Low-Cost Radar Cross Section Measurement With a Resin-Made Model Coated With Conductive Paste

This study aims to provide practical information regarding the development of a cost-effective test model for low-cost radar cross Section (RCS) measurements. Two major possible approaches for reducing the manufacturing cost of a model are replacing the molding material with an inexpensive one and measuring a scaled-down model with a higher operating frequency. In this study, the former was achieved using a resin-made model coated with a conductive paste to imitate a complete metallic model. According to the scaling laws, the entire geometry, including the surface roughness of the model because of the conductive paint, must be appropriately scaled down. Because this roughness scaling is practically infeasible, we experimentally prove that, if an appropriate molding material is selected, shape-only scaling is valid without roughness scaling. As a preliminary study, two types of polyurethane resins with different densities were tested for their suitability for RCS measurements, and the resin with the higher density was shown to be appropriate for the RCS test model. Based on this preliminary study, 1/20- and 1/40-scale aircraft models were manufactured using aluminum or polyurethane resin, and their RCS values were measured to demonstrate that almost equivalent RCS patterns could be obtained. Therefore, low-cost RCS testing was achieved using the resin-made scaled model.


I. INTRODUCTION
M EASURING the radar cross section (RCS) at a low cost is important for the cost-effective design and development of low-RCS products such as stealth aircraft. For low-cost RCS measurements, the following two aspects must be considered. First, the RCS testing facility should be as small as possible to reduce its construction and operating costs. Second, if the final product is unavailable at the time of RCS measurement, a mock-up must be used; therefore, the manufacturing cost of the test model should be reduced.
In this study, we primarily focused on the latter. Two major approaches for reducing the manufacturing cost of a model are: replacing the molding material with an inexpensive one [16], [17], [18], [19] and measuring a scaled-down model with a higher frequency [1], [2], [20], [21], [22], [23], [24], [25], [26]. In [18] and [19], resin-made aircraft models, whose sizes were approximately 1/32 and 1/64 of the full-scale model, were manufactured by stereolithography 3-D printing and highly conductive silver painting. The measured RCS of these models was in good agreement with the numerically calculated RCS of a perfectly conducting model with identical geometry.
Similar to the aforementioned studies [18], [19], this study aims at demonstrating the effectiveness of a resin-made scaled aircraft model as a substitute for a complete metallic model for RCS measurement. The major contribution of this study in light of previous studies is that, in addition to two resin-made models with different scales, we developed complete metallic models of identical shapes and scales for comparison. Therefore, we experimentally proved the equivalence of the metallic and resin models with different scales, without relying on comparisons of the measurement results with numerical simulations. According to scaling laws [2], the entire geometry, including the surface roughness of the models owing to the conductive paint, must be appropriately scaled down. Because this roughness scaling is practically infeasible, we show that shape-only scaling is valid without roughness scaling. Table I compares the present study and those reported in [18] and [19].
Moreover, we experimentally demonstrated that the molding material of a resin-made model must be considered for the accurate emulation of a metallic model. A model manufactured using very coarse resin material exhibited a considerably different RCS pattern compared to that of a metallic model even if the resin model was coated with conductive paste. We believe that this result will benefit researchers and engineers working in the field of RCS measurement and evaluation. The remainder of this article is organized as follows. Section II describes the methodology of the RCS measurement used in this study. Because of the spatial limitation of the measurement facility, we employed an NFFFT algorithm based on synthetic aperture image formation, referred to as the "imagebased NFFFT." As a preliminary study, Section III investigates the appropriate molding material for building a low-cost resin model for RCS measurements. Small sample models were manufactured using two types of polyurethane resin, and the RCS was measured to evaluate the equivalence of the resin models to a model manufactured using a perfectly conductive material. Section IV discusses the RCS measurement results for aircraft models developed based on a preliminary study in which resin and aluminum models were compared. Finally, Section V summarizes the conclusions of this study.

II. METHODOLOGY
In this section, we explain the methodology of the RCS measurement employed in this study. The RCS is usually defined in the far-field region of the test object. The spatial limitation of the measurement facility necessitates the use of an NFFFT algorithm [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15] to convert the near-field data into the far-field RCS. Moreover, we review the scale-model testing, which can reduce the manufacturing cost of the model.

A. Image-Based NFFFT
We briefly describe RCS determination using an image-based NFFFT [10], [11], [12], [13], [14]. Fig. 1 illustrates the concept of the image-based NFFFT. As shown in Fig. 1(a), the target under test is placed at the center of the (x, y) plane, and an antenna moves along an arbitrary scanning curve enclosing it. The main antenna beam was directed toward the target throughout the scanning. The antenna repeatedly transmitted an electromagnetic wave to the target, and collected the scattered wave from the target at the same antenna location (monostatic configuration). Assuming that the target comprised a cloud of equivalent point scatterers with unknown reflectivity, as shown in Fig. 1(b), we reconstructed the reflection coefficients from the received signal. After determining the spatial distribution of the reflectivity, we determined the scattered field at an arbitrary antenna location, including the far-field region of the target in which the RCS can be defined.
The formulation of image-based NFFFT is provided below [13], [14]. The arbitrary antenna scanning curve shown in Fig. 1(a) is denoted by the vector r 0 (u) parametrized by the variable u while an arbitrary spatial location on the 2-D (x, y) plane is represented by the vector r. These vectors are defined as follows: where x and y are unit vectors along xand y-directions, respectively. Under the aforementioned experimental configuration, the received signal at antenna location r 0 can be given by where k = ω/c represents the wavenumber with the angular frequency ω and wave propagation speed c; S is the spatial domain occupied by the target; P(k, r 0 , r) is the antenna amplitude pattern in the direction from r 0 toward r; and C(r) denotes the continuous reflectivity at the location r. Considering reflectivity as a cluster of point scatterers, as shown in Fig. 1(b), the reflectivity C(r) takes the following form: where δ(·) is the delta function. In this discrete case, the received signal in (2) can be expressed as follows: Therefore, after determining the ith scatter location r i and the reflection coefficient C i , we can reconstruct the scattered field at an arbitrary location using (4). The scatterer location r i represents the known predetermined location of the ith image pixel (usually a point on a uniform Cartesian grid), and the reflection coefficient C i is the unknown quantity to be reconstructed.
To determine the reflection coefficient C 1 at location r 1 = (x 1 , y 1 ), we consider the following ideal image according to (3) with i ∈ {1}: The received signal corresponding to (5) can be defined by (4) with i ∈ {1} as follows: Our objective was to recover the ideal image given by (5) from the received signal of (6). This can be accomplished by the following integral transformation of the received signal E s (k, r 0 ) with a weighting function F(k, r 0 , r): where D represents the domain of the integration with respect to the parameter u, and F(k, r 0 , r) is called the "focusing factor," which is defined as follows: In (8), the function g(k, r 0 , r), which is referred to as the "correction factor," is expressed as follows [13], [14]: where α denotes the azimuth angle of the antenna viewed from the local coordinate centered at location (x, y). It is given by where If the scanning curve (x 0 , y 0 ) is a simple circle of radius a centered at the origin of the coordinate system, the curve is expressed as In this case, the explicit form of the correction factor in (9) can be exactly derived as follows [13]: By reconstructing a spatial image with the aforementioned focusing factor F(k, r 0 , r) defined in (8) using the correction factor g(k, r 0 , r) of (9), we can obtain the ideal image defined in (5). Therefore, the scattered field at an arbitrary location can be calculated using (4).
The final step of the image-based NFFFT computes the RCS from the reconstructed image. The RCS is defined as where r 0 = |r 0 |, k r = 2k r 0 /r 0 , and E i 0 (k, r 0 ) is the incident field. Assuming that the incident field E i 0 (k, r 0 ) is given by and considering the far-field distance r 0 → ∞, the RCS in (14) can be expressed as follows [13], [14]: Therefore, after reconstructing the spatial image ψ(r), the Fourier transform of the image yields the far-field RCS. This completes RCS determination using an image-based NFFFT.

B. Scale-Model Testing
As is well known, the most effective way to reduce the cost for building RCS models is by using a scaled model [1], [2], [20], [21], [22], [23], [24], [25], [26]. If a full-scale test object is assumed to be perfectly conducting, the scaling law is simple [2]. We consider an original (1/1-scale) model with RCS σ measured at frequency f . If this model is scaled down to 1/s, where s represents the scaling factor, then the RCS σ ′ of the scaled model measured at the scaled frequency f ′ = s · f and original RCS σ are related by Therefore, if the test object is assumed to be highly conductive and an appropriate signal source with an operating frequency f ′ is available, low-cost RCS measurements can be performed using the 1/s-scale test model. This study attempts to experimentally prove the validity of scale-model testing with a resin-made model coated with conductive paste, namely, to demonstrate the equivalence between full-scale metallic and scaled-down resin models. The resin model exhibited surface roughness owing to the metallic particles in the conductive paste. Therefore, the surface roughness must be properly modified to scale the entire model geometry. As the scaling of the surface roughness is practically infeasible, it must be ignored. This is expected to be valid when the surface roughness is considerably smaller than the operating wavelength. We experimentally demonstrate the validity of this shape-only scaling approach in Section IV.

III. PRELIMINARY STUDY
In this section, as a preliminary study, we consider the appropriate molding material of a resin-made model for the RCS measurement. Simple test models were built with different polyurethane resins, and their RCS models were measured after painting their surfaces with a conducting paste. To examine the validity of the models, the resultant RCS patterns were then compared with the RCS simulated using a numerical electromagnetic solver.

A. Molding Material and Conductive Paste
As listed in Table II, two types of polyurethane resins (SANMODUR MAX and SX, Sanyo Chemical Industries, Ltd., Japan) with different densities were considered in this preliminary study. In the following discussion, we refer to materials with densities of 0.64 and 0.26 g/cm 3 as highand low-density resins, respectively. The surface roughness values of the high-and low-density resins are 8 and 15 µm, respectively.
Stereomicroscopic photographs of the machined surfaces of the materials are shown in Fig. 2. Because these materials are  foamed resins, many air bubbles are observed on the machined surfaces; the air bubbles on the high-density resin shown in Fig. 2(a) are smaller than those on the low-density resin depicted in Fig. 2(b). According to the manufacturer's information, the average cell sizes of these air bubbles were 58 and 80 µm for the high-and low-density resins, respectively. The cell size affected the surface roughness after machining and painting.
The surfaces were painted with conductive paste (DOTITE FE-107, Fujikura Kasei Company Ltd., Japan) containing silver-copper (Ag-Cu) particles as the conductive filler. The electrical conductivity of this paste was 5 × 10 −4 · cm. A spray gun was used to apply the conductive paste, and paint was repeatedly sprayed to completely coat the surface. Fig. 3 shows the stereomicroscopic photographs of the painted surfaces. The thickness of the paint layer was estimated to be 35-45 µm. Because the high-density resin has a smoother surface than the low-density resin, the painted surface of the high-density resin shown in Fig. 3(a) looks finer than that of the low-density resin shown in Fig. 3(b).
Moreover, the painted surface of the low-density resin shown in Fig. 3(b) has a porous structure compared to the high-density resin shown in Fig. 3(a). As illustrated in Fig. 4, the machined surfaces are porous owing to the presence of air bubbles inside the material. As shown in Fig. 4(a), the surface of the high-density resin was expected to be smoothly covered with the conductive paste. However, for the low-density resin that contained relatively large air bubbles, the paste was unable to fill up the bubbles to smoothly cover the machined surface as only a small amount of conductive paste could reach inside the bubbles through the pores on the surface. Consequently, the surface of the low-density resin was rougher, and the thickness of the paint layer was more likely to be nonuniform than that of the high-density resin.
Although the surface roughness itself might have minor effects on the backscattering behavior of the painted surfaces because the roughness was smaller than the wavelength at microwave frequencies, the nonuniformity of the thickness of the conductive paint is likely to cause another problem:   Photographs of the test models. The surfaces with the circular cross section were unpainted in these photographs to show the bare material, but painted as well before the RCS measurement. (a) High-density resin. (b) Low-density resin. the penetration of an incident wave through the painted layer. We investigate the effect of the difference in the surface roughness and the resulting paint nonuniformity on the measured RCS in Section III-D.

B. Test Models
We developed test models for RCS evaluation using the two types of polyurethane resins discussed in Section III-A. The model geometry is illustrated in Fig. 5. Based on the 3-D model shown in Fig. 5, we machined the resin-made test models by precise computer numerical control and then applied the conductive paste mentioned in Section III-A on their surfaces. Fig. 6 shows the photographs of the manufactured test models. It is difficult to recognize the difference in the surface roughness with the naked eye. In Fig. 6, the surfaces with the circular cross section were unpainted to show the bare material. However, they were completely painted when the RCS measurements were performed.

C. Overview of the Measurement
We measured the RCS of the test models as described previously. The experiment was conducted in an anechoic chamber, and the data were processed using the image-based NFFFT algorithm explained in Section II. Fig. 7 depicts the measurement setup and Table III lists the experimental parameters. As shown in Fig. 7, the test model was placed on a styrene foam support mounted on a turntable. The model was aligned with its axis of rotational symmetry along the x-axis and the hemispherical part was directed toward the positive x-direction.
Two identical horn antennas (Model 640 or 638, Narda-MITEQ, USA, for X -or K -band measurements, respectively), where one was for transmission and another was for reception, were fixed at 2 m from the center of the rotation. Horizontally polarized transmission and reception (HH-polarization) were used in this experiment. These antennas were connected to a vector network analyzer (VNA, N5225B, Keysight Technologies, USA) through microwave coaxial cables, which was used as the transmitter and receiver. A 25 dB-gain power amplifier (GT-1040A, Giga-tronics, USA) was placed in the receiving path.
As listed in Table III, two measurement frequencies, namely, the X -band (10.2 GHz-center frequency) and K -band (22 GHz-center frequency), were used for the consistency of the parameters for the scale-model testing discussed in Section IV. The frequency bandwidths were 4 GHz for the X -band and 8 GHz for the K -band, which was twice as wide as the X -band bandwidth. The common angular sample spacing 0.4 • was employed; thus, the image radius for the K -band was approximately half that of the X -band. Similarly, the pixel spacing for the K -band was half of that of the X -band to avoid the aliasing effect in the spatial frequency domain.
To suppress undesired reflections from the chamber walls and fixtures used in this measurement, the empty chamber was measured using the exact same experimental parameters as those used in the measurement with the target placed, and the background data were coherently subtracted from the measured data of the target under test [2], [3], [27], [28], [29]. If we denote the complex-valued data with and without the target by E s (k, r 0 ) and E b (k, r 0 ), respectively, the background-subtracted data E s (k, r 0 ) to be passed to the NFFFT algorithm are defined as follows: We employed a 15 and 7.5 cm trihedral reflector as the RCS calibration targets [2], [3] for the X -and K -bands, respectively. The precise RCS of the trihedral reflectors was known in advance because they were computed using a numerical electromagnetic solver. Both the test and calibration targets were measured, and the collected data were processed using the NFFFT algorithm presented in Section II. Then, by comparing the precomputed RCS and measured RCS of the trihedral reflector, the required calibration factor was computed and applied to the measured RCS of the test model. If we denote the true RCS of the trihedral reflector by σ c , measured RCS of the trihedral reflector and test model by σ c and σ , respectively, the calibrated RCS σ can be obtained using the following equation: To evaluate the validity of the test model in terms of the similarity to a complete metallic model, the reference RCS was calculated using a numerical electromagnetic solver, assuming  that the shapes of the model built using a perfect electric conductor and the resin model were identical (shown in Fig. 5).

D. Results and Discussion
Figs. 8 and 9 show the reconstructed radar images for the X -and K -bands, respectively, where (a) and (b) correspond to the models consisting of the high-and low-density resin, respectively. The shape of the model is clearly imaged around the origin of the coordinates in the images shown in Figs. 8 and 9. Because the symmetrical axis of the model was along the x-axis, as shown in Fig. 7, the obtained images were expected to be symmetrical about the x-axis. As shown in Figs. 8(a) and 9(a), the radar images of the high-density resin exhibit the expected symmetry whereas the lower halves of the images of the low-density resin shown in Figs. 8(b) and 9(b), appear disturbed compared to Figs. 8(a) and 9(a).
As the disturbance of the image was clearly separated from the location of the model, some form of multiple scattering possibly occurred inside the model. As discussed in Section III-A, because the bare surface of the low-density resin was rougher than that of the high-density resin, the thickness of the paint was more likely to be nonuniform for the lowdensity resin. Therefore, one possible reason for this multiple scattering is the penetration of the incident wave through the conductive painting layer, followed by multiple reflections inside the resin and repenetration through the painting layer to be scattered back to the receiving antenna. We created other models with identical shapes, as shown in Fig. 5 with the same molding materials and conducting paste, and confirmed that the disturbance was observed again, only in the radar images of the model made of the low-density resin.
We determined the RCS of the models from the reconstructed radar images using (16). Figs. 10 and 11 represent the RCS calculated from the radar images shown in Figs. 8 and 9, respectively, plotted with the RCS of a perfectly conducting   model whose shape is identical to that in Fig. 5, computed using a numerical electromagnetic solver. Moreover, we used the following mean error to evaluate the validity of the resin model in terms of similarity to the metallic model: where N , σ dB [n], and σ dB [n] denote the total number of discrete samples of the calculated RCS pattern, nth sample of the RCS obtained from a numerical electromagnetic solver, and nth sample of the RCS determined from the radar image, respectively. The RCS values are defined in dBsm. The calculated mean errors σ e are shown in Figs. 10 and 11. For the high-density resin, the measured RCS exhibited excellent agreement with the simulated RCS over the entire azimuth angle. As shown in Figs. 10 and 11, the mean errors were 0.6 and 0.9 dB for the X -band and K -band, respectively. For the low-density resin, the measured RCS agreed with the simulated RCS for both the X -and K -bands within  Figs. 10 and 11. These differences were expected based on the radar images shown in Figs. 8(b) and 9(b), where the lower-half parts of the images show the disturbance indicating that the multiple scattering occurred when the antennas were located in the upper-half of the coordinate system. The mean error for the low-density resin is 2.0 dB for both the X -and K -bands.
Therefore, this preliminary study indicates that the molding material must be carefully selected to ensure that the painted surface is sufficiently smooth, and the conducting paste is uniformly painted on the machined surface. Thus, polyurethane resin with a higher density is a better choice for building RCS test models. The low-density resin may be used by polishing the machined surface before painting the conductive paste; however, this increases the manufacturing cost of the RCS test model.

IV. MEASUREMENT OF AIRCRAFT MODELS
In this section, we discuss the measurement results of aircraft models to demonstrate the validity of scale-model testing with a resin-made model coated with conductive paste.

A. Aircraft Models
We developed 1/20-and 1/40-scale aircraft models, and aluminum and resin models were manufactured for each scale. Based on the preliminary study presented in Section III, highdensity resin was used as the molding material for the resin models, and their surfaces were painted with the conductive paste that was also used for the test models in Section III. The geometries of the aircraft models are shown in Fig. 12, and all the four models, scales, measurement frequencies, and relative costs (cost factor) in terms of the manufacturing cost of Model-AX are listed in Table IV. Note that the 1/20-scale metallic model is the same as that investigated in [14]. We labeled each model for convenience of discussion, where the letter "A" or "R" represents "Aluminum" or "Resin," respectively, and the letter "X" or "K" represents the measurement frequency, i.e., "X -band" or "K -band," respectively.
A 1/40-scale model can be considered a 1/2-scale model of a 1/20-scale model. Therefore, according to the discussion in Section II-B, using the scaling factor s = 2, the RCS of a 1/40-scale model at 20 GHz (K -band) is equivalent to that of a 1/20-scale model at 10 GHz (X -band). In the following discussion, the K -band RCS was converted into the X -band RCS using (17) with the scaling factor s = 2. Fig. 13 shows the photographs of the aircraft models, where Fig. 13(a) and (b) depict the painting process of the conductive paste applied to the resin-made model and the final products, respectively. For the resin-made models (Model-RX and Model-RK), the thin needle-like parts attached to the front of the aircraft models were made with metal to ensure mechanical strength.  Table V lists the measurement parameters. These parameters are the same as those presented in Table III except that fully polarimetric measurements are performed in this experiment. Fig. 15 shows the photographs of the experiment. As shown in Figs. 14 and 15(a), the aircraft model was placed on a styrene foam support mounted on a turntable, and the model was rotated to create a circular synthetic aperture. The antennas used in Section III were also employed in this experiment, but the antennas could be arbitrarily rotated about the lineof-sight axis to measure different polarimetric channels. Other experimental descriptions are the same as those presented in Section III.

B. Overview of the Measurement
As described in Section III, we computed the reference RCS of the aircraft models using a numerical electromagnetic solver. The measured and calibrated RCS values were compared with the reference RCS based on (20) to evaluate the validity of the measurement models.

C. Results and Discussion
Figs. 16 and 17 show the reconstructed X -band radar images for HH-and VV-polarization, respectively. Here, (a) and (b) correspond to the 1/20-scale metallic (Model-AX) and resin model (Model-RX), respectively. All the following radar images were normalized by the HH-polarized maximum image intensity of the trihedral reflector used for the RCS calibration, whose dimensions were 15 and 7.5 cm for the X -and K -bands, respectively. The shapes of the aircraft models are shown in Figs. 16 and 17. A comparison of (a) and (b) indicates that the metallic and resin-made model produced the almost same images. For the VV-polarized images in Fig. 17, one can identify the strong responses at the locations indicated by an arrow in the images. This was because of backscattering from a horizontal aperture mounted on the front of the aircraft  model [14] [see Fig. 13(a)]. A close investigation of these responses in Fig. 17(a) and (b) reveals that the image intensity of the resin-made model shown in Fig. 17(b) is slightly weaker than that of the metallic model shown in Fig. 17(a). As discussed below, this difference affects the RCS values computed from radar images. Fig. 18 shows the RCS reconstructed from the radar images shown in Figs. 16 and 17, plotted with the reference RCS simulated by a numerical electromagnetic solver, where the RCS was computed at 10 GHz. Fig. 18(a) compares the HH-polarized RCS of the metallic (Model-AX) and resin-made model (Model-RX). Both RCS patterns are consistent with the simulated RCS. Similarly, Fig. 18(b) compares the VV-polarized RCS. Unlike in the HH-polarized case, a relatively large difference between the measured and simulated RCS exists around ±150 • . Although the reason for this difference is unclear, the RCS patterns of the metallic (Model-AX) and resin-made (Model-RX) models agree with these angles. Therefore, the difference can be attributed to some characteristics of the numerical electromagnetic solver or image-based NFFFT algorithm used in this experiment. A close investigation of the RCS curve around 0 • revealed that the RCS of the resin-made model (Model-RX) was approximately 3 dB lower than that of the metallic model (Model-AX). As previously discussed, the backscattering around 0 • is mainly attributed to the horizontal aperture mounted on the front of the aircraft model, and the backscattering intensity of the aperture of the resin-made model (Model-RX) is weaker than that of the metallic model (Model-AX). Although some differences exist, the RCS patterns of both models are consistent with each other at most azimuth angles, indicating the validity of the resin-made model.   Fig. 16 with Fig. 19 and Fig. 17 with Fig. 20, the almost identical images are obtained by scaling down the 1/20-scale model to 1/40scale (1/2-scale of 1/20-scale model) and using the K -band frequency which was approximately twice as high as the Xband frequency. Therefore, we can expect that RCS patterns similar to those presented in Fig. 18 are obtained from the K -band radar images shown in Figs. 19 and 20. Fig. 21 displays the reconstructed RCS from the K -band radar images shown in Figs. 19 and 20. As previously mentioned, the K -band RCS was converted to an equivalent X -band RCS using (17) with the scaling factor s = 2. A discussion similar to that for Fig. 18 can be made for Fig. 21. Comparing Figs. 18 and 21 reveals that the almost identical   RCS patterns were reconstructed from K -band radar images, indicating the validity of the scale-model testing.
Finally, Fig. 22 represents the mean RCS errors between the measured and simulated RCS for each polarization. From Fig. 22, the RCS of the metallic models shows better agreement with the simulated RCS for all the cases. However, the difference between the errors corresponding to the metallic and resin-made models is less than 1 dB for all cases. Therefore, we can conclude that the developed resin-made models can be used as substitutes for metallic models, and low-cost RCS testing can be accomplished using a 1/40-scale resin-made aircraft model.

V. CONCLUSION
In this study, we demonstrated the validity of a resin-made scaled model coated with conductive paste as a substitute for a metallic model for RCS measurements. As a preliminary study, we experimentally demonstrated the importance of the molding material of a resin-made model. The RCS pattern of an extremely coarse resin material was considerably different from that of a metallic model, owing to the paint nonuniformity caused by the roughness of the machined surface.
Based on the preliminary study, we manufactured four different aircraft models, where the 1/20-and 1/40-scale models were made with aluminum and polyurethane resin for each scale. The RCS of these models were determined using image-based NFFFT. Although there were slight differences between the RCS patterns of these models, the RCS patterns of the resin and metallic models were almost identical, indicating that low-cost RCS measurements can be performed without using a high-cost metallic model.