Sage Journals: Discover world-class research

Abstract

In the field of 5G mobile communication networks, coverage performance is an important indicator that can affect the user’s terminal experience. The current method for coverage detection is drive test, which is based on mobile terminals. Although the application of the technology in practice has matured, the implementation of drive test requires additional time and manpower. However, in the test phase of a 5G mobile communication network, the detection should be on-the-spot, real time, and repeatable. To overcome these challenges, this article proposes 5G network wireless coverage area detection technology that is based on wireless sensor networks. First, sensor nodes are deployed to collect the received signal strength. Second, the algorithm performs Gaussian filtering on collected data. Then, an interpolation estimation algorithm is adopted to estimate the interpolation of the area. Finally, the data collected by sensor nodes and estimated by interpolation points are integrated to generate the effective coverage area status. To overcome the human subjectivity of the traditional model fitting when performing the variation function fitting in the interpolation estimation, a support vector regression algorithm is employed. The simulation results indicate that the algorithm can rapidly and correctly detect the coverage area of a 5G mobile communication network.

Keywords

5G mobile communication network wireless sensor network coverage detection interpolation estimation support vector regression

Introduction

5G is a new generation of mobile communication systems for mobile communication needs after 2020. According to the development of mobile communications, 5G will have high spectral efficiency and energy efficiency. In the transmission rate and resource utilization, 5G mobile communication will increase by an order of magnitude or higher compared with 4G,¹ and the wireless coverage performance, transmission delay, system security, and user experience will be significantly improved. 5G mobile communication and other wireless mobile communication technology will be closely integrated to form a new generation of ubiquitous mobile information networks. A 5G mobile communication network has become an inevitable trend and research hotspot in the communication field.

Numerous countries are conducting research, development, and testing to significantly improve the network’s wireless coverage performance, transmission latency, system security, and user experience.² At the beginning of 2013, the EU launched the METIS (mobile and wireless communications enablers for the 2020 information society) project for 5G research and development in the 7th Framework Program, which was shared by 29 participants.³ Simultaneously, Korea and China also set up a 5G technical forum and the IMT-2020 (5G) promotion group. The China 863 program also launched a major 5G project that consisted of two research projects in June 2013 and March 2014.⁴ On 8 May 2014, the Japanese telecommunications operator NTT DoCoMo officially announced it will work with Ericsson, Nokia, Samsung and six other manufacturers to test the high-speed 5G network, whose transmission speed is expected to increase to 10 Gbps. Expanded outdoor testing, which is expected in 2015, is expected to begin operation in 2020.

5G is the main direction of the development of next-generation mobile communication technology, and it is an important part of the next generation of information infrastructure. Compared with 4G, 5G will not only enhance the network experience of users but also satisfy the future needs of all interconnections. For the user experience, 5G will offer a higher rate and a wider bandwidth. The speed of 5G is expected to be approximately 10 times higher than that of 4G to satisfy consumer demand for virtual reality, ultra-high-definition video, and other higher network experience needs. From the perspective of industry applications, 5G has higher reliability and lower latency and can satisfy the smart manufacturing, automatic driving, and other industry application needs. From the development trend, 5G remains in the research stage of the technical standards; thus, 4G will remain dominant in the next few years. However, 5G is expected to be officially commercial in 2020.

To achieve these business support capabilities, 5G will present a new breakthrough in wireless transmission technology and network technology.⁵ Key technologies of 5G mobile communication are primarily reflected in ultra-high performance wireless transmission technology and high-density wireless network technology. In wireless transmission technology, 5G will introduce a technology that can tap the potential of spectrum efficiency, such as advanced multi-access technology, multi-antenna technology, coding and modulation technology, and new waveform design technology. In wireless networks, 5G will adopt a more flexible and intelligent network architecture and networking technology, such as a unified self-organizing network (SON) and heterogeneous ultra-intensive deployment.

As shown in Figure 1, specific key technologies of a 5G mobile communication network are large-scale MIMO technology, multi-carrier technology, full-duplex technology, ultra-dense heterogeneous network technology, SON technology, soft-defined networking, and a content distribution network.^6–15

Figure 1.

Key technologies of a 5G mobile communication network.

With the development of these new technologies, the 5G mobile communication network is in a critical stage of research and testing. The network performance indicators that are primarily considered during the test include the user experience rate, connection density, end-to-end delay, mobility, flow density, user peak rate, and network coverage performance, which are listed in Table 1.

Table 1.

Performance indicators of 5G mobile communication networks.

Performance indicators	Definition
User experience rate (bps)	Lowest transmission rate available to users in a real network environment
Connection density (/km²)	Sum of the online devices supported per unit area
End-to-end delay (ms)	Time at which packets are transmitted from the source node to the destination node
Mobility (km/h)	Maximum relative speed between the recipient and the sender
Flow density (bps/km²)	Total data flow per unit area
User peak rate (bps)	Maximum transfer rate available to a single user
Coverage performance	Effective coverage of network signals within the target area

A network coverage performance evaluation must be based on wireless coverage area detection technology. Therefore, we propose a new wireless coverage area detection technology that is based on a wireless sensor network (WSN). The main contributions of our work are as follows: first, we evenly deploy distributed sensor nodes to collect the received signal strength (RSSI) of a 5G base station. Second, we perform Gaussian filtering on the collected data to ensure the accuracy of the data. Third, we divide the target area and select the interpolation points by the Delaunay triangulation technique. Then, we propose an improved Kriging interpolation algorithm to estimate the received signal strengths of interpolation points. Finally, by integrating the data collected by sensor nodes and estimated by interpolation points, we generate the effective coverage area status of a 5G mobile communication network.

To improve the Kriging interpolation algorithm, we primarily improved the curve fitting of the variation function in the algorithm. In the traditional Kriging interpolation algorithm, the fitting method of the variation function curve is human subjective. To overcome this shortcoming, we employ the support vector regression (SVR) to complete the curve fitting.

The remainder of this article is organized as follows: we place our work within the context of related research in section “Related work.” The basic theoretical knowledge of the article is elaborated in section “Basic knowledge.” The algorithm architecture and the steps of the algorithm are presented in section “Coverage detection technology of 5G mobile communication network.” Detailed simulations of the proposed algorithm are presented and the signal contour line of the 5G mobile communication network is generated in section “Simulation experiments analysis.” We present the conclusions of our work in section “Conclusion.”

Related work

The wireless coverage performance of a 5G mobile communication network is an important indicator that must be considered in the test and trial process. The mobile network optimization test, which is represented by the network coverage, has become a daily function for operators. Wireless coverage area detection technology has formed a set of standard processes, and the current technology requires a number of well-trained optimization engineers to complete these processes. However, part of the work of engineers will likely be replaced with new technology and new methods from current technological developments. This trend can be attributed to intelligence and integration. Many types of tools are available for network coverage detection, and a number of the commonly employed detection tools are presented in Table 2.^16–18

Table 2.

Network coverage detection tools and functional classification.

Detection tools	Features
Drive test	Record received downlink level, call quality, signaling, and location information using dedicated test equipment
Signal analysis software	Count and analyze the signaling part of the test data
Geographic information system	Show the network coverage performance in a real map via a geographic information system platform
Traffic statistics software	Automatically calculate the defined performance metrics and provide test analyses in graphical form

A mature and extensive method of coverage area detection in wireless communication networks is the drive test.^19,20 Because of variations in the beam direction of smart antennas, the drive test is not appropriate for a 5G mobile communication network. Based on the repeated measurements that require wireless network coverage, the intensive time and labor consumption of the drive test in the test and trial phase of a 5G mobile communication network, a high-precision, all-round, all-process, all-weather, and all-day technology scheme is necessary to achieve coverage area detection for a 5G mobile communication network. According to these demands, a feasible technology scheme that can realize the coverage detection of a 5G mobile communication network via WSN is proposed in this article. The principle of the scheme is to collect data for an RSSI by the distributed sensor nodes deployed in the communication network and integrate the data to generate the effective coverage area of the network. To obtain the effective wireless coverage status of the entire test area of the 5G mobile communication network with the RSSI data collected by the finite sensor nodes, the blind region outside the sensor nodes must be estimated when performing data processing. Therefore, estimating the RSSI is an urgent problem that must be resolved in the technology scheme of the paper.

The two main RSSI estimation methods are signal propagation model estimation and interpolation estimation.²¹ The former involves estimations based on the RSSIs of sensor nodes combined with an appropriate propagation attenuation model of the signal. Although the model has low computational complexity, it also has low accuracy; moreover, it cannot accurately match the actual geographical environment of the target area. A mature model that can be employed for a 5G mobile communication network has not been developed. Based on the characteristic attributes of the sensor nodes in the neighborhood, the latter’s accuracy of interpolating point attribute estimations is relatively high.²² Common interpolation estimation methods include classification interpolation, inverse distance weight (IDW) interpolation, and Kriging interpolation. Although IDW improves the accuracy of the algorithm by estimating the RSSIs of the interpolation points, the interpolation accuracy requires further improvement because only the distance between the interpolation point and sensor nodes is considered in Mrinmoy et al.²³ The Kriging interpolation employed in Evangelos et al.²⁴ is an unbiased RSSI estimation method of the interpolation point based on the spherical model fitted by variation function values among RSSIs collected by sensor nodes. However, the choice of variation function model is subjective, which causes a lack of credibility in certain cases.

To overcome the shortcomings of the algorithms applied in wireless coverage area detection, this article proposes an improved Kriging interpolation algorithm. The algorithm fits the variation function curve via SVR to overcome the human subjective factors of traditional methods and improve the reliability and accuracy of coverage detection.

Basic knowledge

Kriging interpolation

Kriging interpolation is a generalized grid-based statistical method for geological surveys.²⁵ In mathematics, this method has been proven to be linear unbiased.²⁶ The principle is to estimate the attributes of the interpolation point by the characteristic attributes of the sensor nodes in the neighborhood of the interpolation point. In this article, the characteristic attribute is the RSSI.

Assuming that the RSSI of the interpolation point is $Z (x_{0})$ and the RSSIs after preprocessing the m sensor nodes in the neighborhood are $Z (x_{i}) (i = 1, 2, \dots, m)$ , then $Z (x_{0})$ can be estimated by the weighted sum of $Z (x_{i})$

$Z (x_{0}) = \sum_{i = 1}^{m} λ_{i} Z (x_{i})$ (1)

where $λ_{i}$ represents the weights of $Z (x_{i})$ and $\sum_{i = 1}^{m} λ_{i} = 1$ is derived from the eigenvalue condition. The key to estimating $Z (x_{0})$ is to calculate $λ_{i}$ by equation (1).

According to the second order of $Z (x_{i})$ , the following expressions can be obtained

${\begin{matrix} E [Z (x_{i}) - Z (x_{j})] = 0 \\ Var [Z (x_{i}) - Z (x_{j})] = E {[Z (x_{i}) - Z (x_{j})]^{2}} \end{matrix}$ (2)

Assume that $Z^{*} (x_{0})$ is the unbiased estimate of $Z (x_{0})$ , where the estimated variance of the interpolation point $x_{0}$ is the smallest value

$Va r_{min} (x_{0}) = Var [Z (x_{0}) - Z^{*} (x_{0})] = E {[Z (x_{0}) - Z^{*} (x_{0})]^{2}}$ (3)

The Lagrange multiplier μ is introduced, and it can be expressed as follows

$\frac{\partial}{\partial λ_{i}} E {{[Z (x_{0}) - Z^{*} (x_{0})]}^{2} - 2 μ \sum_{i = 1}^{m} λ_{i}} = 0$ (4)

where $i = 1, 2, \dots, m$ . The following equations are obtained by calculating and simplifying equation (4)

${\begin{matrix} \sum_{i = 1}^{m} λ_{i} γ (x_{i} - x_{j}) + μ = γ (x_{0} - x_{j}) \\ \sum_{i = 1}^{m} λ_{i} = 1 j = 1, 2, \dots, m \end{matrix}$ (5)

where $γ (x_{i} - x_{j})$ represents the variation function value between $x_{i}$ and $x_{j}$ . As the core of Kriging interpolation, the variation function is proposed to solve the spatial characteristics of interpolation points. With the variation function, the characteristic attribute of the interpolation point can be deduced by the rule in which the characteristic attributes of the sensor nodes change with the separation distance among the sensor nodes. The variation function value can be calculated by the following equation

$γ (h_{j}) = \frac{1}{2 {N_{h}}_{j}} \sum_{i = 1}^{{N_{h}}_{j}} {[Z (x_{i} + h_{j}) - Z (x_{i})]}^{2}$ (6)

where $h_{j}$ represents the separation distance of a pair of sensor nodes in the target area and ${N_{h}}_{j}$ represents the number of sensor node pairs in the distance $h_{j}$ . The variation function values of different separation distances can be calculated by equation (6), and these variation function values can be fitted with the variation function curve $γ (h)$ . According to the curve, the variation function values between the attribute of the interpolation point and the attributes of the sensor nodes in the neighborhood can be solved. The weight $λ_{i}$ and the Lagrange multiplier μ can be obtained by substituting them into the equation.

In the process of fitting the variation function curve, a variation function model is usually employed. Then, variance function curve fitting is performed by the least squares method.²⁴ Common classical variation function models include the linear model, Gaussian model, spherical model, and exponential model. When solving practical problems, the model matching method cannot exclude subjective randomness. Moreover, the model may not match the data set to be fitted. To break through the limitations of the model matching method, this article starts with the SVR to fit the variation function curve.

SVR

A support vector machine (SVM) has a better generalization ability for solving the classification problem and can obtain better classification results with limited sample data.²⁷ SVR can be obtained by introducing the $ε$ insensitive loss function in the SVM. The basic idea of SVR is to find an optimal classification surface to minimize the error of the training sample from the classification surface.²⁸

The function curve is predominantly nonlinear when a function is fitted, whereas linear regression fitting is applied in a SVR function fitting. Thus, the sample data must be mapped to a high-dimensional feature space by the nonlinear mapping function for linear regressions.

For the training set $T = {(x_{1}, y_{1}), (x_{2}, y_{2}), \dots, (x_{k}, y_{k})} \in (R^{n} \times R)$ with k samples, $x_{l}$ ( $l \in (1, k)$ ) represents the input vector of the lth training sample and $y_{l}$ is its corresponding output value. Assume that the linear regression function in the high-dimensional feature space is expressed as follows

$f (x) = w Φ (x) + b$ (7)

where $Φ (x)$ is a nonlinear mapping function. The following equation is obtained by introducing an $ε$ linear insensitive loss function

$L (f (x), y, ε) = max (0, | y - f (x) | - ε)$ (8)

where $f (x)$ is the predicted value of equation (7) and $y$ is the true value of the sample.

By introducing the slack variables $ξ_{l} \in (ξ_{1}, ξ_{2}, \dots, ξ_{k})^{T}$ and $ξ_{l}^{*} \in (ξ_{1}^{*}, ξ_{2}^{*}, \dots, ξ_{k}^{*})^{T}$ , the linear SVR problem can be expressed as follows

$min_{w, b, ξ, ξ^{*}} {\frac{1}{2} {‖ w ‖}^{2} + C \sum_{l = 1}^{k} (ξ_{l} + ξ_{l}^{*})}$ (9)

$s . t . y_{l} - [w Φ (x) + b] \leq ε + ξ_{l}, l = 1, 2, \dots, k$ (10)

$w Φ (x) + b - y_{l} \leq ε + ξ_{l}^{*}, l = 1, 2, \dots, k$ (11)

$ξ_{l} \geq 0, ξ_{l}^{*} \geq 0$ (12)

where C is the penalty parameter. Substituting w and b obtained by solving equations (9) to (12) into equation (7) generates a linear regression function in high-dimensional feature space. For convenience, the Lagrange multipliers $α_{l}$ and $α_{l}^{*}$ can be introduced to obtain the dual form of the original equations (9) to (12), which can be expressed as follows

$min_{α, α^{*}} {\frac{1}{2} \sum_{l = 1}^{k} \sum_{j = 1}^{k} (α_{l}^{*} - α_{l}) (α_{j}^{*} - α_{j}) K (x_{l}, x_{j}) + ε \sum_{l = 1}^{k} (α_{l}^{*} + α_{l}) - \sum_{l = 1}^{k} (α_{l} - α_{l}^{*}) y_{l}}$ (13)

$s . t . \sum_{l = 1}^{k} (α_{l}^{*} - α_{l}) = 0$ (14)

$0 \leq α_{l} \leq C, 0 \leq α_{l}^{*} \leq C$ (15)

where $K (x_{l}, x_{j}) = Φ (x_{l}) Φ (x_{j})$ is referred to as the kernel function of SVR. The solution of equations (13) to (15) is a convex quadratic optimization problem.²⁹ Assuming that $\bar{α} = [{\bar{α}}_{1}, {\bar{α}}_{2}, \dots, {\bar{α}}_{k}]$ and ${\bar{α}}^{*} = [{\bar{α}}_{1}^{*}, {\bar{α}}_{2}^{*}, \dots, {\bar{α}}_{k}^{*}]$ are the optimal solutions for the problem, the parameters w and b can be expressed as follows

$w^{*} = \sum_{l = 1}^{k} ({\bar{α}}_{l}^{*} - {\bar{α}}_{l}) Φ (x_{l})$ (16)

$\begin{matrix} b^{*} = \frac{1}{N_{NSV}} {\sum_{{\bar{α}}_{l} \in (0, C)} [y_{l} - \sum_{x_{i} \in SV} ({\bar{α}}_{i} - {\bar{α}}_{i}^{*}) K (x_{i}, x_{l}) - ε] \\ + \sum_{{\bar{α}}_{l}^{*} \in (0, C)} [y_{l} - \sum_{x_{j} \in SV} ({\bar{α}}_{j} - {\bar{α}}_{j}^{*}) K (x_{j}, x_{l}) + ε]} \end{matrix}$

where $N_{NSV}$ is the number of support vectors (SVs). For the sensor node $x_{i}$ , if the corresponding $(α_{i} - α_{i}^{*})$ is not 0, then the sensor node is a SV. Substituting equations (16) and (17) into equation (8) generates the linear regression function $f (x)$

$\begin{array}{l} f (x) = w^{*} Φ (x) + b^{*} \\ = \sum_{l = 1}^{k} ({\bar{α}}_{l}^{*} - {\bar{α}}_{l}) Φ (x_{l}) Φ (x) + b^{*} = \sum_{l = 1}^{k} ({\bar{α}}_{l}^{*} - {\bar{α}}_{l}) K (x_{l}, x) + b^{*} \end{array}$ (17)

Coverage detection technology of 5G mobile communication networks

Based on a WSN, this article proposes a new coverage detection technology of a 5G mobile communication network. Because the sensor nodes can collect information in real time, the purpose of network coverage detection can be achieved anytime and anywhere, and it can satisfy the needs of network multiple coverage detection in the test and trial phase of a 5G mobile communication network. The proposed algorithm simultaneously performs estimations based on RSSIs received by sensor nodes; thus, the requirements of the base station antenna are low. For a conventional omnidirectional antenna or a smart antenna with a narrow beam of adjustable direction, the algorithm has good applicability.

To ensure the completeness of 5G wireless coverage area detection, the coverage detection technology proposed in this article applies Kriging interpolation to realize the network coverage detection in the target area. The algorithm architecture is shown in Figure 2.

Figure 2.

Architecture of wireless coverage detection technology.

The algorithm is primarily composed of three modules: data collection and processing, Kriging interpolation, and network coverage performance generation. Data collection and processing primarily includes performing the target area division, interpolation points selection, RSSI data collection, and data preprocessing. Kriging interpolation primarily includes calculating the variation function values of the sample data set and performing the variation function curve fitting and the interpolation points RSSIs estimation. Network coverage performance generation primarily includes combining the data that were collected by the sensor nodes and estimated by the interpolation points to generate the signal contour line of a 5G mobile communication network. The signal contour line is similar to the contour line in geography. This article defines this line as a closed curve of each point on a topographic map, where the signal strengths are equal. The map can intuitively display the coverage performance of a 5G mobile communication network. The main algorithm modules shown in Figure 2 will be described.

Data collection and processing

RSSI data collection and data preprocessing

In RSSI data collection, if collected only once, then the acquisition results may present errors with the actual values. To ensure the accuracy of the RSSI collected by the sensor nodes, data must be collected many times and the geometric means must be employed as the collected value. Because each collected datum is independent, the RSSIs that are collected many times are subject to a Gaussian distribution.³⁰ Thus, we employ a Gaussian filter for data preprocessing. The small probability interference can be processed by Gaussian filtering to obtain an accurate and stable RSSI. The specific method is as follows.

For multiple collections of RSSIs in each sensor node (RSSI_m is employed to represent the result of the mth collection), the density function of the result x is expressed in equation (19), where n is the number of collections, $μ$ is the mean of the data set, and $σ^{2}$ is the variance

$f (x) = \frac{1}{\sqrt{2 π} σ} \exp [- \frac{{(x - μ)}^{2}}{2 σ^{2}}]$ (19)

$σ^{2} = \frac{1}{n - 1} \sum_{m = 1}^{n} {(RSS I_{m} - μ)}^{2}$ (20)

$μ = \frac{1}{n} \sum_{m = 1}^{n} RSS I_{m}$ (21)

According to the probability density function calculated by equation (19), the corresponding RSSIs in the range of the high probability occurrence area $f (x) \geq 0.6$ (empirical value) are selected. Their geometric mean is used as the RSSI of the sensor node after preprocessing.

Target area division and interpolation point selection

To facilitate the selection of interpolation points and their neighborhoods in the Kriging interpolation, the target area must be divided. Based on the Delaunay triangulation scheme, the area can be divided into several closed triangular meshes according to the sensor nodes positions. The sensor nodes are located at the vertexes of the triangle after division.³¹ The selected interpolation points are located in the triangular mesh. The sensor nodes in the neighborhoods of the interpolation point constitute the three vertexes of the triangle. Because of the space limitations of this article, only a brief explanation is provided.

Improved Kriging interpolation

According to the principle of SVR in section “SVR,” the kernel function $K (x_{l}, x)$ and the penalty parameter C are determined in SVR. Commonly employed kernel functions are the sigmoid kernel function, polynomial kernel function, and radial basis function (RBF) kernel function. The RBF kernel function in equation (22) is superior when the variation function curve is fitted³²

$K (x_{i}, x_{j}) = \exp (- σ {‖ x_{i} - x_{j} ‖}^{2}), σ > 0$ (22)

where $σ$ is the kernel parameter of the smooth kernel function. Therefore, the parameters $σ$ and C must be determined in the SVR fitting of the variation function curve. In this article, we apply the selection method of the RBF kernel function parameter proposed in Ming-shan et al.³³ A method for obtaining parameters with high precision is proposed via a uniform experiment design. Because of the space limitations of this article, only a brief explanation is provided.

The steps of the improved Kriging interpolation are as follows. The algorithm flow is shown in Figure 3.

Step 1: Calculate the distance $h_{j}$ of all pairs of sample data and the corresponding variance function values $γ (h_{j})$ according to equation (6). Then, generate the data set $[h_{j}, γ (h_{j})]$ .

Step 2: Randomly select most of the data in the data set to generate the training set T, $T = {[h_{1}, γ (h_{1})], [h_{2}, γ (h_{2})], \dots, [h_{k}, γ (h_{k})]}$ ; the remainder of the data constitute the test set.

Step 3: Use the SVR training $T$ and fit the variation function curve $γ (h)$ according to the training results.

Step 4: Use the test set generated in step 2 to evaluate the performance of the variance function curve $γ (h)$ and then output the results if the expected requirement is reached; otherwise, modify the parameters and return to step 3 to re-fit.

Step 5: Obtain the variation function values between the interpolation point and the sensor nodes in its neighborhood according to the $γ (h)$ that satisfies the expected requirement. Substitute the values into equation (5) to obtain the weights $λ_{i}$ .

Step 6: Substitute the weights $λ_{i}$ obtained in step 5 into equation (1) to estimate the RSSI of the interpolation points.

Figure 3.

Architecture of the improved Kriging interpolation.

Simulation experiments analysis

Construction of simulation environment

To verify the performance of the 5G wireless coverage detection technology proposed in this article, we choose an 8000 m × 8000 m hilly area as the experimental simulation environment. Assume that the 5G mobile communication network in this area contains four base stations. The special simulation software Atoll developed by France FORSK company is used to draw the signal coverage of the base stations as shown in Figure 4. The parameter settings are shown in Table 3. In the simulation, the signal coverage of the base stations is assumed to be unknown.

Figure 4.

Simulation environment.

Table 3.

Simulation parameter settings.

Simulation parameters	Configuration value
Target area size	8000 m × 8000 m
Simulation scene	Hilly area
Number of sensor nodes	40
Network model	Three sector model
Number of users in each cell	100
Path loss model	Okumura-Hata
Standard deviation of shadow fading	10 dB
Carrier frequency	3.4 GHz

Assume that 40 sensor nodes are uniformly deployed in the area. To compare and analyze the accuracy of the interpolation, 34 sensor nodes are randomly selected as sampling points, whereas the remaining 6 nodes are verification points. To verify the performance of the algorithm, this article simulates three experiments from three aspects: the performance analysis of the variation function fitting, the performance analysis of the improved Kriging interpolation, and the influence of the number of sensor nodes and BS on the performance of the algorithm. The signal contour line in the target area is generated based on these aspects.

Performance analysis of variation function fitting

As the core step of this algorithm, the accuracy of the variation function fitting determines the accuracy of the Kriging interpolation. The variation function data set is generated based on the RSSIs of the selected 34 sensor nodes. Eighty percentage of the set is randomly selected as a training set, and then SVR is used to fit the training set in MATLAB. The results are shown in Figure 5.

Figure 5.

Variation function SVR fitting: (a) fitting curve of the training set and (b) results of the test set.

Figure 5(a) shows the results of the training set fitting, and Figure 5(b) shows the test results of the test set. To compare the performance of the algorithm, the spherical model, the Gaussian model, and the exponential model are used to fit the training set. The comparison results are shown in Figure 6.

Figure 6.

Comparison of different models for variation function fitting.

The root mean square error (RMSE) of each model is separately calculated. The Gaussian model is 27.256, the exponential model is 29.257, the spherical model is 29.245, and the SVR is 24.345. The comparison shows that the SVR fitting presents the lowest RMSE and the highest precision.

To effectively measure the fitting accuracy of each model, 5000 independent random drawings are performed to the 34 sensor nodes. A considerable number of fitting experiments were performed. The average RMSE and average decision coefficients (R²) of each model fit in multiple experiments are listed in Table 4.

Table 4.

Performance comparison of different model fittings.

Performance parameters	SVR	Exponential model	Spherical model	Gaussian model
RMSE	23.469	29.457	28.478	26.684
R²	0.6975	0.5255	0.5790	0.6947

SVR: support vector regression; RMSE: root mean square error.

Table 4 shows that the proposed algorithm fitting curve presents a lower RMSE than the other models and a higher R². Therefore, the variation function curve fitted by the algorithm has a high degree of coincidence.

Performance analysis of the improved Kriging interpolation

To verify the accuracy of the improved Kriging interpolation proposed in this article, the proposed algorithm, the ordinary Kriging interpolation, and the IDW interpolation are interpolated using the six verification points selected in section “Construction of simulation environment” as examples. The comparison results of the interpolations and the true values are shown in Figure 7.

Figure 7.

Comparison of the interpolation algorithms.

By separately calculating the RMSE of the three algorithms, the ordinary Kriging interpolation is 4.8778, the IDW interpolation is 7.7424, and the proposed algorithm is 2.9033. The comparison shows that the proposed algorithm has the lowest RMSE and the highest precision of the interpolation.

Influence of the number of sensor nodes and BS on the performance of the algorithm

First, to verify the effect of the number of sensor nodes on the performance of the algorithm, we select certain nodes from the 40 sensor nodes to participate in the operation of the algorithm. Second, we calculate the RMSE of the generated BS. The signal contour lines and results after multiple simulations are shown in Figure 8.

Figure 8.

Influence of the number of sensor nodes.

As shown in Figure 8, as the number of sensor nodes decreases, the RMSE exhibits an increasing trend, which becomes increasingly distinct. Thus, the results indicate that the number of sensor nodes will have an impact on the performance of the algorithm and few sensor nodes will produce a larger algorithm error. Second, to verify the effect of the number of BSs on the performance of the algorithm, we establish a different number of BSs for the experiment, and the results are shown in Table 5.

Table 5.

Influence of the number of BSs.

Number of BSs	RMSE
1	9.2353
2	13.2734
3	11.6134
4	10.2936
5	12.1346

RMSE: root mean square error.

As shown in Table 5, as the number of BS changes, the RMSE values undergo minimal change. Thus, the number of BSs has a minimal effect on the performance of the proposed algorithm.

Signal contour line generation of the 5G mobile communication network

The interpolations of the target area are estimated using the three algorithms compared in section “Performance analysis of the improved Kriging interpolation.” Then, the signal contour line based on these three algorithms is generated as shown in Figure 9.

Figure 9.

Signal contour line: (a) IDW interpolation results, (b) ordinary Kriging interpolation results, and (c) proposed algorithm interpolation results.

One hundred positions are randomly selected from the target area, and then the RMSE of the estimation results interpolated by these three algorithms at the 100 positions is calculated. To ensure the credibility of the accuracy comparison, and the 5000 sets of experiments randomly selected in section “Performance analysis of variation function fitting” are used as examples, the three algorithms are applied for the 100 positions in the 5000 experiments and the mean RMSE is calculated. The mean RMSE of the IDW interpolation is 25.9589; the ordinary Kriging interpolation is 14.6908; and the proposed algorithm is 10.2936. The improved Kriging interpolation algorithm proposed in this article produces the most accurate signal contour line, which can better reflect the true coverage of the 5G mobile communication network.

Conclusion

To obtain the wireless coverage area status of the 5G mobile communication network in this test and trial phase, the article proposes a new technology for wireless coverage area detection based on a WSN. The signal contour line generated by the proposed technology can visually display the signal coverage of the 5G mobile communication network and has a certain reference value for measuring the network coverage performance. The validity of the signal contour line is verified by simulation experiments. Although the WSN is deployed, the proposed algorithm will generate an all-weather and all-round signal contour that can satisfy the special requirements of being on-the-spot, real time, and repeatable in the test and trial phase of the 5G network.

Footnotes

Handling Editor: Francisco Vasques

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research,authorship,and/or publication of this article: This work was partially supported by the National Natural Science Foundation of China (grant no. 61273302).

References

Prados-Garzon

Ramos-Munoz

Ameigeiras

et al . Modeling and dimensioning of a virtualized MME for 5G mobile networks. IEEE T Veh Technol 2017; 66(5): 4383–4395.

Zhang

Imran

et al . Energy efficient hybrid satellite terrestrial 5G networks with software defined features. J Commun Netw 2017; 19(2): 147–161.

METIS. Mobile and wireless communications enablers for the 2020 information society (EU 7th framework programme project), http://www.metis2020.com

Wen

Zhu

PY.

5G: a technology vision. Huawei, 4 November 2013, http://huawei.com/en/about-huawei/publications/winwin-magazine/hw-329304.htm

Wang

Haider

Gao

et al . Cellular architecture and key technologies for 5G wireless communication networks. IEEE Commun Mag 2014; 52: 122–130.

Marzetta

. Noncooperative cellular wireless with unlimited numbers of base station antennas. IEEE T Wirel Commun 2010; 9: 1436–1449.

Gao

You

Jiang

et al . MIMO-GMC wireless transmission technology for beyond 3G mobile communications. Chin J Commun 2004; 32: 105–108.

Cheng

Zhang

. Optimal dynamic power control for full-duplex bidirectional-channel based wireless networks. In: Proceedings of IEEE international conference on computer communications (INFOCOM), Turin, 14–19 April 2013, pp.3120–3128. New York: IEEE.

Ahmed

Eltawil

Sabharwal

. Rate gain region and design tradeoffs for full-duplex wireless communications. IEEE T Wirel Commun 2013; 12: 3556–3565.

10.

Aggarwal

Duarte

Sabbarwal

et al . Full- or half-duplex? A capacity analysis with bounded radio resources. In: Proceedings of IEEE information theory workshop, Lausanne, 3–7 September 2012, pp.207–211. New York: IEEE.

11.

Lim

Kim

et al . Full duplexity in beamforming-based multi-hop relay networks. IEEE J Sel Area Comm 2012; 30: 1554–1565.

12.

Osseiran

Braun

Hidekazu

et al . The foundation of the mobile and wireless communications system for 2020 and beyond: challenges, enablers and technology solutions. In: Proceedings of IEEE vehicular technology conference (VTC Spring), Dresden, 2–5 June 2013, pp.1–5. New York: IEEE.

13.

NGMN. NGMN recommendation on SON and O&M requirements, 2008, http://www.mendeley.com/research/ngmn-recommendation-son-om-requirements/

14.

McKeown

Anderson

Balakrishnan

et al . OpenFlow: enabling innovation in campus networks. SIGCOMM Comput Commun Rev 2008; 38: 69–74.

15.

Peng

. CDN: Content distribution network, http://arxiv.org/abs/cs/0411069v1

16.

Raymond

Cytil

. Downlink scheduling optimization in GSM networks. IEEE Commun Lett 2004; 8(10): 611–616.

17.

Yan

. Cellular mobile communication system network optimization. Telecommun Sci 1999; 15(2): 6–9.

18.

Zhang

Peng

. Design and optimization of GSM mobile communication network planning. Beijing, China: Posts & Telecom Press, 2003, pp.108–116.

19.

Mohammad

Walid

Tapan

. Further validation of an electromagnetic macro model for analysis of propagation path loss in cellular networks using measured driving-test data. IEEE Antenn Propag M 2014; 56(4): 108–129.

20.

Sun

Feng

Jin

et al . Handoff algorithm and PoSC model under high-speed railway environment. J Zhejiang Univ 2012; 46(9): 1553–1559.

21.

Vincent

Alexandre

Yves

. Theoretical bit error floor analysis of 16-QAM OFDM signal with channel estimation using polynomial interpolation. IET Signal Process 2016; 10(3): 254–265.

22.

Arshad

Augustine

Bamidele

et al . Location prediction optimisation in WSNs using Kriging interpolation. IET Wirel Sens Syst 2016; 6(3): 74–81.

23.

Mrinmoy

Indrajit

Mainak

et al . A node deployment mechanism accounting into received signal strength and frequency diversity for a wireless sensor network. In: 2016 IEEE conference on sensors, Orlando, FL, 30 October–3November 2016, pp.1–3. New York: IEEE.

24.

Evangelos

Dimitris

Adrian

et al . Multiterminal source coding with copula regression for wireless sensor networks gathering diverse data. IEEE Sens J 2017; 17(1): 139–150.

25.

Koya

Takeo

. Kriging-based interference power constraint: Integrated design of the radio environment map and transmission power. IEEE T Cogn Commun Netw 2017; 3(1): 13–25.

26.

Song

Mihai

Jan

. Adaptive weighted expected improvement with rewards approach in Kriging assisted electromagnetic design. IEEE T Magn 2013; 49(5): 2057–2060.

27.

Zhou

. SVM soft margin classifiers: linear programming versus quadratic programming. Neural Comput 2005; 17(5): 1160–1187.

28.

Liu

Jiang

. Adaptive support vector regression method for structural system reliability analysis. J Zhejiang Univ 2015; 49(9): 1692–1699.

29.

Norikazu

Jun

Tetsuo

. Global convergence of SMO algorithm for support vector regression. IEEE T Neural Networ 2008; 19(6): 971–982.

30.

Jiang

Liu

et al . A probabilistic radio map construction scheme for crowdsourcing-based fingerprinting localization. IEEE Sens J 2016; 16(10): 3764–3774.

31.

Kurt

Milos

. Wireless sensor networks—node localization for various industry problems. IEEE T Ind Inform 2015; 11(3): 752–762.

32.

Shahaboddin

Dalibor

Hossein

et al . Sensor data fusion by support vector regression methodology—a comparative study. IEEE Sens J 2015; 15(2): 850–854.

33.

M-S

Wang

Z-M

Zhang

. New method for selecting parameters of support vector machine regression based on uniform design. J Syst Simul 2008; 20(8): 2195–2199.

Research on wireless coverage area detection technology for 5G mobile communication networks

Abstract

Keywords

Introduction

Related work

Basic knowledge

Kriging interpolation

SVR

Coverage detection technology of 5G mobile communication networks

Data collection and processing

RSSI data collection and data preprocessing

Target area division and interpolation point selection

Improved Kriging interpolation

Simulation experiments analysis

Construction of simulation environment

Performance analysis of variation function fitting

Performance analysis of the improved Kriging interpolation

Influence of the number of sensor nodes and BS on the performance of the algorithm

Signal contour line generation of the 5G mobile communication network

Conclusion

Footnotes

Declaration of conflicting interests

Funding

References