Delivering cost-efficient 5G high-speed internet in rural areas

The expectation that internet connectivity will be available anywhere at any time is one of the key drivers of demand for mobile broadband (MBB). Thus far, however, communication service providers (CSPs) have focused on delivering MBB services in densely populated areas that offer a relatively fast return on investment, and more sparsely populated rural areas have often been left behind.

If the disparity in service between densely populated and rural areas persists, regulators may feel compelled to impose requirements on CSPs to invest in improvements to rural coverage that cannot be motivated by a business case. While non-terrestrial networks, where satellites deliver blanket coverage over large areas, have been proposed as a solution to serve rural areas, the served capacity per area falls an order of magnitude short of what terrestrial networks are able to achieve [1]. It is therefore reasonable to assume that terrestrial networks will remain important cornerstones of the rural digitalization journey in the years ahead.

Ericsson’s approach to deploying terrestrial networks in rural areas

Ericsson has developed a cost-effective approach to deploying terrestrial networks in rural areas with population densities below 50 people per square kilometer that involves mounting advanced 5G radios on high towers to expand the service area per site [2]. The rationale behind this approach is that the total cost of ownership is closely tied to the number of deployed sites. By increasing the service area, the number of subscribers per site increases, which will boost return on investment even when the higher cost per site for high-tower construction is taken into account.

It is possible to enhance the capacity and user experience in rural deployments by upgrading existing sites, such as replacing existing equipment with larger antennas to improve coverage and/or adding new frequency bands to increase capacity. This approach may be adequate if the existing network already provides some rudimentary base coverage. The alternative is a greenfield deployment, where capable radios are mounted on newly built high towers that at least partially replace the legacy deployment, but with a lower site count.

To evaluate the potential benefits of different high-tower deployment options and multiband radio configurations, we have run a variety of radio network simulations. The advanced site-specific radio propagation model we employed considers the terrain profile as well as land use, including both vegetation and buildings, which was derived from digital maps.

Understanding the requirements on coverage and capacity

The key objective of our research has been to meet the demand for MBB connectivity that is expected in 2028 in sparsely populated areas characterized by a population density (ρ) of up to 50 people/km2, with as few sites as possible. To achieve satisfactory network connectivity the following requirements must be met:

• User experience: The ability to deliver 20Mbps on the downlink (DL) and 1Mbps on the uplink (UL) for at least 95 percent of all users [3]
• Capacity: The ability to serve an average traffic demand (v) of 40GB/month/subscriber on the DL (v_DL= 40) and 4GB/month/subscriber on the UL (v_UL= 4).

The capacity (c) requirement per subscriber translates to the requirement on the network capacity per unit area, with the placeholder x denoting the traffic and capacity demands for the DL and UL respectively:

In the case of a wireless network that can serve traffic density on the DL and UL of up to c ̂_DL and c ̂_UL , the upper boundary for the population density that this network would be able to support can be calculated using the following equation:

Assuming a subscriber share (S) for the considered CSP of S_CSP = 30% and time (T) of 500 active hours per month, the population density is upper bounded by ρ ≤ min (1.876 c ̂_DL , 18.76 c ̂_UL). For satellite systems with 30MHz bandwidth, with performance requirements of c ̂_DL = 8 and c ̂_UL = 1.5kbps/km² taken from the ITU-R (International Telecommunication Union – Radiocommunication Sector) Report M.2514-0 [4], the population density may not exceed 0.0150 per square kilometer. This is more than two orders of magnitude below the population density target considered in our research, highlighting the importance of terrestrial networks in rural areas.

Basic mobile connectivity requires control channel coverage on both the UL and DL on at least one frequency band. Given the imbalance in transmit power between the base station (BS) radio with 4W/MHz and mobile terminal with a maximum transmit power of 200mW, it is apparent that coverage will always be UL limited. This DL-UL imbalance is exacerbated for higher frequencies that rely on time division duplex (TDD) with a TDD ratio of 4:1 in favor of the DL, leaving only 20 percent of the transmit slots for UL transmissions.

Impact of terrain and foliage on radio propagation: A site-specific modeling approach

Terrain profile and foliage have a major effect on radio propagation in rural scenarios. The site-specific propagation model [5] we used in our research complements stochastic models by allowing for more detailed and realistic deployment scenarios. Digital maps provide terrain heights and information about land use, including the location of trees and buildings. Propagation predictions of the pathloss involve representing the surface profile comprising terrain, buildings and vegetation in the vertical plane with a set of absorbing half-screens [6]. Radio waves may partly diffract at the edges of obstacles or traverse through them, giving rise to multiple transmission paths that impinge at the receiver. We obtained the resulting pathloss at the receiver by adding multiple paths that take different routes above the surface profile and through foliage.

Figure 1 illustrates the surface profile on the vertical plane between a transmitter (Tx) and receiver (Rx) about 8km apart. The transmitter is mounted on a tower of height h. The direct line-of-sight (LoS) propagation path is shown for three different tower heights. At a tower height of h=60m the direct path is blocked by a hill at a distance of 3.5km−4km from the transmitter. Increasing the tower height to 120m grossly reduces the shadowing caused by the hill, and at h=240m the LoS path passes the hill unobstructed. This illustrates how high towers can help to enhance the chance of a direct unobstructed transmission path and may help to cover larger areas.

Figure 1: Site-specific propagation model with on-surface and through-foliage propagation

An individual propagation path with carrier frequency f_c of length d along the surface profile is composed of the free space loss and the excess pathloss due to diffractions L_D, as well as foliage loss L_F and building loss L_B, with c = 299,792km/s denoting the speed of light:

Diffractions are caused by obstacles that block the direct LoS path, as indicated in Figure 1. The diffraction loss L_D is modeled by knife-edge diffraction over intermediate halfscreens of the surface profile [5]. We added a fixed building loss of L_B = 15dB for users located indoors. The foliage loss, caused by propagation paths that radiate through vegetation, is approximated by a loss-per-meter model [7] in which dF accounts for the foliage depth in meters:

It is sometimes argued that employing a sufficiently large antenna is enough to compensate for losses at growing carrier frequencies; a claim that may hold true for free space propagation. However, other propagation effects also tend to grow with the carrier frequency. As propagation paths that diffract around and/or traverse through obstacles further penalize higher carrier frequencies, low band maintains the lowest pathloss toward users in the least favorable channel conditions.

Optimizing rural network coverage: Antenna configurations and ISD strategies

We considered a CSP with a subscriber share of S_CSP = 30% and the frequency assets listed in Figure 2. Base stations are mounted on towers at height h, each equipped with three sectors per site. We assume a BS transmit power of 4W/MHz and a mobile Tx power of 0.2W for frequency division duplex (FDD) bands and 0.4W for TDD bands.

The BS antenna configuration is denoted by (V_E×H_E)×(V_A×H_A). The vertical and horizontal number of cross-polarized antenna elements, V_E and H_E, constitutes one subarray, which is characterized by a fixed relative phase offset between adjacent antenna elements. This means that the antenna can be tuned to point at a certain direction but cannot be adjusted during operation. Several subarrays may be assembled to constitute the antenna array, with V_A and H_A denoting for the vertical and horizontal number of subarrays. The relative phases between adjacent subarrays may be arbitrarily adjusted, allowing for beamforming over time and frequency [8]. The number of Tx and Rx chains are given by NTR = 2V_A H_A and the total number of antenna elements amount to N_AE= V_EH_EV_AH_A.

The antenna configurations on mid-band FDD and TDD in Figure 2 are in line with antennas deployed in urban and suburban environments. For low band, on the other hand, a larger antenna is deemed appropriate, having a two-column array with 10 elements each, which increases the antenna area while reducing the vertical half-power beamwidth. Since the flat terrain implies relatively small angular variations of a dominant propagation path, most of the energy of impinging signals will fall within the main lobe of the antenna, so that most users will benefit from the larger antenna gain at low band. Moreover, the large form factor of the low-band antenna with a length of about 2.6m may be less of an issue.

Performance evaluation in two environments

To assess the effects of vegetation on rural coverage, we selected two different rural environments in the midwestern region of the United States for our performance evaluations. The first is located in northern Iowa between the cities of Des Moines and Minneapolis, around the East Fork and the West Fork of the Des Moines River. This area is characterized by a flat, open topology with only 3 percent foliage. The second environment is in northwestern Wisconsin, northeast of Minneapolis and south of Lake Superior. While this second environment has similarly flat terrain, 70 percent of its area is covered by forest. The average population density in both environments is well below 50 people per square kilometer. However, outside the main population hubs, the population density is much lower, often below 10 per square kilometer.

In our simulation, we placed BS towers on a hexagonal grid with a given inter-site distance (ISD) in both environments. We determined the exact site locations by identifying nearby hilltops and using that information to maximize site coverage.

We tested different realizations of the mobile network with ISDs ranging from 4km to 30km as well as BS tower heights of 60m, 120m and 240m. The main objective was to serve the expected traffic demand while maintaining the desired user experience with as few sites as possible. This translates into finding the largest ISD that is able to satisfy the generated traffic that corresponds to a given population density.

Mobile traffic is generated according to user location. Since subscribers are mainly indoors, we assumed that 80 percent of the traffic offered is generated by indoor users. The remaining 20 percent is distributed uniformly over the entire outdoor area. A major part of the mobile traffic is thus clustered around populated places, while coverage is to be maintained over the entire area.

Uplink coverage

Maintaining UL coverage on low-band FDD is of critical importance. Figure 3 shows the UL cell-edge user throughput measured at the fifth percentile for various networks with different ISDs. The left side of Figure 3 shows that it is possible to maintain a cell-edge throughput of around 1Mbps for ISDs up to 30km for tower heights down to 60m in flat, open environments. Increasing the tower height improves UL throughput, due to reduced probability of shadowing. Interestingly, increasing the antenna area on low band from 4TR to 8TR has an effect that is similar to doubling the tower height. An 8TR antenna may be assembled by mounting two 4TR side-by-side, giving an array of dimension (10×1)×(1×4).

Figure 3: UL cell-edge user throughput as a function of the ISD in Iowa (at left) and Wisconsin (at right)

A comparison of the cell-edge throughput for a given ISD on the left (Iowa) and right (Wisconsin) sides of Figure 3 illustrates the detrimental effect of foliage on UL coverage. To compensate for the reduced coverage due to foliage, both a larger low-band antenna area (8TR) and higher towers are necessary to reach a network ISD of 20km.

Network capacity

As the ISD increases, the area covered by each site increases accordingly. Given the average area (A) per site on a hexagonal deployment (A= 2√3 (ISD/2)²) at an ISD of 20km, one site covers an area close to 350km². Hence, capacity may become an issue, even in sparsely populated areas. Once base coverage is established, capacity may be added in the form of mid-band spectrum.

Figure 4 shows the maximum supported population density according to the equation

against the network ISD for various antenna and multiband configurations. If the required user experience of 20Mbps on the DL and 1Mbps on the UL is not met for at least 95 percent of all users, the population density is set to zero.

We used the methodology described previously to derive the maximum population density for given traffic demands per subscriber. A tower height of 60m was deemed sufficient for the flat, open Iowa scenario (left), whereas a tower height of 120m was chosen for the Wisconsin scenario with foliage (right). Towers as high as 240m could be a viable option to approach an ISD of 20km in areas with substantial vegetation, but we have not considered that in this work due to the high construction cost.

Given the baseline 4TR low-band antenna, the UL cell-edge throughput requirement limits the ISD to 20km in Iowa and short of 10km in Wisconsin. Doubling the low-band antenna area by deploying an 8TR antenna mitigates the UL throughput limitation (indicated by black triangles) and makes it possible to extend the ISD to 30km in Iowa and close to 15km in Wisconsin. Given the larger low-band antenna, the capacity is increasingly DL limited (indicated by white upside-down triangles), owing to the tenfold increase in traffic demand per subscriber. Adding additional frequency bands on mid-band FDD and TDD makes it possible to tailor the site equipment to a given population density, starting with a single low-band radio for extremely sparsely populated areas and adding additional bands as population density increases.

Interestingly, in flat and open scenarios with ISDs exceeding 20km, such as the environment in Iowa, a triple-band configuration is needed to serve a population density of 10 per square kilometer. Mid-band TDD is the most effective option to boost performance on the DL, due to the DL-heavy TDD pattern, while mid-band FDD is more suitable for increasing capacity on the UL. The value of Massive MIMO (multiple-input, multiple-output) on mid-band TDD is shown by comparing the capacity with a configuration where the 64TR Massive MIMO array is replaced by an 8TR classic radio.

In summary, improving UL coverage by employing a larger low-band antenna and/or increasing the tower height enables the coverage of larger areas, which in turn requires more spectrum to serve the increased amount of traffic per site. Our research indicates that a large 8TR low-band antenna provides little added value on the DL, especially for multiband configurations. The deployment of a 4T8R low-band antenna configuration with four Tx and eight Rx chains is a more viable, cost-efficient option. A 4T8R antenna can be realized by mounting two ordinary 4TR antennas next to each other and using only one of them for DL transmission, thereby making it possible to relax the requirements on antenna calibration.

Conclusion

Topography and foliage have a major impact on the attainable coverage and capacity of wireless networks in rural areas. Even small hills can create shadow regions with severe negative effects on coverage. The uplink (UL) is always the bottleneck in terms of coverage, due to the transmit power imbalance between base station and mobile transmit power. Our research indicates that the most effective way to improve UL coverage is to increase the antenna area on low band, which opens up for networks with a larger inter-site distance (ISD). The second most efficient way to improve UL coverage is to build high cell towers. The operation of high-power user equipment on low-band frequency division duplex (FDD) has the potential to further enhance the ISD. As the coverage area increases, capacity demand per sector will also increase.

Due to the larger traffic demand, capacity is often limited on the downlink (DL). Mid-band FDD and time division duplex (TDD) are effective means to utilize capacity demand by offloading the traffic of users with good link quality, thereby leaving precious low-band resources for users with poor channel conditions on the cell edge. Mid-band TDD is particularly attractive, due to the large amount of available bandwidth and the DL-heavy TDD pattern.

Acknowledgements

The authors are grateful to Henrik Asplund, Martin Johansson, Jason Chen and Anders Ericsson for their help in extending the simulation tool to model areas exceeding 100x100 km², which was necessary to evaluate network performance with ISD exceeding 20km. They would also like to thank Eva Englund for her help in extracting the key findings presented in this article.