EMF Assessment Utilizing Low-Cost Mobile Applications

Abstract

This study introduces a low-cost alternative method for mapping the electric field strength from 4G LTE base stations and identifies areas where this mapping is more accurate. A drive test campaign was conducted in the urban environment of Thessaloniki, Greece, using data obtained from three identical smartphones, each connected to a different mobile operator and an exposimeter. The smartphones used a mobile application to record Reference Signal Received Power (RSRP) values, while the exposimeter measured the electric field strength in selected frequency bands. In the first part, the variability of the received power over different periods within certain areas was studied, and the reasons for this variability were identified. In the second part, a linear factor was calculated to convert RSRP values into electric field strength using data from both the application and the exposimeter. The converted RSRP values were subsequently compared with the exposimeter data for validation. The results indicate that in areas where the variability of the received power is lower, the linear relationship between smartphone and exposimeter data is statistically stronger resulting in calculated electric field strength values are closer to the measured.

1. Introduction

The evolution of wireless communication technology, as well as the expansion of mobile networks in urban environments, has led to increasing public concern about electromagnetic exposure and its potential health effects [1]. While studies monitoring exposure with the evolution of wireless technology [2,3] and comparing exposure levels among different countries have been conducted [4], extensive measurement campaigns are essential to ensure that exposure limits are not exceeded [5] and that public health is protected [6]. However, the specialized equipment and highly trained personnel necessary for such measurement campaigns make the latter quite costly.

Many studies have explored alternative methods for estimating electromagnetic exposure, such as artificial intelligence models using information about base stations [7], utilization of mobile metrics from smartphones for identifying points of interest [8], placement of low-cost sensors near base stations [9] and the development of machine learning models to convert data obtained from mobile applications into electric field strength [10].

This study aims to provide a method for mapping the electric field strength emitted by the base stations of 4G cellular networks in outdoor urban environments using low-cost tools, such as smartphones equipped with a specific application. It is presumed that a greater level of uncertainty is acceptable in electric field mapping than in compliance testing since the former aims to identify areas of high and low exposure. Moreover, in this work, we aim to identify areas where the estimation is likely to be more accurate based on the variability of received power over different time periods. Specifically, three smartphones with an application that provides Reference Signal Received Power (RSRP) values, along with an exposimeter, were used in a drive-test measurement campaign. The exposimeter data were initially used to determine a factor for converting RSRP values to electric field strength and subsequently to validate the proposed method.

This method offers an immediate estimation of the actual downlink electric field strength from 4G networks without requiring high computational power used in simulations (e.g., with ray tracing) or expensive equipment used in compliance testing (e.g., frequency selective radiation meters). It can serve as a mapping tool to identify points of interest, in terms of exposure, for further measurement with specialized equipment.

This study is organized in the following sections: Section 2 describes the measurement methodology, equipment used, and data processing procedures. Section 3 presents the analysis results, while Section 4 discusses the findings, highlights the advantages and limitations of the proposed method, and suggests directions for future research. Lastly, Section 5 shows the conclusions of this study.

2. Materials and Methods

2.1. Area of the Measurement Campaign

The measurement campaign was conducted using an electric scooter on two different 10-km routes, covering central streets in the city of Thessaloniki, Greece (Figure 1). The first route was measured 21 times, from 5 to 14 September 2023, namely, three times per day (morning, noon, and evening) on every day of the week (Monday to Sunday). The second route was measured six times, from 2 to 6 October 2023, namely, at noon of each weekday (Monday to Friday), plus an additional time on Thursday night. The average speed of the scooter was approximately 15 km/h, with each route repetition taking about 40 min.

2.2. Measurement Equipment

The measurement equipment included three identical Xiaomi 12 Pro 5G smartphones, each connected to a different mobile operator with mobile data enabled, and the EME Spy Evolution exposimeter (Figure 2a).

For data collection, the smartphones utilized the G-NetTrack Pro application (Sofia, Bulgaria, version 30.1) [11], a mobile app capable of capturing and recording certain Key Performance Indicators (KPIs) such as signal strength (e.g., RSRP), signal quality (e.g., Reference Signal Received Power, RSRQ), downlink bitrate, and other parameters like timestamp, latitude, and longitude. The full list of parameters can be found in [12]. The app records data from the main cell and up to 17 neighboring cells and does not require device rooting. Notably, a neighbor cell may refer to a different 4G band within the same Base Station (BS) rather than a different BS. The app’s sampling rate was 1 s, resulting in approximately 2500 measurements per route repetition. The exported file also included a timestamp and geographic coordinates for each measurement.

The EME Spy Evolution exposimeter (Plouzane, France) can isotropically capture and record electromagnetic radiation with a measurement uncertainty of ±1.5 dB (for frequencies below 4 GHz) in specific bands selected by the user [13]. The chosen bands for our scenario included all downlink (DL) bands of mobile networks (791–821, 925–960, 1805–1880, 2110–2170, 2570–2620, 2620–2690, 3300–3800 MHz) for the first route, resulting in a minimum sampling period of 4 s and a total number of about 600 measurements per route repetition. For the second route, all mobile network bands, including uplink (UL), were included, resulting in a sampling rate of 6 s and a total of about 400 measurements per route repetition. The exported file included a timestamp.

Equipment placement involved two bags: a bag with three compartments positioned below the scooter’s handlebars containing the three mobile phones and a second bag placed at the front of the driver holding the exposimeter (Figure 2b). The instrument placement was consistent for both routes and all route repetitions.

2.3. Measured Data Processing

2.3.1. Data Preprocessing

The focus of this study was solely on 4G LTE networks. This is due to the app’s limitations in providing information about neighbor cells in 5G networks and because 4G coverage is nearly 100% along the selected route for all three operators, whereas 5G coverage is much lower. Furthermore, the 5G network in Greece follows Non-Standalone Architecture (NSA), meaning the 4G network is utilized even when a smartphone is connected to 5G [14,15].

The parameter used by the app to determine the received electromagnetic power on the smartphones was the RSRP. RSRP is an indicator used in handover and cell reselection decisions and is calculated as the average power of the resource elements carrying reference signals [16]. The utilized RSRP values concerned only the 4G LTE network and were available for the main cell and each of the 17 neighbor cells for every smartphone, i.e., mobile operator in this study. The total received (reference) power for each smartphone was calculated by summing the RSRP value for the main cell and all 17 neighbor cells after converting from decibel milliwatts (dBm) to milliwatts (mW). The total received reference power for all smartphones was the sum of these calculations for all three smartphones. Similarly, the received power from the exposimeter was calculated as the sum of the power flux densities of all 4G bands, which were 791–821 MHz, 1805–1880 MHz, and 2620–2690 MHz in September 2023, with the 2110–2170 MHz band also added for 4G in October 2023 based on app data and spectrum allocation information [17,18].

To minimize fluctuations in the received power data from both the exposimeter and the app resulting from multipath propagation, shadowing, and similar effects, a sliding window smoothing technique was applied. Specifically, a moving average sliding window of size 2 was applied to the exposimeter data, and a window of size 8 was used for the app data to cover the same distance, given the difference in sampling rates of the two types of equipment.

An important step for comparing the two received power values was to synchronize the four devices: the three smartphones and the exposimeter. Due to different sampling rates (four or six seconds for the exposimeter and one second for the smartphone application), achieving a common starting point using the timestamps resulted in one exposimeter data point corresponding to four (or six) measurement points from the application. As the exposimeter provides nearly instantaneous readings compared to the smartphone app, only one out of every four smartphone app data points was retained. The choice of which measurement point to retain was made by aligning the normalized received power from the exposimeter and the application. An example of this alignment for a portion of randomly selected route repetition of the first route is presented in Figure 3.

The final stage of the data processing procedure involved filtering out excessively high and low values from both the smartphones and the exposimeter data. Specifically, the mid-90% of the data was retained after eliminating values that fell between the 5th and 95th percentiles. This filtering was applied to every route to remove outlier samples.

2.3.2. Correlation of Power Measured with the Two Types of Equipment

In a previous study, a method for mapping the exposure from 4G LTE cellular networks using a mobile app was introduced [19]. This method was possible due to the correlation coefficient between the received power of the exposimeter and that of the app being above 0.5 for the entire dataset, indicating a strong linear relationship. Specifically, on days with low road traffic, the correlation coefficient exceeded 0.65, while on high-traffic days, it dropped below 0.5.

Additionally, it was concluded that considering only the main and up to three neighbor cells out of the 17 available was sufficient for exposure mapping.

By establishing this linear relationship, a cost-effective alternative for mapping electromagnetic radiation exposure using smartphones and an application was introduced. However, this estimation method had limitations due to variability in road traffic and communication load.

Figure 4 displays the correlation coefficient results between the received power of the exposimeter and the app for several repetitions from the first route. The app’s received power calculations varied based on the number of considered neighbor cells.

From this figure, two key points can be confirmed: First, three neighbor cells were considered sufficient for the mapping, as the correlation coefficient did not significantly improve when all 17 neighbor cells were considered. Second, some route repetitions showed higher correlation coefficient results than others due to lower traffic and communication load. For instance, during low traffic conditions, such as Tuesday evening corresponding to Repetition 1, the correlation between the exposimeter and app data was higher compared to high traffic conditions, such as Friday noon corresponding to Repetition 6. This is because the exposimeter captures power from the entire 4G spectrum, while the app records only reference signals, i.e., RSRP. Specifically, during rush hours, the power emitted by BSs increased [20], affecting only the exposimeter data. Moreover, high street traffic led to multipath propagation in certain areas, affecting the exposimeter and mobile device differently, thereby disrupting the linear relationship. These effects reduced mapping accuracy on high-traffic days.

The goal of this study was to reduce these limitations by identifying areas with high variability in received power over route repetitions. This will allow us to predict where our estimations are reliable and where there is a high likelihood of misestimation, as it is observed that high variability can disrupt the linear relationship between the received power measured by the two types of equipment.

2.3.3. Routes Segmentation

Since the exact locations of the measurements differed between route repetitions (due to variations in travel speed, traffic conditions, etc.), it was necessary to define fixed segments within each selected route and assign each measurement point to one of these segments. By doing so, the data could be analyzed for their behavior over different route repetitions within the same segments.

The creation of these segments was achieved by using one of the route repetitions as a prototype and creating circles with a constant radius to cover the entirety of the route. This method resulted in 167 segments for the first route using 25-m radius circles, as presented in Figure 5, and 120 segments for the second route using 30-m radius circles, due to the higher sampling rate of the exposimeter, ensuring that all segments included at least one data point per repetition.

It is worth noting that some of the route repetitions were not included in the calculation hereafter due to slight route changes that could be accommodated by the above segmentations, leading to a total of 17 repetitions for the first route and 5 repetitions for the second.

The analysis of the variability of the received power for each selected route was performed with five separate calculations: one for the exposimeter and four for the app (one for each operator separately and one for the sum of all three operators). In every route repetition, these five calculations were made for each measurement point, and the result for each point was allocated to the corresponding route segment. Subsequently, the average value of the calculations was determined for each segment. This process resulted in 5 × 17 × 167 values for the first route and 5 × 5 × 120 values for the second. Based on the average value for each segment and route repetition, we identified the segments with high and low variability.

2.3.4. Conversion Factor

Since the aim of this study was to estimate the actual electric field strength, it was essential to extract both the power flux density and, by extension, the electric field strength from the smartphone’s parameters.

Given that all measurements were taken in the far field, the power flux density was related to the received power, as shown in Equation (1) [21].

$$S={\displaystyle \frac{P_R}{A_{eff}}}$$

(1)

where $S$ is the power flux density in Watts per square meter, $P_R$ is the received power in Watts, and $A_{eff}$ is the effective aperture in square meters.

The effective aperture is given by Equation (2) [21].

$$A_{eff}={\displaystyle \frac{{\lambda }^2}{4\pi }}\ast G$$

(2)

where $\lambda$ is the wavelength of the received electromagnetic wave in meters and $G$ is the antenna gain.

The wavelength can be calculated from the absolute radio-frequency channel number (ARFCN) provided by the app. The ARFCN can be converted to frequency, and thus to wavelength.

We propose that the received power can be related to the RSRP value as shown in Equation (3):

$$P_R=RSRP\ast Correction Factor$$

(3)

where $RSRP$ is the strength of the reference signal in Watts, and the $Correction Factor$ adjusts the reference signal power to represent the total power received by the smartphone. It is important to note that since our goal was to estimate the received power across the entire frequency band (not just the user-specific signal), the Correction Factor must consider network load and traffic. Thus, it is not a constant value but varies depending on network conditions.

Combining Equations (2) and (3) into Equation (1), we can rewrite the expression for the power flux density as follows:

$$S={\displaystyle \frac{4\pi }{{\lambda }^2}}\ast RSRP\ast Conversion Factor$$

(4)

where the $Conversion Factor$ is a linear term that includes both the antenna gain and the $Correction Factor$.

This $Conversion Factor$ was calculated using a portion of the dataset via the least squares method, with no intercept forcing the fit through the origin. Specifically, the goal was to determine the value $x$ that minimizes:

$$mi{n}_x{\displaystyle \sum }_{i=1}^n{(S_{RSRP,i}\ast x-S_i)}^2$$

(5)

where $S_{{RSRP,i}_i}$ is the estimated total received power from the app, including both the main cell and three neighbor cells, divided by the corresponding effective aperture and $S_i$ is the power flux density from the exposimeter for measurement $i$. $n$ represents the total number of measurements from the selected portion of the dataset.

Equation (5) is minimum when:

$$x={\displaystyle \frac{{{\displaystyle \sum }}_{i=1}^n\left(S_{RSRP,i}\ast S_i\right)}{{{\displaystyle \sum }}_{i=1}^n\left(S_{RSRP,i}^2\right)}}$$

(6)

3. Results

In this section, the results of the variability of the received power within each segment are presented, areas that are more likely to have high variability are identified, and the conversion of the RSRP to power flux density in these areas is evaluated. The first subsection examines the variability results, and the second the conversion results.

3.1. Received Power Variability

The variability of the total received power was evaluated using the coefficient of variation (CV) due to significant differences in received power values between segments. For each of the five received power calculations, the standard deviation and mean of the received power table (5 × 17 × 167) were calculated across all route repetitions for each segment. The CV was then obtained by dividing the standard deviation by the mean value [22].

To understand why some areas exhibited higher variability than others, Figure 6 visualizes the segments with the top 15 highest CV values for each received power calculation for the first route.

Although there were some scattered points with high CV along the route, four distinct areas with high CV values were observed for both the exposimeter and the app received power (a–d). To identify common factors in these areas, we first examined the traffic conditions. According to the “TrafficThess” website [23], all four locations are known to experience traffic jams or are in close proximity to traffic jams during rush hours (Figure 7).

These areas are shown in Figure 8 with images obtained from Google Street View. In these images, taken randomly and not during rush hours, we observed high traffic congestion.

Although high traffic loads are common in high CV areas, other areas, such as the street between locations (a) and (b), also experienced high traffic during rush hours despite having lower CV values. This suggests that there must be other key factors as well that contribute to the high variability of received power. One such key factor is the height of the installed antennas. This is because when the line of sight (LOS) is interrupted only on high-traffic days, high CV values are expected.

This indicates that in high-traffic areas with Small Cell Antennas (SCA) installed at low heights (usually 3–10 m), rather than macro cell antennas placed on rooftops (above 10 m), a high CV value is expected due to changing LOS conditions. Observing the area of high traffic load between (a) and (b) in Figure 9, we noted that the mobile antennas are placed on the rooftop of a tall building, reducing the likelihood of LOS interruption and thus resulting in lower CV values.

Figure 10 illustrates the BSs in Thessaloniki, categorized by height, with areas of high CV values also marked. It is evident that in areas (b), (c), and (d), there is a high concentration of SCAs, leading to changes in the LOS situation during high traffic loads.

Conversely, in area (a), only macro cell antennas are present, positioned at greater heights. The high CV values in this area resulted from interruptions in LOS during heavy traffic conditions between the measurement point and the BSs located between areas (a) and (b) (as shown in Figure 9). This is due to the small angle between the road and the BSs, making the LOS easily obstructed by large vehicles.

It is worth noting that more segments had high CV values within the first route; only some with the highest values are highlighted here. These findings also applied to the second route, indicating that areas with frequent traffic jams and small cell antennas installed at low heights exhibit significant variability in received power values.

3.2. Conversion of RSRP to Electric Field Strength

3.2.1. Calculation of the Conversion Factor

The $Conversion Factor$ was calculated using Equation (6) and a specific portion of the dataset. Specifically, the selected dataset consisted of measurements taken on a day with high road traffic and in positions with constant LOS. This approach ensured that the converted electric field strength values represented the worst-case scenario in terms of exposure. The selected measurements were from a repetition of the first route corresponding to a Friday night. The $Conversion Factor$ was determined to be $5.33\ast {10}^3$.

With this $Conversion Factor$, we calculated the power flux density using only the mobile app data. This calculation involved multiplying the total RSRP value divided by the corresponding effective aperture, as described previously, for each measurement point by the constant $Conversion Factor$. This resulted in a database containing the calculated power flux density for each measurement point.

We evaluated the calculated power flux density results by comparing them with the actual power flux density obtained by the exposimeter. Specifically, we allocated each measurement point for both the calculated and obtained power flux density to the corresponding segment and calculated the mean squared error ($MSE$) for each segment based on Equation (7) [25].

$$MSE={\displaystyle \frac{1}{n}}{\displaystyle \sum }_{i=1}^n{(S_{RSRP,i}\ast x-S_i)}^2$$

(7)

where $n$ is the number of measurement points from the entire dataset within each segment for each selected route, $S_{RSRP,i}\ast x$ is a measurement point of the calculated power flux density and $S_i$ is a measurement point of the obtained power flux density within each segment. This resulted in 167 $MSE$ values for the first and 120 for the second selected route.

Figure 11 presents the $MSE$ results between the calculated and obtained power flux density in each segment.

It is evident that the mean squared error had high values in areas known to have high variability compared to other areas. It is worth noting that the chromatic scale was not linear, so the blue spots had MSE values ranging from 10⁻¹⁰ to 10⁻⁷, while all other colors correspond to values above 10⁻⁷.

3.2.2. The Impact of Variability

To evaluate whether the linear relationship between the RSRP and the power flux density was stronger in “low variability” areas, we compared the results based on variability. Each segment was identified as either a “high-” or a “low-variability” segment based on the combined CV values of the power flux density measured by the exposimeter and the total RSRP value from the app. We introduced the combined CV value for each segment as a metric for this analysis, calculated using the combined standard deviation and mean values, as described in Equation (8).

$$C{V}_{combined}={\displaystyle \frac{\sqrt{{\sigma }_{exposimeter}^2+{\sigma }_{application}^2}}{{\mu }_{exposimeter}+{\mu }_{application}}}$$

(8)

where ${\sigma }_{exposimeter}$ and ${\sigma }_{application}$ is the standard deviation of the average received power obtained by the exposimeter and the app, respectively, for each segment, while ${\mu }_{exposimeter}$ and ${\mu }_{application}$ is the corresponding mean value.

After the calculation of the combined CV for each segment, the segments were sorted, with the first half classified as low-variability segments.

Each data point from all route repetitions was then allocated to its corresponding segment. Once allocated, the data points within each segment were smoothed, and any outlier values were removed. The entire dataset was subsequently split into low- or high-variability datasets based on the corresponding segment to which each data point was assigned.

The low-variability dataset had a maximum CV value of 0.74 for the exposimeter and 1.2 for the app data. It is important to note that lower individual CV values may still occur in the high-variability dataset, as classification was based on combined CV values rather than individual ones.

Assuming the measurements were taken in the far field, the power flux density was converted to electric field strength using Equation (9) [26].

$$E\left[V/m\right]=\sqrt{S\left[W/m^2\right]\ast 120\pi }$$

(9)

The measures used to compare the calculated (with the $Conversion Factor$) and measured electric field strength values were the Mean Absolute Error ($MAE$) and the Mean Absolute Percentage Error ($MAPE$), calculated based on Equation (10) and Equation (11), accordingly [27,28], and the linear relationship between them was evaluated using Pearson’s correlation coefficient.

$$MAE={\displaystyle \frac{1}{n}}{\displaystyle \sum }_{i=1}^n\left|E_{calculated,i}-E_{measured,i}\right|$$

(10)

$$MAPE={\displaystyle \frac{1}{n}}{\displaystyle \sum }_{i=1}^n\left|{\displaystyle \frac{(E_{calculated,i}-E_{measured,i})}{E_{measured,i}}}\right|$$

(11)

where $n$ is the number of data points in each variability dataset, $E_{calculated,i}$ is the calculated electric field strength and $E_{measured,i}$ the obtained electric field strength.

To illustrate the results, we used a scatter plot where the x-axis represents the measured electric field strength values, and the y-axis represents the calculated values. Figure 12 shows the scatter plot for the variability datasets corresponding to the first route. The plot also includes the $y=x$ line and the regression line. Additionally, the correlation coefficient, $MAE$, and $MAPE$ results between the calculated and measured values are displayed.

It is evident that the calculations in the low-variability dataset were more accurate than in the high variability, having a greater correlation coefficient and lower $MAE$ and $MAPE$ values. Additionally, the angle between the regression line and the x-axis was closer to the ideal 45° in the low-variability dataset.

The results could further improve in the low-variability areas by setting a lower acceptable CV value rather than separating the dataset in half.

A more practical method of identifying low-variability areas, also considering that the exposimeter data may not be available, was by manually selecting these areas based on previously mentioned criteria, such as avoiding areas with high traffic and dense SC antenna placement.

An example of manually selected low-variability areas for both routes based on these criteria is presented in Figure 13.

The number of the selected low-variability segments for the first route was 85, which is greater than half the number of all segments, and for the second, it was 60, equal to half of the total number of segments.

The results of the manual selection of low-variability segments for the first route are presented in Figure 14.

The results in the manually selected low-variability segments were better than the previous ones, which were separated in half based on the combined CV value, with the correlation coefficient being greater and the $MSE$ and $MAPE$ being lower.

The results for the second selected route are presented in Figure 15, following the manual separation of data into low and high-variability areas, as shown in Figure 13.

The previous conclusions also applied here, with calculations in the low-variability segments being more accurate than in the high-variability segments. Additionally, the $Conversion Factor$ obtained from the first selected route applied well to the second route, with an $MAE$ smaller than 0.12 V/m in the low-variability segments and an $MAPE$ value close to 33%. It is worth noting that the data points from these two routes that were measured in the same locations were less than 20% of the entire dataset.

4. Discussion

In this study, a conversion factor was calculated and used to estimate electric field strength from RSRP values obtained by a smartphone app. The estimated electric field strength values were then compared with actual measurements taken by an exposimeter during a drive test campaign that included multiple repetitions of two different routes. The accuracy of the estimation was found to be higher in segments where the variability of the received power over repetitions was lower.

It is important to note that this variability is not directly related to the accuracy of the electric field estimation from smartphones but instead helps identify areas where phenomena likely to cause poor estimation may occur. Specifically, poor estimation became evident when changes in received power were not consistently captured across all instruments. For example, if the received power of the exposimeter increased by 1.5 times, but the corresponding power from the app increased only by 1.1 times, this discrepancy indicates a problem.

These inconsistencies may result either from differences in instrument placement or from changes in the network traffic load that affect only the exposimeter. In the first case, phenomena such as scattering or obstacles causing shadowing can create different LOS conditions across instruments. In the second case, network traffic load variations do not affect the reference signals, which are transmitted at constant power. Both cases lead to high variability in the received power over repetitions, especially in the exposimeter data.

Since this method aims to estimate the received electric field strength using solely smartphones, the areas with high variability can be identified manually without the need for data from the exposimeter, by considering the road traffic conditions and the BS height. High road traffic areas combined with a high concentration of SCA or low macro cell antenna masts in relation to the location of the measurement can be identified as high-variability areas.

The results presented in the previous section showed that in manually selected low-variability areas, the calculated electric field values were very close to the actual measured values, with $MAPE$ results close to 30% and $MAE$ values less than 0.12 V/m. It is worth noting that although in low-variability areas, most phenomena described earlier were less evident, other challenges, such as the exposimeter being partially shadowed by the driver, the smartphone’s proximity to the scooter chassis, and the imperfect isotropy of the devices can also cause discrepancies. Additionally, the lack of calibration of the three smartphones, due to the absence of proper equipment, such as a wireless network emulator, suggested that uniformity in received power among the devices cannot be assured, decreasing the accuracy of the estimation.

Various studies have utilized mobile apps to estimate exposure. In [29], the Received Signal Strength Indicator (RSSI), which is an indicator that estimates the overall signal strength while considering network traffic load, was used after being adjusted to match the dynamic range of fluctuations in the mobile network, to estimate the actual electric field strength in stationary measurements. While this method takes into consideration the mobile traffic load, it requires the antenna factor of the smartphone or a conversion factor, to convert the RSSI to electric field strength, as well as the network load conditions to adjust the dynamic range of RSSI. Our study addressed the communication load impact by providing a worst-case scenario conversion factor. This approach offered a simple and effective initial estimation of actual electric field strength, which remained accurate even under more challenging measurement conditions, such as during a driving test.

In another study [30], an app was developed to measure the received power of smartphones. Due to a lack of information regarding the smartphone’s antenna gain, the authors normalized the received power from app measurements taken on a bicycle and used the kriging interpolation method to create an exposure heat map. A similar heat map was generated from measurements made with a broadband meter over a 6 min interval at several locations within the same area. The results showed a correlation between the heat maps produced by both methods, demonstrating that the smartphone-based approach can effectively generate exposure maps for electromagnetic field assessment. This method has the advantage of using interpolation to cover a wider area with heat maps. Moreover, because the received power is obtained by a custom app, the exact parameters can be customized. However, unlike our study, a conversion factor was not calculated, meaning that actual electric field strength values could not be estimated.

This previous method has also been applied in other studies. For example, study [31] used a similar approach, utilizing RSRP combined with kriging interpolation to map exposure. The results were compared with actual electric field strength values obtained by a frequency-selective field strength meter. The advantage of this method over ours is that interpolation allows for estimating exposure in areas where no measurements were taken. However, as in other studies, the actual received power was not estimated, unlike in our research.

Other studies have employed mobile apps to estimate exposure in a more indirect manner. For instance, in [32], RSRP was used to determine and validate the path loss distribution of electromagnetic field radiation emitted by small cell antennas by comparing measured RSRP values with theoretical calculations. This path loss distribution can then be used to estimate exposure at specific locations, given knowledge of the BS power emissions. Study [33] used RSRP to compare downlink exposure between small and macro cell antennas, with total exposure comparison (both uplink and downlink) made possible by incorporating the transmission (TX) power provided by the app and the throughput. Study [34] extrapolated electromagnetic field exposure levels from RSRP using a physics-informed machine learning model.

In other studies, RSRP has been used to evaluate signal coverage [35], assess network performance [36,37], and compare 4G and 5G network performance [38].

The advantage of the method introduced in this study is that it estimates actual electric field strength using only a linear conversion factor applied to low-cost app metrics. This approach requires one smartphone per mobile operator and an app capable of recording the RSRP values from the main and at least three neighbor cells, as well as the frequency of the received signal. Although not essential, knowledge of road traffic conditions and BS placement can help assess the accuracy of the results.

Considering that the RSRP value ranged from −140 to −44 dBm [39], the estimated electric field using the 800 MHz and 2600 MHz frequency bands ranged from approximately 0.00008 to 31 V/m. This range covers extreme cases where the main cell and three neighbor cells have RSRP values of −44 dBm at the upper limit in the 2600 MHz frequency band, while at the lower limit, only the main cell had an RSRP value of −140 dBm in the 800 MHz frequency band.

Future research could involve conducting confirmation measurements in both low- and high-variability areas with extended stationary data collection and instruments located closer together. Additionally, calculating the conversion factor in various environments could generalize this method. This study could also be expanded to other networks, including 2G, 5G, Wi-Fi, and Bluetooth, to provide a more comprehensive picture of actual exposure.

5. Conclusions

This study presented an effective, low-cost method for mapping actual power flux density and, by extension, electric field strength in 4G mobile networks. The method required multiple identical smartphones—one per mobile operator—and an app capable of recording RSRP values from the main and up to three neighboring cells, along with the signal frequency. Using a worst-case conversion factor, the RSRP values were converted into power flux density and subsequently into electric field strength.

Additionally, the study introduced a technique for identifying areas with low variability in received power over time by excluding high-traffic areas with dense concentrations of small cell antennas installed at low heights. This exclusion improved the accuracy of the calculations.

While the method required a smartphone for each mobile operator, a predetermined conversion factor, and knowledge of small cell antenna density and road traffic to assess accuracy, its immediate applicability without the need for machine learning models, high computational power for computational electromagnetic simulations, or expensive frequency selective equipment offers a significant advantage. This makes it highly effective for quickly estimating actual electric field strength and mapping downlink exposure with ease. The main limitation of this method is that exposure to electric fields may be overestimated in some areas since we use the worst-case conversions factor, as mentioned above. However, even with this overestimation inherent in the electric field estimation of all areas, the method can still distinguish regions of interest for further investigation.