Universität Duisburg-Essen Home
German

How to find us
Staff
Students@NTS
Research
Publications
News / Conferences
External Links

Intranet

Studies / Lectures
Final theses /
Project works: Topics


Services
Heinz-Luck-Fire Lab

AUBE '17 / SUPDET 2017
ICOF 2016

EUSAS e.V.

Electrical and Electronics
    Engineering

Alumni Engineering Sciences
Engineering Sciences
Universität Duisburg-Essen

Mobile radio:
Smart Antennas
Adaptive MIMO
     communication systems

Filter bank multicarrier
    transmission systems

Adaptive multicarrier
    transmission


Radar-Technique:
Wideband GHz and THz
     radar sensors


Fire detection:
Automatic Fire Detection
Test devices for
     dust and water fog tests

Polarised
     Aerosol Charactersiation



NTS-Poster

Wideband GHz and THz radar sensors

Stand: 11.08.2016
Universität Duisburg-Essen, Campus Duisburg
  • UWB-Radar

  • Motivation
    In recent years detection and imaging using UWB radar systems have been attractive for researchers. Compared to other sensing technologies such as optical or infrared systems UWB radar does not need a visual line-of-sight making it suitable for smoke and dust filled emergency scenarios. The fine time resolution due to the huge bandwidth provides accuracy and resolution in the high millimeter range. In addition, a UWB radar has notable robustness against narrowband interferers, hence minimizing the effect of multipath propagation as the propagating waves in many cases do not overlap. Thus, a certain degree of multipath immunity is provided. Therefore, robots equipped with radar systems and UWB sensors are attractive for detection and imaging applications in emergency situations.

    Therefore Radar technologies are a major and very successful topic for research at the department NTS. Over 45 scientific papers in most cases for IEEE conferences have been published from NTS on this topic since 2007, two of them and a PhD dissertation have been awarded (amongst others by the IEEE in the form of best paper awards).


  • THz imaging and THz material characterization

    Topics for upcoming research at NTS are focused on applications in the THz range in order to extend the expertise of the department in high-resolution imaging and ellipsometric permittivity estimation of the target surface to frequencies above 100 GHz. Compared to UWB radar, the extended bandwidth of THz signals will contribute to a significantly increased spatial resolution. Significantly smaller antennas in the THz range also open up new fields of application. Thus, the application of the high resolution UWB algorithms in time and frequency domain has to be adapted with respect to the higher frequency and larger bandwidth of the THz systems.

    Key effects like specular surfaces in UWB applications present at THz radar to a much lesser extent, while the roughness of the target surface at THz frequencies have to be accounted for. The methods for material characterization investigated at NTS using the depolarization of the material surface provide new evaluation options. In addition, it is possible to optimally explore surfaces with both specular as well as scattering elements by means of a method developed at NTS.




    Wavefront detection

    Although the large bandwidth provides to a certain degree of multipath immunity the separation of multiple wavefronts under severe interference conditions with no a priori information still keeps a challenging issue in the field of super-resolution applications. For this purpose two correlation based wavefront detection algorithms have been developed at the department NTS namely the "Dynamic Correlation Method"(DCM) and its polarimetric version the PDCM.

    DCM
    The DCM algorithm is based on the classical correlation algorithm where a cross-correlation between a reference pulse and the investigated pulse is performed. However, in the DCM a set of synthetic wavefront patterns consisting of a superposition of two differently delayed reference pulses is used as a new synthetic reference for the correlation. From the correlation pulse two wavefronts are extracted, instead of only a single wavefront. The extracted wavefronts are then windowed and subtracted from the investigated pulse. The method is iterated until a termination condition is fulfilled. As a termination condition in the DCM a dynamic power detector is employed. Further details about the DCM can be found in in [1].
    The following figure shows the Radargramm and the extracted wavefronts of a test object that has been scanned over 360°.

          

    PDCM
    The second wavefront detection algorithm designed at the department of NTS is the polarimetric DCM or PDCM. In situations where weak reflections interfere with strong ones (e.g. strong double bounce reflections masking weak reflections from edges) the DCM algorithm fails at detecting and separating the masked reflections. Thus, to separate the strong from the weak echoes the Pauli scattering matrix decomposition is applied onto the radar data [2].
    After that two new measurements of the object under test are obtained, where m MUT,S(t) exhibits only the contributions from single-bounce scatterers and m MUT,D(t) exhibits reflections only from double-bounce scatterers. In the following Figure two sample measurements from a test object m MUT,HH(t) and m MUT,VV(t), the resulting single bounce m MUT,S(t), and double bounce m MUT,D(t) measurements and the positions of the reflections for the four measurements are shown.



    From the figure, it can be seen that in this example after combining the impulse responses m MUT,VV(t) and m MUT,HH(t) to m MUT,S(t) and mMUT,D(t), three echoes can be extracted for the single-bounce scatterers and two echoes from the impulse response for the double-bounce scatterers resulting in a total of five echoes. This is not the case if the two impulse responses m MUT,VV (t) and m MUT,HH(t) are examined separately. The following figure shows the test object, the 360° Radargramm and the extracted wavefronts using the PDCM.

          

    Kirchhoff-Migration

    Migration based imaging algorithms are well documented in the literature [3]. They are used in many scenarios, due to the simplicity of their implementation, lack of a need of a priori information and rather accurate images. Applying migration on the raw radar data an image of a test object can be obtained without any pre-processing of the data. Thus, the image of the object represents a matrix, which is generated by the spatial superposition of multiple impulse responses from different measurement directions (see the following figure). In the figure three different phases of the Kirchhoff Migration method are shown. The left image shows the results when only a single impulse response is migrated. In the next two images multiple impulses responses from multiple angle of views are considered (45 impulses responses for the middle image and 135 for the one on the right)..

                 

    Improvement of the Kirchhoff-Migration
    At the department NTS an improved Kirchhoff-Migration algorithm was designed that consists of the following methods.

    Fully polarimetric Radar Data and Wiener filter
    An optimization of the Kirchhoff-Migration is achieved by introducing the polarimetric scattering matrix (see Wavefront detection). Furthermore an improvement of the radar data used for the migration algorithm can be achieved by deconvolution of the raw data. The deconvolution with a reference pulse results in the channel impulse response. However, simple deconvolution using the transfer function of a reference pulse may distort the signal most notably at lower SNR. Hence a modified Wiener filter is applied at NTS. The transfer function of this Wiener filter is given by



    where Href(f) is the transfer function of a reference pulse.

    However, deconvolution is not only limited to deconvolution of the measured impulse response. In image processing deconvolution is often employed for deblurring of an image. Thus, the modified Wiener filter can be directly applied on the migrated image, where the reference transfer function is the Fourier transform of the point spread function of the image. By applying this method a deblurring of the object contour is achieved.

    k-means Segmentierung
    An enhancement of the migrated image can be achieved by applying a segmentation method. The kmeans segmentation is used for the reduction of the image noise level. The migration algorithm generates the image of the target as well as the area in which the target is located by a superposition of echoes from the target. Using this method for radar imaging all pixels from the digitalized image at a distance equal to the acquired echoes are weighted with the amplitude of the echoes. These pixels that are not part of the contour of the target exhibit higher amplitude resulting in a higher image noise level. However, by segmenting the image by intensity levels a suppression of the noise can be achieved. In the following figure a photo of a test object and the resulting Kirchhoff-Migration and improved Kirchhoff-Migration images are shown.

    The image in the middle at the first line is the resulting image of the single bounce scatterers from the Kirchhoff-Migration algorithm by only applying the scattering matrix. The image on the right of the first line represents the same scatterers, but after noise reduction with the k-means segmentation method while the left image at the second line is the final deblurred image using the improved Wiener filter. Finally, the results of the improved Kirchhoff-Migration algorithm of the double bounce scatterers are shown in the image on the right at the second line.

                 

                              

    Super-Resolution Radar Imaging

    A second method for the generation of a radar image is based on the distance measurements (wavefronts). At the department NTS two "Super-Resolution" radar imaging methods have been designed and investigated - the PCA and the PDMA methods.

    Polarimetric Convergence Algorithm (PCA)
    The PCA is a Radar imaging algorithm based on fuzzy logic for the calculation of the angle of arrival of every extracted wavefront (see Wavefront Detection) and thus assigns (combined with time of arrival data) each wavefront a scattering point. The PCA algorithm consists of three parts:
    1. The wavefronts are extracted using the PDCM method (see Wavefront Detection).
    2. The wavefronts are then optimized using a post-processing algorithm where the wavefronts are combined into clusters. The clusters are then approximated using a polynomial curve fitting algorithm and interpolated.
    3. The angle of arrival for each wavefront is then calculated, thus a scattering point can be determined.
    This calculation is based on the following procedure. Due to the bistatic antenna configuration, the scatterer for the wavefront under test is positioned along an ellipse. The two focal points of the ellipse are the positions of the transmitter Tx and received Rx antennas. The major and the minor axis of the ellipse are function of the wavefront under test data. The exact position of the scattering point can then be determined by the angle of arrival. For this purpose the information from the neighboring measurements are considered as shown in the following figure.



    In the figure above the resulting ellipse from the n-th wavefront is shown in red and the resulting ellipse from the m-th wavefront in blue. The intersection points of the two ellipses can then be determined using the formulas [5]. Thus, an angle of arrival for this scenario can be calculated and a Gaussian function for this angle of arrival is generated as shown in [6]. By adding the Gaussian functions for different neighboring measurements an optimal angle of arrival is determined. The image of a test object using the PCA is shown in the following figure..

          

    Polarimetric Direct Mapping Algorithm (PDMA)
    A second method for a generation of a radar image of an object using detected wavefronts based on measurements with two receiver antennas has been presented by NTS. Similarly to the PCA the PDMA consists of three steps. The first and the second part of the algorithm are identical to the PCA, but here each step is performed twice, once for each receiver antenna. The resulting image is also based on the determination of the coordinates of the test objects scatterers. In this method the scatterers are determined by the intersection points of paired ellipses from the two receiver antennas [5]. In the following figure one typical scenario is shown.



    Using the PCA the radar image of a test object is shown in the following figure. The diagram in the center shows the PCA result, the significantly better PDMA result is shown right.

                 

    Object recognition

    At the department NTS an object recognition algorithm was developed based on seven translation and rotation invariant image features:
    1. Fitting Circle – The radius of circle that fits the entire imaged object.
    2. Form Factor – The Form Factor represent the ratio between the radius of the fitting circle and the number of detected scattering centers of the test object (see „ Super-Resolution Radar Imaging).
    3. Moment of Inertia – The moment of inertia of the scattering centers results from the radar image by



      Here rs,k is the distance between the center mass of the imaged object and the determined scattering centers.
    4. Curvature-Scale-Space (CSS) – CSS is a feature of a graph defined by its curvature. In the next figure the normalized CSS feature resulting from the Gaussian filtered contour of a test object and the contour itself are shown. The peak values of the CSS represent the position of the edges and corners of the contour of the object. The edges result in a positive peak value, where the corners are represented by a negative peak.



    5. Fourier Descriptors (FDs) – the FDs of an object contour represent the pixel coordinates of the given contour after their Fourier transformation.
    6. Image Moments – Seven image moments are defined by Hu in [7]. These moments are both defined as translation and rotation invariant.
    7. Eccentricity - The eccentricity feature of an object is a scalar value invariant to translation and rotation, which specifies the deviation of a certain section from the circular shape. 0 describes a circular shaped object, values between 0 and 1 describe an ellipsoidal shaped object and if the eccentricity of an object equals 1 then the object is a straight line.
    The recognition of a test object is based on the minimum mean square error between the determined seven object features of the object under test and a set of reference features. In the following figure 12 test objects and the probabilities of correct detection of the 12 test objects using the method designed at NTS are shown.





    Ellipsometry

    Microwave-UWB-Ellipsometry
    In the field of security applications, e.g. for fire detection, not only the information on the object shape (e.g. in case of a burning object or a hot-spot) is of great importance, but also the information on its material characteristics. For example it can be deciding if a hot-spot is located on a carpet floor or on a stone floor.
    Against this background an optical material characterisation technology was advanced towards the Microwave-UWB-Ellipsometry. The advantages of analyses performed via the Microwave-UWB-Ellipsometry are e.g.:
    • The possibility of analysing semi-hidden objects, as the method is based on contactless reflection and scattering measurements.
    • Due to the larger wavelength of the incident radiation compared to the optical ellipsometry, there is almost no influence of aerosols like smoke or dust present in the ambient air (e.g. due to a fire) on measurements.
    • Due to the larger wavelength of the incident radiation and its higher penetration depth also sub-surface analyses become feasible.
    The estimation of the material characteristics by the Microwave-UWB-Ellipsometry is based on the inverse application of the Fresnel equations. The information necessary for the characterisation is obtained by polarimetric reflection measurements performed at one or several angles.



    The left figure shows a schematic sketch of the setup. Tx and Rx are the transmitting and receiving antennas respectively and θi is the angle of reflection and incidence. In the right figure a realised measurement setup mounted on a 6-axis positioning table is shown. The DUT here is a MDF-board.
    The next figure shows the measured parallel and orthogonal reflectance of the DUT (MDF-board) at different reflection angles θi. The theoretical ideal values obtained by simulations are shown as solid lines.



    Based on the measured values of the reflectance, the dielectric constant of the DUT can be estimated by the application of the inverse Fresnel equations. The shown DUT exhibits an estimated dielectric constant of εr ≅ 2.7, which agrees well with the values known from literature.

    Microwave-UWB-Ellipsometry for small objects
    The object size has a strong impact on the permittivity estimation, as diffraction effects of the edges of small objects superimpose on the reflection measurements. This effect is illustrated by the following figure, showing the estimated dielectric constant εr based on reflection measurements performed at different distances of the object edges (blue curve). The red dotted curve shows the nominal value εr ≅ 2.7) 7 of the DUT. The strong fluctuations of the estimated dielectric constant in the first 30 cm distance to the border of the DUT arise from the diffraction effects of the object edge.



    To counteract this problem a combined imaging and material characterisation method was developed, which incorporates the information of the object geometry in the estimation of the dielectric constant. The permittivity of typical indoor objects with a size of approximately 20 cm x 20 cm could be estimated with an uncertainty better than ± 6% in the range of 4 - 13 GHz.

    Microwave-UWB-Ellipsometry of rough surfaces
    Different from plane surfaces, rough surfaces also scatter radiation in directions different from the ideal angle of specular reflection θi. As the classical ellipsometry is based on specular reflection measurements of perfectly plane surfaces, the analyses of rough surfaces of e.g. bulky materials are strongly adulterated.
    In order to allow the characterization of naturally rough surfaces or even bulky materials a method was developed which combines specular and diffuse reflection by a combination of the Fresnel and the Lambert equations.



    First investigations performed on bulky materials show promising results, as seen in the following figure for the analysis of sand-lime bricks.



    The analysis of sand-lime bricks were performed in the range of 43.5° to 46.5°. The red curve shows the estimated dielectric constant by applying the original UWB-Ellipsometry. Due to the impact of the diffuse scattering, the estimated value of the dielectric constant εr is about 2. However, the nominal value measured at a plane surface is about 4 (pink curve). The developed material characterisation based on the combination of specular reflection and diffuse scattering (blue curve) minimises the influence of the surface roughness, allowing a correct material characterisation..

    Wideband-Ellipsometry at higher frequencies
    Due to the continuous developments of radar systems for security applications (e.g. access control) or automotive applications (e.g. anti-collision systems) single chip transceivers become more and more energy efficient and cheaper. In addition, due to the development of radars for even higher frequencies (e.g. 24 GHz or 60 GHz), their size and weight also shrink continuously. As the Microwave-UWB-Ellipsometry is not restricted to a certain frequency range, these developments open up new areas of application.
    Due to the smaller wavelength of the applied radiation, typical surface textures become rougher. A typical plaster wall can be approximated by a plane surface in the range of 5 GHz. At 60 GHz the diffuse scattering may dominate and strongly impact on the material estimation. These effects stress the importance of the investigations on the ellipsometry for rough surfaces, which combines the specular reflection (Fresnel equations) with the diffuse scattering (Lambert equation).


    Ellipsometry with single chip radar systems
    The reflection measurements for the ellipsometry are usually performed at a reflection angle around 30°-45°, which is a trade-off between the needed high difference in the reflectivity of the parallel and the orthogonal polarisation and the antenna cross-talk. The usual antenna set-up is therefore of bi-static type not allowing the implementation of single chip radar systems. The disadvantage of a bi-static configuration can be overcome by a novel set-up approach implementing a retro-reflector.
    A corner cube retro-reflector reflects an incident radiation to the source, in a limited range, independently of the angle of incidence. The principle of the described corner reflector is shown in the next figure.



    The novel NTS designed ellipsometric set-up with a retro-reflector is shown in the next figure.



    A first measurement campaign with a low cost FMCW-radar chip working in the range from 57 GHz to 64 GHz showed very promising results, opening up the door for low-cost mobile ellipsometry platforms. Although the principle is strikingly smart and simple and also the antenna arrangement is easy to realise much more research is needed as a precise super-resolution ellipsometry measurement has to take into account the different antenna and retroreflector footprints on the objects surface.

    1. R. Salman, I. Willms
      Joint Efficiency and Performance Enhancement of Wavefront Extraction Algorithms for Short-Range Super-Resolution UWB Radar
      The 7th German Microwave Conference (GeMiC), Ilmenau, Germany, 12-14 Mar. 2012.
    2. D. Damyanov, R. Salman, I. Willms, T. Schultze
      A Super-Resolution Polarimetric Wavefront Extraction Algorithm for UWB-Radar under massive Interference Conditions
      The 11th European Radar Conference, EuRAD, Rome, Italy, 5-10 Oct. 2014.
    3. R. Salman, I. Willms, L. Reichardt, W. Wiesbeck
      On Polarization Diversity Gain in Short Range UWB-Radar Object Imaging
      IEEE International Conference on Ultra-Wideband (ICUWB), Syracuse, USA, 17-20 Sept. 2012.
    4. S. Kidera, T. Sakamoto, T. Sato
      Accurate UWB Radar Three-Dimensional Imaging Algorithm for a Complex Boundary Without Range Point Connections
      IEEE Transactions on Geoscience and Remote Sensing, 2010
    5. D. Eberly
      Intersection of Ellipses
      Technical report, Geometric Tools, LLC, 2000.
    6. D. Damyanov, R. Salman, I. Willms, T. Schultze
      Super-Resolution Feature Extraction Imaging Algorithm for complex Objects
      IEEE International Conference on Ultra-Wideband (ICUWB, Paris, France), 1-3 Sept. 2014.
    7. Ming-Kuei Hu
      Visual Pattern Recognition by Moment Invariants
      Transactions on Information Theory (IRE), 8(2):179187, 1962.