*2.1. The RL Technique*

The in-house developed 3D-RL simulation tool has been developed in Matlab programming environment. The detailed operating mode of the algorithm has been previously published [32] and validated in complex urban environments [33]. The principle of the RL approach is to approximate the full wave methods based on Maxwell's equations into a set of equations based on GO and UTD. The in-house implemented 3D-RL code has been optimized in order to decrease computational cost when consider large, complex scenarios, such as indoor locations, by means of hybrid simulation approach. In this sense, neural network interpolators, electromagnetic diffusion equation for diffraction estimation, and deep learning database assistance by means of collaborative filtering have been implemented within the code. The algorithm basis has three steps:


The first step consists in the creation of the 3D environment under evaluation. For that purpose, all the details of the environment are considered, taking into account its real dimensions, morphology, topology, and material properties (by means of the conductivity and dielectric permittivity) for all the obstacles within the scenario at the frequency band under analysis. In the simulation procedure, a set of rays are launched from the transmitter and electromagnetic propagation phenomena such as reflection, refraction, and diffraction are considering along all the path rays. Parameters such as transmitters location, angular resolution of rays, cuboids size of the scenario, frequency of operation, and number of reflections are considered as input parameters in the algorithm. A trade-off between angular resolution of launching rays, cuboids size of the scenario, required computational time, and results accuracy must be achieved during the simulations. A convergence analysis in terms of number of reflections and launching rays of the algorithm has been performed and it is presented in [32], as well as the optimal spatial resolution for large scenarios, which is presented in [34]. These parameters are used in the simulations and are presented in Section 2.2.

Finally, the third step consists in the results analysis, where different outcomes can be obtained. The 3D-RL tool is based on a modular structure, where the user can select the results of interest. The different results that can be selected are large-scale propagation parameters such as received power or path loss analysis, or small-scale parameters such as power delay profile (PDP), delay spread, coherence bandwidth, or doppler spread, among others. In this work, interference analysis has been implemented as a new module for the network performance assessment. In this library, once the power level and signal to interference noise ratio (SINR) results have been obtained for all the spatial points of the scenario, different modulations can be assessed for different communication links, presenting evaluation of EVM within the complete volume of the scenario under analysis.

### *2.2. Scenario Description*

The selected scenario is an auditorium placed in an open free city area surrounded by inhomogeneous vegetation. Figure 1 presents the real and schematic view of the considered scenario, which is part of the Campus of Tecnologico de Monterrey, Monterrey, Mexico. The considered scenario is a complex scenario in terms of radio wave propagation characterization as it is a combination of an outdoor and indoor environment rich in multipath trajectories due to the large quantity of obstacles and people within it. The different workspaces of the auditorium environment have been recreated in the simulation algorithm, taking into account the inhomogeneous vegetation, trees, tables, chairs, the auditorium area, the cafeteria area, corridors, and a random distribution of people, both in the outdoor and indoor areas of the auditorium. A generic human body model design created specifically to be embedded in the 3D-RL algorithm has been used to enhance a more realistic scenario, as users have a

significant influence in radio wave propagation in this type of complex environments [35]. The specific details of the developed human body model and its integration with the 3D-RL tool can be found in [36].

User density within the scenario, in terms of inherent transceivers density, has been performed by simulation. The considered scenario has a user capacity of 150 persons within the auditorium, and 40 more approximately, when all the tables around the auditorium are occupied. Thus, for a high-node density within the complete scenario, it has been considered that one per four persons has a wearable that can increase overall interference levels, which lead to a sensor network of 75 wearables. A medium node density has been considered with 38 wearables (one device per eight persons), and a low node density of 19 wearables (one wearable over 16 persons). All the wearables have been considered at 1.2 and 0.8 m height, emulating smart glasses or wrist-worn devices in the case of seated persons. Figure 2 represents an aerial view of the scenario (ceiling has been removed for illustration purposes) with the wearable's location for the three different nodes density cases. For the simulations, two different frequencies have been considered, 2.4 GHz and 5.8 GHz, considering the later for infrastructure operation (i.e., not as wearable devices). For the network performance analysis, a ZigBee system has been considered at 2.4 GHz frequency band, as well as Bluetooth V4.0 transceivers within the scenario. This election is based on two modes of operation: Data gathering (from WSNs or users) and wireless transport networks (given by 5.8 GHz WLAN devices). ZigBee offset-quadrature-phase-shift-keying (O-QPSK) modulation, with a bandwidth of 3 MHz and a bit rate of 250 kbps, whilst Bluetooth V4.0 employs at higher rates, differential 8-level phase-shift keying (8-DPSK) modulation at a bit rate of 3 Mbps. Simulation parameters are summarized in Table 1.



(**a**)

**Figure 1.** *Cont.*

**Figure 1.** Urban considered scenario where, (**a**) is the real view of the scenario (retrieved from: http://www.rdlparquitectos.com/es/proyectos/pabellon-la-carreta/), (**b**) is a 3D aerial rendered view, and (**c**) detailed view of the indoor part of the scenario.

**Figure 2.** Aerial view of the scenario with the wearable's location for the three different node density cases. Auditorium ceiling has been removed for illustration purposes, (**a**) high-node density, (**b**) medium-node density, and (**c**) low-node density.

### **3. Simulation Results**

### *3.1. Received Signal Strength*

Firstly, the Received Signal Strength for the full volume of the scenario has been obtained by means of the 3D-RL simulation tool. Although the whole volume of the scenario has been analyzed in simulation, only the bi-dimensional planes of received power for a selection of relevant heights have been presented in Figures 3 and 4. The considered heights have been 1.2 m, which is the same height as the transmitters, emulating smart glasses receivers for a seated person (Figure 3), and 0.8 m height, emulating wrist-worn receivers for a seated person (Figure 4).

Figure 3 presents the RSS maps at 1.2 m height considering different node density setups for two different frequencies, 2.4 GHz and 5.8 GHz, which are the typical frequencies used in generic wearables, as well as in wireless sensor network infrastructure (i.e., 802.15.4 networks and 802.11 networks). It can be seen that the morphology and topology of the scenario, as well as node density distribution, have a great impact in radio wave propagation. It is observed a high RSS variability within the whole scenario due mainly to the presence of different scatterers such as people, furniture, walls, vegetation, and trees, among others, which cause a rich multipath propagation in the environment under analysis. It can also be remarked that RSS is around 20–30 dBs higher when high-node density is considered. Moreover, it is worth noting that there are localized regions with higher power levels, i.e., hot spots, within the scenario, caused by individual transceivers, that tends into power clusters when the number of nodes increases in a region. This is the case of the auditorium indoor area, where there is a higher concentration of wearables, and therefore it also has a higher concentration of received power (around 5–10 dB more), being more remarkable in the case of high-node density due to the higher number of wearables involved. The higher intensity in this area is also given by the body shielding effect, which enhances the signal to be concentrated in the auditorium indoor area, where more users are presented. In addition, different regions can be identified within the simulation scenario where power levels differ, such as the outdoor and indoor part of the scenario, or the cafeteria area in the right part of the scenario. These regions are delimited by different types of walls (glass in the auditorium and concrete/metal in the cafeteria), which gives rise to different associated propagation phenomena, such as the appearance of hot-spots in the aforementioned localized areas, regardless node density setups. Frequency also plays an important role showing that for all the analyzed cases, higher power levels are received at lower frequencies, as it was expected. The presence of hot-spots is also more remarkable at lower frequencies (notice that the power level scale has been maintained for both frequencies for better comparison).

**Figure 3.** Bi-dimensional planes of received power [dBm] at 1.2 m height at 2.4 GHz and 5.8 GHz frequency bands. Different scenario setups have been considered: (**a**) high-node density scenario, (**b**) medium-node density scenario, and (**c**) low-node density scenario.

**Figure 4.** Bi-dimensional planes of received power [dBm] at 0.8 m height at 2.4 GHz and 5.8 GHz frequency bands. Different scenario setups have been considered: (**a**) high-node density scenario, (**b**) medium-node density scenario, and (**c**) low-node density scenario.

Figure 4 presents the same results as Figure 3, but in this case, for a 0.8 m height, emulating wearables such as smart watches operating in the wrist of a seated person. As it can be seen, a high variability of RSS is also observed in this case, showing slightly lower values of received power because the bi-dimensional plane representation is not at the same height of the transmitter's wearables. In these cut-planes, the presence of hot-spots is still visible but with lower intensity for all the different node density setups. This specific behavior is caused due to bigger link distances with the transmitters comparing with the previous cases analyzed. Besides, there are relevant differences between the operating frequencies, with more losses at the higher frequency, as it was expected.

In order to gain a better insight into the differences of received power for different heights and frequencies in the three nodes density cases, the radial distribution of RSS in the considered scenario has been assessed and it is represented in Figure 5. The considered radial line is depicted in Figure 5d, a representation of the aerial view of the scenario (ceiling has been excluded for illustration purposes), which correspond to Y = 20 m along the X-axis. As it was expected, in the comparison for the low-node density case, the trend of the obtained received power values is lower than in the other cases. Hence, the distribution of received power levels depends strongly on the node's distribution in the considered scenario. In addition, there is a slightly difference between RSS values at the different heights, showing higher values in general for the 1.2 m height (the same cut-plane as the transmitters). However, this trend is not homogeneous for all the spatial points, because as stated above, the morphology and topology of the scenario plays an important role in radio wave electromagnetic propagation for complex environments. Regarding the different frequencies, at the 5.8 GHz frequency band the losses are 5–10 dB approximately higher than the 2.4 GHz frequency band.

**Figure 5.** Linear distribution of received power [dBm] for different heights and frequency bands for Y = 20 m along the X-axis, (**a**) high-node density scenario, (**b**) medium-node density scenario, (**c**) low-node density scenario, and (**d**) radial line representation.

### *3.2. Signal to Interference Noise Ratio*

Effective signal and interference levels are determined by received power level distributions, which in turn are determined by the number of nodes as well as their topological distribution, in relation with the surrounding environment. In order to have clear insight into the interference levels in the proposed sensor network, SINR volumetric estimations have been obtained for the whole scenario. An acceptable communication link formation in terms of quality of service is given by the fulfillment of the following condition [37]:

$$P\_{RX}(\stackrel{\rightarrow}{d}, RX\_{\text{Inv}}) \ge SENS\_{RX}(SINR\_{\prime}, R\_{b\prime}, m\_{\in}) \tag{1}$$

where *PRX* is the received power for each transceiver, as a function of spatial location <sup>→</sup> *d* and receiver hardware parameters *RXhw* (e.g., antenna gain, noise factor) and *SENSRX* is the receiver sensitivity, determined by the required *SINR* threshold (or *Eb*/*N*<sup>0</sup> in the case of digital systems), transmission bit rate *Rb* and the modulation and coding scheme *mc*. In this context, the determination of useful received power and detected interference levels provides coverage/capacity relations as a function of service requirements and density of nodes within the scenario. The scenario location of all transmitting, as well as receiving elements, is a fundamental parameter, due to the large variability in power distribution in the complex environment under consideration.

Once the wireless channel characterization has been performed for different transceiver densities, as described in previous section, interference analysis in terms of SINR can be obtained, leading to system coverage/capacity estimations. For that purpose, the worst-case conditions have been considered, in terms that SINR analysis are provided when the interconnecting device operates with in-band inter-system interference. For the different node density cases, one interconnecting device has been considered as the transmitter and the rest as in-band inter-system interference.

Figures 6 and 7 show the bi-dimensional plots at 1.2 m height of the SINR distribution for two different locations of the transmitter, indoor and outdoor, considering different sensors densities distributed non-uniformly in the scenario, as interferers (see Figure 2 to have insight into the different node distribution densities). These plots represent an upper bound in relation with quality degradation in terms of simultaneous operation of the transceivers. The provided SINR values can be mapped afterwards to *Eb*/*N*<sup>0</sup> ratios, where modulation scheme as well as transmission bit rate can be explicitly considered.

Figure 6 presents the SINR values considering the transmitter a wearable (i.e., smart glasses) of a seated person inside the auditorium (X = 20.35 m, Y = 22 m) (i.e., Tx 33) and the simultaneous operation of the rest of transceivers in different node densities setups. It can be observed that the highest SINR values appear in localized areas nearby the transmitter in the indoor part of the auditorium. The differences between nodes densities are pronounced, around 10–15 dB approximately between the low- and high-node density cases for the indoor area of the scenario. Besides, for all density cases, the SINR values for the outdoor region of the scenario are very low. These results show the high dependence of node density and represent an upper bound in terms of quality degradation when other transceivers are operating simultaneously. Regarding the differences between different frequency bands, it can be seen that there is not a lot of variability in the SINR values between different frequencies, but for the 5.8 GHz frequency band, SINR values are slightly lower, depending of the spatial considered point due to the morphology of the scenario.

To gain insight into the significance of considering the whole three-dimensional scenario, the same SINR analysis has been obtained for an outdoor located transmitter node, a wearable of a seated person placed in the outside tables of the auditorium, in the left down corner of the scenario (X = 11.7 m, Y = 5.8 m) (i.e., Tx 5). These results are presented in Figure 7, where a bigger area than the previous case of high values of SINR around the selected transmitter can be observed. This is caused because of the smaller number of scatterers presented in the outdoor part, which allows a higher area with high SINR values, for all different node density cases. Besides, the outdoor-indoor communication is deeply affected by the scenario boundaries, such as the auditorium walls, being the worst case for the high

node density case. For all node density cases, it can be observed that the SINR values in the indoor area of the auditorium are very low, which means that the communication link will not be feasible with the indoor region of the auditorium. As in the previous case, there are not many differences in SINR values between frequencies, with slightly lower SINR values for the higher frequency band (5.8 GHz).

**Figure 6.** Bi-dimensional planes of signal to interference noise ratio (SINR) [dB] for Node 33 (wearable placed at 1.2 m height, same as a seated person in the indoor area of the auditorium, X = 20.35 m, Y = 22 m) at 2.4 GHz and 5.8 GHz frequency bands. Different scenario setups have been considered: (**a**) high-node density scenario, (**b**) medium-node density scenario, and (**c**) low-node density scenario.

**Figure 7.** Bi-dimensional planes of SINR [dB] for Node 5 (wearable placed at 1.2 m height, same as a seated person in the outdoor area of the auditorium, X = 11.7 m, Y = 5.8 m) at 2.4 GHz and 5.8 GHz frequency bands. Different scenario setups have been considered: (**a**) high-node density scenario, (**b**) medium-node density scenario, and (**c**) low-node density scenario.
