LabVIEW based Platform for prototyping dense LTE

Networks in CROWD Project

Rohit Gupta and Thomas Vogel

National Instruments

Dresden, Germany

Email: {rohit.gupta, thomas.vogel}@ni.com

Nikhil Kundargi and Amal Ekbal

National Instruments

Austin, USA

Email: {nikhil.kundargi, amal.ekbal}@ni.com

Arianna Morelli

INTECS

Pisa, Italy

Email:arianna.morelli@intecs.it

Vincenzo Mancuso and Vincenzo Sciancalepore

Institute IMDEA Networks

Madrid, Spain

Email:{vincenzo.mancuso, vincenzo.sciancalepore}@imdea.org

Russell Ford and Sundeep Rangan

NYU Wireless

New York, USA

Email:{russell.ford, srangan}@nyu.edu

Abstract—Next generation wireless networks (5G) have to

cope with significant traffic increase due to high quality video

transmission and cloud-based applications. Such requirements

create the need for a revolutionary change in architecture rather

than a series of local and incremental technology updates. A

dense heterogeneous deployment of small cells such as pico/femto

cells in addition to high power macro cells is foreseen as one

of the potential solutions to achieve these requirements. While

there is significant amount of research in this area that relies on

simulations at PHY, MAC and higher layers, it is still necessary to

validate the algorithms for next generation systems in a real-time

testbed. However, the ever increasing complexity in all layers of

current and future generations of cellular wireless systems has

made an end-to-end demonstration of the network limited to

industrial research labs or large academic institutions. In this

paper, we show a LabVIEW1 based PXI platform in which LTE-

like SISO OFDM PHY Layer is integrated with an open source

protocol stack to prototype PHY/MAC cross layer algorithms

within CROWD2 Software Defined Networking (SDN) framework

as a solution to tame dense deployment of wireless networks.

I. INTRODUCTION

Mobile data traffic demand is growing exponentially andthe trend is expected to continue for the near future, es-pecially with the deployment of 4G networks. In order tocope with such rapid explosion of traffic demand, mobilenetwork operators have already started to push for denser,heterogeneous deployments. Current technology needs to steertowards efficiency to avoid unsustainable energy consumptionand network performance collapse due to interference andsignaling overhead. In fact, interference due to uncoordinatedresource sharing techniques represents a key limiting factorin the design of dense wireless networks, where resourcesare limited due to either the costs for licensed bands orthe proliferation of hot spots in license-exempt bands. Thissituation calls for the deployment of network controllers witheither local or regional scope, with the aim of orchestrating

1LabVIEW is graphical system design tool for prototyping and is a

trademark of National Instruments Corp.2CROWD is an EU FP7 Project that stands for ’Connectivity management

for eneRgy Optimised Wireless Dense networks’

the access to wireless and backhaul resources of the variousnetwork elements. In the FP7 CROWD project [1] we showhow a Software-Defined Networking (SDN) based approachcan be suitably adopted to design the next generation of densewireless mobile networks. This approach enables the requiredlevel of flexibility and reconfigurability, while at the same timeproviding energy-efficient network infrastructure for both theradio access and the backhaul.

In order to demonstrate the performance of PHY/MACcross layer algorithms within a testbed, we utilize NationalInstruments’ (NI) PXI-based FlexRIO SDR platform for rapidprototyping [2]. This SDR platform provides a rich heteroge-neous environment including multi-core Windows/Linux PCand real-time operating system (RTOS) running on high-performance general purpose processors (GPP) such as Intelprocessors and FlexRIO FPGA modules containing XilinxVirtex-5 and Kintex-7 FPGAs. It also provides a rich set ofRF, Digital to Analog Converter (DAC) and Analog to DigitalConverter (ADC) modules that can meet the bandwidth andsignal quality requirements of 5G systems.

SDR prototyping engineer faces several challenges whicharise due to the use of different software design flows toaddress different components of the system (i.e., RF, baseband,and protocol stack). In addition, the components may lack acommon abstraction layer. This can result in complicationsand delays during system development and integration. TheNI LabVIEW graphical system design software is able toaddress these challenges by providing a common developmentenvironment for all the heterogeneous elements in the NISDR system (i.e., the GPP, RTOS, FPGA, converters and RFcomponents), with tight hardware/software integration and agood abstraction layer [3]. The environment is also compatiblewith other software design tools/languages such as VHDL,C/C++ etc. This integrated design environment is the primaryreason we chose this SDR platform for prototyping and itenabled us to quickly reach an initial working version of ourdemonstration system and rapidly iterate on that design.

In this paper, we show preliminary integration results onhow an LTE based PHY layer is integrated with the opensource NS3 LTE LENA stack [4]. We also propose an initialconcept on how this testbed can be used to demonstrate en-

Fig. 1: Network control architecture.

hanced Inter-cell Interference Coordination (eICIC) algorithm,for example Almost Blank Sub-Frame (ABSF) [5], which willbe fine-tuned using SDN-based approach in the later stageof the project. However, the implementation of this testbedis generic enough to prototype small-cell dense LTE networkwithin lab environment and demonstrate the performance ofeICIC algorithms.

II. CROWD ARCHITECTURE

The proposed CROWD architecture [1] aims at leveragingthe heterogeneity of dense wireless deployments, both interms of radio condition and non-homogeneous technologies.It offers tools to orchestrate the network elements to mitigateintra-system interference, improve performance of channel-opportunistic transmission/reception techniques, and reduceenergy consumption. An extremely dense and heterogeneousnetwork deployment comprises two domains of physical net-work elements: backhaul and Radio Access Network (RAN).The latter is expected to become increasingly heterogeneousnot only in terms of technologies (e.g., 3G, LTE, WiFi),cell ranges (e.g., macro-/pico-/femto-cells), but also at densitylevels (e.g., from macro-cell Base-Station (BS) coverage inunderpopulated areas to several tens or hundreds of potentiallyreachable BS in hot spots). Such heterogeneity also createshigh traffic variability over time due to statistical multiplexing,mobility of users, and variable-rate applications. In order toachieve optimal performance most of the times, reconfigurationof the network element is required at different time intervalsfrom very fast (few 10s of milliseconds) to relatively long (fewhours), affecting the design of backhaul and the RAN compo-nents. In order to tackle the complex problem of reconfigu-ration, we propose to follow an SDN-based approach for themanagement of network elements as shown in Fig. 1. Networkoptimisation in the proposed architecture is assigned to a setof controllers, which are virtual entities deployed dynamicallyover the physical devices. These controllers are technology-agnostic and vendor-independent, which allow full exploitationof the diversity of deployment/equipment characteristics. Theyexpose a northbound interface, which is an open Application

Fig. 2: Testbed architecture

Program Interface (API) to the control applications. We definecontrol applications as the algorithm that actually performsthe optimization of network elements, for example AlmostBlank Sub-Frame (ABSF). The northbound interface doesnot need be concerned with either the details of the dataacquisition from the network or the enforcement of decisions.Instead, the southbound interface is responsible for managingthe interaction between controllers and network elements.

We propose two types of controllers in the network (seeFig. 1): the CROWD Regional Controller (CRC), which is alogical centralized entity that executes long-term optimisations,and the CROWD Local Controller (CLC), which runs short-term optimizations. The CRC only requires aggregate datafrom the network, and is in charge of the dynamic deploymentand life cycle management of the CLC. The CLC requiresdata from the network at a more granular time scale. For thisreason, CLC only covers a limited number of Base Stations [1].The CLC can be hosted by a backhaul/RAN node itself, e.g.,a macro-cell BS, so as to keep the optimisation intelligenceclose to the network. On the other hand, the CRC is likelyto run on dedicated hardware in network operator data centre.Such an SDN-based architecture allows the freedom to runmany control applications which can fine tune the networkoperation with different optimization criteria, e.g. capacity,energy efficiency, etc. The CROWD vision aims to providea common set of functions as part of southbound interfacewhich can be used by the control applications, for exampleLTE access selection [6] and LTE interference mitigation [7]to configure network elements of a dense deployment.

III. TESTBED ARCHITECTURE

Fig. 2 shows the general overview of testbed architecture.The functions of the MAC and higher-layer protocols (in-cluding CLC) run on a Linux computer. The protocol stackcommunicates with PHY layer running on NI/PXI system

TABLE I: LTE-like OFDM system parameters

Parameter Value

Sub-carrier Spacing (∆f) 15 kHz

FFT Size (N) 2048

Cyclic Prefix (CP) length (Ng) 512 samples

Sampling frequency (Fs) 30.72 MS/s

Bandwidth 1.4, 3, 5, 10, 15, 20 MHz

Number of used sub-carriers 72,180,300,600,900,1200

Pilots/Reference Symbols (RS) spacing Uniform (6 sub-carriers)

over Ethernet using an L1-L2 API that is based on SmallCell Forum API [8]. We have implemented the complexhigh throughput baseband signal processing for an “LTE-like” OFDM transceiver for the eNB and UE in LabVIEWFPGA using several FlexRIO FPGA modules due to the highthroughput requirements. We use NI 5791 FlexRIO AdapterModule (FAM) as RF transceiver. This module has continuousfrequency coverage from 200 MHz to 4.4 GHz and 100 MHzof instantaneous bandwidth on both TX and RX chains. It fea-tures a single-stage, direct conversion architecture, providinghigh bandwidth in the small form factor of an NI FAM.

A. Introduction to LabVIEW based LTE-like PHY

The current PHY implementation has only one antenna port(i.e., SISO) supported per node with FDD operation. We havechosen to implement only the downlink transmitter/receiverand we plan to show the performance of our algorithms indownlink direction. The PHY modules have been designedto loosely follow 3GPP specifications, and hence referredto as “LTE-like” system, since our testbed is intended forresearch instead of commercial development. We describemain LTE OFDMA downlink system parameters in Table I.However, some components and procedures of a commercialLTE transceiver (for example, random access and broadcastchannel) have deliberately been omitted because they fall out-side the scope and requirements of our testbed. Only essentialdata and control channel functions are implemented.

We use Xilinx Coregen library for the channel coding,FFT/IFFT and filter blocks, while custom algorithms havebeen developed in LabVIEW for all the other blocks. Onthe transmitter side (see Fig. 3), Physical Downlink SharedChannel (PDSCH) and Physical Downlink Control Channel(PDCCH) Transport Blocks (TB) are transferred from theMAC layer and processed by each subsystem block as theyare synchronously streamed through the system. Handshakingand synchronization logic between each subsystem coordinateeach module’s operations on the stream of data. The fields ofthe Downlink Control Information (DCI), including parametersspecifying the Modulation and Coding Scheme (MCS) andResource Block (RB) mapping, are generated by the MAClayer and provided to the respective blocks for controlling thedata channel processing. Once the PDCCH and PDSCH dataare appropriately encoded, scrambled and modulated, they arefed to the Resource Element (RE) mapper to be multiplexedwith RSs and Primary and Secondary Synchronization Se-quences (PSS/SSS), which are stored in static look-up tableson the FPGA. Presently, the Resource Block (RB) pattern isfixed and supports only one user. However, multi-user anddynamic resource allocation will be included in later versions.OFDM symbols are generated as shown in Fig. 3 and convertedto analog for transmission over-the-air by the NI 5791 RFtransceiver.

Resample

PDSCH

buffer

Modulation

mapping

PSS/SSS

LUT

RE map

IFFT + CP

insertionNI 5791 FAM

PDCCH

buffer

LabVIEW FPGALabVIEW FPGA

LabVIEW RT

Turbo

encoding

Convolutional

encoding

Pilot

LUT

from MAC

Scrambling

Modulation

mappingScrambling

DMA

DMA

Fig. 3: LTE-DL transmitter FPGA block diagram.

Resample

CP removal +

OFDM Symbol

segment

CFO

CompensationFFT

Pilot

Extraction

Channel

EstimationEqualization

PSS/SSS

Sync

Mod symbol

extractionto Decoder

Module

NI-5791 FAM

P2P FIFO

Fig. 4: LTE-DL receiver FPGA block diagram.

SCH turbo

decode

LLR

Calculation

SCH/CCH

RE extraction

CCH decode +

unscramble

CRC

DCI

parse

HARQ

buffers

to MAC

Coded

CCEs

UE-specific

CCEs DCI

LLR values

Coded data

REs Decoded TBs

DMA

DMA

PDCCH

buffer

ded TBs

PDSCH

buffer

P2P FIFO

from RX

Fig. 5: LTE-DL PDCCH and PDSCH decoder FPGA block

diagram.

Fig. 4 shows the high-level block diagram of our OFDMreceiver implementation. The NI 5791 RF transceiver receivesthe analog signal and converts it to digital samples for process-ing by the FPGA. This is followed by time synchronizationbased on the LTE PSS/SSS. The cyclic prefix (CP) is thenremoved, OFDM symbols are segmented out and the CarrierFrequency Offset (CFO) Compensation module corrects forCFO impairments. Fast-Fourier Transform (FFT) is then per-formed on the samples and reference symbols are extracted forchannel estimation and equalization. The equalized modulationsymbols for the data and control channels are then fed to aseparate decoder FPGA, which is connected to the receiverFPGA using a Peer-to-Peer (P2P) stream over the PCI Expressbackplane of the PXI chassis. Fig. 5 shows the implementa-tion of the downlink channel decoder. The first stage of thedecoding process is to demultiplex the symbols belonging tothe PDSCH and PDCCH. The Downlink Control Information(DCI) is then decoded from the PDCCH Control ChannelElements (CCEs) and passed to the SCH turbo decoder moduleto decode the PDSCH data, which are finally sent to the MACusing the Small Cell Forum API [8].

B. Introduction to Open Source NS3 LENA LTE Stack

We have adopted the NS3 LTE LENA simulator [4] as theframework for implementing the upper-layer LTE stack for ourtestbed platform. We leverage the protocols and proceduresprovided in the robust LTE LENA model library, which aregenerally compliant with 3GPP standards. Fairly completeimplementations of the data-plane MAC, RLC, and PDCPprotocols are provided along with simplified versions of thecontrol-plane RRC and S1-AP protocols. Some core networkprotocols such as GTP-U/GTP-C and the S1 interface are

Fig. 6: Almost Blank Sub-Frame (ABSF) overview.

partially implemented, as well. Though used primarily byresearchers as a discrete-event network simulator, NS3 canalso be configured to function as a real-time network emulatorand interfaced with external hardware. By creating instancesof LTE nodes on top of the NS3 real-time scheduler, we caneffectively emulate the eNB and UE stack. Also, by utilizingthe provided Message Passing Interface (MPI) support, wecan exploit parallelism to boost the performance of NS3. Byrunning each node instance in a separate thread or process, wecan conceivably emulate multiple eNB, UE and core networknodes all on the same multi-core host, as represented in Fig.2.

C. ABSF as an example CLC Application

A major drawback of the LTE multi-cellular system is inter-cell interference. A very promising mechanism to cope withthis is called Almost Blank SubFrame (ABSF) [9]. ABSFas shown in Fig. 6 mitigates the inter-cell interference byassigning resources such that some base stations produce blanksubframes, thus preventing their activity when the interfer-ence exceeds a threshold. Several centralized and distributedsolutions have been already suggested in the literature toexploit the ABSF mechanism. Our proposal is designed tobe implemented in the CLC, which is in charge of acquiringthe user channel conditions and computing an optimal basestation scheduling pattern in the available subframes usingSDN framework.

We provide here a brief overview of the proposed ABSFalgorithm, called Base-Stations Blanking (BSB) [5], which hasbeen designed for a particular scenario: content distributionover the whole network, such as road traffic or map updates.Specifically, the algorithm takes care of base station schedulingduring the content injection phase. The base stations delegatea few mobile users to carry and spread content updates to theother users, through the device-to-device (D2D) interface, dur-ing the content dissemination phase. The content disseminationphase is deemed outside the scope of the algorithm, given thatin a dense scenario, the content injection phase becomes themost critical part. The algorithm exploits the ABSF techniqueto minimize the time required by the base stations to achievehigh spectral efficiency when performing packet transmissionsduring the injection phase.

The BSB algorithm provides a valid ABSF pattern byguaranteeing a minimum SINR for any user that might bescheduled in the system. The algorithm is implemented asa part of the control application LTE eICIC running on theCLC. The CLC filters per-user data and provides global per-eNB statistics to the control application. The CLC queries thedownlink channel state information of any user through its

Fig. 7: Testbed Setup for eNB-UE downlink scenario.

serving LTE eNB. Based on the collected user channel stateinformation, the application computes the appropriate ABSFpatterns for each eNB, and sends them back to the CLC, whichdispatches the patterns to eNBs via the southbound interfacerunning on top of the standard X2 interface.

IV. PRELIMINARY RESULTS

We illustrate here the initial results of our progress re-garding the testbed integration and the simulation-based eval-uation of ABSF algorithms since work began with the NIPXI/LabVIEW platform. Figure 7 shows the overall setup ofour proposed prototype which contains SISO OFDM downlink(eNB + UE). There are five basic elements in our testbed, asdescribed below:

1) Laptop running LabVIEW for developing the SISOOFDM/LTE transmitter/receiver and also deployingthe PHY layer code to the PXI systems.

2) Laptop running the NS3 eNB protocol stack in Linux3) PXI System running the eNB transmitter baseband

PHY on Virtex5/Kintex 7 FPGAs with a NI 5791front-end module functioning as the transmitter DACand RF upconverter.

4) PXI System running eNB receiver baseband PHY onVirtex5/Kintex 7 FPGAs with a NI 5791 FAM usedas the downconverter and ADC.

5) Laptop running the NS3 UE protocol stack in Linux

In the future, we will have several more PXI systems toemulate multiple eNBs/UEs in a dense wireless lab networksetting to demonstrate the performance of PHY/MAC crosslayer interference mitigation algorithms like ABSF within theCROWD framework.

A. Testbed Integration

We currently have a working, validated implementationof the LTE PDSCH channel running on the testbed. Fig. 8shows a screen shot of the LabVIEW Virtual Instrument (VI)

Fig. 8: Snapshot of TX/RX VI running SISO OFDM LTE

PHY

running the transmitter and receiver with QPSK modulationand a fixed transport block size. We have also been successfulin adapting the NS3 LENA LTE protocol stack so that it canbe interfaced with the SISO OFDM PHY. The encoding ofNS3 MAC messages for over-the-air (OTA) transmission isproprietary and already completed. The current adaptation ofthe LENA stack supports either placing SISO OFDM PHYin a loop-back mode with eNB and UE running within asingle instance of NS3 or eNB and UE running in two separateinstances of NS3.

B. ABSF Simulations

We now illustrate some initial results of the proposed LTEeICIC application effect on the performance of D2D-assistedcontent update distribution. Let us assume that the contentsare split into several chunks. Thus, LTE eNBs adapt theirtransmission based on any single chunk. In Fig. 9 we comparethe performance of the proposed control application againstthe standard scheduling approach. The base station activitiesare depicted in the graph as function of the chunks delivered inthe network. The base stations remain active for about 10% ofthe total number of subframes, showing an efficiency gain upto 10 times against the standard scheduling approach withoutany kind of MAC enhancements. Note that the first chunksare used by the control application to reach a steady-state forthe next subframes. When network changes occur, the controlapplication will adjust the ABSF patterns accordingly.

V. CONCLUSION

In this paper, we showed how NI/PXI platform based onLabVIEW can be used for advanced wireless LTE prototypingto demonstrate end-to-end link. Since, the design of the testbedis based on open source protocol stack and modular hardwarefrom NI PXI Platform, it is much more cost effective com-pared to other commercial LTE testbed solutions, especiallyfor demonstrating research ideas for next generation wirelesssystems. The graphical system design based on LabVIEW andits integrated environment were one of the key reasons thatwe were able to achieve significant results and make goodprogress with testbed integration. We also showed preliminaryresults for testbed implementation of LTE PHY layer andits integration with NS3. The next step is to complete theimplementation of PDCCH and also the implementation of

10 20 30 400

10

20

30

40

50

60

70

80

90

100

Chunks

Base S

tation S

ubF

ram

e U

tiliz

ation [

%]

Approach with BSB

Base Station 1

Base Station 2

Base Station 3

Base Station 4

Base Station 5

10 20 30 40 500

10

20

30

40

50

60

70

80

90

100

Chunks

Base S

tation S

ubF

ram

e U

tiliz

ation [

%]

Approach without BSB

Base Station 1

Base Station 2

Base Station 3

Base Station 4

Base Station 5

Fig. 9: A comparison of subframe utilization per base station,

between the standard scheduling approach without eICIC

mechanism and the CLC approach, utilizing the BSB

algorithm.

small cell forum API, so as to allow integration of LTE PHYlayer with NS3 in real-time. In the end, we aim to demonstratethe performance of our proposed ABSF algorithms in a labsetting using multiple cells emulated on several PXI systems.

ACKNOWLEDGMENT

The research leading to these results has received fundingfrom the European Union Seventh Framework Programme(FP7/2007-2013) under grant agreement no. 318115 (Con-nectivity management for eneRgy Optimised Wireless Densenetworks - CROWD). We would like to thank Malik Gul forinitial design of SISO OFDM code in LabVIEW, which wasa very useful starting point for this project.

REFERENCES

[1] H. Ali-Ahmad, C. Cicconetti, A. de le Oliva, V. Mancuso, , M. R. Sama,

P. Seite, and S. Shanmugalingam., “SDN-based Network Architecture for

Extremely Dense Wireless Networks,” IEEE Software Defined Networks

for Future Networks and Services (IEEE SDN4FNS), November 2013.

[2] “NI FlexRIO Software Defined Radio.” [Online]. Available:

http://sine.ni.com/nips/cds/view/p/lang/en/nid/211407

[3] “Prototyping next generation wireless systems with software defined

radios.” [Online]. Available: http://www.ni.com/white-paper/14297/en/

[4] “Overview of NS3 based LTE LENA Simulator.” [Online]. Available:

http://networks.cttc.es/mobile-networks/software-tools/lena/

[5] V. Sciancalepore, V. Mancuso, A. Banchs, S. Zaks, and A. Capone,

“Interference Coordination Strategies for Content Update Dissemination

in LTE-A,” in The 33rd Annual IEEE International Conference on

Computer Communications (INFOCOM), 2014.

[6] 3GPP, “3GPP TS 24.312; Access Network Discovery and Selection

Function (ANDSF) Management Object (MO),” Tech. Rep.

[7] A. Daeinabi, K. Sandrasegaran, and X. Zhu, “Survey of intercell inter-

ference mitigation techniques in LTE downlink networks,” in Telecom-

munication Networks and Applications Conference (ATNAC), 2012 Aus-

tralasian, Nov 2012, pp. 1–6.

[8] “LTE eNB L1 API Definition, October 2010, Small Cell Forum.”

[Online]. Available: http://www.smallcellforum.org/resources-technical-

papers

[9] 3GPP, “Evolved Universal Terrestrial Radio Access Network (E-

UTRAN); X2 application protocol (X2AP), 3rd ed,” 3GPP, technical

Specification Group Radio Access Network, Tech. Rep.