Effects of Crowd Density on Radio Propagation at 24 GHz in a Pedestrian Tunnel for 5G Communications

In this paper, we report the results of radio propagation characterisation in a pedestrian tunnel with different crowd densities at 24 GHz using commercial ray-tracing software called Wireless InSite. The 3D empty tunnel and human body models we created using computer-aided design software and imported into Wireless InSite. The tunnel model is based on a pedestrian tunnel connecting Suria and KLCC, which is located in the heart of Kuala Lumpur. Five three-dimensional (3D) human body models with different levels of detail were developed and tested. The crowd densities investigated were 0, 0.05, 0.1, 0.15 and 0.2 people/ $\text{m}^{2}$ which correspond to 0, 25, 50, 75, and 100 people, respectively, in the study area. The results show that the path loss exponent, log-normal shadowing’s standard deviation, and fluctuation in received power increase as the number of people increases. When the crowd density is above 0.1 people/ $\text{m}^{2}$ , the path loss exponent of the large-scale path loss model is higher than that of the empty tunnel. The results of this study are also useful for understanding the effects of human crowds on millimetre wave propagation in indoor tunnel-like environments such as hallways, enclosed corridors, mines, and transportation tunnels. The findings contribute to increasing the effectiveness of network planning and deployment for 5G communication, especially in pedestrian tunnels.


I. INTRODUCTION
Connectivity is an important aspect of constant communication, regardless of where we are. Devices including mobile phones, tablets, smartwatches, and laptops are always connected to the Internet. Much significant research has been done to study millimetre wave propagation inside an indoor simulation. These include measurement and simulation inside multiple indoor environments ranging from small area to large area such as conference room, typical living room, office area and indoor research centre [1], [2], [3], [4]. However, throughout normal day-to-day activities, people tend to move from place to place, whether they are walking on the The associate editor coordinating the review of this manuscript and approving it for publication was Jjun Cheng . road, passing through alleys and passageways, or even inside a pedestrian tunnel. A tunnel is an important man-made structure that is useful in many ways, such as for mining and transporting vehicles and pedestrians. Thus far, investigations on the propagation characteristics of radio waves inside tunnels have mostly focused on different types of tunnels [5], [6], [7]. A recent study involving pedestrian tunnels with human crowds was reported in [8] and [9]. The results show that human blockage is significant, especially in the millimetrewave band. However, the work in [9] only shows the measurement results of a crowd with approximately 20-30 people while in [8] only 4 human bodies were considered. It is still unclear how specific crowd densities may affect the millimetre-wave propagation in a tunnel. Hence, the aim of this work is to close the gap by investigating the effects of human crowd density on millimetre wave propagation inside a pedestrian tunnel using a realistic simulation environment of the human crowd model inside a pedestrian tunnel.

A. MILLIMETER-WAVE PROPAGATION IN TUNNEL
Different types of tunnels will have different effects on the characteristics of radio propagation owing to the shapes and materials of the tunnel. Two major factors determine the propagation model of the tunnel. The first tunnel cross section. The three major cross-sectional shapes are rectangular, circular, and semi-circular [5], [10]. The received power in a circular tunnel is larger than that in semi-circular tunnel. Tunnel dimensions also have strong impact on the radio wave propagation too. The attenuation constant differed for each different dimension. The second factor was the tunnel path. They can be differentiated into straight and curved tunnels [10]. The received power decreases with an increase in the tunnel curvature, as larger curvatures mean much more reflection loss and propagation loss, which results in lower received power.

B. SIMULATION-BASED RADIO PROPAGATION CHARACTERIZATION METHODS FOR TUNNEL
In the literature, 3 different methods have been proposed to characterize the tunnel propagation model which is geometrical optics (GO) [11], ray frustum techniques and fast Fourier transform (RFT-FFT) [12], as well as ray tracing [13]. GO analytically determines the paths between the transmitter and receiver which eventually lowers the complexity of computing. For the RFT-FFT, this method directly estimates the impact of tunnel structures and surface roughness on the performance of millimetre waves. Therefore, this method is numerically efficient. Ray-tracing employs three propagation mechanisms: reflection, transmission, and diffraction. The fundamentals of ray tracing are based on the uniform theory diffraction (GO/UTD) Wireless InSite offers a variety of propagation model to be selected from. This includes Full 3D, X3D, Hata, TPG, OPAR, et-cetera. However, Wireless InSite suggested to use either Full 3D or X3D propagation model for indoor environments. Since Full 3D has no restriction on object shape such as flat or sloped roofs, this propagation model is more suitable for the tunnel environment as it has curved circular ceilings. For Full 3D propagation model, two ray-tracing methods are used: the shooting and bouncing ray (SBR) method and the eigenray method. The SBR method was first used to draw ray routes across the three-dimensional building geometry without considering for the position of individual field points. The rays were then traced from the source locations, with the rays secularly reflecting off the building walls. The rays that strike the building walls are secularly reflected and tracked until the maximum number of reflections is reached, or until the rays strike the study area limit. Fig. 1 shows how diffracting edges are identified using rays shot from the transmitter. Fig. 1 (a) shows two rays reflected from facet 3, but only one reflected from facet 55. Therefore, the diffracting edge is between the two rays. Thus, a diffraction points on edge 55-56 is located and the path followed by the incident field is constructed. In Fig. 1 (b), the edge 10-11 would be identified as the diffracting edge for the incident field as it reflected from facet 13. On the other hand, the eigenray method entails an explicit creation of the ray routes between each transmitter (Tx) and receiver (Rx) that fulfil Fermat's principle of least time, except for transmissions, where there is no refraction. The eigenray technique can only be used with ray pathways that contain up to three reflections and diffractions. Similar to the SBR approach, it can detect up to 30 transmissions per ray path. However, this technique imposes limits on the pathfinding.

C. HUMAN MODELLING FOR MILLIMETRE WAVE PROPAGATION STUDY
Several human models have been used in the previous studies. In this work, five human models are tested, and the model that is suitable for Wireless InSite is then used in the investigation of crowd effects. In [14], a simple rectangular block was presented, which was then simplified to a double-knife-edge model. In [15], a knife-edge model was proposed. Another type of human model is a cylindrical model [16]. In addition to simplified models, simple full-body models can be found in [17]. In this study, a new full-body complex model is developed and tested to understand whether increasing the details of the 3D model shape will improve the result.
In [15], the authors suggested that the dielectric properties of human skin should not be considered because the conductivity is sufficiently high to be assumed as a perfect electrical conductor. However, the implementation of clothing materials in human models could increase the model accuracy. Some previous studies have shown that the knifeedge model is a better choice than all other models because it is more compact and concise. Nonetheless, in this work, it is necessary to study different human models because existing work on human simulation using Wireless InSite is lacking. The limitations of Wireless InSite in 3D model creation must be considered.

II. SIMULATION SETUP
This section is divided into 3 subsections: empty tunnel modelling, human modelling, simulation scenario, and path loss VOLUME 11, 2023   modelling. We first list all the notations of parameters in Table 1 A. EMPTY TUNNEL MODELLING The simulation environment was taken from a pedestrian tunnel connecting Suria KLCC and KL Convention Centre, located in the heart of Kuala Lumpur. The dimensions of the tunnel were obtained from [9]. Fig.2 shows a cross section of the tunnel. The tunnel has different cross-sections consisting of a horseshoe shape and rectangular box shape that alternate with each other. The tunnel length is 94m and has a width of 5.3m. Due to the limitation of Wireless InSite in creating a floorplan, importing a DXF file from other drawing tools is required. AutoCAD was used to construct the tunnel model. The properties of the tunnel materials are listed in Table 2. To launch a 3D-ray inside Wireless InSite, Full 3D propagation model is chosen inside the simulation software during setting up the study area. This is because the objective of this research is to study and characterize the effect of  crowd density on propagation model rather than to improve or compare ray tracing method. Furthermore, the FULL 3D propagation model method used is considered an advance method since 3D modelling is not easy. The FULL 3D method used is also based on a tested SBR method which shows reliability and consistency across literatures. Fig. 3 shows the 2D and 3D views of the 3D-ray launching during simulation. A total of 9001 receivers were placed along the red line with 0.01m spacing between each receiver to fully cover the entire distance of the tunnel. Graphics of propagation rays for Rx no.3000 were shown in Fig. 3. (b) and (c).

B. HUMAN MODELLING
Human body modelling plays a critical role in characterising millimetre-wave propagation inside tunnels. The five 3D human models created in AutoCAD and imported to Wireless InSite are shown in Fig. 4. A ray-tracing simulation was then conducted according to the setup in [18] for comparison and validation purposes. The human model was placed in between the Tx and Rx at a 7.5 m distance. The human model is moved to the right and left with a 0.02 m increment up to 0.5 m from the initial position, as shown in Fig. 5. The frequency used was 26 GHz, the same operating frequency in [18]. The antenna is omnidirectional with a gain of 24.30 dBi. The Tx power is set to 24 dBm and 3 dB noise figure for Rx. The study area was set to Full 3-D. The simulation floor is set to concrete as in Table 1 with a thickness of 0.3 m while the human surface material is cotton with ε = 1.7, σ = 0.3, thickness = 0.01m.

C. SIMULATION SCENARIO FOR CROWD INSIDE TUNNEL
The complete simulation scenario was established by integrating the 3D models developed in Section 2.1 and Section 2.2. The tunnel model and human body model were combined in Wireless InSite as different features for simulation. Uniformly distributed random human location coordinates were generated using MATLAB to simulate a human crowd. A cylindrical human model was used for the crowd simulation. The crowd density was set to 0, 0.05, 0.1, 0.15 and 0.2 people/m 2 (which corresponding to 0, 25, 50, 75 and 100 people respectively) in the simulation scenario considered. The minimum distance between the human models is set to 0.5 m to avoid overlapping. The area of the tunnel was 498.2 m 2 (length = 94m, width = 5.3m). For each crowd density, 10 different realizations of random coordinate sets were simulated. Fig. 6 shows a sample simulation scenario for the Wireless InSite. The waveform used was a 24 GHz sinusoid. The frequency is the same operating frequency when the measurement inside the tunnel in [9] to maintain consistency and for comparison purposes. The antenna was horn type with a gain of 20 dBi. The horn antenna aperture was set to 0.404 × 0.0319m (width x height). The feed dimension was set to 0.0101 × 0.0062 × 0.07m (width x height x length). The Tx and Rx heights were both set to 1.52m. The transmit power for the Tx was 20 dBm and the noise figure for Rx was 3 dB. The study area was set to Full 3-D.

D. PATH LOSS MODELLING
The ray-tracing simulation results were then compiled and analysed using MATLAB. The measurement and simulation are done with respect to 1 Tx and 1 Rx at every interval as the focus of this paper is to characterize path loss. Two main parameters are observed: the received power and path loss for each data set. The received power for an empty tunnel was compared with the measurement and simulation performed in [8]. The large-scale path loss model using a close-in free-space reference distance is given by where PL(d) is the average path loss at distance d, d o is the reference distance, n is the path loss exponent, X σ is a Gaussian distributed random variable with zero mean and standard deviation σ . Typically, d o is set to 1 m. Based on the simulation results, the path loss exponent, n, as well as the σ can be estimated. The value of n and σ were tabulated for each dataset and the average value was calculated.

A. HUMAN BODY BLOCKAGE
According to paper [18], the simulation environment used for studying the human blockage effect only considered the floor where the human model was standing. In order to replicate similar environment from [18], study area was setup without putting up walls and ceilings. This is done to provide references for comparison as well as verifying the accuracy of our model using the measurement data in [18]. Therefore, only ray reflections from the human body blockage and floor were considered. Human attenuation was calculated by determining the difference between the received power with and without a human blocker in the study area of the simulation [18]. All the 3D human models considered have a width of 0.5 m and a height of 1.65 m. Fig. 7 shows the results of the human attenuation for all five human models. For the simple rectangular model, the maximum attenuation was approximately 15 dB. A significant increase in attenuation is seen from -0.25m to +0.25m where the human body is blocking the line-of-sight (LoS) of the Tx and Rx. The maximum attenuation of the cylindrical model was approximately VOLUME 11, 2023 8 dB. The simulated attenuations from the cylindrical, simple rectangular and knife-edge models are similar to the human blockage measurement presented in [18]. The existence of the arm, legs and head in the full-body simple model and full-body complex model caused additional propagation rays to arrive at the Rx when the human was moving across the Tx-Rx. This can be seen in the larger fluctuations of the attenuation when a human blocker exists compared with that of an empty space. The knife-edge model exhibited a maximum attenuation of approximately 19 dB. Human model validation was done using data from literature [18] and [19]. In this study, the cylindrical model was considered for the simulation of a human crowd in a tunnel due to three main reasons. First, the result from the cylindrical model simulation is close to the measurement result taken in [18] in which the cylindrical model has the least attenuation among all other human models that were simulated. Second, among 5 human models, cylindrical and simple rectangular are the ones that suitable to be simulated in human crowd due to the compatibility of human model type in Wireless InSite. In density 0.2 people/m 2 , there will be 100 human bodies to be created, complex human body structures such as the full body simple and full body complex will take too long to be created. While for knife edge model, the infinite height of this model will affect the ray propagation on the ceiling which is not practical. Third, similar cylindrical model was used for Wireless InSite in a previous study in [19]. Fig. 8 illustrates the received power versus distance with a distance resolution of 0.01 m. For validation purposes, the result is compared to the measurement result in [9] in the form of received power vs Tx-Rx distance. The simulation results agree very well with the measurement results in terms of the received signal fluctuations. The statistical comparison between measurement in [9] and our simulation result are shown in Table 3. The statistics observed and calculated are Standard Deviation, Root Mean Squared Error (RMSE),  Mean Absolute Percent Error (MAPE), and Mean Absolute Error (MAE). There are two areas in the graph where there is shows an increasing trend in the received power at 30-35 m and 60-70 m. Notably, these two areas are box-shaped tunnel section. The increase in received power is because more rays are being received by the receiver antenna owing to the curved shape ceiling of the tunnel before the box-shaped tunnel. The transmitted rays were reflected towards the middle part of the tunnel owing to the curved ceiling. A similar observation was given in [6], where a circular shaped tunnel had larger received power than a normal rectangular box-shaped tunnel. Fig. 9 shows the received power versus distance for crowd densities of 0.05, 0.1, 0.15 and 0.2 people/m 2 . The fluctuation of the received power is the least for the smallest crowd size, which is 25 people (or 0.05 people/m 2 ). Starting from a crowd of 50 people, the fluctuation in the received power becomes significantly different from that of the empty tunnel. The 100 people crowd scenario shows the largest fluctuations in the received power. As the number of humans in the tunnel increases, there are more surfaces for the rays to be reflected before reaching the receiver. Owing to the variation in the phase shifts of the direct and reflected beams, constructive or destructive addition occurred, and a larger variety of received power is observed [20]. The deep fade may also be due to the blockage of humans in between Tx and Rx [15]. More humans caused more blockages between the Tx and Rx; hence, the received power fluctuated more.   Fig. 10. shows the estimation of the path loss exponent for the empty tunnel. Fig. 11 shows the probability density function (PDF) of fading for an empty tunnel, and by fitting a normal distribution, the standard deviation σ can be determined. Based on the received power, the path-loss exponent can be obtained by least-squares fitting. The results show for the empty tunnel, the path loss exponent is 1.52, which is lower than that of the free-space path loss, which is 2. This is because of the waveguide effect in the tunnel [10].  The estimated log-normal shadowing standard deviation for the empty tunnel is 5.29. This result is consistent with the measurement results reported in [9]. Fig. 12 shows a sample result for the path loss exponent characterization for different crowd sizes (or densities). The estimated path loss exponent increased as the crowd density increased. The path-loss exponents for all crowd densities considered were lower than the free-space path-loss exponent of 2. This finding is supported by [1], [2], [4], [9], [21], [22], and [23], where an indoor scenario environment generally has a path loss exponent lower than that of free space. The average values of the estimated path loss exponent and log-normal shadowing standard deviation based on ten simulation runs for different crowd sizes are listed in Table 4. Fig. 13 shows plots of the data given in TABLE 4. For 25 people and above, the path loss exponent increases linearly with the crowd size. However, a smaller crowd size (or density) such as 25 people (or 0.05 people/m 2 ) may lead to a path loss exponent smaller than that of the empty tunnel. This is because there are rays being reflected and diffracted from the human body which caused variations in the constructive and destructive addition of the directed and reflected rays. In this case, there are more rays reaching the Tx due to the reflection from the human body, thus lowering the path loss exponent. Furthermore, when the crowd density is small, the effect of reflected and diffracted rays on the human body is more significant than the effect of human blockage itself.

C. CHARACTERIZATION OF HUMAN CROWD EFFECTS
In Fig. 14, the simulation result for the front part of receiver number 2400 for 25 people crowd size is shown. The human model circled in red reflected the transmitted rays. The receiver number 2400 has a total of 25 paths ray in 25 crowd size with received power of -24.44 dbm. This is slightly higher than in empty tunnel scenario measured at the same receiver point which recorded 17 ray paths and -24.7 dBm received power. As the crow density increases, the outcome follows the expectation as more bodies are covering the line-of-sight of Tx-Rx. Furthermore, even if there is human blocking the LoS, the signals may get reflected by the  walls and other human bodies before reaching the Rx. For the log-normal shadowing standard deviation, the value increases steadily from 0 to 100 people.

IV. CONCLUSION
In this study, the effect of a human crowd on millimetre wave propagation inside a pedestrian tunnel has been investigated. Specifically, the characteristics of received power, path loss, and crowd shadowing are determined through a 3D raytracing simulation. As the number of humans inside the tunnel increases, the fluctuation in the received power along with the tunnel also increases. The crowd density (or size) inside a pedestrian tunnel has a significant effect on the path loss exponent of the large-scale path loss model and the standard deviation of log-normal shadowing. Based on the simulation results of a realistic pedestrian tunnel model, as the number of humans (or crowd density) increases, the path-loss exponent also increases. Thus, a higher crowd density leads to a faster decrease in received signal strength as the Tx-Rx distance increases. However, it is interesting to note that a smaller crowd size (or density) such as 25 people (or 0.05 people/m2) may lead to a path loss exponent smaller than that of the empty tunnel. On the other hand, the standard deviation of log-normal crowd shadowing increases as crowd density increases. This project considered a specific pedestrian tunnel environment in which the accuracy of Wireless InSite simulation is validated through accurate comparison between real measurement and simulation results. The comparison shows little disparity thus the method used is valid. Wireless InSite simulation can work independently without measurement, therefore, the developed methodology can be implemented in similar indoor tunnel-like environments such as hallways, enclosed corridors, shopping malls, and mines as the structure of these indoor tunnel-like environments can be drawn in other 3D drawing software and imported into Wireless InSite for simulation on propagation prediction.

APPENDIX
The abbreviations used in this paper are listed as: