Path Loss Propagation Evaluation and Modelling based ECC-Model in Lowland Area on 1800 MHz

Propagation modeling is the most important part of mobile wireless network planning. Wireless network planning requires an accurate calculation of the path, which depends on different environmental conditions. It requires accurate path loss modeling of the characteristics of a specific region. The study aimed to obtain a path loss propagation model by modifying the ECC model and using linear, logarithmic regression in lowland areas. The measurement used drive test method, located in the Jakabaring area that represented the lowland area. This research used four existing path loss models, namely Okumura-Hatta, COST-Hatta, Ericsson Model, and ECC Model. It was found that the Okumura-Hatta model had the largest RMSE value, 34.90, followed by the Ericsson model, 27.07, while the ECC model had the smallest RMSE value, 8.43. The ECC model required to be modified using logarithmic, linear regression to obtain the proposed model. The results of the evaluation showed that the proposed model improved with RMSE 4.93, MAPE 2.71, and MAD 3.91, whereas the values of the existing ECC Model before modification were 8.43 for RMSE, 4.72 for MAPE and 7.09 for MAD. The proposed model provided an accurate prediction of the path loss propagation in a lowland environment. The results of the study can be used for planning engineers to plan, design, and implement the wireless communication networks in lowland area conditions. Keywords— ECC Model, Drive Test, Regression Liner Logartimik, RMSE, MAPE, MAD


INTRODUCTION
Wireless communication has been experiencing phenomenal growth both in terms of technology and data usage [1]. Internet data usage traffic continues to increase and is expected to increase over the next few years. Demand for service and service delivery has been increasingly heterogeneous and continues to grow, which causes the quality and capacity requirements to become essential needs [2]. Reliable and high-efficiency wireless network planning, therefore, requires accurate path loss planning [4].
Path loss is attenuation caused by radio propagation, namely frequency, the distance between TX and RX, ground operation, and antenna height [5], [6], [7]. The path loss is also induced by losses of free space, refraction, diffraction, reflection, and others [8], [9]. A signal from the transmitter antenna will experience multipath fading to the receiving antenna, which can cause constructive or destructive signals [7]. Path loss is one of the most important factors in link budget analysis or wireless network planning [10].
Propagation modeling is one of the most important aspects of mobile wireless network planning [11], [12], [13]. Various environmental conditions cause a different path loss resulting from radio wave propagation [14], [15 ], [16]. Accurate path loss modeling will be useful for planning, designing and implementation of telecommunications networks and for analyzing network coverage [1], [17], [18]. Several types of propagation are used by network design engineers, namely Okumura, Hatta, Cost231, Cost231-Hatta, ECC model, and Ericsson model [15]. Propagation modeling today is not accurate due to some differences in environment, terrain, and climate conditions [19]. There is no propagation modeling acceptable to all regions [2]. The selection of a propagation model suits in a particular area is not an easy task. It depends on the description of the terrain in a specific area.
Some researchers have researched modeling path loss propagation in various conditions. Pinto et. al. examined path loss determination using linear and cubic regression inside a classic tomato greenhouse [20]. They studied the propagation models for tomatoes greenhouses area, operated in 2.4 GHz. This study carried out direct modeling using Cubic Regression. Aljadid [21] researched the modeling of propagation in dense urban areas using 900 MHz frequency. It modified the Hatta model using the Linear Regression method. Nadir and Ahmad developed the propagation model with the modification of the Okumura-Hatta model using Cubic Regression [11]. Majed et. al. investigated path loss models indoor environment at frequencies 4.5 GHz, 28 GHz, and 38 GHz. They produced a proposed model and improved it while comparing it with the CI, FI, and ABG models [22].
This research focused on the evaluation of four existing propagation models, namely Okumura-Hatta, Cost231-Hatta, ECC, and Ericsson, as well as on the propagation modeling modification to obtain a modeling adapted to the characteristics of the region and the climate in swampy areas.

II. METHODS
The research method consisted of several phases, namely field measurement process using drive test method, simulation of the existing propagation model, propagation model modification, evaluation, and analysis of the proposed model.

a. Field Measurement process
Palembang is one of the cities in Indonesia, dominated by swampy areas. The conditions and characteristics of the swampy areas will have different propagation path losses. Terrain and contour conditions affect the characteristics of the propagation path loss.
The measurement process in the field was carried out in lowland areas by measuring the receiving signal level in the area and convert it to the path loss value. Drive test is the measurement of the received strength signal to the EU (User Equipment) of the power emitted by a BTS (Base Transceiver Station) on a moving car to determine the signal's strength and level [23]. Drive testing is usually performed by operators or regulators to check the operator's signal coverage and to determine the quality of the signal received.  Figure 1 illustrates the drive test measuring equipment. Hardware media consist of HP laptop, Samsung S5 & cable data, and GPS. The software measurement used was the GENEX Prob. In this study, the sample measurements were carried out using a driving test at the predetermined location and route, according to the objectives of this study.

b. Simulation of the Existing Propagation Model
This phase determined one of the most realistic existing models to be modified, namely by determining the value of the smallest path loss and the measured path loss results in the four existing model scenarios used. The existing models used were the Okumura-Hatta model, the COST-Hatta model, the ECC model, and the Ericsson model.

ECC Model
In this model, the path loss is given by the following equation: [28], [29]. where: For big cities: is the distance (km), is the frequency (MHz), ℎ is the height of the BTS, and ℎ is the height of the receiver antenna (m).

Modification of Propagation Models
Propagation modeling that has been evaluated and is closest to the conditions in Lowland areas would be modified using logarithmic, linear regression to be used as the Y value. The linear, logarithmic regression formula is [21]: where is the constant, is the regression coefficient (slope), and is the distance as an independent variable. Values and may be determined as [8]: Where is the difference between the path loss value from the drive test measurement and the modeling.

Evaluation of the Existing Propagation Model
This study adopted three index evaluation parameters to evaluate the accuracy of propagation modeling, namely: Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Mean Absolute Deviation (MAD).
MAPE is the absolute value of the percentage error data against the mean value, or it can be formulated as [30]: RMSE is the sum of the error squares or the difference between the actual value and the predicted value, then divides the amount by the amount of time forecast data and then draws its roots, or can be formulated as [23], [31].
MAD is the absolute value of the data deviation from the mean, or it can be formulated as: [30] = ∑ ( _ − _ ) (23) where: = the number of sample data.

_
= the values obtained by measuring the device. _ = the value obtained using the equation.

III. RESULT AND DISCUSSION
This chapter analyzes the results of the modification process of the existing model channel and the evaluation of the resulting model.

a. The Comparison Analysis of Existing Path Loss Models
The four propagation models used in this study were Okumura-Hatta, Cost231-Hatta, Ericsson Model, and ECC Model. The simulation used the type of suburban area, adjusted to the area condition measured in the lowland Jakabaring area. The frequency used in this research was 1800 MHz, and the transmitter was a 4G network. The results of the comparison of simulation path loss can be seen below.  Figure 3 shows the results of the test drive measurements that the path loss value is directly proportional to the distance between the BTS and users. At the distance of 0-1500 m path loss measurements, it resulted in a higher value. The ECC model had the closest trend value to the measurement results at a distance of < 1000 m, whereas the Ericsson model had the most detailed path loss value at a distance of > 1000 m.
This research used RMSE, MAPE, and MAD to obtaining the value of delta path loss accumulation. Evaluation of propagation modeling was completed using equations (18)(19)(20). The results of the calculation are as follows. Table 2 presents the evaluation paremeter values on the four existing models  The ECC formula model is as in (10)(11)(12)(13), where these study measurements were made in sub-urban or mediumurban areas, specifically in Low Land areas, using equations (14). Logarithmic, linear regression formula used equations (21)(22)  Results formulation of the proposed model is simulated using MatLab to be compared to the ECC Model. The two simulation results are compared to the trend of the measurement results by Drive Test.  Figure 4, the path loss generated by the proposed model has a higher value than the path loss in the previous ECC model and the distance value is constant. The value of the proposed model was closer to the measurement results than the previous ECC model. The values of RMSE, MAPE, and MAD were compared to obtain the values of delta path loss accumulation between measurement results and path loss models. The results of the calculation are shown in Table 3.