In this section, we describe the methodology for the FWA network characterization and present the proposed network planning algorithm. Furthermore, we present how FWA network data are obtained via simulations, and we discuss the scenarios that are considered for the network characterization and the validation of the network planning algorithm.
3.1. Network Analysis
We use graph theory to analyze FWA networks. A graph
g consists of a collection of vertices
v that represent the devices in the network and a collection of edges
e that connect two vertices [
54]. The FWA network topology is represented by a graph where devices are represented by vertices and where edges indicate that a line-of-sight path exists between two devices, i.e., no buildings are obstructing the direct path between the two devices. The vertices have attributes, e.g., indicating the device type and location of the device that it represents. The edges have the link distance as an attribute. We analyzed the following metrics for the different networks.
- Average cpe vertex degree :
The average number of links of all cpe devices;
- POP eccentricity:
Maximum of the shortest distance from the pop to all other cpe devices in the graph;
- Median link length d:
Median distance in meters between two devices;
- Average path length l:
Average hop count of the shortest path length of all cpe devices towards a pop device;
- Total network capacity:
The total capacity of the network that is available on all wireless links.
These metrics influence the performance and quality of service (QoS) of the network. With a higher average cpe vertex degree, the network density increases. With more possible links between devices, the total network capacity increases, and fewer edge devices are required to obtain a route from each cpe device towards a pop device. The average link length, measured in meters, also gives an indication of the network capacity, as wireless links with a smaller link distance have an increased signal-to-noise ratio (SNR) and more complex MCSs can be used, which results in a higher capacity. The average path length, measured in hop count, influences the network performance on a higher level. As radio propagation in free space travels at the speed of light, propagation delays are minimal, and latency and jitter are mainly caused by the MAC and network layer settings. With a higher number of hops on the path, the packet latency will also increase due to an increased processing time at the hops. Therefore, the latency in dense networks is expected to be smaller than for the field trial and early adopter scenarios.
The analysis was conducted using Python’s igraph package, and the validation of our analysis scripts was performed using a simplified small FWA network for which we can easily calculate graph statistics manually. The results of the network analysis are presented in
Section 4.1.
3.2. Network Planning
The goal of the network planning algorithm is twofold. First, the locations of edge devices need to be defined in order to get all cpe devices connected to the FWA network. Second, each cpe device needs to have a route with a sufficient capacity towards a pop device. Edge devices do not connect any customers directly and are added to the network for two reasons. First, they can be used to create a wireless mesh, i.e., connect cpe devices that can otherwise not connect to the FWA network. Second, they increase the network capacity, e.g., when the capacity of a wireless link is not sufficient for transferring the required data.
During the network planning phase, a route is defined for each cpe towards a pop, given all cpe and pop device locations, and a predefined QoS requirement, i.e., we need to allocate a certain data rate for each cpe on the wireless links that are used to reach the pop device.
3.2.1. Prerequisites
The required input data for the network planning algorithm is a database containing the LOS links between all devices, as well as the link distances in meters. The link budget parameters from
Section 2.3 are configured, including a constant antenna gain (independent of the beamforming angle), the rain rate, and possible vegetation obstructing the LOS path. No reflected paths are considered. Furthermore, each
cpe device has an associated data rate requirement. In
Section 3.3, the methodology for obtaining the database with wireless link info is described.
3.2.2. Preparation
From the database, a graph was constructed where vertices represent devices, with device type (
cpe,
pop, or
edge), data rate requirement (for
cpe devices), and physical location as attributes. In the graph, edges represent that a LOS path is present between two devices, with the distance (in meters) as an attribute. From the distance attributes, wireless link capacities were calculated based on the link budget calculation presented in
Section 2.3 and added as an additional attribute. One additional (artificial)
pop device was added, which was connected to all other
pop devices. This allows for network topologies with multiple
pop devices. The capacity of a link connecting a
pop device to the artificial parent
pop is the sum of all capacities of the child
pop device.
Feasibility checks were performed on the input data before running the network planning algorithm. A first check consists of verifying that the input graph is connected, i.e., there is a path from any vertex to any other vertex in the graph. If not, it is impossible to serve the unconnected
cpe devices as no path towards a
pop exists. A second feasibility check consists of summing the required data rates of all
cpe devices and verifying that this is smaller than the sum of capacities of the links towards the artificial parent
pop. If this is not the case, there is a capacity constraint at the
pop devices and not all
cpe devices will have their required data rate at peak moments. If one of these feasibility checks fails, a manual interaction is required, during which
edge devices are added to the network, as we will describe in the next section. This manual interaction in network planning is also required to apply with (local) regulations [
55]. If the capacity of the wireless links of the
pop is not sufficient, additional
edge devices can be added near the
pop device, or additional
pop devices can be added. This is again a manual decision based on the available infrastructure.
3.2.3. Algorithm
Algorithm 1 describes the network planning algorithm based on an input graph with vertices that represent
cpe,
edge, and
pop devices and edges that represent LOS links between devices. In the first step, the
cpe devices are sorted. For all vertices with a
cpe type attribute, the shortest paths towards the artificial
pop are calculated via Dijkstra’s algorithm [
54] using link distances as weights. The vertices are sorted in the following order:
Algorithm 1: Network routing for predefined throughput requirements. |
![Sensors 23 02280 i001]() |
We first performed routing for the vertices with the highest data rate requirement in order to prevent first optimizing routes of other vertices and having no more link capacity available to serve the high-demanding customers. For vertices with identical data rate requirements, routing was first performed for vertices with the lowest number of shortest paths towards the (parent) pop device. Lastly, for vertices with equal data rate requirements and the number of shortest paths, we performed routing first for the vertices with the highest number of hops on the shortest path.
After having a sorted list of vertices for which we need to define routing towards the pop device, we again used the shortest path algorithm to define the routing. An attribute was added to the vertex with the path that needs to be followed, and the available link capacity attribute was updated on all edges along that path, i.e., the “available” remaining data rate of the wireless link was lowered by the throughput requirement of the vertex. If the available data rate on an edge was lower than the data rate requirement of the network, we ran the shortest path algorithm again after temporarily removing the edge from the graph. If the available data rate on an edge was lower than the minimum data rate requirement of the network, we removed the edge to prevent this edge from being used for routing traffic of other vertices. Therefore, the graph gets updated and the shortest path algorithm will result in other paths compared to the first time we ran the algorithm.
If the preparation or network planning failed, e.g., due to the graph non-connectivity or due to a throughput bottleneck on one of the wireless links, a manual intervention was required, in which additional
edge devices were placed in the network. The determination of the location of the
edge devices is a manual task that is difficult to automate due to the high number of legal and practical restrictions where
edge devices might be placed, i.e., the placement of base stations is subject to regulations [
56]. Furthermore, the number of
edge devices in the network is expected to be limited. The locations of the additional
edge devices were added to the input database, as well as the LOS links, and their corresponding distances to other devices in the network.
The algorithm was implemented in Python. A presentation of the network characterization and network planning tool is provided in the
Appendix A.
3.3. Network Data Acquisition
We obtained FWA network data from simulations using the green radio access network design (GRAND) tool, which is a deployment tool for wireless radio access networks [
57]. The tool is designed for the network planning of cellular networks, i.e., to define base station locations considering mobile users, and has been adjusted to enable the simulation of FWA networks [
58].
The starting point of the tool is a map of the considered deployment environment. The map consists of three parts: the building locations, the street locations, and the area limits. All of the buildings obtain a unique building identifier. In Belgium, this map is freely available from the government or from OpenStreetMap [
59]. A configurable number of
cpe devices were added to this map, with the constraints that there can be at most one device per building and that the device is positioned at the building’s facade near the street. The Cartesian coordinates of these devices were randomly defined via a uniform distribution using the area limits from the map. For each device, new coordinates were generated as long as the coordinates did not correspond with a building that has no
cpe device already assigned. Once the coordinates corresponded to a building, they were modified so that the
cpe was located at the street-level facade of the building at a height of 4 m above the ground. The result of the
cpe device placement algorithm is that a predefined number of devices are randomly allocated to different buildings on the map, at the facade closest to the nearest street. From the GRAND tool, we obtained a database with identifiers and coordinates of all
cpe devices. In the second step, the possible radio links between different
cpe devices were defined by searching for possible LOS paths between two devices based on the location of devices and the map with building locations. From this second step, we obtained a database with all LOS links between the different
cpe devices. For each link, we obtained the identifiers of the two devices, as well as the link distance in meters.
By selecting different input maps, network data for different environments were generated. By adjusting the number of cpe devices that are present in the network, different scenarios were simulated.
3.4. Considered Scenarios
In this paper, we considered the scenarios for different FWA networks listed in
Table 3. For each scenario, we ran 50 simulations with the GRAND tool to consider the randomness of the scenarios. Each simulation resulted in a database with
cpe device locations and all of the LOS links between the devices. Three environments were compared, each having a surface area of 1 km
: downtown Ghent as an urban city, the village Leest, and a rural area in the neighborhood of Leest. The environments are shown in
Figure 2. The number of
cpe devices in the area, i.e., the internet subscribers to the network, ranged from 50 to 600. For the rural environment, only 10 and 50
cpe devices were considered, as there were fewer than 100 buildings. In the centralized
pop scenario, one
pop device was located centrally at a location where a cabinet currently exists. In the decentralized
pop scenarios, two random
cpe devices were replaced by a
pop device. The weather condition changes from sunny to heavy rain, with a rain rate of 25 mm/h, and the influence of vegetation was analyzed assuming that 10% of the link distance is covered by vegetation.
For the user requirements, the two use cases presented in
Table 4 were considered. In the first use case, it was assumed that all customers require a peak data rate of 300 Mbps. In the second use case, it was assumed that 30% of the customers subscribe to an economy plan (with a data rate of 30 Mbps), 30% of the customers subscribe to the standard plan (with a data rate of 100 Mbps), and 10% of the customers need a peak data rate of 500 Mbps.
Three different wireless communication technologies were considered: mmWave 5G at 28 GHz, IEEE Std. 802.11ad at 60 GHz, and then a future wireless communication system at 140 GHz. MmWave 5G operates from 26.5 GHz to 29.5 GHz and supports channel bandwidths up to 400 MHz. The minimum SNR ranges from 2.2 dB for binary phase shift keying (BPSK) modulation up to 25.2 dB for 256-QAM [
60]. The maximum data rates that can be achieved were calculated via (
6), with DR the data rate in Mbps, Q the modulation order that depends on the MCS, R the code rate, F the scaling factor (set to 1), N the maximum number of allocated resource blocks (set to 264, which corresponds to a subcarrier spacing of 120 kHz), T the symbol duration (calculated via
with n the numerology), and OH the overhead (set to 0.18 for frequency range 2 of the 5G specification) [
61].
In this equation, a SISO system with a single layer, i.e., a single data stream, is considered, in order to compare the results with the IEEE Std. 802.11ad technology. The selected code rates range from 0.5 (for BPSK) to 948/1024 (for 256-QAM) [
61]. The numerology for a subcarrier spacing of 120 kHz is n = 3, and the corresponding data rates range from 145 Mbps for BPSK to 2.155 Gbps for 256-QAM for a channel bandwidth of 400 MHz.
At 60 GHz, IEEE Std. 802.11ad radios were considered, with channel bandwidths of 2.16 GHz and a single carrier physical layer. With code rates ranging from 0.5 for BPSK (MCS 1) to 0.75 for 16-QAM (MCS 12) and 0.81 for QPSK (MCS 9), the data rates range from 375 Mbps for BPSK to 4.62 Gbps for 16-QAM modulation, which requires an SNR of 12.6 dB [
21]. An orthogonal frequency division multiplexing (OFDM) physical layer is also available with higher data rates, but was not considered for the remainder of this paper.
Currently, no wireless communication systems exist at 140 GHz. In order to compare future wireless communication systems at 140 GHz with existing technologies at 28 and 60 GHz, the channel capacity for the different frequencies was analyzed, which was calculated via (
7), with C the channel capacity in bits/s, B the channel bandwidth in Hz, and SNR the signal-to-noise ratio.
The channel capacity provides a theoretic upper bound of the spectral efficiency. For the comparison, the transmit power and antenna gains were kept constant, i.e., the same EIRP was considered for the three frequencies, and the difference in channel capacity was only caused by the different bandwidth and PL models for the three frequency bands. In reality, the EIRP of future wireless communication systems may differ because exposure regulations are subject to change, because the achievable transmit power of future systems is not yet determined, and because the used antenna systems may have larger antenna arrays with a high directivity.