# Infrastructure planning and topology optimization for reliable mobile big data transmission under cloud radio access networks

- Xiang Hou
^{1}, - Bin Lin
^{1}, - Rongxi He
^{1}Email author and - Xudong Wang
^{1}

**2016**:119

https://doi.org/10.1186/s13638-016-0625-1

© Hou et al. 2016

**Received: **31 December 2015

**Accepted: **24 April 2016

**Published: **3 May 2016

## Abstract

With the development of user-centered and environment sensing technology of 5G, large capacity and ubiquitous coverage and massive data collection and processing will bring new challenges in wireless networks. The cloud radio access network (C-RAN) has been envisioned to provide a new wireless architecture for reliable transmission of mobile big data. In this paper, we focus on network planning deployment issue based on the optical mixed diet (OMD) technology. Specifically, the ring and spur topology optimization (RSTO) problem under the C-RAN architecture is investigated. The RSTO problem is formulated as a generic integer linear program (ILP) which can optimally (i) minimize the network deploying cost; (ii) identify the locations of Remote Radio Units (RRUs) and optical add-drop multiplexers (OADMs); (iii) identify the association relations between RRUs and OADMs; and (iv) satisfy the mobile coverage requirements so as to allow the mobile big data to be transmitted through the RRUs. We propose a new heuristic algorithm based on C-RAN architecture. Numerical results validate the ILP formulation and show the performance benefits of the proposed algorithm in terms of efficiency and effectiveness against Gurobi, which is an ILP solver. In numerical studies, we also demonstrate the performance benefits of the incorporation of CoMP technology in terms of total deployment cost reduction.

## Keywords

## 1 Introduction

Nowadays, the information society has stepped into a new era, in which the mobile Internet, cloud computing, big data, and Internet of Things are the leading technologies [1]. The big data has main features: volume, variety, velocity, and value [2]. Thus, future wireless networks are required to provide the ability to store and process the large data.

Wireless access network is facing new challenges, such as network deployment, resource management, mobile data security, and network consumption. To deal with these challenges, a new type the radio access network architecture named cloud radio access network (C-RAN) has been proposed in [3].

The C-RAN has four “C” characteristics, i.e., clean, centralized processing, collaborative radio, and real-time cloud radio access network. C-RAN is viewed as an architecture evolution based on distributed base stations (BSs). The C-RAN architecture breaks down the traditional BS into a base band unit (BBU) and Remote Radio Units (RRUs). The BBU is a digital unit that implements the MAC, PHY, and antenna array system functionality. The BBU is often located in a remote site such as a central office (CO) and thus the energy, real estate, and security costs are minimized. The RRU can be mounted either outdoor or indoor such as on poles, sides of buildings, or anywhere power and a broadband connection exist, making installation less costly and easier. The RRU is typically connected to the BBU using fibers. C-RAN adopts CoMP technology to reduce interference, lower power consumption, and improve spectral efficiency in [4] and [5].

Due to the requirement of a high-volume, variety, and non-uniform distribution of the big data in time domains and space domains, wireless access networks must have high reliability. For example, many applications of online banking, mobile payment, and remote business are generated on smart phones. Thus, the networks are required to transmit data much more reliably.

In order to increase the utilization of the optical fiber, optical mixed diet (OMD) mode is proposed. In the OMD mode, the BBU pool supplies a pair of specific wavelengths to transmit the signal to the OADM. The OMD integrates the function of WDM. The BBU and a certain RRU are using a specific wavelength module and a plurality of RRU can share a pair of optical fibers; thus, the OMD mode can increase the utilization of the optical fiber.

In this paper, we focus on the ring and spur topology optimization (RSTO) problem under the C-RAN architecture. A generic integer linear programming (ILP) model is proposed to capture the nature of the RSTO problem. In this paper, we focus on network planning deployment issue based on the OMD technology.

The RSTO problem under the C-RAN architecture is investigated. The contributions of this paper are summarized as follows. (1) The RSTO problem is formulated as a generic ILP model which can optimally:(i) minimize the network deploying cost; (ii) identify the locations of RRUs and OADMs; (iii) identify the association relations between RRUs and OADMs; and (iv) satisfy the mobile coverage requirements so as to allow the mobile user access through the RRUs. (2) We also propose a new heuristic algorithm under C-RAN architecture. The algorithm can deploy the C-RAN architecture in small- and large-scale scenarios. (3) Compared with the result of Gurobi in different scenarios, simulation results verify that the algorithm is correct and effective.

The remainder of the paper is organized as follows. In Section 2, we review the previous related work. In Section 3, the problem formulation is given, including the network model, given inputs, variables, constraints, and objective of the RSTO problem. The heuristic algorithm is presented in Section 4. Computational results and case studies are presented in Section 5. Finally, we conclude the paper in Section 6.

## 2 Related work

We review the related studies on network planning and placement. In [7], three types of architectures, namely optical fiber direct-drive, optical transmission network (OTN)/wavelength division multiplexing (WDM), and unified passive optical network (Uni-PON) are proposed. These three models have been further elaborated in [8–10], respectively.

Our previous work has studied on the network planning and placement, which have direct influence on the future network deployment and long-term performance [11–15]. The problem formulated to determine the locations of communication network equipment is discussed as follows. Site planning of relay station GA-based solution in green wireless access network is proposed in [11]. An ILP model is formulated to maximize the throughput of an LTE network in [12]. Cascaded splitter placement and layout in long-reach PON are discussed in [13]. A dimensioning and site-planning optimization framework was proposed for integrated PON and wireless cooperative networks in [14]. An ILP model for the cascaded LRPON topology and network equipment placement, and network dimensioning are proposed in [15].

Other related works have studied on topology and placement optimization, such as [16] and [17]. A model for combined virtual core network function placement and topology optimization is proposed in [16]. In [17], the author presents some optimization problems in WMNs and different heuristic methods such as local search, genetic algorithms, and tabu search for solving them near-optimally. Relayed-Based Heterogeneous Wireless Access Networks (RHWAN) is envisioned as the promising network architecture proposed in [18]. A hybrid wireless mesh network infrastructure which connects the smart meters of each consumer with the data aggregation points (DAP) is proposed in [19]. An optimal deployment model on Universal Data Aggregation Point (UDAP) is proposed in [20].

In this research, we choose the ring and spur topology optimization based on the OMD technology under the C-RAN architecture. The fiber can bidirectionally transmit the signal between the BBU and the OADMs; thus, it has high network survivability. Wireless access network is demanded to provide higher reliability in the big data era. To the best of our knowledge, there are few studies on the deployment and optimization for C-RAN infrastructure under the ring and spur topology, which is just the focus of our research.

## 3 Network model and problem formulation

### 3.1 Network model

The C-RAN network model consists of four network entities: BBU, OADM, RRU, and mobile stations (MSs), as shown Fig. 1. The OADMs and the RRUs are eligible to be deployed at certain outdoor potential sites (PSs) where uninterrupted power supply can be provided [5].

- Term 1:
Inflow: the directed path which represents the flow from BBU to OADM

_{ k }. - Term 2:
Outflow: the directed path which represents the flow from OADM

_{k}to BBU. - Term 3:
Inflow path: the path starts from BBU and ends at OADM

_{ k }. - Term 4:
Outflow path: the path starts from OADM

_{ k }and ends at the BBU.

Without loss of generality, we assume that the MSs are homogeneously distributed in the area of interest (AOI), which yields a uniform distribution on the traffic demand in the AOI. We define that a site is covered under the C-RAN architecture if the received radio signal strength at the site is above a given threshold level (receiver sensitivity). Small-scale fading is not explicitly included in the system model since a long-term planning and design is targeted. To test the mobile coverage, we define test points (TPs) within the AOI. The TPs are also used for testing the received power intensity for MSs from the associated RRU. In Fig. 1, the locations of TPs and PSs for both OADM and RRU are also illustrated.

### 3.2 Problem statement

The RSTO problem can be stated as follows. Let *G* = (*Ω* : *E*), where *Ω* is the set of the nodes and *E* is the set of the direct edge/links. The BBU is denoted as the node *B. Ω* is partitioned into four parts, i.e., the set of TP, the set of OADM, the set of RRU, and the set of BBU, which are denoted as *Ω*
_{TP}, *Ω*
_{OADM}, *Ω*
_{RRU}, and *Ω*
_{B}, i.e., *Ω* = *Ω*
_{TP} ∪ *Ω*
_{RRU} ∪ *Ω*
_{OADM} ∪ *Ω*
_{B}.

#### 3.2.1 Given

- (1)
The set of fixed TPs, which are the customer premises

- (2)
The set of PSs for deploying RRUs and OADMs

- (3)
The location of the BBU

- (4)
The cost per unit length of fiber (

*$*/km)*C*^{ f }, including fiber purchase and deployment cost. - (5)
The Manhattan distance between node

*i*and node*j*(*d*_{ ij }). - (6)
The cost of RRU (

*C*^{R}) and OADM (*C*^{O}). - (7)
The maximal transmit power of a RRU

- (8)
The minimal signal-to-noise-ratio (SNR) requirement for an MS

### 3.3 Variable

- (1)
The PS

_{ m }is selected to place a RRU if*a*_{ m }= 1; otherwise,*a*_{ m }= 0. - (2)
The PS

_{ t }is selected to place an OADM if*b*_{ t }= 1; otherwise,*b*_{ t }= 0. - (3)
The MS at TP

_{ n }is associated with the RRU if*z*_{ mn }= 1; otherwise,*z*_{ mn }= 0. - (4)
There is a directed link from node

*i*∈*Ω*_{B}∪*Ω*_{OADM}to node*j*∈*Ω*_{B}∪*Ω*_{RRU}∪*Ω*_{OADM}if*e*_{ ij }= 1; otherwise,*e*_{ ij }= 0. - (5)
The TP

_{ n }is covered if*q*_{ n }= 1; otherwise,*q*_{ n }= 0. - (6)
*e*_{ ij }is on the directed flow from the BBU to OADM_{ k }if \( {g}_{ij}^k=1 \); otherwise, \( {g}_{ij}^k=0 \). - (7)
*e*_{ ij }is on the directed flow from the OADM_{ k }to BBU if \( {h}_{ij}^k=1 \); otherwise \( {h}_{ij}^k=0 \).

### 3.4 Constraints

- (1)
The mobile coverage requirement should be satisfied, i.e., the coverage ratio should be larger than a predefined value.

- (2)
The SNR for an MS located at each TP should be larger than a SNR threshold.

- (3)
Each TP is covered by at least one RRU.

- (4)
The OADM ring consists the BBU and the selected OADMs.

- (5)
There must be at least a RRU to connect to the selected OADM.

- (6)
The RRU can cover the MS with CoMP technology.

- (7)
The double-flow control is used to establish the mathematical model.

- (8)
If the PS is selected to place an OADM on the OADM ring, there must be an input path and output path.

### 3.5 Objective

Definitions of Symbols in the RSTO problem

Symbol | Definition |
---|---|

| The set of the BBU |

| The set of the OADMs |

| The set of the RRUs |

| The set of the TPs |

| The number of OADMs in the network |

| The number of RRUs in the network |

| The number of TPs in the network |

| The distance between node |

| The maximal transmit power of a RRU |

| The average normalized thermal noise in AOI |

| The path loss exponent |

| The minimal SNR requirement for each TP |

| The location incidence vector, |

| The location incidence vector, |

| A TP coverage incidence vector, |

| RRU-TP association matrix |

| BBU-OADM-RRU association matrix, |

| The maximal distance of OMD-RRU that can be deployed in the network |

| OADM |

| BBU − OADM |

### 3.6 Problem formulation

Equation (1.1) is to minimize the total deployment cost of the C-RAN network, including the cost of fibers, the cost of RRUs, and the cost of OADMs.

*a*

_{ m }= 1. Constraint (1.4) ensures that the PS is selected to place an OADM if

*b*

_{ i }= 1.

*a*

_{ j }= 1. Constraint (1.10) ensures that RRUs can be connected to the OADM, if

*b*

_{ i }= 1.

*e*

_{ ij }= 1,

*e*

_{ ij }probably exists in the inflow path from BBU to OADM

_{ k }. Constraint (1.13) stipulates that only when

*e*

_{ ij }= 1,

*e*

_{ ij }probably exists in the outflow path from the BBU to OADM

_{ k }.

_{ i }and OADM

_{ j }exist firstly, then there probably exists the path between the BBU and OADM

_{ k }.

_{ k }, its inflow path must end at OADM

_{ k }, and its outflow must start from OADM

_{ k }.

_{ k }is selected to place on the OADM ring, then OADM

_{ k }has an inflow path and an outflow path.

_{ j }is on the path from BBU to OADM

_{ k }, then OADM

_{ j }has an input path and an output path. Also, there is an inflow path that belongs to the path from BBU to OADM

_{ j }. Constraint (1.23) stipulates that if OADM

_{ j }is on the path from BBU to OADM

_{ k }, then OADM

_{ j }has an input path and an output path. Also, there is an outflow path that belongs to the path from BBU to OADM

_{ k }.

_{ k }, constraint (1.24) stipulates that the same path cannot exist in the inflow path and outflow path at the same time. Constraint (1.25) stipulates that the path is directional, which means only one direction can be chosen. Constraint (1.26) ensures that the TP satisfies the constraint of the minimum SNR requirement. Constraints (1.27)–(1.29) stipulate the definition of Q.

Constraints (1.30)–(1.34) state the each entry *A*, *B*, *Q*, *Z*, *E*, *H*
^{
K
}, and *G*
^{
K
} is binary.

## 4 Heuristic algorithm

### 4.1 Algorithm description

The problem formulation is given in the previous section, including the network model, given inputs, variables, constraints, and objective of the RSTO problem. However, with the continuous expansion of network size, solving the above RSTO formulation is not an easy task especially for large-scale networks due to the exponential growing nature of ILP. To make the design more scalable for large-size networks, in this section, we propose a new heuristic algorithm.

- (1)
Initialization of system parameters.

- (2)
RRU selection. This step achieves the coverage TP through candidate RRU.

- (3)
OADM selection.

- (4)
Construct the C-RAN architecture based on the ring and spur topology.

- (5)
C-RAN layout (including the location of selected RRUs and OADMs, the optical link between RRUs and OADMs, the optical ring of OADMs and BBU, the association between RRUs and OADMs), total deployment cost.

Now, we introduce the pseudocode of this the heuristic algorithm, and then we give an application based on this heuristic algorithm.

- Step 1
(Lines 1–2): Initialization. Firstly, a plurality of TPs, potential sites for deploying RRUs, and OADMs are set in an AOI after an actual survey. A two-dimensional coordinate graph is established on the basis of the latitude and longitude of the actual positions of TPs, and RRU candidates and OADM candidates are import into this two-dimensional coordinate graph based on their actual location of the latitude and longitude. The distances between TPs and RRU candidates (i.e., parameter

*d*_{T − R}), RRU candidates and OADM candidates (i.e., parameter*d*_{R − O}), OADM candidates and OADM candidates (i.e., parameter*d*_{O − O}) are calculated and stored, respectively. The parameters of*N*_{BBU},*Ω*_{TP},*Ω*_{RRU},*Ω*_{OADM},*ϒ*_{threshold},*D*_{max(T − R)},*D*_{max(R − O)},*D*_{max(O − O)}are set according to the requirement of actual deployment scenarios. - Step 2
(Lines 3–22): RRU selection. The goal of this step is to make each TP covered by one candidate RRU. First, we calculate the SNR between the TPs and candidates RRUs. If the received SNR at a TP form a candidate RRU that satisfies the minimum SNR threshold, then we select the RRU which covers the maximum amount of the TPs, then add this RRU into

*Ω*_{RRU − selected}.We will delete this already selected RRU and these TPs which are covered by selected RRU. The TPs will select the RRU in turn according to the calculated TP numbers covered by the candidate RRU in descending order The candidates can be selected until all TPs are covered by the candidate RRU. - Step 3
(Lines 23–34): OADM selection. The goal of this step is to set up the optical links between the candidate OADM and all the selected RRUs at step 2. We calculate the distances between the selected RRUs and candidate OADMs. If the distance satisfies the requirement of the selected RRU and candidate OADM limitations (i.e., parameter

*D*_{max(R − O)}), we select the candidate OADM by a proximity principle. Then, we will add this OADM into*Ω*_{OADM − selected}. If each selected RRU could select an OADM which satisfies the requirement of the distance limitations, then these OADMs will be added into*Ω*_{OADM − selected}. The loop will end until all the selected RRU at step 2 could connect to an OADM. The optical links between the selected RRU at step 2 and OADM can be set up. - Step 4
(Lines 35–51): Construct the C-RAN architecture based on the ring and spur topology. The goal of this step is to construct the OADM ring. The ring will be constructed from the BBU as the start point. Firstly, we calculate the distance (i.e., parameter

*d*_{BBU − O}) between the BBU and the selected OADMs (i.e., parameter*Ω*_{OADM − selected}). Secondly, we find the minimum \( {d}_{\mathrm{BBU}-{O}_j} \) for all*O*_{ j }and connect the BBU to the corresponding*O*_{ j }, then remove this*O*_{ j }in*Ω*_{OADM − selected}. Thirdly, we calculate the distance (\( {d}_{O_j-{O}_i} \)) between*O*_{ j }and*O*_{ i }(*O*_{ i }∈*Ω*_{OADM − selected}−*O*_{ j }), then we also find the minimum distance and connect*O*_{ j }to*O*_{ i }. The loop will end until the*Ω*_{OADM − selected}has only one element*O*_{ i }. Finally, we connect this*O*_{ i }to BBU, and the OADM ring can be constructed.

Finally, in lines 52–53, the heuristic outputs the layout solution (including the location of the selected RRUs and OADMs, the optical link between RRUs and OADMs, the optical ring of OADMs and BBU, the association between RRUs and OADMs) and total deployment cost.

### 4.2 An example for the proposed fast heuristic algorithm

*P*

_{RRU}, and the candidate OADMs denote

*P*

_{OADM}(see Fig. 3).

We defined a generic cost unit (gcu) [21] to simplify the evaluation of deployment costs in all the case studies.

Step 1: Initialization.

Step 2: Achieve all of the TPs that can be covered by the candidate RRUs. In Fig. 3, the set of the candidate RRUs denotes *P*
_{RRU} = {*P*
_{RRU}0, *P*
_{RRU}1, *P*
_{RRU}2, ⋯, *P*
_{RRU}19}. The set of the candidate OADMs denotes \( {P}_{\mathrm{OADM}}=\left\{{P}_{{}^{\mathrm{OADM}}}0,{P}_{{}^{\mathrm{OADM}}}1,\cdots, {P}_{{}^{\mathrm{OADM}}}9\right\} \). The set of TPs denotes \( {P}_{{}^{\mathrm{TP}}}=\left\{{P}_{{}^{\mathrm{TP}}}0,{P}_{{}^{\mathrm{TP}}}1,{P}_{{}^{\mathrm{TP}}}2,\cdots, {P}_{{}^{\mathrm{TP}}}34\right\} \). All of the TPs can be covered by the four RRUs, \( {P}_{{}^{\mathrm{RRU}}}10,{P}_{{}^{\mathrm{RRU}}}0,{P}_{{}^{\mathrm{RRU}}}3,{P}_{{}^{\mathrm{RRU}}}6 \). In specific, 17 TPs can be covered by \( {P}_{{}^{\mathrm{RRU}}}10 \); 6 TPs can be covered by \( {P}_{{}^{\mathrm{RRU}}}0 \); 6 TPs can be covered by \( {P}_{{}^{\mathrm{RRU}}}3 \); and 6 TPs can be covered by \( {P}_{{}^{\mathrm{RRU}}}6 \).

Finally, output the optical link between RRUs and OADMs, the optical link between OADM and BBU, the association between RRUs and OADMs (see Fig. 5).

Step 5: Compute the total deployment cost, computing time in this ring and spur topology.

In scenario 0, the total deployment cost is 1824.43 gcu. The topology running time is 1.308 s. The layout and total deployment cost results of the ring and spur topology under C-RAN is the same as the result obtained with Gurobi which will be introduced later in next section.

## 5 Numerical analysis

### 5.1 Parameter settings

We implement the optimization model of the previous section and solve the RSTO problem using Gurobi Optimizer [22], which is a state-of-the-art ILP solver. Gurobi is designed to exploit modern multi-core processors and the performance is proved to be superior to CPLEX [23]. Without loss of generality, the Manhattan distance of two nodes is used to represent the corresponding edge length of the fiber deployment route, and all the fiber segment links have the same weighted deployment costs.

C-RAN component cost

Component | Cost (gcu) |
---|---|

Fiber (/km) | 1 |

Burying fiber (/km) | 50 |

RRU | 16 |

OADM | 50 |

Simulation parameter settings

Parameter | Value |
---|---|

| 2 |

| 8 W |

| 29 dB |

| 2 mW |

| 10 km |

Scale | 20 km |

Comparison of RSTO results with and without CoMP in scenario 1

Scenario 1 | Without CoMP ( | With CoMP ( |
---|---|---|

Selected number of PSs for RRU placement | 4 RRUs | 3 RRUs |

Obj. value (gcu) | 1760.8 | 1543.8 |

Compute time (s) | 0.02 | 0.03 |

Optimality Gap | 0 % | 0 % |

### 5.2 Numerical analysis of small-scale scenarios

Problem size of simulated scenarios

Scenario | Number of nodes | |||
---|---|---|---|---|

TP | RRU candidates | OADM candidates | ||

I | 100 | 40 | 20 | |

II | 500 | 120 | 50 | |

III | 900 | 190 | 80 | |

Constraints | Variables | Nonzero | Ave. memory | |

I | 3058 | 2419 | 8423 | 127.59 MB |

II | 9371 | 6843 | 26418 | 736.74 MB |

III | 24873 | 18397 | 72629 | 7851.38 MB |

Comparison analysis of small-scale scenarios

Scenario | I | II | III | |||
---|---|---|---|---|---|---|

Scheme | Gurobi | Heuristic | Gurobi | Heuristic | Gurobi | Heuristic |

Obj. value (gcu) | 3458.12 | 3614.47 | 4672.35 | 4928.84 | 6028.96 | 6375.14 |

Computing time (s) | 17.24 | 2.07 | 46.39 | 4.15 | 130.12 | 6.34 |

Optimality gap | 4.52 % | 5.48 % | 5.74 % |

### 5.3 Numerical analysis of larger-scale scenarios

Problem size of simulated large scenarios

Scenario | Number of nodes | ||
---|---|---|---|

TP | RRU candidates | OADM candidates | |

IV | 1500 | 300 | 150 |

V | 2500 | 450 | 230 |

Comparison analysis for large scenarios

Scenario | IV | V | ||
---|---|---|---|---|

Scheme | Gurobi | Heuristic | Gurobi | Heuristic |

Obj. value (gcu) | No result | 8127.67 | No result | 10342.14 |

Computing time (s) | 7.15 | 8.63 |

From Table 8, Gurobi can hardly find the optimal result with the increase of the network scale; however, the heuristic algorithm can still obtain the solution effectively and efficiently.

Therefore, we can conclude that the RSTO-formulation-based Gurobi solution scheme are more suitable for smaller-scale C-RAN network layout planning, and our proposed heuristic algorithm could be suitable for real-world large-scale C-RAN network layout planning in practice.

### 5.4 Comparison of the tree and branch and ring and spur topology

- (i)
The deployment cost of the ring topology is relatively higher than that of the tree topology due to the deployment of bidirectional optical fiber.

- (ii)
In respect to the computing time, the ring topology is shorter than the tree topology. In the tree topology, it needs to make restrictions to every path from the BBU to the RRUs, but in the ring topology, it only needs to make restrictions to the path in the OADM ring.

The network scale of scenarios IX, X, and XI

Scenario | IX | X | XI |
---|---|---|---|

PS for OA/OADMs | 5 | 5 | 5 |

PS for RRUs | 20 | 30 | 50 |

The number of TPs | 20 | 40 | 100 |

Comparison of the tree and ring topology

Scenario | IX | X | XI | |||
---|---|---|---|---|---|---|

Scheme | Tree | Ring | Tree | Ring | Tree | Ring |

Total cost (unit, gcu) | 1834.46 | 2208.12 | 2436.54 | 2895.03 | 3269.07 | 3890.01 |

Computing time | 2.09 | 0.34 | 5.41 | 0.87 | 7.65 | 1.21 |

## 6 Conclusions

C-RAN provides a promising wireless architecture for reliable transmission of mobile big data. In this paper, we study the fundamental problem of topology optimization based on the more reliable string and spur topology. We formulate the RSTO problem to minimize the total deployment cost of the C-RAN network. We have examined the proposed RSTO problem via extensive case studies in term of its feasibility and scalability. Furthermore, we propose a new heuristic algorithm to obtain the near-optimal C-RAN network layout effectively and efficiently, especially suitable for a large-scale C-RAN network scenario in practice. In numerical studies, we also demonstrate the performance benefits of incorporation of CoMP technology in terms of total deployment cost reduction.

## Declarations

### Acknowledgements

This study is sponsored by the National Science Foundation of China (NSFC) No. 61371091, No. 61171175, and No. 61301228; the National Science Foundation of Liaoning Province No. 2014025001; the Program for Liaoning Excellent Talents in University (LNET) No. LJQ2013054; and the Fundamental Research Funds for Central Universities under grant No. 3132015045.

**Open Access**This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

## Authors’ Affiliations

## References

- P Zhang, Q Cui, Y Hou, J Xu, Opportunities and challenges of wireless networks in the era of mobile big data. Journal of China Science
**60**, 433–438 (2015)Google Scholar - The 34th Statistical Report of the Department of Internet in China. Chinese Internet Network Information Centre, 2014.Google Scholar
- Q Lei, Z Zhang, F Cheng, 5G radio access network architecture based on C-RAN. Telecommunication Science
**1**, 1–10 (2015)Google Scholar - X. Wang. C-RAN: the road towards green RAN. China Communication Journal, 2010Google Scholar
- K. Chen. C-RAN White Paper, version 2.5. China Mobile Research Institute,2011Google Scholar
- MA Esmail, H Fathallah,
*Fiber Fault Management and Protection Solution for Ring-and-Spur WDM/TDM Long-Reach PON*. 2011 IEEE Global Telecommunications Conference, 2011, pp. 1–5Google Scholar - http://www.huawei.com. Accessed 1 Oct 2011
- A design algorithm for ring topology centralized-radio-access-network. Paper presented at the Optical Networks Design and Modelling (ONDM) Conference, Brest, France, 2013Google Scholar
- K. Ramantas, K. Vlachos, G. Ellinas, A. Hadjiantonis. A converged optical wireless architecture for mobile backhaul networks. Paper presented at the Optical Networks Design and Modelling (ONDM) Conference, Brest, France, pp. 155-160, 2013Google Scholar
- L Kazovsky, SW Wong, T Ayhan, KM Albeyoglu, MRN Ribeiro, A Shastri, Hybrid optical-wireless access networks. Proc. IEEE
**100**(5), 1197–1225 (2012)View ArticleGoogle Scholar - B Lin, L Lin, Site planning of relay station in green wireless access networks: a genetic algorithm approach, in
*Soft Computing and Pattern Recognition (SoCPaR), Dalian, China*, 2011, pp. 167–172Google Scholar - B Lin, PH Ho, L Xie, X Shen, J Tapolcai, Optimal relay station placement in broadband wireless access networks. IEEE Transaction on Mobile Computing
**9**(2), 259–269 (2010)View ArticleGoogle Scholar - B Lin, Y Tian,
*Energy and Radiation-Aware Base Station Placement in Eco-Sustainable LTE Networks*. 2nd International Conference on Electronics, in Communication and Control (ICECC2012), 2012, pp. 3162–3165Google Scholar - B Lin, L Lin, PH Ho,
*Cascaded Splitter Topology Optimization in LRPONs*. International Conference on Communications (ICC), Ottawa, Canada, 2012, pp. 3105–3109Google Scholar - L Lin, B Lin, PH Ho,
*Power-Aware Optimization Modeling for Cost-Effective LRPON Infrastructure Deployment*. 21st International Conference on Software, Telecommunications and Computer Networks (SoftCOM), 2013, pp. 1–5Google Scholar - A Baumgartner, VS Reddy, T Bauschert,
*Mobile Core Network Virtualization: A Model for Combined Virtual Core Network Function Placement and Topology Optimization*. 2015 1st IEEE conference on Network Softwarization (NetSoft), London, 2015, pp. 1–9Google Scholar - A Barolli, F Xhafa, M Takizawa,
*Optimization Problems and Resolution Methods for Node Placement in Wireless Mesh Network*. 2011 14th International Conference on Network-Based Information Systems (NBiS), Tirana, 2011, pp. 126–134Google Scholar - B Lin, J Guo, R He, T Yang,
*Infrastructure Deployment and Dimensioning of Relay-Based Heterogeneous Wireless Access Networks for Green Intelligent Transportation*. The 2nd conference on Catalysis (ICC 2014), Suzhou, 2014, pp. 447–460Google Scholar - E Inga-Ortega, A Peralta-Sevilla, RC Hincapie,
*Infrastructure Deployment and Dimensioning of Relay-Based Heterogeneous Wireless Access Networks for Green Intelligent Transportation*. 2015 IEEE PES on Innovative Smart Grid Technologies Latin America (ISGT LATAM), Montevideo, 2015, pp. 30–35Google Scholar - A Peralta-Sevilla, E Inga, R Cumbal, R Hincapie,
*Optimal Dimensioning of FiWi Networks Over Advanced Metering Infrastructure for the Smart Grid*. 2015 IEEE Colombian Conference on Communications and Computing (COLCOM), Popayan, 2015, pp. 1–6Google Scholar - DAR Chaves, EA Barboza, CJA Bastos-Filho, A Multi-Objective Approach to Design All-Optical and Translucent Optical Networks Considering CapEx and QoT. 14th International Conference on Transparent Optical Networks (ICTON)
**49**(1), 1–4 (2012)View ArticleGoogle Scholar - Gurobi Optimizer 4.6, Gurobi Optimization Inc., 2012.Google Scholar
- Gurobi Optimizer, State of the Art Mathematical ProgrammingN Solver, [online], http://www.gurobi.com/.