Integrating Wireless BBUs with Optical OFDMFlexible-Grid Transponders in a C-RAN Architecture

Avishek Nag, Yi Zhang, Luiz A. DaSilva, Linda Doyle, and Marco RuffiniCONNECT Centre, University of Dublin, Trinity College, Ireland

naga@tcd.ie, zhangy8@tcd.ie, dasilval@tcd.ie, linda.doyle@tcd.ie, marco.ruffini@tcd.ie

Abstract: We propose a case study on hardware-level virtualisation of C-RAN BBUs and opticalflex-grid OFDM transponders, showing cost savings of integrating fixed and mobile network devicesin a realistic converged network scenario.c© 2017 Optical Society of America

OCIS codes: (060.4256) Networks, network optimisation; (060.4258) Networks, network topology

1. Introduction

Traditional radio-access networks (RAN) are evolving towards cloud-RAN (CRAN) by pushing the digital basebandunits (BBUs) far way from (up to 40 km for latency requirements) their remote radio heads (RRHs) towards a cloud-based BBU pool [1, 2]. At the same time, there has been a recent push towards consolidating the network by mergingthe access and metro transport network through optical access technologies with longer reach (40 km and above) toreduce cost and power consumption [3]. However, this trend becomes questionable in the context of C-RAN. This isbecause longer reach fiber-based access [3] may conflict with the strict latency requirements of C-RAN. Furthermore,there is also a trend towards flex-grid optical transport networks to handle the interconnection between BBU clouds[2]. With these technology trends in view, this paper proposes a network architecture that combines the concepts ofmetro/access node consolidation, C-RAN and flex-grid transmission. The architecture is shown in Fig. 1, where a long-reach access network directly serves an optical OFDM-based flex-grid core network that has consolidated metro-corecentral offices (MCCO). Note that the MCCO will only house those BBUs for which the RRHs are within 40 km ofthe MCCO. For all other RRHs, the corresponding BBUs will be housed in the remote node (RN): these remote nodeshousing BBUs are called active remote nodes (ARN).

MCCO

MCCO(with BBU)

MCCO

MCCO

MCCO

OXC

RN

RN

ARN(with BBU)

Fiber link

Up to 10 kmLess than 30 km

ARN(with BBU)

More than 30 kmUp to 10 km

RRH

Flexible-gridoptical network

PON

PONARN

(with BBU)

More than 30 km Up to 10 km

Fig. 1. Network Architecture.

RRH 3

Integratable BBU (at MCCO)

FFT Channel Estimation

Frequency Offset Compensation

IFFTAdding Cyclic Prefix

ADC

Removing Cyclic Prefix

RF

Antenna

Para

llel t

o Se

rial

DAC

DAC

I

Q

MZM

MZM90°

PON and CPRI

Optical OFDM Integratable Transponder (at MCCO)

RRH 1

ADC RF

Antenna

Optical OFDM digital processing

Wireless BBU pool

RLCPDCP MAC

RRH 2

ADCRF

CoreNetworks

BBU 1

BBU 2

BBU 3

Antenna

GPP

ASIC

Fig. 2. MCCO Architecture (Uplink).

When we place some of the BBUs in the MCCO, we can go a step further and integrate some of the hardwarefunctionalities of the BBU pool and the OFDM optical transponders in the MCCO using hardware-level virtualisation.Therefore, the MCCO can house programmable general purpose processors (GPPs) running common processing units[5] of the BBU pool and flex-grid transponder pools, and other standalone functions on application specific integratedcircuits (ASICs) which can coordinate with each other to perform all the hardware functions of the BBU pool andthe optical transponder pools. Figure 2 depicts functional blocks of the MCCO, with yellow blocks being the GPP-implemented blocks and the pink blocks being the ASIC-implemented blocks. This hardware-level virtualisation canreduce hardware complexities and lead to CapEx reduction.

The motivation for such an architecture comes from the concept of network function virtualisation (NFV). Byfacilitating the network functions to run on distributed hardware, NFV allows the hardware itself to be used in a time-shared manner for multiple device-level functions. In fact, hardware-level virtualisation of optical transponders has

been recently reported in [4]. The main contributions of this paper are: (a) A modified C-RAN architecture with co-location of some BBUs all the way in the MCCOs with flex-grid transponders; and (b) cost optimisation of hardware-level virtualisation, to study how the CapEx is affected by integrating BBUs and transponders in the MCCO.

The first-stage optimisation problem uses the location of the RNs (with the RRHs connected to them) and optimallyplaces the MCCOs that serve all the RNs, minimising fiber costs. The second stage then integrates the BBUs for thenearer RRHs with the optical transponders in the MCCO and solves a traffic engineering problem that efficiently (min-imising spectrum and cost) maps the mobile traffic to the optical channels corresponding to two types of transponders:(1) integrated with BBUs and (2) non-integrated (Fig. 2). Our results show that there are significant CapEx savings byhaving MCCO-integrateable BBUs.

2. Problem Description and Formulation

The first stage problem of placing the optical nodes and determining the topology is solved according to modelsdescribed in [6, 7]. With the inputs from the first stage solutions we identify the set Si, the set of RNs which are within30 km of the ith MCCO. We then solve a multi-layer (mapping traffic flows to optical channels and optical channels tophysical fibers) optimisation problem, the output of which gives: (a) A minimised cost of transponders in the network,(b) the set of transponders and their bandwidth that can be integrated and cannot be integrated with the BBUs in theMCCO, and (c) A virtual topology in terms of routing and capacity assignment on top of the physical core networktopology. The complete integer linear program (ILP) is described in Tables 1 and 2. The objective function minimisesthe total cost of transponders. Constraints (1)-(2) are flow conservation and virtual topology assignment constraintsthat map the traffic flow on optical channels or lightpaths and the lightpaths on the physical links. Constraints (3)-(4)count the number of integrateable transponder ports. Constraint (5) ensures all the integrateable transponder ports in anode should match in terms of their capacities with the total capacities of all the integrateable BBUs. Constraints (6)and (7) are the logical and physical capacity constraints respectively.

Table 1. Variables and Constants- Si: Set of intermediate COs which are within 30 km ofa core node, for each node i.- NSi

: Number of RRHs served by the RNs in Si.- degi: Nodal degree of node i.- r: Sustained data rate per RRH.- N , E: Set of nodes and edges, respectively, in the net-work.- x(t)

ijk: Variable denoting a lightpath from node i to nodej of rate Rk supporting traffic from type-t BBUs.- f

sd(t)ij : Traffic flow from source node s to destination

node d, employing the lightpath from node i to node jfor traffic originating from type-t BBUs.- FGB: Filter guard band, in GHz.- Fmn: Number of fibers on link (m,n).

- BW , BWk: Capacity of each fiber and spectrum re-quired to support data rate Rk, respectively.- Λ

(t)sd : Traffic demand aggregated from type-t-BBUs

from node s to node d [t = 1: MCCO-BBUs; t = 2:ARN-BBUs].- Ck

INT , CkNINT : Unit costs of a transponders of rate Rk

that can be integrated or not integrated with a MCCOBBU, respectively.- T s

ik, Tdjk: Variables which counts the number of

transponder ports in source node s and destination noded at rate Rk.-P ijk(t)

mn : Variable denoting whether the lightpath fromnodes i to j at rate Rk, is routed through fiber link (m,n)for type-t traffic.

Table 2. Complete ILP Formulation

Minimize:∑

k

∑i

∑j C

kINT · x

(t=1)ijk +

∑k

∑i

∑j C

kNINT · x

(t=2)ijk

subject to:

(1)∑

j fsd(t)ij −

∑j f

sd(t)ji =

{ Λ(t)sd , if i = s

−Λ(t)sd , if i = d

0, otherwise(3) T s

ik ≥∑

j x(t=1)ijk , ∀s = i,∀k

(4) T djk ≥

∑i x

(t=1)ijk , ∀d = j, ∀k

(2)∑

n Pijk(t)mn −

∑n P

ijk(t)nm =

{ x(t)ijk, if m = i

−x(t)ijk, if m = j

0, otherwise(5)

∑k Rk · T s

ik +∑

k Rk · T dik = degi · r ·NSi

∀s = i,∀d = i.(6)

∑s

∑d f

sd(t)ij ≤

∑k x

(t)ijk ·Rk ∀i, j ∈ N,∀t

(7)∑

t

∑k(P

ijk(t)mn ·BWk) + (FGB ·

∑t

∑k P

ijk(t)mn ) ≤ BW · Fmn ∀i, j ∈ N, ∀(m,n) ∈ E

3. Results and Discussion

We run our optimisations in CPLEX with 1121 RN positions obtained from a real dataset for Ireland obtained froma service provider. The dataset consists of geographical locations of the RNs along with the number of customersthey serve. The RNs are served by 18 MCCOs and each of these MCCOs are connected in a mesh topology with 29

bi-directional links generated using models in [7]. We assume 3 types of optical-OFDM transponders (100G, 400G,and 1000G) each occupying a spectrum of 50, 75, and 175 GHz and relative cost of 1.44, 1.76, and 2 units respectively(assuming that technology enhancement by 2018 will enable such volume discount)[8]. The total fiber spectrum isassumed to be 4000 GHz (12.5 GHz ×320 slots) and the filter guard band FGB to be 12.5 GHz. The sustaineddata rate per RRH is assumed to be 50 Mbps and 100 Mbps for two reference cases of traffic volumes. The traffic isaggregated for all RRHs served by MCCO node i and then the inter-nodal traffic matrix is generated using the gravitymodel [9]. We also test our model with different values of discount obtained due to integration of BBUs with MCCOtransponders, from 10% to 50%. The results are presented in Figs. 3 and 4 which plot the CapEx normalised to thetransponder cost units and CapEx savings, respectively, with different discounts due to MCCO integration comparedto the benchmark case with no BBUs in the MCCO. We can observe from Fig. 3, as expected, that the overall CapExdecreases starting from the benchmark case where no BBUs are put in the MCCO to the cases where BBUs areintegrated in the MCCO with different discount values. The slope of the CapEx is steeper with higher traffic volume(i.e., with 100 Mbps per RRH). Fig. 4 shows the percentage savings in CapEx over the benchmark case (all BBUs inARN) and we find that the CapEx savings have heavy dependency on the discounts obtained by integrating BBUs withoptical OFDM transponders.

Cap

Ex N

orm

alis

ed to

Tra

nspo

nder

U

nit C

osts

2000

3600

5200

6800

8400

10000

No

MC

CO

With

MC

CO

, dis

c =

10%

With

MC

CO

, dis

c =

15%

With

MC

CO

, dis

c =

20%

With

MC

CO

, dis

c =

25%

With

MC

CO

, dis

c =

30%

With

MC

CO

, dis

c =

35%

With

MC

CO

, dis

c =

40%

With

MC

CO

, dis

c =

45%

With

MC

CO

, dis

c =

50%

50 Mbps per RRH 100 Mbps per RRH

Fig. 3. CapEx for different cases.

% C

apEx

Sav

ings

0

4.5

9

13.5

18

With

MC

CO

, dis

c =

10%

With

MC

CO

, dis

c =

15%

With

MC

CO

, dis

c =

20%

With

MC

CO

, dis

c =

25%

With

MC

CO

, dis

c =

30%

With

MC

CO

, dis

c =

35%

With

MC

CO

, dis

c =

40%

With

MC

CO

, dis

c =

45%

With

MC

CO

, dis

c =

50%

50 Mbps per RRH 100 Mbps per RRH

Fig. 4. CapEx Savings by putting BBUs in MCCO compared tobenchmark case of all BBUs in ARN.

While it is evident that the percentage of BBUs that can be placed in the MCCOs is small (7.5% in our casestudy, obtained from the first-stage optimisation), these BBUs mostly correspond to very dense urban RRHs and theaggregated traffic they serve is higher in percentage compared to the percentage of BBUs (19% of overall traffic forour case). Therefore, integrating these BBUs with the flex-grid transponders can save up to 16% of overall CapEx.

4. Acknowledgments

This publication has emanated from research supported in part by a research grant from Science Foundation Ireland (SFI) andis co-funded under the European Regional Development Fund under Grant Number 13/RC/2077 (CONNECT), Grant Number14/IA/2527 (OSHARE) and also European Union’s Horizon 2020 for research, technological development, and demonstrationunder grant agreement no. 688941 (FUTEBOL), as well from the Brazilian Ministry of Science, Technology and Innovation (MCTI)through RNP and CTIC.

References1. F. Musumeci et al., “Optimal BBU Placement for 5G C-RAN Deployment Over WDM Aggregation Networks," IEEE/OSA J. Lightwave

Technology, vol. 34, no. 8, Apr. 2016.2. J. Zhang et al., “Baseband Unit Cloud Interconnection Enabled by Flexible Grid Optical Networks with Software Defined Elasticity," IEEE

Communication Magazine, Sept. 2015.3. M. Ruffini et al., “DISCUS: An End-to-End Solution for Ubiquitous Broadband Optical Access," IEEE Communication Magazine, Feb. 2014.4. M. Jinno et al., “Virtualisation in Optical Networks from Network Level to Hardware Level [Invited]," IEEE/OSA J. Opt. Commun. Netw.,

vol. 5, no. 10, Oct. 2013.5. Chih-Lin I. et al., “Recent Progress on C-RAN Centralization and Cloudification," IEEE Access, Sept. 2014.6. A. Nag et al., “N:1 Protection Design for Minimizing OLTs in Resilient Dual-Homed Long-Reach Passive Optical Network," IEEE/OSA

Journal of Optical Communication and Networking, February, 2016.7. DISCUS, Deliverable 7.6, Report on the Core Network optimisation and Resiliencey Strategies, Technical report, The DISCUS Project (FP7

Grant 318137), 2015.8. FP7 IDEASLIST D1.1 “Elastic Optical Network Architecture: Reference Scenario, Cost and Planning," June 2013.9. P. Tune and M. Roughan, “Internet Traffic Matrices: A Primer", in H.Haddadi, O. Bonaventure (Eds.), Recent Advances in Networking,

(2013).