Material Characterisation and Radio Channel Modelling at D-Band Frequencies

As the throughput requirements for wireless communication links keep rising, characterization of sub-THz radio channels is necessary. This paper presents the results of a radio channel measurement campaign in which we characterize the full D-band, ranging from 110 to 170 GHz, for distances up to 5 m. We measured penetration and reflection loss for a broad set of materials that are commonly used in indoor environments, including wood, glass, acrylic and concrete, and measured corner diffraction losses. Measurements over the full 60 GHz bandwidth reveal frequency selectivity as well as a periodic variation of both penetration and reflection loss, which is attributed to the thin film effect. Based on measurements in a conference room and outdoors, we create a spatio-temporal channel model for the conference room and an outdoor path loss model. The channel models show that the radio channel is extremely sparse with respect to multipath components, containing only a Line-of-Sight path with signal attenuation close to path loss in free space, and first-order reflections with a measured attenuation that corresponds to the sum of the path and reflection loss.


I. INTRODUCTION
I N the last decade, research on wireless communication at mm-wave frequencies up to 100 GHz made fifth generation (5G) communication possible [1]- [7]. Nevertheless, exploration of a new radio spectrum is needed to enable beyond 5G applications, requiring high-throughput wireless connectivity. Some of these future high-capacity applications, such as wireless backhaul and fixed wireless access, require longrange wireless communication, whereas other applications require high data rates at lower distances. Examples of the latter include close proximity data kiosks, augmented and virtual reality (AR/VR), and holographic displays. Sub-THz communication is considered a key technology for sixth generation (6G) applications [8], [9]. In the D-band, ranging from 110 to 170 GHz, enough bandwidth is available, and this frequency range is apt for short-range high-throughput communication.
Channel characterization at D-band frequencies is ongo-ing, but more research is needed to obtain a unified Dband channel model, as existing channel models consider only a sub-band around the center frequency 140 GHz, or use small antenna separations. Cheng et al. performed Lineof-Sight (LOS) path loss (PL) measurements up to 0.86 m and compare 30 GHz, 140 GHz, and 300 GHz frequency bands [10], concluding that measured PL is close to free space path loss (FSPL) for all three frequency bands, with frequency-dependent fluctuations caused by the environment or reflections on the measurement equipment. Path loss, spatial and temporal characteristics at 140 GHz and 28 GHz are compared by Nguyen et al. for a shopping mall environment [11]. They confirmed the sparsity of the 140 GHz channel and found a high spatial correlation between the channels corresponding to the two frequencies. Pometcu et al. use a vector network analyzer (VNA) based channel sounder with a larger bandwidth of 30 GHz to characterize LOS and non-Line-of-Sight (NLOS) radio propagation in a laboratory VOLUME XX, 2021 setting and NLOS propagation in an office environment [12], [13], reporting a PL exponent below 2 for a LOS channel, and body and wall attenuation up to 27 dB. Kim et al. consider the full D-band channel [14], proposing a D-band PL model for LOS, obstructed LOS and reflected NLOS communication for a distance up to 90 cm. Al-Saman et al. performed channel measurements at frequency 108 GHz in an industrial environment and with antenna separations up to 5 m, reporting PL exponents ranging from 1.6 to 2 [15]. Dupleich et al. created spatio-temporal channel models for a conference room at 190 GHz, using a channel sounder with a bandwidth of 7.5 GHz [16]. Diffraction is well-studied at mm-wave frequencies [17]- [19] but not yet at sub-THz frequencies.
As channel modeling via ray-tracing has proven to be an alternative for stochastic channel models at mm-wave frequencies and beyond [20]- [25], propagation characteristics of different materials should be investigated. Piesiewicz et al. provide the refractive index and absorption coefficients for frequencies up to 350 GHz for plaster, glass, wood, and wallpaper via time-domain spectroscopy [26]- [28]. Correia and Frances estimate material characteristics based on power measurements at frequency 60 GHz [29], and several papers present penetration and reflection loss measurements, but it is clear that for a lot of materials the propagation characteristics above 100 GHz are not yet known [30]. Penetration loss is the attenuation when a signal penetrates through a blocking material and is well-studied at mm-wave frequencies [17], [31]- [33]. Xing et al. provide guidelines for measuring penetration loss [34] and present penetration and reflection loss measurements using a 140 GHz channel sounder with a 4 GHz bandwidth for common materials such as drywall and glass [35]. Reflection, transmission, and scattering measurements are studied for the same set of materials [36]. Penetration through a plasterboard wall and a door is investigated by Pometcu [12]. Penetration and reflection losses for an incident angle of 45 • are provided by Olsson et al. using a VNA based channel sounder with a bandwidth of 7 GHz and a center frequency of 140 GHz [37]. Kim et al. reported that the measured PL for the reflected NLOS case is close to FSPL when the reflector is an aluminum plate and the incident angle equals the reflection angle [14].
We have designed a VNA based D-band channel sounder with a 60 GHz bandwidth for characterizing the full D-band radio channel for distances up to 5 m. The goal of this paper is to fill existing research gaps on D-band propagation channels. In particular, we present reflection and penetration loss measurements for a broad set of materials commonly used in indoor environments, and fit the refractive index to the measurement data. The results of these reflection and penetration measurements can be used in ray-tracing applications. Furthermore, we present indoor and outdoor propagation measurements with the goal of creating a D-band channel model that can be used for network performance evaluation. To the best of the authors' knowledge, this is the first paper presenting reflection and penetration loss measure-FIGURE 1. Channel sounder schematic overview. A vector network analyzer (VNA) creates a radio frequency source (RF SRC) at port 1, which is up-converted via frequency multiplication and down-converted using a local oscillator (LO) signal generated by an external signal generator. The down-converted reference and measurements signals are analyzed by the VNA. ments in the D-band for different materials such as acrylic and tabletop wood, as well as presenting corner diffraction and both indoor and outdoor channel models valid for the full 60 GHz bandwidth.
The paper is structured as follows. Section II presents the methodology, including the channel sounder design and measurement setups. The results are presented in Section III and the conclusions follow in Section IV.

A. CHANNEL SOUNDER DESIGN
A schematic overview of the channel sounder setup is presented in Fig. 1. A two-port VNA generates a radio frequency (RF) source input from 9.167 to 14.167 GHz that is multiplied by a factor 12 via frequency multiplication using an external frequency up-convertor, which results in an RF output in frequency range 110 to 170 GHz. The frequency convertor contains a harmonic mixer for down-conversion, which is used to generate the reference signal measured by the VNA. The harmonic mixer has a multiplication factor 10 and uses a 10.972 to 16.9782 GHz local oscillator (LO) input that comes from an external signal generator. At the receiver side, the obtained signal is down-converted using the harmonic mixer of the frequency down-convertor that uses the same LO input and the down-converted signal is sent to the measurement port of the VNA. The VNA measures the phase and amplitude difference between the reference signal at port 1 and the measured signal at port 2.
Standard gain pyramidal horn antennas with a gain increasing from 22.2 dBi for 110 GHz to 23.3 dBi for 170 GHz are used as transmit (TX) and receive (RX) antennas and are connected to the frequency convertors' WR-6 waveguides. The antennas have a H-plane half power beam width (HPBW) ranging from 13.2 • for 110 GHz to 12 • for 170 GHz and an E-plane HPBW ranging from 12 • for 110 GHz to 8.8 • for 170 GHz. The Fraunhofer far-field distance d F of these The cables connecting the signal generator, VNA, and convertors are characterized with respect to signal attenuation and group delay. The group delay of the cables carrying the LO signal is less than 100 ps and the attenuation at 17 GHz is less than 10 dB. From the measured transfer function H(f ), we get PL via (2), with N the number of frequency sweeps for the considered scenario, G a (f ) the frequency-dependent antenna gain and C(f ) a correction term based on reference measurements that are performed in the center of the lab, without nearby reflectors [38].
The inverse discrete Fourier transformation (IDFT) results in the channel impulse response (CIR) from which the averaged power delay profile (PDP) is found, as can be seen in (3), with W the Hann window.  [38].

B. PENETRATION AND REFLECTION LOSS CHARACTERIZATION
For radio channel characterization via ray-tracing it is of utmost importance to have accurate reflection and penetration loss data for different materials. Therefore, we characterized reflection and penetration loss for the materials listed in Table 2. These materials are selected as they are assumed to be used in the most common objects present in indoor environments.
For the penetration and reflection loss measurements, both antennas are at the same height of 1.3 m above the ground and leveled horizontally. The height of the antenna, in combination with the antenna's narrow HPBW, ensures that no ground or ceiling reflections are received. A laser pointer is used for the antenna alignment. For each measurement scenario, 5 co-polarized vertical (VV) and horizontal (HH) measurements are performed with a 1 mm distance variation, in order to spatially average over the small scale fading (SSF) area. At the midband frequency 140 GHz, 1 mm corresponds to approximately half a wavelength. The 1 mm distance variations cause FSPL variations less than 0.01 dB. We obtain averaged measured PL via (2), with C(f ) the frequencydependent correction factor that is obtained from a reference measurement with antenna seperation 2 or 3 m and without any material under test (MUT) obstructing the LOS path.

1) Penetration loss
We performed measurements with a distance of 2 m as well as a distance of 3 m between the antennas. For both distances, the MUT is placed right in the middle between the two antennas. The material dimensions are large enough to ensure that no diffraction occurs and all received power has penetrated the MUT. Due to the narrow HPBW of the used antennas and the absence of objects nearby the measurement setup, no environmental reflections are received. This is confirmed by an analysis of the PDP, which only shows one peak. Subtracting theoretic FSPL from the measured PL (corrected based on the reference measurement) results in measured penetration loss as a function of frequency. Dividing the attenuation (in dB) by the thickness of the measured material results in penetration loss as a function of frequency and thickness.
A second processing approach is also considered and yields similar results. In this approach, we first split up the 60 GHz frequency band into 10 GHz sub-bands. For every subband, we get the PDP via (3). One sub-band contains 500 frequency points, which results in a time domain resolution of 0.10 ns. The penetration loss is then obtained by subtracting the power in the (only) peak of the PDP from the power in the peak of the reference LOS measurement with the same distance.

2) Reflection loss
For the reflection measurements, the TX and RX antennas point towards a reflection point on the MUT with an angle varying from 15 • to 60 • with respect to the surface normal VOLUME XX, 2021 in the reflection point. The distance from the antennas to the material ranges from 1 m to 1.5 m. We performed copolarized VV and HH measurements for every material, angle and polarization setup. Similarly to the penetration loss measurements, we define reflection loss as the added loss compared to the FSPL corresponding to the distance equal to the reflected path length. Based on the reflection loss measurements, the refractive index of the different materials is estimated via a minimum mean squared error (MMSE) estimation of (4). In this equation, θ is the incident angle, n is the refractive index of the MUT and r is the reflection coefficient for the perpendicular polarization, which is obtained from the co-polarized VV reflection loss measurements.
The antenna cross-polarization discrimination is measured via the methodology outlined by Xing et al. [34] and found to be 21.7 dB. For an incident angle of 30 • , cross-polarized measurements with a vertically polarized TX antenna and horizontally polarized RX antenna are performed.

C. CORNER DIFFRACTION
Next to penetration and reflection, diffraction is another propagation mechanism that needs to be considered for NLOS communication. We performed co-polarized VV diffraction measurements around a 90 • concrete corner of a corridor. Figure 2 shows a picture of the measurement setup. Both antennas are placed at the same height of 1.3 m above ground level at 0.5 m from the corner, with the angle between the antennas and the wall ranging from 45 to 22 • . Multiple measurements are performed and averaged for every TX-RX angle combination. We obtain diffraction loss as a function of frequency by subtracting FSPL corresponding to a distance of 1 m from the measured PL which we derive via (2).

D. SPATIO-TEMPORAL CONFERENCE ROOM CHANNEL MODEL
We performed angular measurements for two separations, 2.6 m and 3.75 m, in a conference room measuring 4 m by 4.6 m of which a picture is shown in Fig. 3. One wall is made of glass, another is made of wood and the remaining two walls are made of layered drywall. Both antennas are vertically polarized, placed at the same height of 0.8 m and rotated around the antenna aperture in steps of 12 • , which corresponds to the antennas' HPBW at 140 GHz. We measure physical received power for every TX and RX angle, which results in a power angular profile (PAP) from which we get angle of arrival (AoA) and angle of departure (AoD) information with an angular resolution of 12 • by integration over all TX and RX angles, respectively. For every TX and RX angle combination, we obtain the power delay profile (PDP) by performing an inverse discrete Fourier transform after applying a Hann window. This results in a spatio-temporal model with received power as a function of delay and AoA by integrating the PDPs over all TX angles. From the PDPs and PAPs, root-mean-square (RMS) delay and angular spread values are obtained.

E. OUTDOOR PATH LOSS MODEL
We performed outdoor LOS PL measurements for distances ranging from 0.8 to 5.0 m in steps of 0.1 m. Both antennas are at the same height of 1.3 m above ground level, and the TX antenna is pulled away from RX antenna along two tracks that are both parallel to a building and shown in Fig. 4. Track 1 is next to the building's window whereas track 2 is next to the wall. The distance from the measurement track to the building is 1.2 m. Both antennas are levelled and are aligned to each other using a laser.
The measurement data is averaged over 1 GHz subbands and fitted to the PL model from (5), with d the distance in meter between the antennas, P L 0 the reference PL in dB at 1 m, n the PL exponent and χ the shadow fading term based on a zero-mean normal distribution with standard deviation σ. We considered both a close-in (CI) model for which the reference PL is calculated via Friis formula and the PL exponent is the only regression parameter, as well as a floating-intercept (FI) PL model for which both the reference PL and PL exponent are fitted to the measurement data.
In addition to LOS measurements, we also define the attenuation of the best NLOS path by measuring PL corresponding to the reflected path for distances ranging from 1.5 to 5.0 m in steps of 0.5 m. For track 1, the signal reflects on glass whereas for track 2, it reflects on the building facade. At every distance we perform 3 measurements with a slight difference in reflection angle. We compare the averaged measured attenuation of the reflected path to geometry-based calculations. The distance of the NLOS path is calculated via (6), with d wall the distance to the wall and d LOS the LOS distance between TX and RX. The incident angle is calculated via (7). Table 3 lists the measured penetration loss values for different frequencies, materials and polarizations, averaged over 10 GHz subbands and over the two measurement distances. The median difference between the results of measurements with 2 and 3 m distance is 0.2 dB and a maximum difference of 2.1 dB occurs for the horizontally polarized measurements of MDF wood above 160 GHz. This confirms the accuracy and methodology of the measurement. Penetration loss as a function of frequency and per cm thickness is shown in Fig. 5 for the different materials from Section II-B except stainless steel and the concrete slab. For these materials, we could not measure the penetration loss as the measurement was within the noise floor of the channel sounder. Given the distance between the TX and RX antennas and the maximum measurable PL of 135 dB, the penetration loss is higher than 50 dB. Both the time-domain and frequency-domain analysis give the same penetration loss results. For acrylic and PVC we clearly see a periodic behaviour which is caused by multipath propagation and confirmed by a thin film analysis. We consider the generalized scenario from Fig. 6 where the incident plane wave is not perpendicular to the surface of the MUT. An incoming plane wave hits the MUT at incident angle θ. The reflected plane wave has a 180 • phase shift as the dielectric constant of the MUT is assumed to be higher than that from air. Using the same  assumption, no phase shift occurs at the reflection within the MUT. There is positive interference when the length of the reflected path inside the MUT is an integer multiple of the wavelength, but as the permittivity inside the MUT differs from free space, the wavelength is shortened. The wavelength within the MUT is λ air /n MUT . Therefore, the condition for constructive interference is given by (8), with m an integer value, n MUT the refractive index of the MUT, λ air the wavelength in the air, t the thickness of the MUT and φ the angle with respect to the normal within the MUT, which relates to the incident angle θ via Snell's law.

A. PENETRATION LOSS
Based on (8) we can calculate the frequency offset in between two frequencies for which constructive interference occurs via (9), given a fixed thickness t and known refractive index n MUT .
In our measurements, the incident angle θ and therefore also φ is zero. Based on (9) and the frequency interspacing from Fig. 5 we calculate the refractive index n MUT . The averaged frequency offset, also called the free spectral range, for PVC is 9 GHz, which corresponds to a n MUT of 1.66. As a validation, we use this refractive index to find the minimum penetration loss values, corresponding to constructive interference, and the maximum penetration loss values, corresponding to destructive interference, via (10). For m ranging from 13 to 17 the maxima and minima of the penetration loss measurements correspond to the calculated frequency.
The free spectral range for acrylic is 32 GHz, corresponding to an refractive index n MUT of 1.557 as the thickness of the acrylic sheet is 3 mm. Using (10) we get constructive interference for frequencies 128 and 160 GHz and destructive interference for 112 and 144 GHz, which corresponds to the minima and maxima in Figure 5. It should be noted that even though the periodic behaviour is clearly visible for acrylic and PVC, it is also present for the other materials. The thicker the material, the smaller the free spectral range. For MDF with a thickness of 18 mm the spectral range becomes so small that the frequency variation is not visible due to the averaging. This is illustrated in the zoomed region of Fig. 5a which shows the penetration loss of MDF wood per cm thickness. The variation in penetration loss for cardboard is caused by the irregular internal structure of the cardboard, i.e., two layers with a third one woven in between.
From Fig. 5, we conclude that there is no significant dependency on polarization, compared to findings in the 30-50 GHz mm-wave frequency band [17] where an additional attenuation up to 2 dB/cm is reported dependent on the polarization. For MDF, tabletop, PVC and glass there is a linear relationship with frequency. Therefore, we fit the averaged penetration losses per cm thickness to the linear model of (11), with L 110 GHz the penetration loss in dB at 110 GHz, f the frequency in GHz, F the frequency-dependent term and t the material thickness in cm. Table 4 contains the values for L 110 GHz and F for the different materials.
With a loss of 5.7 dB/cm at 110 GHz, penetration loss for wood is clearly higher in the D-band compared to mm-wave frequencies, e.g., the penetration loss ranges from 2.4 dB/cm to 4.2 dB/cm at 45 GHz [17] and the penetration loss of a wooden door is 4.6 dB/cm at 73 GHz [32]. Compared to these measurements, the absorption coefficient of 1.7 dB/cm reported in [26] seems to be rather low. Comparing our measurement results to time-domain spectroscopy, the measured penetration loss of acrylic is slightly higher than the reported loss of 2.1 dB/cm [28]. The measured penetration loss of glass is 5.7 dB/cm at 110 GHz which is higher than the loss of 4.1 dB/cm from [28], but 3 dB lower than the D-band measurements performed in [35]. Our measured penetration loss of glass is similar to other measurements that were performed at 73 GHz [32]. From this, we conclude that there  is a large variation in penetration loss of glass, which we believe largely depends on the composition and structure of the glass panel, e.g., whether it is coated glass, safety glass or regular glass. The penetration loss of cardboard is in line with literature [37]. Figure 7 shows the measured reflection loss as a function of frequency for the different polarizations for a fixed incident angle of 30 • with respect to a line normal to the MUT. Similarly to the penetration loss measurements, we see a periodic behaviour for both PVC and acrylic, with the same frequency offset of 9 GHz for PVC and 32 GHz for acrylic. In contrast with the penetration loss, there is no clear dependence of reflection loss with frequency. The median reflection loss values for the different materials, angles and polarizations are presented in Table 5.

B. REFLECTION LOSS
For incident angles up to 45 • there is a trend that the reflection loss decreases with increasing angle, which is in line with the Fresnel reflection coefficients. As expected, the lowest reflection loss is measured for stainless steel (1.2 dB), followed by the wooden tabletop (4.3 dB), which has the highest measurable penetration loss. As acrylic and PVC have lower penetration losses, it follows that they have a higher reflection loss (respectively 8.1 and 8.5 dB). Similar to the penetration loss measurements, there is a periodic variation for acrylic and PVC which is attributed to the constructive and destructive interference of multipath propagation within the medium. The higher reflection losses for horizontal co-polarized measurements compared to vertical co-polarized measurements expected as for the VV measurements the MUT is parallel to the plane of incidence.
The reflection loss for drywall reported in [36] is 7.5 dB at 30 • , which is similar to a tabletop material, lower than wood, acrylic, PVC and cardboard but higher than stainless steel. At 10 • , the reported RL of drywall is 9.8 dB. For wood, PVC and acrylic the measured reflection loss at 15 • is also higher compared to 30 • , but the difference is smaller than for drywall. The reflection loss of 6.5 to 8 dB for glass with incident angle 45 • reported by Olsson et al. corresponds well to our measured reflection loss of 7.8 dB for vertical co-polarization, whereas the reported reflection loss for a wooden door is considerably higher than our measured reflection loss of wood and tabletop materials.
For materials acrylic, MDF wood and the wooden tabletop, the measured reflection loss for cross-polarized antennas is   within the noise floor of the channel sounder. For cardboard, the cross-polarized reflection loss is 37.4 dB, which is 17.1 dB higher than the co-polarized reflection loss. For PVC, the cross-polarized reflection loss is 38.0 dB and for stainless steel it is 19.6 dB. As the difference between the cross-and co-polarized reflection measurements is smaller than the antenna XPD for cardboard and stainless steel, we conclude that depolarization effects occur. Table 5 also lists the results of the MMSE estimation of the refractive index. The fitted refractive index value for acrylic corresponds to the result we obtained from the penetration loss measurement and free spectral range. For PVC, there is a small difference of 0.13 between the two methodologies. Compared to an estimation of the refractive index based on time-domain spectroscopy, the fitted refractive index of wood is slightly higher than the reported value of 1.4, while the refractive index of glass is slightly lower than the reported value of 2.6 [26]. The index of refraction of PVC corresponds to the value of 1.6 that is reported in [28]. Given that a Fresnel curve does not provide an optimal fit to reflection coefficients above 100 GHz [34], the fitted refractive indices correspond well with previous research. Figure 8 visualizes measured diffraction loss as a function of frequency for the different TX-RX angle combinations and Fig. 9 shows the diffraction loss as a function of RX angle for fixed TX angle and a fixed frequency. The diffraction loss calculated via a knife-edge diffraction (KED) model is also shown. As the Fresnel diffraction parameter ν increases with frequency, the diffraction loss is also expected to increase with frequency, which does not correspond to our measurements. For diffraction angles higher than 23 • the signal-tonoise ratio (SNR) was too low. We expect the diffraction loss at both TX and RX angles 45 • , i.e., diffraction angle 0 • , to be 6 dB, whereas the measured loss is 5 dB. This can be caused by antenna misalignment, with the corner not exactly at the antenna's boresight. Compared to our measurements the KED model over-estimates diffraction loss with a difference up to 10 dB for TX and RX angles 37 • . The over-estimation was also reported at lower frequencies (10 and 26 GHz) [18]. The lower diffraction loss was caused by penetration through the corner and multipath propagation. Both effects are negligible in our measurement setup, but due to the smaller wavelength the corner can not be modeled by a perfect straight edge. Also, the KED model assumes the edge to be a sharp obstacle, at which a secondary electromagnetic source is defined according to the Huygens principle. The wavefront of this secondary source propagates in the geometric shadow area. However, the considered corner is not a sharp obstacle and due to the 90 • concrete corner, no wavefront propagates  in this region. It should also be noted that our measurement setup did not have millimeter-level accuracy, which explains the difference between different measurement runs. Figure 10 shows the PAP for the two considered distances in the conference room, using a 10 GHz bandwidth around center frequency 140 GHz. The spatio-temporal model is shown in Fig. 11. As expected, there is a strong LOS component in both figures, with an AoA/AoD of 0 • , PL corresponding to free space PL and a delay corresponding to the distance between the two nodes. Next to the LOS component, two wall reflections are present with a slightly higher delay and higher PL. Compared to lower frequency bands, the channel is extremely sparse with respect to multipath components.    path depends on the free space PL corresponding to distance d NLOS as well as a reflection loss that depends on the incident angle θ, as can be seen in Fig. 4. With increasing distance, the incident angle increases and the reflection loss decreases.

D. SPATIO-TEMPORAL CONFERENCE ROOM CHANNEL MODEL
Due to the small distance to the building, the path length difference also decreases. Both effects contribute to a lower relative NLOS PL for higher distances compared to lower distances, which explains the low value of the fitted PL exponent. Therefore, the difference of attenuation between the LOS path and the reflected NLOS path decreases for larger distances, which can also be seen in Fig. 12. The rootmean-squared error between the measurements and the fitted LOS model is 0.3 dB, which increases to 3.0 dB for the NLOS PL model.

IV. CONCLUSIONS AND FUTURE WORK
In this paper, material characteristics and radio channel models for D-band frequencies are presented. For the first time, the full 60 GHz bandwidth is characterized for distances up to 5 m, allowing measuring frequency selectivity. Measuring over the full band reveals that the thin film effect should be considered for both reflection and penetration when material thickness is in the order of 1 to 10 mm. Based on the periodicity and material thickness of the penetration loss measurements, refractive index values are obtained for acrylic and PVC. A second methodology for obtaining the refractive index at D-band frequencies is via an MMSE estimation based on the reflection coefficients. Both methods have similar results, which are in line with previous research using time-domain spectroscopy. Reflection loss depends on the polarization, i.e., whether the material is parallel or perpendicular to the field of incidence, whereas penetration loss does not depend on polarization. On the other hand, penetration loss increases with frequency, which is not the case for reflection loss. The highest measurable penetration loss was found for tabletop wood (50.5 dB), which shows a lower reflection loss. Materials with a higher reflection loss, such as acrylic and PVC, have a lower penetration loss. Diffraction loss around a concrete corner is measured for diffraction angles up to 20 • , showing that a knife-edge diffraction model overestimates diffraction loss at D-band frequencies. Lineof-sight floating-intercept path loss models for a conference room and outdoors show a path loss close to free space. The conference room channel model includes a LOS component and first-order reflected components, from which the delay and amplitude correspond well to calculations based on the distance of the reflected path and the reflection loss. For the outdoor channel model, the additional attenuation of the reflected path is about 10 dB, so when there is a nearby reflector, the reflected path forms a fallback in case the LOS path is blocked. Future work includes the implementation of the channel models in a ray tracer solution and implementing the unified channel model in a framework to automatically generate channel impulse response realizations. DAVID PLETS received a M.Sc degree in Electrical Engineering from Ghent University (Belgium) in 2006 and joined the WAVES research group that same year. He obtained his PhD degree in 2011 and has been Assistant Professor since 2016. His current research interests include localization techniques and IoT for both industry-and healthrelated applications. He is also involved in the optimization of wireless communication and broadcast networks, with a focus on coverage, exposure, and interference.
CLAUDE DESSET is a senior researcher at imec, Leuven, Belgium since 2001. His main expertise is on high-throughput wireless communication systems and signal-processing, focussing on modelling and optimizing physical-layer performance and power consumption. He is especially targeting MIMO-OFDM systems and algorithms, error-correction coding, and the impact of hardware implementation constraints. He is investigating Massive MIMO and mm-wave communication systems, but also emerging mm-wave radars.
EMMERIC TANGHE received the M. Sc. degree in Electrical Engineering from Ghent University (Belgium) in July 2005 and joined the WAVES research group that same year. He obtained his PhD degree in 2011 with a dissertation on the modelling of indoor and outdoor propagation through field measurements. He became a part-time professor in medical applications of electromagnetic fields in and around the human body in 2015. He continues his work on propagation modelling, covering indoor and outdoor propagation as well as propagation for wireless body area networks and medical implants.
ANDRÉ BOURDOUX received the M.Sc. degree in electrical engineering from the Université Catholique de Louvain, Belgium, in 1982. In 1998, he joined IMEC. He is currently a Principal Member of Technical Staff with the Advanced RF Research Group of IMEC. He is also a System Level and Signal Processing Expert for both the mmwave wireless communications and radar teams. He has more than 15 years of research experience in radar systems and 15 years of research experience in broadband wireless communications. His research interests include advanced architectures, signal processing and machine learning for wireless physical layer, and high-resolution 3D/4D radars.