Cost- and energy-efficient backhaul options for heterogeneous mobile network deployments

Heterogeneous network (HetNet) deployments are becoming more and more popular among mobile operators. The key rationale behind a HetNet deployment is to provide user coverage via macro base stations (BSs) and to guarantee high capacity to costumers only where needed via the ad hoc deployment of small cells, i.e., micro, pico, and/or femto BSs. HetNets have a number of advantages, which include reduced energy consumption both in indoor [1] and in outdoor [2] deployments. In fact, small cells consume less power than macro BSs. In addition, by using small cells to cater for capacity hot spots and to serve costumers residing indoors it is possible to reduce both the number of macro BSs needed in a given area and their transmission power, i.e., by serving mainly outdoors users macro BSs do not need to transmit anymore at very high power to penetrate walls to reach indoor customers.

The advantages of HetNet deployments come at the expense of a more complex and (possibly) more costly backhaul network, compared to the case of homogeneous mobile networks. For example, it has been shown that the total power consumption of a HetNet deployment to a large degree is affected not only by the presence of the backhaul [2] but also by its specific technological and architectural choices [3]. This means that backhaul networks take a share of the cost and of the energy consumption of mobile networks. How significant is this impact depends on the BS deployment scenario as well as on the technology and architectural choice for the backhaul itself. Currently backhaul networks are largely based on microwave, copper, and fiber. These technologies offer different advantages depending on the deployment scenario [4]. Fiber-based alternatives have a relatively high deployment cost (CAPEX) but provide long-term support in terms of offered capacity. A microwave-based backhaul is attractive in terms of short time to market, low investment in infrastructure, and simple deployment. Digital Subscriber Lines (DSLs) might be still appealing for backhauling indoor small cells in the presence of an existing copper infrastructure, bearing in mind their capacity limitations. All the above aspects become crucial for mobile operators the moment they need to make decisions on when and how to deploy their backhaul infrastructure. More specifically, operators need to understand in advance what will be the cost and the energy consumption impact of each one of these technologies (i.e., fiber, microwave, and copper) when having to deploy a completely new backhaul segment (i.e., Greenfield) or when having to upgrade an already existing one (i.e., Brownfield).

This paper presents a comprehensive methodology that can be used to analyze the total cost of ownership (TCO) of a number of backhaul network options based on fiber, copper, and microwave. The use case under examination considers an European urban scenario with both outdoor and indoor users. The latter are served by a layer of femto cells deployed inside the buildings where they reside, while the former are catered by macro BSs. The mobile broadband traffic levels considered in the study include values derived from the past 5  years (2010–2015) and also a short-to-medium-term future outlook, i.e., requirement envisioned for the year 2016–2025. In order to consider a wide range of options, the TCO analysis presented in the paper looks at both Greenfield and Brownfield scenarios, and it also considers the case in which a new fiber-based backhaul segment can be either deployed (i.e., fiber trenching option) or rented as needed from a third-party entity (i.e., fiber leasing option). To the best of our knowledge this work is among the first presenting such a detailed TCO analysis where both Greenfield and Brownfield deployments are considered. The results derived from the study highlight that it is not possible to find a “one size fits all” backhaul solution. Even if there are no doubts that both microwave and fiber will play a predominant role in future backhaul networks, the possible migration paths might vary. This can be attributed to a number of factors, such as the presence of an existing infrastructure, spectrum and license costs, the availability of equipment, the degree of willingness to invest in a completely new infrastructure, the technological development (possibly reducing cost and power consumption of some equipment), and the quality of service (QoS) to be provided to the end-user. For example, in a Greenfield deployment fiber-based architectures are the most energy-efficient option. However, they lead to relatively high TCO values over a period of 15 years. Leasing the fiber infrastructure might help in reducing CAPEX, but if a mobile operator does not have the possibility to lease dark fibers, the use of a microwave-based backhaul can still provide low TCO values. On the other hand a microwave-based backhaul is not the best in terms of energy efficiency. In a Brownfield deployment, to maximize energy efficiency the best option is to replace immediately all the copper and microwave infrastructure with a fiber-based one. However, this requires a large investment that might not be fully compensated over a 15-year period. In fact, the migration option that leads to the lowest TCO value is based on a gradual migration take-up, where the existing copper and microwave equipment is progressively replaced with fibers.

The first set of papers addressing the design of backhaul architectures for HetNets, i.e., [3, 5, 6], focused mostly on energy consumption evaluation.

The authors in [5] assessed the impact of the backhaul segment on the overall HetNet energy consumption under different wireless traffic levels and considering different backhaul technologies. The results presented in the paper showed that the backhaul network can play a significant role in the total power consumption of a HetNet and, more importantly, that this role becomes even more prominent with increasing values of the wireless traffic. The authors in [3] provided models to compute the power consumed by a backhaul network based on fiber (with point-to-point connections) and microwave (with point-to-point, ring, and star topologies). The results presented in the paper showed how the backhaul network is responsible for a significant part of the overall power consumption of HetNets. The authors also showed that fiber-based backhaul architectures are significantly more energy efficient than their microwave-based counterparts. Finally, the authors in [6] analyzed several backhaul options for indoor small cell based on copper, fiber, and microwave. The objective was to evaluate whether backhaul might become a bottleneck when providing mobile broadband services to indoor users. The results presented in the paper showed that the backhaul power consumption may become an issue in the case of very dense indoor small cell deployments and they highlighted how optical fiber is the most energy-efficient technology option.

While the papers just mentioned mainly focused on energy consumption modeling and estimations, there are other works that has been recently published whose aim is to assess and minimize the TCO of backhaul networks. The authors in [7] presented a technoeconomic model to calculate the TCO of HetNets that accounts for the cost of both the wireless (i.e., macro and femto BSs) and the backhaul infrastructure. The results presented in the paper show that in urban areas it is possible to reach up to 70% cost savings using indoor small cell deployments in place of traditional ones based on macro BSs. The works in [8, 9] studied both homogenous and heterogeneous wireless architectures and assessed the impact of their backhaul network on the entire TCO. The results indicate that in HetNets the backhaul network represents a considerable portion of the TCO, while in homogeneous networks, based on macro BSs only, the backhaul cost does not affect significantly the TCO. In [10] the authors evaluated a microwave-based backhaul network for small cells in terms of cost-efficiency and time required for deployment. Their results showed that point-to-point microwave is a cost-efficient technology to provide high backhaul capacity in short deployment times. Moreover, the authors in [4, 8, 11, 12] assessed the Capital Expenditure (CAPEX) and Operational Expenditure (OPEX) of mobile backhaul networks in several Greenfield deployment scenarios. More specifically, paper [8] assessed the economic impact of fiber and microwave backhaul solutions in the presence of low and high wireless capacity demands, while paper [11] compared the cost of different backhaul deployments based on fiber and copper in a scenario where fixed broadband access is already provided using fiber to the node solutions. Both papers considered HetNets based on small cell deployments. These results show that copper-based backhaul networks can cost up to 70% less than fiber-based backhaul networks. Finally, papers [4, 12] presented a TCO model for backhaul networks that includes a detailed CAPEX and OPEX breakdown. These studies concluded that there is a significant impact in terms of CAPEX during the first years of the wireless network deployments. However, these costs can lead to a good return of investment in the following years when the penetration rate of small cells increases.

To the best of our knowledge, the literature mainly considers Greenfield deployment scenarios where the existence of an infrastructure that can be potentially re-used to backhaul small cells is not accounted for. On the other hand, the majority of the households today are connected via copper cables used to provide Digital Subscriber Line (DSL) services. DSL is a very popular last-mile fixed broadband technology, and up to 2011, about three-fourth of the FTTX (fiber to the x = node/cabinet/building) solutions relied on DSL to connect to end-users [13]. In such a scenario it might become economically convenient to re-use as much as possible legacy DSL solutions for backhauling femto cells, especially the ones deployed indoors. We refer to these cases as Brownfield backhaul deployments.

The contribution of this paper is twofold. It combines the work of [6] and [4] by analyzing both the power consumption and the TCO value of different backhaul architectures in a Greenfield scenario. This allows new important considerations, with respect to the ones already discussed in [6] and [4]. Second, this paper presents and compares the power consumption and the TCO value of different migration options for backhaul networks in a number of Brownfield deployment scenarios. This is a new and important contribution because it shows how different migration paths can lead to very different power consumption levels and TCO values. These new results can help mobile network operators to understand: (i) which is the best way to deploy/upgrade their current backhaul network infrastructure, and (ii) how to choose (when needed) the most energy- and cost-efficient migration paths.

This section provides an overview of the proposed methodology for assessing the TCO of mobile backhaul networks. The methodology is divided into four phases (Fig. 1).

The first phase is the traffic demand estimation. It is used to compute the expected wireless traffic requirements in the deployment area for a given year. The inputs are the population density and other relevant data describing the network and service usage such as: the number of active mobile subscribers, the user profile (e.g., heavy or ordinary), the daily traffic variation, and the penetration rate of different mobile terminals. Using the long-term large-scale traffic model proposed in [6] this phase returns the value of the area traffic demand.

The second phase is the wireless deployment. It is used to determine the number of BSs to be deployed in the area. In this case the inputs are the area traffic demand from the previous step in addition to other parameters related to the wireless network technologies to be used, i.e., the capacity of all BS types and the penetration rate of the small cells [6]. Accordingly, it is possible to obtain the number of macro BSs and small cells required to cover the area and to satisfy the wireless traffic requirements of a given year.

The third phase can be of two types. The first one called backhaul deployment refers to a Greenfield scenario where no pre-existing backhaul infrastructure is present in the area. The second one called backhaul migration refers to a Brownfield scenario where there is a pre-existing infrastructure that can be re-used for backhauling small cells. For the backhaul deployment case the inputs are: the output of the wireless network deployment step, the peak capacity of both macro BSs and small cells, the backhaul architecture, and the transmission/switching capacity of the backhaul equipment. For the backhaul migration case there is an additional input, i.e., the description of the technical details of the already existing infrastructure that can be used for backhauling small cells. For both options the output is the number of network devices (e.g., microwave antennas, optical transceivers, and DSL modems) that will have to be deployed in the area.

We provide a detailed description of the TCO assessment methodology introduced in the previous section, and we present the analytic models adopted in our study. First we describe in details the traffic estimation model, then we present the wireless and backhaul network deployment strategy, and we conclude with the detailed description of the TCO evaluation model. The notations used in this section are summarized in Table 1.

In this section the methodology and power consumption models discussed so far are used to evaluate the TCO of the considered backhaul architectures in the context of an European urban scenario. It is assumed that the area under consideration is \(A = 100\)  [km\(^2\)] with buildings uniformly distributed measuring \(80\times 80\) m and separated from each other by 20 m. The total number of buildings in the area is set to be 10,000, each one with 5 floors and 2 apartments per floor (i.e., the total number of apartments \(N_\mathrm{ap}\) corresponds to 100,000). In addition, it is assumed that the population density is \(\rho = 3000\) users\(/\mathrm{km}^2\) and that only one operator serves the area. Moreover, the distances between the CO, local exchanges, and end-users are calculated using a Manhattan street model [16]. In order to compute the traffic demand \(\tau \), we assume that 16% of the subscribers are active during peak hours (i.e., \(\alpha _\mathrm{max} = 16\%\)) and that the capacity requirement of an ordinary user is 1/8 of the one of a heavy user [14]. Moreover, we assume that a laptop user generates two and eight times more data traffic than a tablet and a smartphone user, respectively. More details about the assumption used to compute the traffic demand can be found in [14, 17].

The study presented in the paper considers a 15-year time period that goes from year 2010 until year 2025. For each year the traffic demand estimation phase described in Sect. 4.1 is used. It is assumed that in the year 2011 the area is served by a homogeneous wireless network based only on macro BSs, i.e., \(\eta =0\). Starting from the year 2010, the wireless deployment evolves toward HetNets (i.e., with the gradual introduction of femto BSs) and there is a linear increase in the femto penetration rate of 5% every year. Regarding the backhaul network, it is assumed that in 2010 the macro BSs are backhauled using microwave point-to-point links.

In the Greenfield scenario, starting from the year 2011, the operator deploys a new backhaul infrastructure to support HetNets. The new infrastructure can either be based on microwave or fiber (it is assumed that the operator will not deploy a new infrastructure based on copper because of its limited capability of supporting high capacity over long distances). As a result, in the Greenfield scenario, the operator (starting from year 2011) can choose either one of backhaul A1, A3, A4 and A5, described in Sect. 4.3.

On the other hand, in the Brownfield scenario the operator is able to leverage on the already existing fiber and copper infrastructure, i.e., the operator owns a backhaul infrastructure like the one described by A2. As a result, in the year 2011 the operator faces two options for network upgrade. The first choice is to continue deploying a microwave-based backhaul (i.e., as the one described by A1), while the second option is to migrate to new network architectures that reuse legacy ducts and trench infrastructure. This migration can be executed according to different models. In the following, we describe some of these alternatives.

Gradual migration take-up (M1) with this migration model, the operator decides to exploit the copper infrastructure to backhaul the indoor femto BSs. As a result, in the year 2011 the operator selects A2. Afterward, the mobile operator gradually migrates from A2 to A5. We assume that the operator starts replacing the copper infrastructure with a PON-based backhaul 3 years before the copper infrastructure is expected to be unable to support the required traffic levels. This occurs when the traffic demand for fixed broadband access networks exceeds 100 [Mbps] per household and is calculated using the traffic forecast model shown in [18]. In addition, the operator also gradually replaces the microwave-based backhaul used for the macro BSs with PONs. This process starts 3 years before the time when the area traffic demand is expected to exceed 1000 [Mbps/km\(^2\)]. During the migration process in the first year only 20% of the old technology is replaced, in the second year 50%, in the third year 10%, and by the fourth year 100% of the old technology has been replaced with the new one.

No take-up (M5) this case is similar to M1, where the operator chooses A2 and then migrates to A5. However, in this case the migration is not carried out gradually. In fact, the operator keeps the VDSL2 copper infrastructure until its capacity reaches the exhaustion point (i.e., fixed broadband access networks capacity exceeds 100 [Mbps]) and only at that point the operator replaces it with a PON-based backhaul. Similarly, the replacement of the microwave-based backhaul with PONs is only carried out after the wireless traffic demand exceeds 1000 [Mbps/km\(^2\)].

No take-up (M6) this migration model is almost the same as M2 with the only difference that the migration from copper-based to PON-based backhaul for the femto cells does not proceed gradually. On the other hand the macro BSs are always backhauled via microwave.

The value of the TCO is evaluated for all the migration options described above to find the most economically attractive. The power consumption and cost values shown in Table 2 are used in our calculations.

With respect to the microwave backhaul technology, it is assumed that the microwave antenna power consumption corresponds to \(P_{\mathrm{low}{\text {-}}\mathrm{c}}\) when the traffic at the antenna is lower than 500  [Mbps], and to \(P_{\mathrm{high}{\text {-}}\mathrm{c}}\) when the traffic is higher. Fast Ethernet connections operating at 100 [Mbps] are used inside the buildings to connect the femto BSs to the GES (in A1 and A3). In addition, the femto BSs are distributed uniformly among the buildings and in the area. As a result, the number of GESs (\(N_\mathrm{GES}\)) is equal to the number of buildings if \(\eta >0\). We assume that the power consumption of a GES linearly scales with the number of ports that are used for backhauling the femto BSs, i.e., \(P_\mathrm{GES} = \dfrac{N_\mathrm{femto}}{N_\mathrm{b} n_\mathrm{ports}^\mathrm{GES}}P_\mathrm{GES}^\mathrm{max}, \forall \eta > 0.1\), where \(P_\mathrm{GES}^\mathrm{max}\), and \(n_\mathrm{ports}^{GES}\) are the maximum power of a GES and the total number of ports of the GES, respectively. Moreover, in the Greenfield scenario, two possible values are considered for the maximum distance between the femto BSs and the local exchanges, i.e., 300 [m] and 1 [km]. On the other hand, in the Brownfield scenario, we consider only a maximum distance of 300 [m] between the femto BSs and the local exchanges. This is because VDSL2 technologies are not able to cope with distances longer than 300 [m], i.e., A2 cannot be considered as a migration option.

For this study, a technician salary is assumed to be 72 [USD/h] for the first year and the energy cost is set to 0.15 [USD/kWh]. After the first year we assume a yearly increase based on the geometric progression \(c_n = c_1 q^{n-1}\), where \(\textit{c}_n\) represents either a technician salary or the energy cost in the year n and q = 1,03 is the increase ratio [4]. Unless stated differently, a fixed yearly depreciation of 5% is also specified for the network equipment cost.

This section presents the numerical results obtained for the case study described before. It is divided into two parts, i.e., Greenfield and Brownfield considerations.

This article proposed a total cost of ownership (TCO) assessment methodology that can be used to evaluate the TCO performance of backhaul networks. The methodology includes four steps: (i) wireless traffic demand estimation, (ii) wireless network deployment, (iii) backhaul network deployment (for Greenfield scenarios) or backhaul network migration (for Brownfield scenarios), and (iv) TCO evaluation. The TCO assessment methodology was employed in a Greenfield scenario to compare the TCO values of four backhaul network architectures and in a Brownfield scenario to compare the TCO values of six backhaul network migration options. In both cases a wireless network composed of indoor femto and outdoor macro BSs was considered.

The analysis of the Greenfield scenario proves that backhaul architectures based on PONs are very energy efficient, but they also require a large investment (due to trenching or leasing of the fiber infrastructure) that leads to high TCO values. This is particularly true when the distance between femto BSs and local exchanges is long (e.g., greater or equal to 1 [km]). Our results prove that using a fiber to the building (FTTB) infrastructure for backhauling femto BSs and a microwave links for backhauling macro BSs becomes the most cost-efficient solution when it is possible to lease dark fiber already deployed in the area.

The analysis of the Brownfield scenario shows that from an energy consumption perspective the best solution is to migrate toward PONs as early as possible (i.e., migration options M3 and M4). However, this also results in very high TCO values. A more cost-efficient solution is to exploit the existing copper infrastructure for backhauling the indoor femto cells and gradually replace it starting from a few years before the copper capacity exhaustion point (i.e., migration options M1 and M2).

Finally, we plan to extend the study presented in the paper and to apply the proposed assessment methodology to evaluate the impact that spatial and temporal traffic variations have on the TCO value of a mobile network (i.e., radio + transport). In this regard we also plan also to investigate potential dynamic energy management schemes specifically tailored for the proposed architectures.