Initial Study of Radio Tomographic Imaging for Human localization by using Simulation Model

This paper explains the details of modelling the simulation works designin setup for the RTI system. Th simulation modelling using software is focused on the interaction of electromagnetic behaviour in a dielectric medium of human inside a monitoring area. The modelling works have involved the criteria of the human, frequency and number of sensor nodes, dielectric properties of the human and last but not least, the configurations of the Radio tomography imaging (RTI) system. The model is then developed in the software to observe and investigate the result.


Introduction
Radio Tomographic Imaging (RTI) technique is currently the most engaging approach for Device Free Localization (DFL) technology [1].DFL is a technology used mainly for human activity detection by practising the passive localization system, meaning to say the impromptu presence or movement of the human can be tracked down because they are not required to cooperatively participate in the localization process by carrying any tagging devices [2].The existing techniques, for instance, employed in the DFL technology are camera-based sensing, thermal sensing, physical excitation based on pressure or vibration detection, passive infrared sensing and radio frequency (RF) sensing [3], [4].The RF sensing principle has led to the establishment of the RTI approach, which eventually evolved as one of the most efficient localization techniques.Across the monitoring wireless sensor network (WSN), this RTI system practically works by exploiting the attenuation of RF signals caused by the presence of targeted objects.The electric characteristics of biological tissue during electromagnetic radiation exposure determine the biological consequences caused by that radiation [5]- [7].Determining the effective permittivity of biological tissue has garnered recurrent attention in recent years due to challenges with measuring biological tissue in vivo and its permittivity, which exhibits nonlinear properties in the frequency domain.[8].The impact of various human body types on RF wave propagation when they disrupt the direct route between a transmitter and receiver.[9].

Received signal strength
Wireless sensor technology is used in several applications.However, RSS used to detect the human by using RF node technology.Therefore, the use of the RSS can determine the changes.The RSS measures the Wi-Fi signal strength's RF value.The Wi-Fi signal strength is the wireless signal power received by the receiver and is measured in decibels (dBm).The RSS is a relative indicator, while the dBm is an absolute figure measuring power levels in mW (milliwatts).Decibels (dBm) are also known as the ratio of RF power, which indicates an electrical device's signal level and radio waves.0dBm corresponds to 1mW of output power.However, when there is an increase of dBm, the output power is double that of the decibel [10].As the decibels increase, the value of the output power will also increase.Can express this using Equations ( 1) and (2).
() = 1.10 () 10 (2) Besides, to determine if there is a change in the monitoring area, can calculate the losses of the signal by using:

Simulation based system design
It is necessary to design a system in order to control the confounding variables and ensure the repeatability of obtaining radio tomography attenuation for each scenario in this work in the context of the overall research work for the proposed method as shown Figure 1.For the finite element modelling, the selected geometrical dimensions of the structure and the RF nodes, the number of nodes, the frequency used and the dielectric properties for the model and electromagnetic behaviour are discussed in the following section.These factors need to be taken into consideration because they can affect the measurements for radio attenuation.In addition, measuring strategies are presented, and the best method to acquire the data in this work is determined.Lastly, the image reconstruction algorithms and image quality assessment are presented.

Configuration of the RTI set-up for human
By utilizing software, an Electromagnetic Waves, Frequency Domain interface available under a branch of the Radio Frequency module is selected to simulate the RTI system.Figure 2 below outlines the procedure to be completed before the results can be attained and analyzed.

Figure 2. Procedure of model builder in COMSOL Multiphysics software
The sequence is started by selecting the Radio Frequency module at a single frequency stationary study until exporting the simulation result for post-processing analysis.By default, COMSOL Multiphysics does simulate the model in 3D even if a 2D space dimension is chosen.Aforementioned, a 2D geometrical RTI system that represents the monitoring area is developed to mimic the actual hardware structure.However, several modelling features such as the size of the RF node and human model are accessed via comprehensive modelling criteria, as discussed previously.The physical geometry and properties of the constructed monitoring area are built according to the specific dimensions as in Table 1.The next step is to add the material settings to each domain entity in the designed model.In this FEM study, the materials feature is related to setting the domain's physical properties: relative permittivity, relative permeability and electrical conductivity.The air that fills the free space and sensor nodes is available and fixed from the global built-in material.The monitoring area itself is defined as made up of square, and its properties are selected from the predefined material [11].
In this study, the Radio Frequency module formulates and solves the differential form of Maxwell's equations with the element of initial and boundary conditions.The boundary condition setting is crucial for ensuring that the electric field can propagate correctly inside the model.The boundary condition for the power source is established at the boundary between the RF sensor node.These boundaries are known as the Port node, where it can launch and absorb electromagnetic energy.On the other hand, the free space's boundary is set to scattering boundary condition to make it transparent for scattered waves.It is only accessible on the external boundaries where it specifies that the propagating waves are absorbed in the boundary and, thus, there is no reflection at the boundary.Subsequently, a mesh is generated and discretised automatically into finite triangular elements.In a simulation process, by using 8 RF node, decrease the simulation complexity, resulting in a shorter solution time.The meshing, in this case, is performed by using the default physics-controlled mesh at uniformly extrafine element size.Figure 3 below displays the result of finite element meshing in the 2D RTI model.

Figure 3 Extra fine meshed geometry
The RTI system is then computed using the default stationary study.The study solver is executed with the assumption that the response of the RTI model does not vary in time.The Frequency Domain in this study is constituted as a single-frequency study step at 2.4 GHz with solution time ranging from 10 s to 15 s.Finally, the visualisation of the 2D performance results is determined according to the selected post-processing tools.The analysis of the simulation results is based on observing the surface plot and generating its numerical data into a file.The surface plot is configured to display the electric field distribution plot, and the numerical grid data is mapped into 600 × 600 square arrays consisting of 360000 pixels of data.

Sensitivity Maps
In addition, the forward problem is also being solved by using the analytical solution of sensitivity maps which construct the sensitivity matrices.The sensitivity map, also known as the Jacobian matrix, , is the projection effects from a particular sensor node to a known receiving node in terms of weighting matrices.The matrices indicate the area where relative permittivity changes would influence the electric field measurement.[12].Different arrays have different sensitivity, depth of investigation and different signal strength [13].
From the local linearization, 190 sensitivity maps are obtained to represent the electric field distribution, (, ).The element,  of this matrix define the electric field, (, ) and the changes at a measurement area due to a slight in the dielectric properties of that element, (, ) of the model [14].In this study, the electric field distribution is computed on the extra-fine tetrahedral mesh.The sensitivity map for the sensor node pairs,  and  at a given position, (, ) is determined as shown in Equation ( 4) below: where  and  are the excitation and measurement sensor node pairs, respectively.Each Jacobian matrix is acquired from the FEM study of the forward model without the human phantoms appearing in the monitoring area.Based on the RTI technique, the signal's attenuation due to the presence of phantoms can be integrated by uniformly dividing the imaging area into small square pixels,  (Q.Wang et al., 2016).Therefore, a weighting model is applied because the attenuation of a link contributes differently to each pixel.The grid size for each Jacobian matrix is 0.1 cm, and the number of grid pixels in the computational domain is 600 × 600.Table 2 shows the examples of sensitivity maps of excitation at RF sensor node,  = 2 and j = 3, with the corresponding receivers at RF sensor nodes  = 6,  = 8 and  = 7.The scale on the images is the same to confirm that as the distance between two sensor nodes increases, the sensitivity decreases.

Results and discussion
The calibration of the RTI system was done in two stages.The first measured the RSS value when the experiment area was empty, and the second measured using obstacle (human).Figure 4 shows the RSS value of both measurement calibration and measurement.The RSS values in the calibration of the system for all 8 RF nodes of the system were in the range of −23dBm ≤ x ≤ −25dBm when the zone was empty, and the RSS values of humans were in the range −42dBm ≤ x ≤ −52dBm.The system can identify the presence of intruders, and it triggers a feedback loop that has a substantially more significant standard deviation and variance in an inhabited room with an obstacle which was human.Even from Table 3, we can deduce that statistics (no moving) like variance and standard deviation are great for detecting intrusions.Furthermore, one of the characteristics to emphasise in this part is the change in variance.The variance was computed so that the system could determine the difference in RSS readings from the average.The variance for the empty zone was approximate −23dBm ≤ x ≤ −26dBm; however, when the human was detected inside the area, the difference in RSS value increased between -15 dBm to -25 dBm.The outcomes that contribute to the research area are the applications of RTI system in detecting and localization multitarget using 8 RF node through simulation.

Conclusion
This paper explains the details of modelling the simulation works designing setup for the RTI system.The simulation modelling using software is focused on the interaction of electromagnetic behaviour in a dielectric medium of human inside a monitoring area.This work begins with the modelling of the RTI system.In order to elaborate on the forward problem of RTI in detail, the corresponding mathematical model for the physical field in the monitoring area is established.Also, the finite element modelling (FEM) of the RTI system has been performed in two-dimensional (2D) through software to solve the partial differential equations and obtain sensitivity maps.

Figure 1 .
Figure 1.Flow chart of the overall simulation study.
Construct a physical model using available geometries RF sensor nodes Human phantoms Define the materials for each domain Define the boundary condition for each sensor node Generate a mesh for the constructed model Compute the study.Define the boundary condition for each sensor node Visualize the 2D result

Table 1 .
Input parameters of the RTI set-up for human model

Table 2 .
Normalized sensitivity maps of an excitation at RF nodes

Table 3 .
Data statistic of RSS for 100 cycles between two nodes.