Encyclopedia of Wireless Networks

Living Edition
| Editors: Xuemin (Sherman) Shen, Xiaodong Lin, Kuan Zhang

Network Slicing-Enabled Green C-RAN

  • Yi-Han ChiangEmail author
  • Yusheng Ji
Living reference work entry
DOI: https://doi.org/10.1007/978-3-319-32903-1_212-1



Network slicing is an emerging technology that allows mobile network operators (MNOs) to slice their physical networks into logical ones, thereby providing independent services to multiple mobile virtual network operators (MVNOs). With the incorporation of network slicing, physical network resources can be allocated in terms of network slices, each of which represents a virtualized and isolated network with reserved network resources.

Cloud radio access network (C-RAN) incorporates the concept of cloud computing into RANs, which generally consists of a baseband unit (BBU) pool and remote radio heads (RRHs). The BBU pool is located at a centralized site with abundant computing resources, while the RRHs are distributed antenna units that are deployed apart from the BBU pool. By pooling the BBUs of RRHs together, lower capital and operation expenditures (CapEx and OpEx) for MNOs can be envisaged.

Remote radio head (RRH) is a set of lightweight antenna units with its processing units partly or totally migrated to the BBU pool, but retaining its location at distant sites. RRHs are connected to the BBU pool by delivering in-phase and quadrature (IQ) data through fronthaul (FH) links. Thanks to the resource pooling feature, the use of RRHs evolves as a cost-effective alternative (due to the simplified hardware architectures as compared to traditional macro base stations) and makes large-scale network deployments easier than before.

Historical Background

The emergence of mobile data tsunami as forecasted in Cisco (2019) has stimulated MNOs to search for next-generation wireless networks. To catch up with the explosive growth of mobile data traffic while keeping CapEx and OpEx at a minimum, network densification has been widely regarded as a promising solution in that it offers denser service coverage and higher capacity due to the massively deployed small cells, and benefits from lower transmit power consumption because of shortened last-mile distances. To manage such denser networks efficiently, MNOs are inspired to pay attention to C-RAN (I et al., 2014; Checko et al., 2015; Saxena et al., 2016; Alimi et al., 2018) in its benefits of better resource optimization and cost effectiveness.

In parallel with the standardization of 5G systems, network slicing (Zhou et al., 2016; Rost et al., 2017; Foukas et al., 2017; Zhang et al., 2017) has attracted growing research attention due to its flexibility and scalability to meet diverse requirements of future mobile networks. By means of network slicing, each physical network (owned by an MNO) can be dynamically sliced into several logical ones, thereby providing reserved network resources to individual MVNOs. In this way, a wide range of services (e.g., data, video and telephony) with distinct requirements (e.g., bandwidth, delay, reliability and privacy) from different MVNOs can be embedded in the same physical network.

In addition to boosting network capacity, plenty of existing works have been devoted to energy-saving issues (Arnold et al., 2010; Auer et al., 2011) in cellular systems, thereby saving CapEx and OpEx for MNOs. Thanks to the centralized architure, C-RAN can be smoothly integrated with network slicing while achieving network greenness. Since BBUs and RRHs will be initiated to serve user equipment (UE) for MVNOs, where BBUs and RRHs can incur power consumption in their computation and communication, respectively, they should be efficiently activated and utilized. One possibility for the energy concervation inspired by Wang et al. (2017), Aqeeli et al. (2018) and Yao and Ansari (2018) is that one BBU can serve a cluster of RRHs, and less active BBUs can lead to lower power consumption in the BBU pool. Another energy-saving opportunity from Ashraf et al. (2011), Holtkamp et al. (2014) and Chiang and Liao (2017) indicates that each RRH consumes power in its operation and radio transmission. As a consequence, the less the active RRHs, the lower the resulting power consumption. Altogether, how BBUs and RRHs can be activated and how BBU-RRH and RRH-UE mappings can be performed, while satisfying network capacity limits and user requirements, are decisive in green C-RAN with network slicing.


A generic C-RAN architecture (as shown in Fig. 1) consists of a set \(\mathcal {L}\) of BBUs (colocated in a BBU pool) and a set \(\mathcal {M}\) of RRHs, where each RRH is connected to the BBU pool through a high-speed FH link, and the BBUs are controlled and managed by the MNO. On top of the C-RAN, there exists a set \(\mathcal {V}\) of MVNOs, each of which has an independent set \(\mathcal {N}_v\) of UE (each with a data rate requirement of Rn, \(\forall n \in \mathcal {N}_v\)) to serve. After receiving service requests from MVNOs, the MNO needs to serve the set \(\mathcal {N} = \bigcup _{v \in \mathcal {V}} \mathcal {N}_v\) of UE.
Fig. 1

An illustration of a network slicing-enabled C-RAN, where MVNO A requests for 2 BBUs and 3 RRHs, and MVNO B asks for 3 BBUs and 3 RRHs. After collecting the demands from the MVNOs, the MNO simply activates 2 BBUs and 2 RRHs, in the way that network capacity limits and user requirements can all be met

In the BBU pool, an active BBU can be realized by initiating a virtual machine (VM), which can serve one or more UE. Clearly, the BBU-RRH mapping is either one-to-one or one-to-many. In addition, each BBU \(l \in \mathcal {L}\) can only support up to \(C^{\mathrm {max}}_l\) RRHs, due to its limited computing resources and FH capacities.

Each BBU/RRH can be activated or deactivated to reduce its operational power consumption. Typically, such a deactivation can save energy effectively, since a non-zero energy dissipation can be attributed to operating in active mode. Therefore, operational power consumption can be lowered whenever a BBU/RRH can be deactivated.

Thanks to the centralized processing of C-RAN, multiple RRHs can cooperatively transmit to UE, thereby improving the received signal qualities. Suppose that universal frequency reuse is adopted by all RRHs to fully utilize the network spectrum of bandwidth B. Denote by SINRn the received signal-to-interference-plus-noise ratio (SINR) at each UE as
$$\displaystyle \begin{aligned} \mathrm{SINR}_{n} = \frac{\sum_{m \in \mathcal{M}} \mathsf{p}_{mn} H_{mn}}{I_{n} + N_0}, \quad \forall n \in \mathcal{N}, \end{aligned} $$
where Hmn refers to the channel power gain from RRH m to UE n, N0 is the noise spectral density at each UE, In aggregates the interfering powers experienced by UE n. In addition, \(\mathsf {p}_{mn} \leq P^{\mathrm {tx}}_m\) is the transmit power allocated to UE n by RRH m, where \(P^{\mathrm {tx}}_m\) is the RF output power of RRH m at its maximum load (namely RRH m’s transmit power budget).
The total power consumption is to aggregate the operational and transmit power consumption of all BBUs and RRHs, which can be expressed as
$$\displaystyle \begin{aligned} \begin{array}{rcl}{} &\displaystyle &\displaystyle p^{\mathrm{tot}} \\ &\displaystyle &\displaystyle = \overbrace{ \sum_{l\in\mathcal{L}} P^{\mathrm{act}}_{\mathrm{R},m} \mathsf{y}_m + \sum_{l\in\mathcal{L}} P^{\mathrm{slp}}_{\mathrm{R},m} \left( 1-\mathsf{y}_m \right) + \sum_{m \in \mathcal{M}} \sum_{n \in \mathcal{N}} \varDelta_{\mathrm{R},m} \mathsf{p}_{mn} }^{\text{RRH power consumption}} \\ &\displaystyle &\displaystyle + \underbrace{ \sum_{m\in\mathcal{M}} P^{\mathrm{act}}_{\mathrm{B},l} \mathsf{z}_l + \sum_{m\in\mathcal{M}} P^{\mathrm{slp}}_{\mathrm{B},l} \left( 1-\mathsf{z}_l \right) + \sum_{l \in \mathcal{L}} \sum_{m \in \mathcal{M}} \varDelta_{\mathrm{B},l} \mathsf{q}_{lm} }_{\text{BBU power consumption}}, \end{array} \end{aligned} $$
where \(P^{\mathrm {act}}_{\mathrm {R},m}\) and \(P^{\mathrm {slp}}_{R,m}\) are the power consumption of RRH m operating in active mode and sleep mode, respectively, and \(P^{\mathrm {act}}_{\mathrm {B},l}\) and \(P^{\mathrm {slp}}_{\mathrm {B},l}\) are those of BBUs. ΔR,m and ΔB,l represent the slopes of the load-dependent power consumption of RRH m and BBU l, respectively. The binary decision variables ym and zl indicate the activeness of RRH m and BBU l, pmn is the transmit power of RRH m allocated for serving UE n, and qlm indicates whether RRH m is associated with BBU l.
Mathematically, the problem of enabling network slicing to green C-RAN (NSGC) can be formulated as (for brevity, \(\boldsymbol {\mathsf {p}} = \left \{ \mathsf {p}_{mn}, \forall m,n \right \}\), \(\boldsymbol {\mathsf {q}} = \left \{ \mathsf {q}_{lm}, \forall l,m \right \}\), \(\boldsymbol {\mathsf {y}} = \left \{ \mathsf {y}_m, \forall y \right \}\) and \(\boldsymbol {\mathsf {z}} = \left \{ \mathsf {z}_l, \forall l \right \}\)):
$$\displaystyle \begin{aligned} \begin{array}{rcl} \begin{array}{cccclcl} &\displaystyle &\displaystyle \underset{\boldsymbol{\mathsf{p}},\boldsymbol{\mathsf{q}},\boldsymbol{\mathsf{y}},\boldsymbol{\mathsf{z}}}{\text{min}} &\displaystyle \quad &\displaystyle p^{\mathrm{tot}}, \\ &\displaystyle &\displaystyle \text{s.t.} &\displaystyle &\displaystyle {\mathbf{C1:} \enspace B \log \left( 1+\mathrm{SINR}_n \right) {\geq} R_n,}\ \forall n \in \mathcal{N}, \\ &\displaystyle &\displaystyle &\displaystyle &\displaystyle {\mathbf{C2:} \enspace \sum_{n \in \mathcal{N}} \mathsf{p}_{mn} \leq P^{\mathrm{tx}}_m \mathsf{y}_m,}\ \forall m \in \mathcal{M}, \\ &\displaystyle &\displaystyle &\displaystyle &\displaystyle {\mathbf{C3:} \enspace \sum_{l \in \mathcal{L}} \mathsf{q}_{lm} = \mathsf{y}_{m},}\ \forall m \in \mathcal{M}, \\ &\displaystyle &\displaystyle &\displaystyle &\displaystyle {\mathbf{C4:} \enspace \sum_{m \in \mathcal{M}} \mathsf{q}_{lm} \leq C^{\mathrm{max}}_l \mathsf{z}_l,}\ \forall l \in \mathcal{L}, \\ &\displaystyle &\displaystyle &\displaystyle &\displaystyle {\mathbf{C5:} \enspace \mathsf{p}_{mn} \geq 0,}\ \forall m \in \mathcal{M}, n \in \mathcal{N}, \\ &\displaystyle &\displaystyle &\displaystyle &\displaystyle {\mathbf{C6:} \enspace \mathsf{q}_{lm}, \mathsf{y}_{m}, \mathsf{z}_{l} {\in} \{ 0,1 \},}\ \forall l {\in} \mathcal{L}, m {\in} \mathcal{M}, \end{array} \end{array} \end{aligned} $$

where C1 states that the achieved data rate should be high enough to meet each UE’s data rate requirement, C2 indicates that each RRH can allocate transmit powers to UE (up to its transmit power budget) only if it is operating in active mode, C3 ensures that each active RRH is associated with exactly one BBU, C4 presents that each active BBU can only accommodate a limited amount of RRHs, C5 and C6 refer to the decision variables of RRH-UE mapping, BBU-RRH mapping, the activeness of RRHs and BBUs, respectively.

In fact, finding an optimal solution to the NSGC problem is essentially NP-hard. Therefore, it is of more practical interests to look for an approximate solution with a polynomial-time complexity and a provably good performance guanrantee.

Definition 1

Consider the NSGC problem as follows: does there exist a solution \(\left ( \boldsymbol {\mathsf {p}}, \boldsymbol {\mathsf {q}}, \boldsymbol {\mathsf {y}}, \boldsymbol {\mathsf {z}} \right )\) such that the constraints C1 to C6 can all be satisfied, and the total power consumption ptot is at most U? The problem instance can be expressed as
$$\displaystyle \begin{aligned} \begin{array}{rcl} \mathsf{NSGC} \Big( \mathcal{L}, \mathcal{M}, \mathcal{N}, B, \left\{ R_n \right\}, \left\{ H_{mn} \right\}, \left\{ C^{\mathrm{max}}_{l} \right\}, \left\{ P^{\mathrm{tx}}_m \right\}, N_0, \\ \hfill U, \left\{ \varDelta_{\mathrm{R},m} \right\}, \left\{ \varDelta_{\mathrm{B},l} \right\}, \left\{ P^{\mathrm{act}}_{\mathrm{R},m} \right\}, \left\{ P^{\mathrm{act}}_{\mathrm{B},l} \right\}, \left\{ P^{\mathrm{slp}}_{\mathrm{R},m} \right\}, \left\{ P^{\mathrm{slp}}_{\mathrm{B},l} \right\} \Big). \end{array} \end{aligned} $$

Definition 2

(Weighted Set Cover) Given a collection \(\mathcal {C}\) of subsets of a finite set \(\mathcal {S}\) with \(\bigcup _{D\in \mathcal {C}}D=\mathcal {S}\) and a positive integer V . Does \(\mathcal {C}\) contain a subcollection \(\mathcal {C}^\prime \subseteq \mathcal {C}\) with \(\sum _{D\in \mathcal {C}^\prime }W_D \leq V\), such that every element of \(\mathcal {S}\) belongs to at least one member of \(\mathcal {C}^\prime \)? The problem instance can be denoted by \(\mathsf {WSC}\left (\mathcal {S},\mathcal {C},V,\left \{W_D\right \}\right )\).

Theorem 1

The NSGC problem is NP-hard.


Consider the restricted case of only one BBU, i.e., \(\mathcal {L} = \left \{ l^\circ \right \}\). Since the NSGC problem involves power allocation whereas the WSC problem does not have the analogous part, it is necessary make the data rate requirements low enough to ensure that C1 can always be held for all UE. By carefully examining C1, it can be observed that the following fact holds: as \(R_n \geq \bar {R}_n\), C1 can be satisfied as long as each UE is served by an arbitrary one RRH with a uniform transmit power (i.e., \({P^{\mathrm {tx}}_m}/{\left \vert \mathcal {N}_{l^\circ }\right \vert }\) for each \(n\in \mathcal {N}_{l^\circ }\)), where Given with \(R_n \geq \bar {R}_n\), it is apparent that the uniform transmit powers are always allocable.
The NP-hardness of the NSGC problem can then be proved by restriction. In particular, the NSGC problem can be restricted by allowing only instances with simply the active operational power consumption (by setting ΔR,m, ΔB,l, \(P^{\mathrm {slp}}_{R,m}\) and \(P^{\mathrm {slp}}_{B,l}\) to zeros) and binary channel states (\(H_{mn}\in \left \{0,1\right \}\)). Other parameters can be chosen freely without affecting the following instance mapping. Now, consider an WSC problem instance of
$$\displaystyle \begin{aligned} \mathsf{WSC}\left(\mathcal{S},\mathcal{C},V,\left\{W_D\right\}\right).\end{aligned} $$
Then, the corresponding restricted NSGC problem instance can be obtained as
$$\displaystyle \begin{aligned} \begin{array}{rcl} \mathsf{NSGC} \Big( \left\{ l^\circ \right\}, \mathcal{C}, \mathcal{S}, 1, \left\{ K_1, K_1,\dots \right\}, \left\{ E_{Cs} \right\}, K_2, \left\{ 1,1,\dots \right\}, \left\vert \mathcal{C} \right\vert, \hfill \\ \hfill V, \left\{ 0,0,\dots \right\}, \left\{ 0,0,\dots \right\}, \left\{ W_D \right\}, \left\{ W_D \right\}, \left\{ 0,0,\dots \right\}, \left\{ 0,0,\dots \right\} \Big). \end{array} \end{aligned} $$
where \(K_1 = \log \left ( {\left \vert \mathcal {S} \right \vert }/{\left \vert \mathcal {S} \right \vert -1} \right )\), \(K_2 = \left \vert \mathcal {C} \right \vert \), EDs = 1 if s ∈ D and 0 otherwise. Since the transformation between the restricted NSGC problem instance and the WSC problem instance can be run in polynomial time, the NP-hardness of the NSGC problem is thus proved.

Key Applications

Network slicing-enabled green C-RAN generally involves the dynamic operations of BBU/RRH activation, BBU-RRH and RRH-UE mappings, which can be adequately designed to support applications (e.g., enhanced mobile broadband (eMBB), ultra-reliable low-latency communications (URLLC) and massive machine type communications (mMTC)) that require reserved network services, while heading towards network greenness.



  1. Alimi IA, Teixeira AL, Monteiro PP (2018) Toward an efficient C-RAN optical fronthaul for the future networks: a tutorial on technologies, requirements, challenges, and solutions. IEEE Commun Surv Tutorials 20(1):708–769CrossRefGoogle Scholar
  2. Aqeeli E, Moubayed A, Shami A (2018) Power-aware optimized RRH to BBU allocation in C-RAN. IEEE Trans Wireless Commun 17(2):1311–1322CrossRefGoogle Scholar
  3. Arnold O et al (2010) Power consumption modeling of different base station types in heterogeneous cellular networks. In: Proceedings of the future network and mobile summitGoogle Scholar
  4. Ashraf I, Boccardi F, Ho L (2011) SLEEP mode techniques for small cell deployments. IEEE Commun Mag 49(8):72–79CrossRefGoogle Scholar
  5. Auer G et al (2011) How much energy is needed to run a wireless network? IEEE Wireless Commun 18(5):40–49CrossRefGoogle Scholar
  6. Checko A et al (2015) Cloud RAN for mobile networks–a technology overview. IEEE Commun Surv Tutorials 17(1):405–426CrossRefGoogle Scholar
  7. Chiang Y, Liao W (2017) Multicell sleeping control and transmit power adaptation in green heterogeneous networks. In: Proceedings of the IEEE GLOBECOMCrossRefGoogle Scholar
  8. Cisco (2019) Cisco visual networking index: global mobile data traffic forecast update, 2017–2022. https://www.cisco.com/c/en/us/solutions/collateral/ser- viceprovider/visual-networking-index-vni/white-paper- c11-738429.pdf
  9. Foukas X et al (2017) Network slicing in 5G: survey and challenges. IEEE Commun Mag 55(5):94–100CrossRefGoogle Scholar
  10. Holtkamp H et al (2014) Minimizing base station power consumption. IEEE J Selected Areas Commun 32(2):297–306CrossRefGoogle Scholar
  11. I C et al (2014) Toward green and soft: a 5G perspective. IEEE Commun Mag 52(2):66–73Google Scholar
  12. Rost P et al (2017) Network slicing to enable scalability and flexibility in 5G mobile networks. IEEE Commun Mag 55(5):72–79CrossRefGoogle Scholar
  13. Saxena N, Roy A, Kim H (2016) Traffic-aware cloud RAN: a key for green 5G networks. IEEE J Select Areas Commun 34(4):1010–1021CrossRefGoogle Scholar
  14. Wang K, Zhou W, Mao S (2017) On joint BBU/RRH resource allocation in heterogeneous cloud-RANs. IEEE Internet Things J 4(3):749–759CrossRefGoogle Scholar
  15. Yao J, Ansari N (2018) QoS-aware joint BBU-RRH mapping and user association in cloud-RANs. IEEE Trans Green Commun Netw 2(4):881–889CrossRefGoogle Scholar
  16. Zhang H et al (2017) Network slicing based 5G and future mobile networks: mobility, resource management, and challenges. IEEE Commun Mag 55(8):138–145CrossRefGoogle Scholar
  17. Zhou X et al (2016) Network slicing as a service: enabling enterprises’ own software-defined cellular networks. IEEE Commun Mag 54(7):146–153CrossRefGoogle Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  1. 1.Information Systems Architecture Science Research DivisionNational Institute of InformaticsTokyoJapan

Section editors and affiliations

  • Yusheng Ji
    • 1
  1. 1.Information Systems Architecture Research DivisionNational Institute of Informatics, JapanTokyoJapan