincidence imaging. One is the correlation method with lower resolution. The other is the parameterized method which can produce much higher resolution but is too sensitive under grid mismatch to give recognizable target images. Furthermore, current algorithms that are applicable to solve the sensitivity to basis mismatch are unfortunately ineffective for the grid mismatch in radar coincidence imaging [4, 5]. This limitation seriously affects the applicability of the imaging method in actual scenarios where the grid mismatch generally exists. This paper proposes a joint correlationparameterization method for image reconstruction which can achieve both high resolution and good imagery quality under grid mismatch.
The paper is organized as follows. Section 2 is devoted to the basic principles of radar coincidence imaging. Section 3 analyzes the impact of grid mismatch and proposes the joint correlationparameterization method for image reconstruction. Section 4 concludes the work.
2. Basic Principles
Coherent signals, which are widely employed by most of the imaging radars, generally produce detecting signals that show significant spatial correlation, as shown in Figure 1(b). However, the essence of radar coincidence imaging is to produce timespace independent radar signals in the detecting area. In other words, the detecting signals at different instants or in different positions are independent of each other. As shown in Figure 1(a), at an arbitrary instant, the timespace independent detecting signals fluctuate sharply in the imaging plane and its instantaneous spatial distribution presents remarkable variety.
(a)
(b)
Generally, the target location can be firstly estimated based on the detection and localization techniques [6, 7], which is defined as the center of the target area. Then a local coordinate system is established in the target area center, as shown in Figure 2. The target area is discretized to a grid consisting of small rectangles of uniform size and shape. Each small rectangle is the grid cell and is approximated by its own center. Thus, targetscattering centers are actually assumed to be initially located at the gridcell centers. The discrete target area is expressed as , where is the position vector of the th grid cell center. There are transmitters and a receiver, whose position vectors are and , respectively. The transmitted signal of the th transmitter is denoted as and the receiving signal is . The detecting signals distributed in the target area are labeled as , where is the position vector of an arbitrary grid cell within .
The ideal transmitted signals for radar coincidence imaging are supposed to be grouporthogonal and timeindependent as denoted in (1), which can produce the timespace independent via single transmitting [1]. Consider where , is the envelope function and is the pulse width. If the antenna number is more than 2, for example, , the detecting signal produced by such transmitted signals will show approximate timespace independence [1], expressed as follows: Then, we express the receiving signal as the superposition of the detecting signals. Consider where is the scattering coefficient of the scattering center located at the th grid cell and for the grid cell without targetscattering center . For the sake of simplicity, the coincidence imaging formulism needs a reference signal [1], which herein can be simply structured using as follows: Consequently, (3) becomes Then, the scattering coefficient of an arbitrary grid cell can be explicitly obtained via the correlation between the receiving signal and , That is,
Note that the detecting signal or can be computed based on the known transmitted waveforms. Thus, the scattering coefficient of each grid cell can be derived via the correlation between the receiving signal and the reference signal. Therefore, the target scene can be recovered via conducting the correlation in (7) for the whole target area grid. This method to reconstruct target images is defined as the correlation method.
The excellent pointtopoint relationship in (7) means a high resolution, but it requires that the transmitted signals have the perfect timeindependence as presented in (1). Unfortunately, microwave transmitted signals are far from this requirement and their timeindependent degree is inadequate to achieve the high resolution represented in (7) via the correlation method [1]. Therefore, to improve the resolution of radar coincidence imaging, the parameterized method is employed which is less constrained by the signal timeindependence. This method uses the relationship between the receiving signal and the reference signal, as shown in (5), to structure a radar coincidence imaging equation, given as follows: where , is the reference signal matrix, Sr is the vector of the receiving signal, and is the unknown vector of the scattering coefficient. The columns and the rows of basically represent the detecting signals in different positions and at different instants, respectively. Hence, the incoherent property of is basically determined by the timespace independence of the detecting signals. Therefore, the independent characteristic of the detecting signal will ensure a fullrank . Consequently, the target scatteringcoefficient vector can be uniquely recovered as .
3. Impact of the Grid Mismatch and Image Reconstruction
The imaging equation presented in (8) is derived under the gridmatch condition. Generally, regardless of how finely the target area is gridded, targetscattering centers may not lie in the gridcell centers. Assume that there exists the grid mismatch. The scatteringcenter position vector actually is instead of the assumed for the lth grid cell. Thus, the actual receiving signal is . Consequently, the imaging equation should be , where is caused by the difference between and . As and are both unknown variables, the above imaging equation can be simplified as where and denotes the modeling error caused by grid mismatch. Then (9) represents the wellknown linear signal model. Thus, the actual scatteringcoefficient vector should be derived as . The previous solution derived from (8) is relabeled as . That is, , where is neglected because of the gridmatch assumption. Certainly, the imaging quality of the parameterized method will be decreased due to the modeling error.
To illustrate the gridmismatch impact, we give an example to show how the position bias affects the modeling error and how the modeling error affects the imaging quality. In the quantitative manner, the imaging quality discussed here is firstly indicated by the relative imaging error, expressed as . Then the modeling error is normalized as the signalerrorratio, denoted as . The position bias is defined as , where is the gridcell perimeter.
The example employs an transmitter 1receiver array, consisting of 5 antennas. The arrangement of the antennas and the target is shown in Figure 3. The target area is 2 m × 2 m and is discretized to 40 × 40 grid cells, each of which is 5 cm × 5 cm. The grouporthogonal and timeindependent transmitted signals herein are generated by modulating the sinusoid signals on amplitude via mutually independent Gaussian noises. The carrier frequency is 9.5 GHz, the pulse width is 50 μs, the bandwidth is 1 GHz, and the sampling frequency is 2 GHz. The results are shown in Figure 4.
In this example, the target image in Figure 4(b) is badly blurred beyond reorganization when . The target image in Figure 4(c) is recognizable when the imaging error is 0.9, which requires that should not be lower than 46 dB. It demonstrates that the parameterized method has a low tolerance to the modeling error. A recognizable image with a small imaging error requires a small modeling error or a high signalerrorratio. Then the question is “will the position bias generally produce a highlevel signalerrorratio to guarantee low imaging error?” Unfortunately, as shown in Figure 3(d), even a very small position bias of 0.001 in this example can cause a large modeling error, which only provides 20 dB singleerrorratio. Moreover, the position bias less than 0.02 will generate a negative . It shows that the modeling error is seriously sensitive to the position bias. The position bias of 0.04 can cause −4 dB signalerrorratio in radar coincidence imaging equation, with no mention of a much larger bias in practice. In this case, modeling error almost overwhelms the receiving signal in the imaging equation. The experiment indicates that the recovery, simply depending on the imaging equation, might impossibly provide a good estimation under such a poor signalerrorratio condition.
Consider another example to compare the correlation method with the parameterized method under grid mismatch. In the example, target images will be reconstructed via the two methods with and without the grid mismatch. Herein, the grid mismatch leads to . Imaging results are shown in Figure 5.
(a)
(b)
(c)
(d)
As shown in Figure 5, the parameterized method provides higher resolution than the correlation method in the case of no grid mismatch. On the other side, when , the imaging quality of the correlation method is almost unchanged but the parameterizedmethod result gets degraded and almost gives no information on the target shape. It is obvious that the correlation method is less sensitive to the modeling error than the parameterized method but has much lower resolution.
In conclusion, neither of the two imagereconstruction methods can achieve both high resolution and good imaging quality in the presence of grid mismatch. According to their respective limitations, there are two ways to solve the image reconstruction under grid mismatch. One way is to improve the resolution of the correlation method. The other one is to decrease the gridmismatch impact in the parameterized method. As stated previously, however, microwave signal does not have adequate timeindependence in nature, resulting in a limited resolution. Thus, the latter might be a possible way to solve the problem of grid mismatch.
A direct idea to decrease gridmismatch impact in the parameterized method is to estimate the modeling error. Note that the modeling error is relevant to the scattering center number, scattering intensity, and the positionbias level of all scattering centers. If there is no prior information of the target shape, the modeling error will be quite difficult to estimate. Then for the targets that are unknown in advance, the obtainable shape information could be sought from the imaging results of the two reconstruction methods. Thus, a possible source to provide the desired knowledge would be the correlationmethod result, considering the seriously distorted images of the parameterized method caused by grid mismatch.
The knowledge that can be used by the imaging equation should be quantitative and precise. However, the result of the correlation method, as observed in Figure 5, presents a blurry target shape. The image cannot precisely indicate the target scatteringcenter location or provide the exact value of the scattering coefficient. Nonetheless, it indeed provides valuable information. That is, the correlationmethod result can denote part of the grid cells which do not contain targetscattering centers. In Figures 5(a) and 5(c), only a minority of image patches, which constitute the target contour, have large values. By contrast, a majority of other image patches have far too small values, which implies the corresponding grid cells have no scattering centers. Hence, these corresponding elements in the scatteringcoefficient vector are estimated to be zero. Such quantitative knowledge will be helpful for the estimation of the modeling error. Detailed implementation of image reconstruction using the idea will be described in the following paragraphs.
Firstly, still under the gridmatch assumption, we estimate the target scatteringcoefficient vector via the two reconstruction methods: is derived via the correlation method and is obtained via the parameterized method. According to and (9), we have the following relationship: where is written as , being the rowvector of . The next step is to select the grid cells without scattering centers according to . In comparison with the average value of , part of the elements in , which have far too small values, are determined to be 0. The index of these zeroelements is denoted as ; that is, . According to , we define , , and . Based on (10), we have As , finally . Therefore, the modeling noise is estimated as where is the pseudoinverse matrix of . By substituting to (9), the new target scatteringcoefficient vector obtained is
If the result remains unsatisfactory, we can repeat the computation of (12) and (13) using the new and . That is, update the receiving signal as . Then derive according to and update the modeling error as . Finally, the estimation is updated as . The update can be performed iteratively until the result is satisfactory. This imagereconstruction method using both the correlation method and the parameterized method is defined as the joint correlationparameterization method. The detailed steps are given in Table 1.

To examine the correlationparameterization method, target images are reconstructed according to the iteration process of Table 1, where the grid mismatch is the same as that in Figure 5(d). The imaging results are given in Figure 6.
(a)
(b)
(c)
As shown in Figure 6, a highquality target image can be obtained through three iterations. The result of the first iteration has presented the target contour, though the contour is light and the imaging error remains remarkable. After the second iteration, the imaging quality is largely improved. By contrast, the imaging quality of Figure 6(b) is much better than the result in Figure 5(d). The third iteration maintains the good imagery quality. Therefore, two iterations actually give a highresolution and finequality target image. It indicates that the joint correlationparameterization method satisfies the requirement for both the resolution and the robustness in the presence of the modeling noise caused by grid mismatch.
4. Conclusions
Radar coincidence imaging can achieve excellent highresolution target images on the condition of grid match, but the imagery quality gets degraded beyond recognition in the presence of the modeling error caused by grid mismatch. Therefore, the paper proposes the joint correlationparameterization method for image reconstruction. The proposed algorithm iteratively modifies the parameterizedmethod result with the estimated modeling error, which is obtained based on the correlationmethod result. Consequently, the gridmismatch impact on the imaging quality is considerably reduced. The example shows that the joint correlationparameterization method can achieve high resolution and maintain the robustness under grid mismatch.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
References
 D. Li, X. Li, Y. Cheng, Y. Qin, and H. Wang, “Radar coincidence imaging: an instantaneous imaging technique with stochastic signals,” IEEE Transactions on Geoscience and Remote Sensing, vol. 52, no. 4, 2014. View at: Publisher Site  Google Scholar
 A. Gatti, E. Brambilla, M. Bache, and L. A. Lugiato, “Ghost imaging with thermal light: comparing entanglement and classical correlation,” Physical Review Letters, vol. 93, no. 9, Article ID 093602, 2004. View at: Publisher Site  Google Scholar
 Y. Shih, “Quantum imaging,” IEEE Journal on Selected Topics in Quantum Electronics, vol. 13, no. 4, pp. 1016–1030, 2007. View at: Publisher Site  Google Scholar
 Y. Chi, L. L. Scharf, A. Pezeshki, and A. R. Calderbank, “Sensitivity to basis mismatch in compressed sensing,” IEEE Transactions on Signal Processing, vol. 59, no. 5, pp. 2182–2195, 2011. View at: Publisher Site  Google Scholar
 M. A. Herman and T. Strohmer, “General deviants: an analysis of perturbations in compressed sensing,” IEEE Journal on Selected Topics in Signal Processing, vol. 4, no. 2, pp. 342–349, 2010. View at: Publisher Site  Google Scholar
 P. Z. Peebles, Radar Principles, John Wiley & Sons, New York, NY, USA, 1998.
 J. S. Thorp, “Optimal tracking of maneuvering targets,” IEEE Transactions on Aerospace and Electronic Systems, vol. 9, no. 4, pp. 512–519, 1973. View at: Google Scholar
Copyright
Copyright © 2014 Dongze Li et al. 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.