**7.2. Data pre-processing**

The raw radar data provided by the M-sequence radar needs some form of data pre-processing to smooth pulse shape, improve dynamic range, minimize the signal to interference plus noise ratio (SINR) by reducing range sidelobes and finally to enhance the

**Figure 34.** Measured gain [dBi] and pattern of antenna element 2 in the E-plane (left) and H-plane (right) .

temporal resolution. The channel impulse response of the radar link can be extracted by deconvoluting with a reference pulse, as we assume the link as an LTI-system. However, it is well known that classical deconvolution by spectral division may drastically distort the result especially at low SNR values. A highly efficient method with low complexity to perform the deconvolution is to apply a simplified Wiener filter with the transfer function

$$H\_{wiener}(f) = \frac{1}{H\_{ref}(f)} \frac{|H\_{ref}(f)|^2}{|H\_{ref}(f)|^2 + 1} = \frac{H\_{ref}(f)^\*}{|H\_{ref}(f)|^2 + 1} \tag{15}$$

where *Href*(*f*) is the Fourier transform of a previously measured offline reference pulse. Hence, the estimate of the deconvoluted channel impulse response *hdeconv*(*t*) is then obtained as *hdeconv*(*t*) = *hwiener*(*t*) ∗ *hmeasured*(*t*) , with *hmeasured*(*t*) being the measured impulse response under test. Depending on the power level, the Wiener filter either acts as an inverse or matched filter for the deconvolution. In Fig. 35 an example of the channel impulse extraction is shown. Note that both pulses are normalized to the same power to enable visual comparison in the plot.

**Figure 35.** Example of deconvoluted pulses normalized to the same power

### **7.3. Material characterization**

34 Will-be-set-by-IN-TECH

**Figure 32.** Schematic illustration of the antenna elements 1 (left) and 2 (right), in [mm].

**Figure 33.** Measured S-parameter (left) and antenna gain in the main beam direction (right).

a similar characteristic. The maximum gain measured is around 15 dBi at 9 GHz.

successful use in polarization diversity systems even above 10.5 GHz, see [45].

**7.2. Data pre-processing**

The gain and the pattern of the antenna were measured in an anechoic chamber. The results for antenna element 2 are presented in Fig. 34. The E-plane corresponds in this case to the azimuth direction, see Fig. 34 left, the H-plane to the elevation one. Antenna element 1 shows

To evaluate the polarization properties, the gain for both co-polarizations (Co-Pol) and both cross-polarizations (X-Pol) in the main beam direction was measured, see Fig. 33 right. The difference (between Co-Pol and X-Pol regarding the antenna gain) provides the information about the polarization purity. The cross-polarization suppression is better than 20 dB at the low frequencies up to 10.5 GHz. Starting from 10.5 GHz, the values of the X-Pol of antenna 2 are increasing (deteriorating). This is due to the current distribution of higher modes which cannot be avoided for higher frequencies. Nevertheless the measured performance allows a

The raw radar data provided by the M-sequence radar needs some form of data pre-processing to smooth pulse shape, improve dynamic range, minimize the signal to interference plus noise ratio (SINR) by reducing range sidelobes and finally to enhance the

is over the biggest portion of the bandwidth better than -25 dB.

The antennas are manufactured with aid of a circuit board plotter on a Duroid RT5880 substrate of a thickness of 0.79 mm. The measured S-parameter, see Fig. 33 left, show a good impedance matching for both polarizations and antenna elements, respectively, between 3.5 GHz and 13.5 GHz (and even higher). The decoupling (*S*21, *S*12) between the two elements

> A material characterization in hostile and pathless scenarios requires a remote measurement. Hence, a method known from optics, the ellipsometry has been adapted to the UWB

36 Will-be-set-by-IN-TECH 214 Ultra-Wideband Radio Technologies for Communications, Localization and Sensor Applications Cooperative Localization and Object Recognition in Autonomous UWB Sensor Networks <sup>37</sup>

microwave range. The estimation of the dielectric characteristics, especially the permittivity and the emissivity are based on the ratio of the reflected power measured at two orthogonal polarizations. The orientation of the polarization is defined with regard to the plane of incidence. The plane of incidence is orthogonal to the surface of the object and is spanned by the incoming and the reflected ray. In Fig. 36 a drawing of the functional principle is given. The ellipsometry method allows an accurate characterization of plane surfaces with the main restriction to measure at a distance of at least 25 cm to any corner. The estimation of the permittivity of a small object or of objects with a complex shape is significantly influenced by effects of diffraction and scattering. As an example in Fig. 37, the deviation of the estimated relative permittivity of a MDF wall (with relative permittivity of approximately 2.9) is plotted as blue line over the distance from the antennas to the corner of the wall. In order to overcome this restriction, the UWB ellipsometry method is used in a combination with the object recognition process using the imaging methods as described later in this chapter. The distortions of the estimated permittivity values, which arise due to the diffraction effects, are then simulated (red curve in Fig. 37) and the calculated patterns of the permittivity curve are then compared with the corresponding measured pattern. For a first investigation, a simplified simulation algorithm (designed for online measurements) to consider the effects of reflection and diffraction for canonical 2 D objects was implemented. The results show that an accurate estimation of the dielectric characteristics of small objects is possible, with an accuracy of about ±3 % for typical indoor objects (e.g. composed of fiberboards or bricks)

Cooperative Localization and Object Recognition in Autonomous UWB Sensor Networks 215

The effect of the object surface structure on the material estimation was analyzed by measurements of bulk materials. For slight roughness, i.e. height deviations much smaller than the wavelength, there is almost no influence on the estimation of the permittivity. For surfaces with a roughness in the order of the wavelength, the estimation of the permittivity has an uncertainty of less than 20 %. The surface roughness can be estimated by the analysis of the depolarization, i.e. measuring cross-polarized to the transmitted polarization. For the measured indoor materials with rough surfaces, the cross-polarized power is at least about

The fundamental problem common to all super-resolution approaches is the precise extraction of the round trip time of UWB pulses. While this approach can easily be performed for single reflection measurements, things become challenging when the distance between multiple scattering centers drops below the range resolution. Constructive and destructive interferences are caused, and the shape of the resulting superposed pulses is distorted massively. Common algorithms for this purpose were analyzed, evaluated and extensively tested under various circumstances. In most cases, they can hardly resolve richly interfered pulses which overlap almost the whole pulse width or have vast computational load. Often, to some extent a priori information is necessary (e.g. the number of pulses to be separated), otherwise these algorithms suffer from inflexible termination conditions or need

Within the CoLOR project a novel wavefront extraction algorithm called Dynamic Correlation Method (DCM) was proposed, [51]. The DCM is based on a correlation search using a set of two differently shifted reference pulses. Thus, the resulting correlation coefficients are no more just a function of one temporal parameter but rather of two parameters which result in a matrix of correlation coefficients. DCM does not ignore the interfering signature of backscattered pulses and provides a pair of pulses taking the interference pattern of them into account. Additionally, it terminates adaptively to different power levels which enables the detection of weak reflections and avoids post-processing. For a further detailed description

with dimensions greater than 10 cm.

15 dB higher than for flat surfaces.

**7.4. Pulse separation**

huge post-processing.

and a comparison with alternative methods, see [51].

**Figure 36.** Schematic representation of the functional principle.

Distance from border in meter −→

**Figure 37.** Permittivity (blue) and simulation values (red) as a function from the distance from the border.

The calculation of the permittivity is performed by the inverse application of the Fresnel-formulas. Assuming a material with a relative permeability *μ*<sup>r</sup> = 1, the expressions for the calculation of the relative permittivity *�*r and the emissivity *e* can be written as follows, with *E* as the received electric field strength for each polarization:

$$\varepsilon\_{\rm r} = \left(\frac{\sin^2 \theta\_l (\frac{E\_{\perp}}{E\_{\parallel}} - 1)}{\cos \theta\_l (\frac{E\_{\perp}}{E\_{\parallel}} + 1)}\right)^2 + \sin^2 \theta\_l \qquad \text{and} \qquad e = 1 - \left|\frac{1 - \sqrt{\varepsilon\_{\rm r}}}{1 + \sqrt{\varepsilon\_{\rm r}}}\right|^2. \tag{16}$$

Here, it is important to note that the given expression for the emissivity *e* is valid for a straight monitoring of the hot spot. The additional information about the hot spot dimension and distance e.g. to a radiometer can be supplied by the UWB radar. So, an estimation of the hot spot temperature is possible.

The ellipsometry method allows an accurate characterization of plane surfaces with the main restriction to measure at a distance of at least 25 cm to any corner. The estimation of the permittivity of a small object or of objects with a complex shape is significantly influenced by effects of diffraction and scattering. As an example in Fig. 37, the deviation of the estimated relative permittivity of a MDF wall (with relative permittivity of approximately 2.9) is plotted as blue line over the distance from the antennas to the corner of the wall. In order to overcome this restriction, the UWB ellipsometry method is used in a combination with the object recognition process using the imaging methods as described later in this chapter. The distortions of the estimated permittivity values, which arise due to the diffraction effects, are then simulated (red curve in Fig. 37) and the calculated patterns of the permittivity curve are then compared with the corresponding measured pattern. For a first investigation, a simplified simulation algorithm (designed for online measurements) to consider the effects of reflection and diffraction for canonical 2 D objects was implemented. The results show that an accurate estimation of the dielectric characteristics of small objects is possible, with an accuracy of about ±3 % for typical indoor objects (e.g. composed of fiberboards or bricks) with dimensions greater than 10 cm.

The effect of the object surface structure on the material estimation was analyzed by measurements of bulk materials. For slight roughness, i.e. height deviations much smaller than the wavelength, there is almost no influence on the estimation of the permittivity. For surfaces with a roughness in the order of the wavelength, the estimation of the permittivity has an uncertainty of less than 20 %. The surface roughness can be estimated by the analysis of the depolarization, i.e. measuring cross-polarized to the transmitted polarization. For the measured indoor materials with rough surfaces, the cross-polarized power is at least about 15 dB higher than for flat surfaces.
