Large-Scale Channel Modeling and Measurements for 10GHz in Indoor Environments

With the advent of 5G mobile communication and researches into the propagation of large-scale channel modeling for frequencies above 6GHz, measurement investigation was performed at 10GHz with horn-type directional antennas in a corridor and a computer room within the Electrical and Computer Engineering Laboratories’ first floor, at Federal University of Pará (UFPA), Brazil. This paper presents data obtained through experimental work, channel modeling with co-polarization V-V and H-H and cross-polarization V-H in line-of-sight (LOS) or non-line-of-sight (NLOS) conditions. The large-scale close-in reference is sustained by a comprehensive analysis, considering propagation mechanisms such as reflection and diffraction. Results demonstrate that the established model had inferior standard deviation in relation to measured data, proving itself more significant to propagation in indoor environments.


Introduction
The considerable quantity of bandwidth available in frequency bands above 6 GHz is an attractive resource to provide multi-Gigabit per second (Gbps) data rates. Such resource would alleviate the traffic flow of mobile data in lower frequency bands, such as those below 6 GHz [1,2]. With the advent of the new generation of mobile communication (5G), there is a great deal of research taking place into the development to provide recommendations. One of its branches of study lies within the millimetric waves, resulting in works above 6 GHz [3][4][5][6]. Frequencies within 3 to 30 GHz (SHF: Super High Frequency) and 30 to 300 (EHF: Extremely High Frequency) bands present similar propagation characteristics, classifying them as millimetric wave (mmWave) bands [4,5]. Therefore, it is fundamental to know the mmWave channel's propagation characteristics for the development of fifth generation mobile and wireless communication systems-5G [7].
These channel propagation characteristics can be defined through path loss models, in large scale, that predict the propagation signal's attenuation according to distance. They are important for developing more efficient communication systems, to optimize the positioning of transmitting antennas and assist on future telecommunication system projects [8].
For indoor environments, there are several structural questions that influence signal behavior, such as construction materials utilized in the building, size of building/rooms/corridors, number of people circulating inside the room, furniture types and placement, and interference with other systems. They confine the waves within the environment, leading to more reflective components and multiple paths for the signal to propagate, as well as crossing of walls (diffraction) and other types of obstacles [9].
A wireless system requires extensive research and comprehension of its propagation channels. Even though a large number of investigations have been undertaken into channels below 6 GHz, there are campaigns aiming at measuring and modeling of 5G channels at 10 to 28 GHz and 30 to 72 GHz bands [5,[10][11][12][13][14][15], presented by institutes such as New York University (NYU) and the Mobile and Wireless Community Enablers for the Twenty-Twenty Information Society (METIS)-the former created with the sole purpose to standardize 5G systems [16,17]. The METIS report identifies probable bands for 5G application; 10 GHz is listed as a promising frequency band.
As an example of a study based in measuring frequencies above 6 GHz, Deng et al. developed models for diffraction and signal strength measurements around sharp-edged objects such as corners, pillars, and irregular surfaces, for the 10, 20, and 26 GHz spectrum. The diffraction measurements were made in closed and exterior environments, using a continuous wave (CW) probe with three pairs of identical horn-type directional antennas at the transmitter and receptor [18].
In 2014, Kim et al. studied large-scale parameters based on various indoor environments at 11 GHz, comprising path-loss polarization characteristics, shadowing, cross-polarization power ratio, delay spread, and coherence bandwidth [19].
Reference [20] presents radio channel characteristics on carrier frequencies 11.2 and 14.6 GHz, with a 1 GHz bandwidth in environments simulated as a corridor, a laboratory, offices, and a conference room inside a building in Beijing, China. There, investigations regarding large-and small-scale fading properties were conducted, based on realistic measurements. Propagation and delay correlations were also discussed.
Ling et al. produced a measurement campaign of channels implementing directional sweeping to capture temporal and spatial propagation characteristics for upper frequencies, from 13 to 17 GHz, in classroom environments [21].
In [13], there are large-scale models and measuring of frequencies 14 and 22 GHz within line-of-sight (LOS) and non-line-of-sight (NLOS) conditions. A dual-slope model is suggested to distinguish the channel-its validity is proven through the close-in free space reference model, and its parameters defined by MMSE (Minimum Mean Square Error). Another approach is a model inspired in waveguide measurements.
Properties of a channel at the 20 GHz band for an office environment are shown in [22]. Measurements were made using a channel probe at the 20 GHz spectrum, with an orthogonal frequency division multiplexing signal and a bandwidth of 50 MHz. Characteristics of omnidirectional antennas were obtained by turning directional antennas through the azimuthal axis and elevated to the receptor's side.
Reference [23] exposes several large-scale measurements utilizing high-gain directional antennas at 28 and 73 GHz. A model that simulates an omnidirectional antenna in which high-gain antennas were rotated a whole 360 degrees in both Tx and Rx is also described in this paper. This took place in offices with line-of-sight (LOS) and non-line-of-sight (NLOS), with copolarization (V-V) and cross-polarization (V-H). For large-scale models, templates were utilized, such as CI, CIX, CIF, CIFX, FI, ABG, and ABGX, in which the path loss exponent (n) is defined through MMSE (Minimum Mean Square Error). The model's random variable has an average value of zero, and standard deviation of measured data is given in decibels. Results show that the models have great accuracy when compared to measured data.
Lei et al. have proposed a measurement system at 28 GHz in 3 different scenarios, including an office, a corridor, and a generic hall, using 26 dBi directional antennas and a VNA to analyze the channel at a maximum distance of 30 meters. The three properties which were analyzed are path loss, RMS delay spread, and the power angular profiles (PAPs). Results illustrate that indoor environments can enhance the received signal power in the case of LOS. However, in case of NLOS, the penetration loss caused by walls and doors can bring considerable attenuation, which implies that minor cells will have an important role increasing the probability of LOS links for future communication systems. Multipath components (MPCs) can be detected from many directions, albeit needing to use high-directivity antennas [24].
MacCartney et al. compiled a comprehensive study on path loss models of simple or multifrequency mmWave, based on layouts for polarization of separate antennas, and combined in LOS and NLOS environments for three common office layouts. The frequency channels used are 28 GHz and 73 GHz [23].
In [25], Geng et al. made propagation measurements at 60 GHz in several indoor environments, with continuous route (CR) and direction-of-arrival (DOA) measurement campaigns. The propagation mechanisms were studied based on DOA data, indicating that direct waves and the first-order reflected waves on smooth surfaces were sufficient in a LOS environment. Whereas in NLOS cases, diffraction was highly significant, and the propagation loss through walls was very high.
The objective of our study is to present a statistical large-scale analysis of measured data in indoor environments at the 10 GHz spectrum. Signal behavior for LOS and NLOS is specified in two different rooms, with antenna copolarization (V-V and H-H) and cross-polarization (V-H).
The paper is organized as follows. Section 2 presents all measurement equipment utilized during the campaigns, the mounted scenarios, and all measurement procedures; Section 3 elucidates the modeling analysis, the proposed large-scale model for path loss and the obtained results in comparison to other models found in the literature; Section 4 provides discussions and conclusions on the obtained data.

Measurement Investigations
This section describes the transmitting and receiving equipment, the scenarios considered for the measurement campaigns, and line-of-sight (LOS)/non-line-of-sight (NLOS) measurements. It also contains information about copolarization (V-V) and (H-H) and cross-polarization for both studied environments. For the first iteration, Tx and Rx were positioned with a distance of 1 meter from each other. And for the subsequent iterations, the distance between Tx and Rx is increased by another 1 meter. It is important to note that these experiments occurred with no objects or persons obstructing the transmitted beam. Table 1 provides specifications of equipment and parameter configurations for the campaigns.

2.2.
Corridor. The corridor, depicted in Figure 1, has dimensions of 15.22 m × 1.37 m. It is built out of brick walls and smooth concrete on the floor, also containing wooden doors, metallic grilles, and containers with power supplies and circuit breakers. The transmitter Tx was fixed at a distance of 1.3 meters from the windowed wall, whereas Rx was progressively moved 1 meter away from Tx for every iteration of the experiment.
For the corridor, approximately 10 power values were measured for each of the 14 measuring points. For each one, 10,001 power data were collected for each meter variation. The procedure was repeated 10 times for each point, resulting in 100.010 power data for each iteration (10 × 10,001). Therefore, 1,400,140 power data were collected at the corridor environment (100,010 × 14 points).
The quantity of measured points by polarization type is distinct for each one. For vertical polarization (V-V) all 14 points were measured. For horizontal polarization (H-H) only 10 measured points were used, and with cross-polarization (V-H) only 8 points were measured. The amount of measured points for H-H and V-V was recorded for each distance until the signal strength reached noise level.

Computer Room.
The computer room's dimensions are 14.97 m × 7.67 m. As configured in the corridor, Tx was placed in a fixed position 1.3 meters away from the wall, also moving Rx away from Tx by 1 meter every iteration. This room is constituted of brick/concrete walls and drywalls, containing glass windows, wood tables, metallic chairs, and evenly disposed desktops, as shown in Figure 2.
The laboratory was chosen because it presents different characteristics comparing with the corridor; the corridor presents a length much larger than the wavelength of the frequencies used, providing greater reflection in this environment; on the other hand the lab presents a smaller length and walls with different compositions; it is formed for masonry walls and wood, for the walls of masonry the reflections are close to that seen in the corridor, and the wooden ones there have less reflection compared to the masonry wall. Therefore, the laboratory is an interesting environment to determine the loss behavior because it is formed by different types of materials and that directly influence the behavior of the signal received in the Rx.
In this environment, a radial component was defined at 3.8 meters away from the lateral walls. In each radial, 10 power values for each of the 12 traced points were measured. Given that 100,010 power data are obtained in every iteration, as previously noted, the total amount of power data extracted reaches 1,200,120. In contrast to the corridor, all 12 points distributed along the room were measured for all polarization configurations (V-V, H-H, and V-H).
For the measurement campaign utilizing copolarized antennas (V-V and H-H), horn-type directional antennas were also used, with 15 dBi gain on both Tx/Rx and a transmission power of 0 dBm. For cross-polarization, the gain is also 15 dBi on Tx/Rx; however, the transmission power is 15 dBm.

Large-Scale Path Loss Models
Large-scale path loss models predict loss according to the distance between the transmitter and the receptor. These models are important for wireless telecommunication projects [8], aiding the analysis of wave propagation through indoor environments, as there are multiple wave deviation events like multipath, reflection, diffraction, penetration, and shadowing-that is, attenuators of received power at Rx [26]. They also provide realistic propagation characteristics for a wireless channel [5,[10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28]. Most models assume loss in a logarithmic scale. Parameters for the models utilized in this study were defined through MMSE, and the close-in free space model (with its extensions) was utilized for approximation.
The close-in (CI) free space path loss model as seen in (1) and given in dB is uniquely parametrized by a sole variable known as the path loss exponent (PLE), or n. This CI model has a physical distance reference represented by d 0 . The loss model parameters are determined by MMSE, minimizing error between measured and simulated data. In this work, the distance reference is set to 1 meter, and the model can be applied to estimate path loss in both cross-and copolarization conditions.
where X CI σ is a random variable of the Gaussian type with an average value of zero and standard deviation in dB, n V-V is the path loss exponent for copolarization and FSPL is the initial loss found in the model defined by (2). λ is the wavelength and d 0 is the previously explained reference distance, set to 1 m.
The CI model possesses an extension for the case of cross-polarization (3), adding a factor called crosspolarization discrimination (XPD). This parameter is defined via MMSE [28][29][30]. The equation then becomes   International Journal of Antennas and Propagation in which XPD is represented in dB, and X CIX σ is a Gaussian random variable with average value of zero and standard deviation in dB. n used in this calculation has to be defined in the measured data in the cross-polarization configuration. Maccartney et al. [23] utilize n V-V for the model that considers cross-polarized antennas (CIX), in this paper, n V-H is used to adjust to a slope of the curve.

International Journal of Antennas and Propagation
Another independent extension for the CI model is the case of copolarized H-H configuration, adding a path loss parameter for H-H copolarization (4). This parameter is similar to XPD and also defined through MMSE.
In (4), n H-H represents the copolarized H-H path loss exponent (PLE), and X CIHH σ is a Gaussian random variable with average value of zero and standard deviation in dB.
The last proposed extension illustrated for the CI model is the obstacle path loss exponent (OPLE), in which an attenuation parameter signifying obstruction by same or distinct obstacles found in the propagation's path.
in which OPLE i is the obstacle path loss exponent, and i corresponds to the quantity of same-type obstacles. This parameter is based on the COST-231 Motley-Keenan model.

Modeling and Result Analysis
In this section, there is an analysis of signal attenuation behavior in relation to the measured path loss for both studied scenarios. The complexity of indoor environments generates multiple paths-reflections, diffractions, penetration effects, and shadowing-that have significant impact on received power [31]. Radio signal attenuation in waveguides (or for instance, tunnels and corridors) is generally much lower than in free space and is inversely proportional to frequency due to the wave orientation effect. About the measured data, PLE, standard deviation, and XPD values were defined through MMSE. Table 2 presents the said values. Table 3 presents values for OPLE for different types of walls. Figure 3 presents data modeled through the CI model for V-V and H-H polarizations and CIX for V-H, in which the PLE is the gradient of the straight line that shows how the signal degrades with distance, also directly influencing the standard deviation values of measured data in the corridor. Figure 4 illustrates the shadowing effect defined by the random variable for directional antennas, where it is observed that the model has a good approximation compared to measured data and checking the simulated signal's variability between Tx and Rx. Figure 5 shows point-to-point standard deviation data, verifying signal variation for each travelled meter in the environment for all 3 polarizations. Figure 3 presents measured data for the 3 polarization types (V-V, H-H, and V-H), and for the close-in model with extensions for the 10 GHz band, data in blue and red represent the values for copolarized antennas, in which PLE values are very close to 1.96 dB and 1.81 dB for V-V and H-H configurations, respectively (the major difference between them is the initial loss). In Figure 3 the data depicted in green are the cross-polarized measured results, where initial loss is substantially higher and lower PLE is observed compared to copolarization.

Corridor Analysis.
The channel's variability is represented by a Gaussian random variable, influenced by the shadowing effect and defined for the close-in free space reference path loss model (1). Figure 4 shows the shadowing effect defined by the random variable for directional copolarized antennas (V-V) in the corridor at 10 GHz. It is possible to denote that the loss model (1) has a very good approximation in relation to measured data, checking the simulated signal's variation between Tx and Rx.
An important approach to measured data is the statistical survey with standard deviation. For this purpose, a point-to-point standard deviation was calculated to verify signal variation for each travelled meter between Tx and Rx for both environments utilizing copolarized and cross-polarized configurations. Figure 5 illustrates point-to-point standard deviation simultaneously for all polarizations.
For the data in red in Figure 5, it is possible to observe a low variability in standard deviation values (on the y-axis) with values between 0 and 0.7 dB. This indicates that there is low variability of received power values for a same point utilizing horn-type V-V copolarized antennas in the corridor. In black, standard deviation values for H-H copolarization are shown, presenting a variability equivalent to V-V configuration and confirming that signal variation is also low in this case. The same approach has been applied for cross-polarization data in the corridor (blue-colored triangles), however yielding higher signal variability ranging from 0 to 3.5 dB.

International Journal of Antennas and Propagation
It is important to acknowledge that the whole behavior's standard deviation presented higher values for copolarization (V-V and H-H) and lower ones for cross-polarization (V-H). This is directly linked to PLE, in which V-H antennas had PLE values below 1, whilst copolarized V-V and H-H presented values above 1. Consequently, point-to-point standard deviation analysis is important, as it reveals how much the received power varies through measured points.

Computer Room Analysis.
Using the same approach as for the corridor of analysis of the corridor, the computer room measurements have modeled data with CI and CIX     The shadowing effect defined by the random variable for directional antennas and signal variability between Tx and Rx is presented in Figure 7; Figure 8 shows point-to-point measured standard deviation data for signal variation    International Journal of Antennas and Propagation checking in each meter travelled in the environment for all three polarizations, respectively. The analysis of the data measured by the standard deviation point-to-point is made with the objective of verifying the variation of the signal for each point measured in all the polarizations. Figure 8 depicts standard variation simultaneously for V-V, H-H, and V-H configurations. The red set of data presents the point-to-point approach at 10 GHz with V-V polarization, and the minimum/maximum observed values in the Y axis range from 0.04 to 0.26 dB. This indicates that there is low variability on received power values for the same point whilst using horn-type V-V copolarized antennas in the computer room. For the black points in the graph, it is the standard deviation for H-H configuration, presenting variability equivalent to V-V, elucidating that measured data vary very little on a point-to-point basis with minimum/maximum values of approximately 0.04-1.05 dB. For the cross-polarization data in the computer room (in green), the standard deviation varies from 0.5 and 5.5 dB, which is evidently greater than the values for copolarized antennas.

Conclusions
This paper had the objective to demonstrate channel modeling with path loss model close-in and its extensions, to characterize future wireless network channels. The close-in model presents good approximations in LOS and NLOS conditions with directional horn antennas in cross-and copolarization for the corridor and the computer room at 10 GHz, one of METIS-determined frequencies. PLE values in both environments utilizing directional V-V antennas are close to the ones seen in the literature [14,27]; however, the values for cross-polarized (V-H) configurations are less than the copolarized values. Another important remark is the analysis of the standard deviation, both in average value or point-to-point. For the average standard deviation, the values favor copolarization (V-V and H-H) in comparison to cross-polarization (V-H) due to low PLE value, as shown in Table 2. When the analysis is point-to-point, the copolarizations have less variation in relation to cross-polarization.
On the point-to-point standard deviation for the corridor environment, it was verified that copolarized antennas V-V and H-H provide low variation, between 0.04 and 1.05 dB, as depicted in Figure 5. For cross-polarization there is more variability, in values ranging from 0 to 3.5 dB, also shown in Figure 5. For the measurements made in the computer room, it is valid to pinpoint that the H-H polarization's standard deviation yields more variability than its V-V counterpart. In a point-to-point analysis, the V-V configuration varies between 0.04 and 0.26 dB, and H-H gives a standard deviation ranging from 0.04 to 1.05 dB, as evident in Figure 8.
The walls also had different influences in each obstacle-its attenuation changes according to the used frequency, being directly proportional to the suffered loss. This is an important point, and future access technologies must work around this problem, given that the new generations of telecommunication systems are bound to evolve to mmWave bands.
Another significant point is the difference in wave propagation within distinct indoor environments. Inside the corridor, wave reflections are heightened due to the proximity of the walls, the material used inside the walls, floor, and ceiling, in addition to grilles and metallic panels. The computer room is a larger environment, with a different composition. It contains drywall and masonry walls as well as chairs, tables, and desktops. These characteristics affect the path loss of electromagnetic waves inside these environments, as this paper displayed its distinctions through the close-in free space loss models and its extensions beyond the standard deviation.
Future works will be carried out in this area, such as large-scale modeling for adjacent frequencies to 10 GHz, such as 8 GHz, 9 GHz, 11 GHz, and 12 GHz, for the same environments cited in this work using different measurement, with the transmitter (Tx) with different heights, with 360°Rx rotation to collect multipath signals using co-and cross-polarization antennas. Moreover, we will carry out small-scale modeling for the 8 to 12 GHz frequency band. Finally, a large-scale and small-scale modeling will be executed for millimeter wave bands in the 28 GHz to 93 GHz band indoors.

Data Availability
The data used to support the findings of this study are included within the article.