3-Dimension Burden Surface Imaging System with T-shaped MIMO Radar in the Blast Furnace

Abstract

This paper presents a T-shaped MIMO radar imaging system for use within a Blast Furnace. A T-shaped MIMO antenna array consisting of 32 antenna elements has been developed. A dielectric loaded waveguide is employed as the antenna element. The antenna has good mechanical hardness and is resistant to dust and high temperature. A cooling system, employing a combination of Nitrogen and water, has been designed for this application. The function of cooling system is to maintain the temperature inside the radar between 30°C and 65°C whilst the outside temperature varies from 80°C to 200°C. A pilot study has been conducted using a prototype version of the radar system installed upon the coke surface. In this way it was possible to obtain measurement results concerned with a real burden surface distribution. These results have been carefully compared with results obtained via computer simulation.

1. Introduction

The term Blast Furnace (BF) is used to describe a huge vessel which is used to produce pig iron. The BF functions as a counter-current heat-exchanger as well as a chemical reactor. Within the interior of the BF preprocessed ores, in the form of sinter or pellets, are mixed with coke in alternate layers. The heat required to drive the iron making process is obtained from burning coke. Consequently the temperature at the top of BF rises to a value somewhere between 300°C to 500°C. An unfortunate by-product of this process is that the BF becomes glutted with gas and dust. The iron-bearing burden is smelted and reduced, in volume, by the hot gases. The burden surface consists of a combination of both solid and molten products and as a result of the heating this assumes the characteristics of an inhomogeneous fluid. In order to ensure that the process is sustained continuously raw material is loaded from the top of the BF, as the bed descends. In metallurgy this process is known as “charging”. Iron production is an energy-intensive and polluting industry. However, according to the principles of metallurgy it is possible to achieve higher combustion efficiency by ensuring that the burden surface is symmetrical and has the optimum shape. In order to achieve this it is necessary to generate an accurate real-time image of the burden surface. Unfortunately the BF is an extremely hostile environment. This coupled with the complexity of the burden surface has made it difficult to create an accuracy image in the past.¹⁾

Microwave/millimeter-wave radar²⁾ has been widely used in imaging applications. It affords a number of important advantages including support for long distance transmission, strong penetrating power, and ability to adapt to harsh environments, etc. Presently the following types of microwave imaging radars are being applied to the task of BF burden surface imaging: single radar, distributed array radar, mechanically-scanned radar, and phased array radar. Single radar can only provide information about the height of a fixed burden surface area within the BF. However it offers the benefits of low cost and high reliability. A distributed array radar³⁾ incorporating 6 spatially separated radars can be used to obtain echo information from larger areas of the burden surface. In this way it is possible to successfully reconstruct a virtual map of the 3-D surface. In a mechanically scanned radar system⁴⁾ the antenna is rotated under the control of a motor. In this way it is possible to build up a 3D image of the burden surface. Unfortunately such systems suffer from disadvantages such as high mechanical failure rate, poor reliability and accumulated error caused by the devices used for mechanical scanning. Phased array radar⁵⁾ is a reliable and popular alternative for burden surface imaging. It employs a group of antenna elements which are fed via a network of individual phase shifters. By controlling the phase shift at each element it is possible to scan the beam of radiation. One of the disadvantages of this approach is that a large array of antenna elements is required in order to provide high angular resolution. This adds to the system complexity and cost.

In recent years, an emerging radar system, based on the MIMO concept,^6,7,8) has begun to attract research interest. With the introduction of space-diversity and multiplexing techniques, MIMO radar is able to operate successfully using fewer antenna elements, in comparison with a conventional array. Compared to the four different radars systems, mentioned above, MIMO radar yields enhance the imaging resolution at significantly reduced cost.^{9,10,11,12,13)} For this reason the technique is particularly suitable for industrial applications such as BF burden surface detection. However, to date, no one has yet advocated that MIMO radar be used in the BF industry or, for that matter, in any other harsh environment.

This paper presents a MIMO radar system for burden surface imaging within a BF. A block diagram of the MIMO radar imaging system is shown in Fig. 1. The radar is operated from 9 GHz to 11 GHz. An angular resolution of 5° is required in order to clearly distinguish between individual particles within the BF. Section 2 describes the hardware which is required as a part of the imaging system. This includes a specially designed antenna that can function reliably within the harsh environment of the BF. This antenna element is used to construct a T-shaped MIMO array antenna. Section 3 describes the installation of the radar along with the design of the cooling system. Section 4 describe the software that is used for signal acquisition and imaging processing. The experimental results are presented in Section 5. Finally conclusions are drawn in Section 6.

Fig. 1.

The block diagram of the MIMO radar imaging system.

2. Industrial MIMO Radar Antenna Design

2.1. Antenna Element Design

A suitable antenna element has been developed for the BF environment. This antenna takes the form of a dielectric loaded waveguide. The temperature at the top of the BF ranges from 300°C to 500°C and the dust density is also reaches very high levels. The waveguide is filled with PTFE which blocks the open-end of the antenna. The use of a PTFE (ε_r = 2.08) filling material leads to antenna miniaturization. It also improved the antennas ability to resist exposure to high temperatures. Furthermore the PTFE filling also prevents dust from entering the open-end of the waveguide. This dust could lead to radar clogging.¹⁴⁾

The antenna is fed from the back using a standard 50Ω SMA connector. A 4 step rectangular ridged waveguide is employed in order to achieve impedance matching between the coaxial feed-line and rectangular waveguide.¹⁵⁾ The configuration of the third and fourth ridges was optimized in CST Microwave Studio, in order to achieve operation over a wider frequency band. The ridged waveguide ensures that there is a good match between the TEM wave, transmitted along the coaxial line, and TE₁₀ mode wave, transmitted along the ridged waveguide. The mode matching segment consists of a 3 mm deep cylindrical cavity located at the end of the ridged waveguide.

To enhance the radiation efficiency, the body of the dielectric is tapered towards the back-end. The central section of the PTFE is wedge shaped, thus ensuring that it remains fixed in placed within the waveguide.

The performance of the antenna element was simulated in CST Microwave Studio. Figure 3 shows the radiation patterns obtained at 9 GHz, 10 GHz, and 11 GHz, respectively. The radiation patterns in the E-plane and H-plane are symmetric. The gain at 10 GHz is 8.8 dB, and the side-lobe level is –13.1 dB. The HPBW is 68.7° in the E-plane and 68.4° in the H-plane.

Fig. 3.

The polar patterns of the antenna in resonant frequency point. (a) 9 GHz, (b) 10 GHz, (c) 11 GHz.

Figure 2(b) shows a photograph of the fabricated antenna. The total length of the antenna is 187 mm. The size of the waveguide cavity is 14 mm×10 mm. Figure 4 shows S₁₁ curves obtained through simulation and measurement. It can be observed that the –20 dB bandwidth is approximately 2.25 GHz.

Fig. 2.

The profiles of the dielectric antenna model (a) and manufactured antenna’s photograph (b).

Fig. 4.

The S₁₁ parameter of antenna with dielectric rod and cylinder dielectric in the front-end.

Fig. 5.

The simulated and measured S₁₁ parameter of the antenna with feed structure.

2.2. MIMO Antenna Array Design

The configuration of the antenna array has a direct impact on the performance of the MIMO radar system. It can affect parameters such as the main-lobe gain, side-lobe level, half power beam width (HPBW), imaging resolution, etc. Generally, in order to achieve high cross-range resolution and narrow beam width one requires a MIMO antenna array having a large aperture size and a certain operational bandwidth.¹⁶⁾ However the area of BF is limited, thus constraining the size of the antenna array. Three kinds of MIMO array, with different element spacing, have been designed. Consequently one must make an engineering trade-off when selecting the best antenna aperture size.

Cross-range resolution is the distance that radar can distinguish along the azimuth direction. It relates to the array aperture size L, the distance between the antenna array and the burden surface R, and the wavelength at the operating frequency λ_c.¹⁷⁾   

$${\delta }_{cr}={\lambda }_c\cdot \frac{R}{L}$$(1)

In order to distinguish different particles within the BF, the minimum distance between two targets should be no less than the cross-range resolution. Here the angular resolution of the antenna array is required to be 5°, which necessitates a cross-range resolution of 3.5 meters at a range of 40 meters. The virtual diameter (denoted L) of the aperture array can be calculated, using Eq. (1), to be approximately 400 mm.

The antenna size is 18 mm×18 mm. When 2 antenna elements, having this separation, are placed along X and Y axes, the simulated parameter S₂₁ is less than –15 dB (see as Fig. 6). In order to minimize the total cost of the radar system it is necessary to minimize the total number of transmit elements (denoted M) and receive elements (denoted N). To this end it is important to ensure a minimum number of total elements, according to Eq. (2).   

$$\{\begin{array}{c}{N}_{equivalent}=M\cdot N\\ N_{total}=M+N\end{array}$$(2)

Fig. 6.

The simulated S₂₁ parameter results of a linear array with 2 elements placed in X and Y direction when d = 0.6.

By spatial convolution of the transmit and receive array it is possible to obtain a virtual array that is equivalent to the original MIMO antenna array. This equivalent array is comprised from elements which are uniformly distributed in a rectangular pattern. For this reason it is easier and more convenient to apply post processing techniques to the equivalent array. A topology and design procedure for a MIMO antenna array has been presented.¹⁷⁾ The technique yields a uniform virtual array having no element redundancy. For that reason this approach has been adopted in this paper. The receive array is uniformly arranged, and the transmit array is divided into sub-arrays in order to reduce mutual coupling.

The virtual element spacing is chosen to be 0.6λ. According our calculations, for the aperture diameter in question (L=400 mm), a uniform array comprising 16×16 elements is required. The number of virtual elements, in the array is 256. The main lobe level is 32.8 dB, the side lobe is –13.1 dB, and HPBW is 5.1°. The gain and angular resolution, therefore meet the requirements for burden surface imaging.

Three different MIMO antenna arrays were designed for this application, as shown in Fig. 7. The number of elements in each array is the same. The rectangular MIMO array incorporates 4 transmit sub-arrays. Each sub-array consists of 4 elements arranged as a 2×2 element rectangular array. The receive array, on the other hand, consists of a 4×4 element uniform rectangular array. The cross-shaped MIMO array is comprised from a transmit array distributed along the X axis and a receive array distributed along the Y axis. Both of these arrays are symmetrical. Finally we have also demonstrated a T-shaped MIMO array antenna. The transmit array, within this antenna, is identical to that of the cross-shaped array. The receive array, on the other hand, is distributed perpendicular to the transmit array.

Fig. 7.

Configuration of 3 different MIMO antenna arrays, incorporating 16 transmit elements and 16 receive elements (dx=dy=0.6λ). (a) rectangular, (b) cross-shaped, (c) T-shaped.

Table 1 compares the element spacing and aperture size of the three MIMO antenna arrays. The rectangular array has a smallest aperture size. It leads to the greatest expansion in aperture size. The rectangular array also incorporates the smallest number of elements. It leads to the lowest mutual coupling between elements. However the structure of the array is complex as the transmit elements are located within the four corners. This leads to difficulties in feeding the array which, in turn, would cause positioning errors when the array is installed in the BF. The cross-shaped and T-shaped MIMO arrays are equal in size. But their transmit and receive arrays are distributed separately. This leads to a reduction in mutual coupling the between transmit and receive elements. Table 1 presents the dimensions of the various MIMO antenna arrays which have been designed for the purpose of this study.

Table 1. Comparison of the different MIMO arrays.

No.	dTx	dRx	dTy	dRy	Array size
a	0.6λ	1.2λ	0.6λ	1.2λ	Lx=6λ, Ly=6λ
b	0.6λ	0	0	0.6λ	Lx=9.6λ, Ly=9.6λ
c	0.6λ	0	0	0.6λ	Lx=9.6λ, Ly=9.6λ

These MIMO antenna arrays are illustrated in Fig. 7. In Table 1 d_Tx and d_Rx denote the spacing, along X-axis, between elements within the transmit and receive antennas, respectively. Whereas d_Ty and d_Ry denote the spacing, along the Y-axis, between elements within the transmit and receive antennas, respectively. L_x denotes the total length of the MIMO array along the X-axis. Whereas L_y denotes the total length of the MIMO array along the Y-axis.

The cross-shaped and T-shaped arrays are easy to feed. However the spacing between the elements is too small in the cross-shaped array. The T-shaped array exhibits a lower level of mutual coupling than the cross-shaped array. For this reason the T-shaped MIMO array antenna is the most suitable for burden surface imaging.

The elements within the T-shaped MIMO antenna array are fed in parallel by a feed network that divides the power equally between 16 different feed lines. The feed network is designed to ensure that amplitude and phase of the signal remains correct at each element. Figure 9(a) shows a prototype of the antenna array. The feed and signal processing circuitry is located behind the antenna array.

Fig. 9.

The radar mounted on the top of BF contains two parts: T-shaped MIMO antenna array (a) and cooling & protective cavity (b). 3D model (c) shows how does water and Nitrogen cooling system work.

3. Radar Installation and Cooling Structure Design

3.1. Radar Installation

Due to the existence of the raw material barrel, rotating chute and gas outlet¹⁸⁾ the available area in which to sight a radar installation is limited to the top of BF. For this reason the T-shaped MIMO radar has been mounted on the repair port. This port is located on top of the BF, see Fig. 8. The radar has a 26.6° inclined angle. This choice of angle enables the radar to image half of the burden surface, at any one instance in time.

Fig. 8.

Schematic of the blast furnace process and position for MIMO radar installation.

3.2. Cooling Modules for BF Radar

The high temperature on the top of BF (300°C to 500°C) combined with high level of dust in the environment has a considerable impact on the stability and service life of the MIMO radar imaging system. A cooling system, using water and Nitrogen, has been designed in order to maintain the temperature of the antenna array within acceptable limits. Figure 9(c) illustrates the design of this cooling structure. The T-shaped antenna array is loaded into the front-end of a cylindrical radome, and the related RF circuitry is fixed behind.

Figure 9(b) shows the cooling and protective cavity which is provided by the radome. The antenna array is located close to the edge of the water cool cavity, in order to enhance the efficiency of heat transfer. Nitrogen gas is blown through the gap between antenna and the T-shaped hole. This helps to prevent dust from accumulating around the antenna elements.

4. Signal Processing and Burden Surface Imaging

4.1. Signal Model of the T-shaped MIMO Radar

As the transmitting element is different from receiving element, the time delay associated with propagation from transmit antenna to target and from target to receive antenna must be recalculated.¹⁷⁾

The imaging geometry is illustrated by Fig. 10. The recorded signal can be modeled as follows:   

$$\begin{array}{l}\text{S(}\mu ,\nu ,k)=\\ \iint f(x,y)\cdot e^{-jk\sqrt{{(x-{\mu }_T)}^2+{(y-{\nu }_T)}^2{\text{+z}}^2}}{e}^{-jk\sqrt{{(x-{\mu }_R)}^2+{(y-{\nu }_R)}^2{\text{+z}}^2}}dxdy\end{array}$$(3)

Fig. 10.

Imaging system geometry for MIMO array.

The transmit antennas are located at (μ_T, ν_T, 0), while the receiver antennas are placed at (μ_R, ν_R, 0). Targets cover the area of (x, y, z).

4.2. Signal Processing

It was shown that the wavenumber-domain approach is the most accurate under near-field conditions. For this reason the range migration algorithm^19,20) has been applied, in this study, to achieve image reconstruction. The spreading loss associated with the propagation is ignored. The interaction between the scatterers, within the scene, is also neglected.

A model for the signal (Eq. (3)) was created by using the method of stationary phase, that is to say the decomposition of a spherical wave into plane wave.^21,22) The signal model can be derived as follows:   

$$\begin{array}{l}{\text{S}}_{\text{point}}(\mu ,\mathrm{\hspace{0.17em} }\nu ,\mathrm{\hspace{0.17em} }\text{k})=\\ \iint \left\{\begin{array}{l}\text{f(x,y)}\cdot \left(\iint e^{-j\left[k_{{\mu }_T}(x-{\mu }_T)+k_{{\nu }_T}(y-{\nu }_T)+\sqrt{k^2-k_{{\mu }_T}^2-k_{{\nu }_T}^2}\right]}d{k}_{{\mu }_T}d{k}_{{\nu }_T}\right)\\ \cdot \left(\iint e^{-j\left[k_{{\mu }_R}(x-{\mu }_R)+k_{{\nu }_R}(y-{\nu }_R)+\sqrt{k^2-k_{{\mu }_R}^2-k_{{\nu }_R}^2}\right]}d{k}_{{\mu }_R}d{k}_{{\nu }_R}\right)\end{array}\right\}dxdy\end{array}$$(4)

Where the dispersion relations [18] are formulated as follows:   

$$\{\begin{array}{l}{k}_x=k_{{\mu }_T}+k_{{\mu }_R}\\ k_y=k_{{\nu }_T}+k_{{\nu }_R}\\ k_z=\sqrt{k^2-k_{{\mu }_T}^2-k_{{\nu }_T}^2}+\sqrt{k^2-k_{{\mu }_R}^2-k_{{\nu }_R}^2}\end{array}$$(5)

The 2-D spatial Fourier transform of Eq. (5) can be expressed as:   

$$S(k_{\mu },k_{\nu },k)=\iint f(x,y)e^{-j(k_{x}x+k_{y}y+k_{z}z)}dxdy$$(6)

According to Eq. (6), scaling of the Jacobian coefficients,¹⁷⁾ formulated in 2-D planar MIMO format, can be written as:   

$$dk=\frac{1}{k}{\left(\frac{1}{\sqrt{k^2-k_{{\mu }_T}^2-k_{{\nu }_T}^2}}+\frac{1}{\sqrt{k^2-k_{{\mu }_R}^2-k_{{\nu }_R}^2}}\right)}^{-1}d{k}_z$$(7)

The wavenumber-domain backscatter data needs to be resampled uniformly. Following interpolation, the final image (or reflectivity) of the target can be formulated as the 3-D inverse Fourier transform of the signal model. This is achieved by substituting for the frequency wavenumber variables from (Eq. (7)).   

$$\begin{array}{l}f(x,y)\text{=}\iiint \text{S(}{k}_{\mu },k_{\nu })\cdot e^{j{k}_z{Z}_1}\cdot \\ \frac{1}{k}{\left(\frac{1}{\sqrt{k^2-k_{{\mu }_T}^2-k_{{\nu }_T}^2}}+\frac{1}{\sqrt{k^2-k_{{\mu }_R}^2-k_{{\nu }_R}^2}}\right)}^{-1}\\ e^{j(k_{x}x+k_{y}y+k_{z}z)}d{k}_{x}d{k}_{y}d{k}_z\end{array}$$(8)

5. Results and Discussion

The performance of the T-shaped MIMO radar was verified through measurement. Two scenarios have been chosen: coke surface in the pilot plant and burden surface in a BF during the production process. The pilot plant was used to simulate the environment of the BF in terms of layout and target composition. However the temperature and dust within the BF were not represented in pilot plant. The burden surface is burning in the BF during the production process.

In both cases the surface was treated as a stationary target. The results from pilot plant measurements are presented first. This data reveals the performance of the proposed radar along with the associated signal processing algorithms, when applied to a real-world application. The second set of measurements was performed during the iron production process within a real BF. These results show that the system can be applied to a real BF system.

5.1. Pilot-plant Experiment

The coke surface in the pilot plant is composed from irregular lumps having an average radius and height of 5 meters and 0.5 meters, respectively. The simulated furnace wall is 2 meters high. It is orientated perpendicular to the coke surface. The wall is made from iron plate. The radar system is mounted 5.66 meters above the surface of the coke and 3.83 meters away from the centre of the cylinder used to represent the BF. The incident angle is the same as the actual installation angle i.e. 26.6 degrees. Initially a set of measurements were conducted using triangular corner cubes (CC) as a target for the imaging system. Subsequently, additional measurements were performed using coke as the target for the imagining system.

Let us first consider the results obtained using the corner cubes. The CCs were placed in specific locations in the pilot plant, as shown in Fig. 11(a). The shortest edge of the triangular CC is 500 mm long. When f₀=9 GHz this leads to an RCS of 23.7 dBsm. The CC can enhance the echo signal intensity and help us to locate features on the surface of the coke. Figure 11(b) shows the echo signal intensity distribution on a 2D plane correspond to the CC location shown in Fig. 11(a). The position of the CC can be clearly identified on inspection of this diagram.

Fig. 11.

Photograph (a) and corresponding echo signal intensity distribution (b) of measurement using static CCs as target.

Now let us consider a second set of measurements that were performed within the pilot plant. During these measurements the coke surface itself was used as an imaging target. The surface of the coke (see Fig. 12(a)) has a convex profile with a height of 0.78 meters and a radius of 1 meter. For the purpose of this test the radar system was mounted at the same location as in the corner cubes measurement. The echo signal intensity data (see Fig. 12(b)) reveals a peak in the centre, as expected. The distribution of echo signal intensity in the XOY plane indicates that the radar system exhibits inferior performance along Y-axis, when compared with X-axis. This is especially true along the edges. The dirty data (shown Fig. 12(c)) was obtained by applying signal processing techniques (see section 4.2) to the received signal data. The noisy data of the coke surface does not provide a great deal of useful information. During the data processing, points at which the reflected power is much lower than average power, in the middle were treated as false values. The height of the edge area along Y-axis was estimated from a knowledge of the peak power on the edge of X-axis. The processed image of the coke surface is shown in Fig. 12(d). The peaks (i.e. the white regions in the middle of processed image) correspond to the apex of the coke surface. It can clearly be seen that the primary features on the surface of the coke can be reconstructed by the imaging system.

Fig. 12.

Measurement using coke surface as imaging target. (a) Photograph of imaging target (coke surface), (b) Echo signal intensity distribution in XOY plane, (c) Dirty data, (d) Reconstructed image after data processing.

5.2. Burden Surface Measurement during Iron Making Process

A prototype of the T-shaped MIMO radar imaging system was mounted on a real BF on April 29th 2012. To be specific it was installed on top of 7th BF within the Wuhan Iron and Steel Corporation (in Wuhan, China). The radar system operated, in this location, for a period lasting more than 6 months.

The radar relayed receive signal from the top of BF to a computer. The software on the computer was used to reconstruct a 3-D image of the coke surface according to signal processing techniques and metallurgy principles. Due to the particular location of the radar system it was only able to generate an image of half of the area of the burden surface. The processing software was then used to generate the remainder of the image. This is achieved by applying principles of metallurgy together with expert knowledge concerning the behavior of BF systems. In this way a processed image of the burden surface can be constructed in about 10 seconds. The software showed the profile of the burden surface in real-time. In this way it is possible to examine the effect of the charging controller during the iron making process. The height, vertical cross-section and 3-D spatial distribution provide valuable real-time information which can be used to control the charging of the BF. Figure 13 shows a screen print of the output from the software system. An image of the burden surface distribution is shown in the upper-left corner of the screen. The echo signal diagram is located in the upper-right hand corner of the screen. In this figure strong reflections are shown in red whilst weak reflections are shown in blue. The lower-left screen provides real-time height information concerning a single point on the burden surface. The vertical cross-section of burden surface is shown in the lower-right hand corner of the screen. These lines are used to illustrate the change in surface shape over a period of tens of minutes. The energy diagram in Fig. 13 indicates that the radar system generates a good image of the centre of the burden surface. The image is also good along the radial direction, parallel to antenna aperture. The software also logs the real-time temperature inside and outside the radar. With the help of the cooling system, the temperature was maintained at 30–60°C inside radar. Meanwhile Temperature outside radar was 80–200°C. The cooling system was, therefore capable of maintaining the working temperature of the electronic circuits between (0–60°C).

Fig. 13.

Print screen of radar system software online.

6. Conclusions

By controlling the shape of the burden surface within a Blast Furnace it is possible to improve the efficiency of the combustion process. This paper presents a real-time imaging system for this application. Measurement results obtained from the pilot plant show that a T-shaped near-field MIMO radar system can create an accurate image of the coke surface. Although the resulting image suffers slightly from the effect of crosstalk, the surface can be mapped accurately. Subsequently the radar system was installed in a real Blast Furnace. The radar was able to operate continuously, in this environment, without fault for a period of 6 months. This suggests that the design of the antenna together with the cooling structure were suitable for use within a high dust, high temperature environment. A good real-time 3-D imaging of burden surface was obtained during the iron making. Interestingly there was little difference in the image quality obtained from the iron making BF and the pilot plant. The factors of high temperature, fire flame, dust and furnace wall may lead to poor image quality. The effects of factors above between propagating waves should be considered in greater detail as part of a future study in this area. Future work will also be required to improve the image resolution.

References

• 1)   J.  Brannbacka and  H.  Saxen: Chem. Eng. Sci., 59 (2004), 3423.
• 2)   D.  Malmberg,  P.  Hahlin and  E.  Nilsson: ISIJ Int., 47 (2007), 533.
• 3)   X.  Chen,  J.  Wei,  D.  Xu,  Q.  Hou and  Z.  Bai: ISIJ Int., 52 (2012), 2048.
• 4)   D. G.  Johnson: Radar Conf. 2008 (RADAR ’08), IEEE, Piscataway, NJ, (2008), 1.
• 5)   T.-Y.  Yun,  C.  Wang,  P.  Zepeda,  C. T.  Rodenbeck,  M. R.  Coutant,  M-Y.  Li and  C.  Kai: IEEE T. Antenn. Propag., 50 (2002), 641.
• 6)   H.  Fu,  H.  Fang,  S.  Cao and  M.  Lu: IEEE Symp. on Electrical & Electronics Engineering (EEESYM), IEEE, Piscataway, NJ, (2012), 478.
• 7)   J.  Li and  P.  Stoica: IEEE Signal Proc. Mag., 24 (2007), 106.
• 8)   A. M.  Haimovich,  R. S.  Blum and  L. J.  Cimini: IEEE Signal Proc. Mag., 25 (2008), 116.
• 9)   C.  Ma,  T. S.  Yeo,  C. S.  Tan and  Z.  Liu: IEEE T. Geosci. Remote, 49 (2011), 3009.
• 10)   C.  Ma,  T. S.  Yeo,  C. S.  Tan,  J.-Y.  Li and  Y.  Shang: IEEE T. Geosci. Remote, 50 (2012), 3189.
• 11)   X.  Zhuge and  A. G.  Yarovoy: IEEE T. Image Process., 21 (2012), 3026.
• 12)   P.  Wang,  H.  Li and  B.  Himed: IEEE T. Signal Proc., 61 (2013), 2282.
• 13)   R.  Feger,  C.  Wagner,  S.  Schuster,  S.  Scheiblhofer,  H.  Jager and  A.  Stelzer: IEEE T. Microw. Theory, 57 (2009), 1020.
• 14)   G.  Armbrecht,  C.  Zietz,  E.  Denicke and  I.  Rolfes: IEEE T. Antenn. Propag., 60 (2012), 5083.
• 15)   Y.  Zhou,  E.  Li,  G.-F.  Guo,  T.  Yang and  L.-S.  Liu: J. Infrared Millim. Te., 32 (2011), 26.
• 16)   I.  Nicolaescu: IEEE T. Geosci. Remote, 51 (2013), 85.
• 17)   X.  Zhuge: Doctoral Thesis, Delft University of Technology, (2010), 217
• 18)   E.  Nilsson and  L.  Baath: IEEE Sens. J., 7 (2007), 1025.
• 19)   J. M.  Lopez-Sanchez and  J.  Fortuny-Guasch: IEEE T. Antenn. Propag., 48 (2000), 728.
• 20)   J.  Fortuny-Guasch and  J. M.  Lopez-Sanchez: IEEE T. Antenn. Propag., 49 (2001), 1434.
• 21)   M.  Soumekh: IEEE T. Image Process., 1 (1992), 64.
• 22)   D. M.  Sheen,  D. L.  McMakin and  T. E.  Hall: IEEE T. Microw. Theory, 49 (2001), 1581.