Simulation Framework for MIMO LTE Network Performance Analysisjournal.kolegija.lt/iitsbe/2013/Aleksiejunas... · the spectrum overlap between LTE and radar signals and can be estimated

Aleksiejūnas et al. Simulation Framework. 9

Simulation Framework for MIMO LTE Network Performance Analysis

Rimvydas Aleksiejūnas a , Kęstutis Svirskas, Jevgenij Krivochiža and Jurgis AleksandravičiusTelecommunications Research Center, Department of Radiophysics,

Vilnius University, Sauletekio 9, bldg. III, LT-10222 Vilnius, Lithuania

Received 10 November 2013, accepted 15 December 2013

Abstract. The paper presents a cloud-based virtual server environment for simulations of MIMO LTEnetworks at Telecommunications Research Center, Vilnius University. Brief information about analyticalmodels for simulating the performance of mobile wireless network is given including antenna analysis,radio propagation channel and radio interference estimation problems. Architecture of software frameworkis discussed, illustrated by LTE network simulation scenario.

Citations: Rimvydas Aleksiejūnas, Kęstutis Svirskas, Jevgenij Krivochiža and Jurgis Aleksandravi-čius. Simulation Framework for MIMO LTE Network Performance Analysis – Innovative Infotechnologiesfor Science, Business and Education, ISSN 2029-1035 – 2(15) 2013 – Pp. 9-13.

Keywords: mobile wireless network; 4G LTE.Short title: Simulation Framework.

Introduction

Widely spread nature of today’s telecommunication networksprovides not only extensive possibilities and many advan-ces in various fields of human life, but also requires cons-tant development of new technologies to support growing de-mand for communications capacity. Telecommunications Re-search Center has been founded at Faculty of Physics, VilniusUniversity in collaboration with telecommunication industrycompanies Huawei technologies Co. Ltd., Omnitel Ltd. andBlue Bridge Ltd. to provide an experimental testbed for re-search and adoption of new telecommunication technologies.

Complexity of wireless network research problems re-quires close comparison of experimental measurements withnumerical simulations. Measurements conducted in isolatedin-lab base station can be extrapolated to network-wide multi-cellular environment. High accuracy of numerical results canbe achieved using long running statistical Monte Carlo simu-lations. For this purpose a cloud-based virtual server environ-ment has been developed allowing to run massive simulationsand interactively share results between team members.

The most important questions under the study at Telecom-munication Research Center are related to 4G LTE (LongTerm Evolution) mobile wireless networks that are current-ly entering the commercial market [1]. Introduction of newnetwork technologies always depends on many technologic-al and economical factors such as limited frequency resourceavailability, the cost and operational conditions of sophistica-ted radio equipment and thorough network planning and opti-mization process. Efficient analytical techniques are requiredfor planning and analysing the wireless networks.

1. Analysis Methods

In order to build reliable analytical models reflecting opera-tion of real mobile networks, multiple effects should be takeninto account as presented below.

1. Transmitting and receiving antenna characteristics (ra-diation patterns and frequency response). In multiple-input multiple-output (MIMO) antenna configurations,mutual coupling between individual antenna elementsbecomes a limiting factor on the large scale MIMO ap-plications and should be taken into account [2].

2. Correlation between MIMO antennas sets the limits formaximum data throughput available for a given trans-mitting and receiving antenna configuration. For thecorrelation analysis, antennas should be modeled not asisolated entities, but in relation to the radio propagationchannel effects such as multipath reflections and scat-tering. The most important radio channel realizationsfor 4G applications have been collected by Europeaninitiated WINNER, WINNER II and WINNER+ pro-jects to define 3GPP related MIMO channel properties- see Refs. [3-5]. Statistically processed results of thesemeasurement campaigns have been used for numericalsimulations of radio propagation effects.

3. Radio interference analysis is another important factorin modern capacity-limited networks operating in densemulti-cell configurations [6-7]. Two basic types of in-terference are considered in this paper, namely, intra-system interference between neighboring LTE cells andexternal interference arising from other radio communi-cation systems, deployed in adjacent frequency bands.

aCorresponding author, email: [email protected]

Innovative Infotechnologies for Science, Business and Education, ISSN 2029-1035 – Vol. 2(15) 2013 – Pp. 9-13.


As an example, the impact of pulsed radar signal interfer-ence on LTE network performance is discussed here.

To fully describe the effects of propagation and interfer-ence on wireless system performance, detailed simulations ofradio modulation, multi-user scheduling and multiple accessscheme should be included into analysis.

The following subsections provide short overview of math-ematical models used in LTE network simulations.

1.1. MIMO Channel Capacity

MIMO antenna systems were introduced with 4G wirelesscommunications and remain promising candidates for up-coming 5G radio technologies [8]. The main goal of MI-MO antennas is to increase data transmission throughput viathe same radio channel bandwidth while at the same timeensuring high quality of service to multiple mobile users.Achieving high MIMO performance requires knowledge ofstatistical properties of radio propagation channel and optimi-zation of transmitter-receiver antennas. By using numericalmethods antenna radiation patterns can be adjusted accordingto the spatial distribution of mobile users.

For MIMO channel model, described by matrix-type equ-ation [9]

y = Hx + n, (1)

where y and x are columns of sizes Nr and Nt, respectively,H is Nr × Nt matrix with Nr and Nt denoting the numberof receive and transmit antennas. The noise can be expressedas

n = σ2nINr

(2)

where σ2n is the variance of additive white Gaussian noise

(AWGN) and INris an identity matrix of the size Nr. The

average signal-to-noise ratio SNR

SNR = P0

σ2n

=E[x2]σ2

n

(3)

where P0 = E[x2] is an average received signal power.

Shannon capacity for MIMO channels is defined by multi-plexing gain - the numberR of independent parallel channels.MIMO channel capacity using modified Shannon’s mutualinformation C of all input covariant matrices Rx:

Rx = E[xxH

], (4)

is expressed as

C = BW · log2 det[INr

+ HRxHH], (5)

where BW is the system bandwidth.For equal power allocation to all transmit antennas,

Rx = P0

NtINt

(6)

and capacity expression becomes

C = BW · log2 det[INr + SNR

NtHHH

]. (7)

Using singular value decomposition, channel transfer mat-rix H can be diagonalized as

H = UDVH , (8)

where

D = diag(√

λ1,√λ2, ...,

√λm, 0, 0

)(9)

and U and V are unitary matrices. Continuing similar ana-lysis to MIMO channels with available channel state infor-mation (CSI) at the transmitter, capacity optimization can beachieved using transmit pre-coding techniques [9-10].

1.2. LTE Related Interference Estimation

Among our recent research topics there is the investigation ofinterference to which new LTE technology will be subjected,such as TV and radar signals operating in the neighboringfrequency bands. To estimate proper conditions under whichnew technology would coexist with established wireless sys-tems is of uttermost importance before beginning implemen-tation of a new network infrastructure.

Most of the previous studies of interference effects onMIMO system capacity takes into account co-channel intra-system interference with AWGN [11-12]. We consider in-terference to LTE downlink transmission consisting of inter-cell and external radar interference. Each sub-carrier k pervictim symbol will be interfered by radar differently accord-ing to the pulse spectrum, giving rise to a number of signal-to-interference-plus-noise ratios (SINR) γm,k of the receivedsignal at the terminal unit m in downlink estimated per eachsub-carrier k, k = 1, ...,K, where K is the total number ofsub-carriers:

γm,k = Ps,kgs,m

IIC + IR + σ2n

, (10)

where Ps,k is transmit output power of the serving base sta-tion (eNodeB) s per sub-carrier k, gs,m is radio channel gainincluding antenna gains and path losses between base stations and mobile user m, IIC is the inter-cell interference, IR

is the external radar interference and σ2n is the variance of

AWGN.The inter-cell interference IIC depends on the traffic load-

ing ρj in neighboring cells, and the loading itself is a functionof local SINR value in neighboring cells therefore posing op-timization problem. Radar interference term IR depends onthe spectrum overlap between LTE and radar signals and canbe estimated from known signal waveforms [13].

A new feature in 4G voice communications is VoLTE(Voice over LTE) representing an evolutionary step over VoIP(Voice over IP) [14]. Recently, we have been performing testsfor network capacity and voice transmission quality using



Fig. 1. Radar to LTE interference geometry.

VoIP technology subjected to various interference effects.Having in mind high data rates and many impairments in thereal world transmission channels, such experiments requiremany well-tuned measurements, especially focusing on timesynchronization between remote network nodes.

The performed experimental measurements of pulsed ra-dar signal impact on live LTE base station have been com-pared with numerical simulations, which enabled us to esti-mate allowable radar signal power levels. Interference geo-metry is depicted in Fig. 1 representing the mobile receiverconnected to the serving base station (eNodeB) and subjectedto external interference from the nearby located radar. Fig. 2represents experimental measurements (PESQ) and theoreti-cal predictions.

The results represent measured VoLTE PESQ (Perceptu-al Evaluation of Speech Quality) and simulated number ofpossible voice calls which reduces to zero when radar signalpower increases - see Fig. 2. The zero points of calls numberrepresent the thresholds of radar power levels which totallyblock VoLTE service.

2. Simulation Framework

A more detailed description of simulation algorithms is avail-able in our recent publications [13,15]. They are based onstatistical Monte Carlo simulations implemented using GNUOctave [16] and Python numerical libraries [17].

Fig. 3. Numerical simulation functions for VoLTE and ra-dar interference analysis.

Fig. 2. Experimental measurements (PESQ) and theoreticalpredictions of maximum number of VoLTE voice calls.

The simulation scenario consists of processes randomlygenerating mobile users over network’s cell area, simulatingradio propagation channel, estimating path losses and multi-path fading effects on received signal from the serving basestation. In addition, neighboring base stations with their ownmobile user distributions are generated which impose inter-cell interference.

The results are statistically averaged over multiple snap-shots of randomly generated mobile user locations withineach network cell.

A schematic diagram of software packages used for assess-ment of external interference from nearby radar transmitterinto LTE downlink service is shown in Fig. 3. It includesfunctions for pulsed radar signal generation (radar_pul-se), a module for radar interference estimation (radar_-interf) with suplementary function calc_fdr, whichcan output graphical results of spectrum overlap between ra-dar and LTE signals.

The main module for simulations lte_sinr accepts userdefined LTE (lte_params) and radar (radar_params)configuration parameters and allows to select path loss model(free_space or okumura_hata). lte_sinr moduleis used to run Monte Carlo simulations over multiple mobi-le user distributions and output results in the form of Mbpsfor data rate or maximum number of voice calls for VoLTEservice.

High-throughput computing software package HTCondordeveloped at University of Wisconsin-Madison [18] is usedfor scheduling long running computing jobs and to provi-de task parallelization on multi-core processors. HTCondorsupports message passing interface (MPI) applications, im-plemented using GNU Octave or Python numerical libraries.

The output of simulations is generated in the form of tablesand graphics which have to be compared against measure-ment results. The algorithm development and experimentalmeasurements in our lab have been performed by different



Fig. 4. Virtual server environment.

team members, therefore effective and timely sharing of thesimulation results has been of high importance. We requiredthe ability for experimenters to modify simulation parame-ters for adapting to real measurement conditions, repeat thesimulations and obtain dynamical prediction results in tabu-lar format suitable for further data processing.

Additionally, the requirement of remote access preferablyvia standard web browser was essential in order to be able tolaunch calculations and check the progress of long runningsimulations remotely.

To meet these requirements we built a cloud-based virtualserver environment on Linux Ubuntu 13.10 OS, loaded withGNU compilers and numerical libraries (Fig. 4.). For MIMOLTE simulations we used algorithms implemented in GNUOctave and Python scripts. Several software interfaces weremade available for users with three different roles.

1. Command line interface via secure shell (SSH) connec-tion is used for algorithm development and testing, sys-tem installation and configuration.

2. Secure FTP connection over SSH is available for dataexchange between local and remote computers to uploadlarge input datasets and download raw simulation resultsfor further post-processing and visualization.

3. IPython Notebook [19] server has been implemented inorder to have the ability to run simulations online with-out the knowledge of Linux shell programming. It al-lows experimenters to connect to the virtual server viastandard web browser (using HTTPS authentication) tomodify simulation parameters, run calculations and pre-view simulation results.

Implemented interfaces allow sharing the same algorithmcode base between a group of researchers without the need ofreinstallation of software libraries on user computer.

Fig. 5. Web view of IPython Notebook user interface.

Web view of one of LTE simulation results is shown in Fig.5. It provides the area for entering and modifying algorithms,launching calculations and previewing simulation results.

Conclusions

Easy to use and maintain virtual server environment has beenimplemented for MIMO LTE communication system model-ing.

The framework allows to perform long running MonteCarlo simulations involving multiple modules of physicalnetwork layer. Simulation algorithms can be shared by ex-perimenters which are able to modify input parameters andrun live simulations corresponding to real measurement con-ditions.

The implemented framework reduces technical workrequired for exchanging results between members of researchteam and minimizes maintenance procedures of the virtualserver system.

Acknowledgement

Virtual server for numerical simulations has been providedby cloud computing services of Blue Bridge Ltd.

References1. P. Bhat, S. Nagata, L. Campoy, I. Berberana, T. Derham, G. Liu, X. Shen, P. Zong, and J. Yang. LTE-advanced: an ope-

rator perspective. – IEEE Communications Magazine50(2) (2012) 104–114 – <http://ieeexplore.ieee.org/lpdocs/epic03/ wrap-per.htm?arnumber=6146489>.

2. F. Rusek, D. Persson, B. K. Lau, E. G. Larsson, T. L. Marzetta, and F. Tufvesson. Scaling up MIMO: opportunities and challengeswith very large arrays. – IEEE Signal Processing Magazine 30(1) (2013) 40–60. – <http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=6375940>.



3. M. Narandzic, C. Schneider, R. Thoma, T. Jamsa, P. Kyosti, and X. Zhao. Comparison of SCM, SCME, and WINNER channel models.– IEEE Apr. (2007) 413–417. – <http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=4212524>.

4. P. Heino, ed. WINNER+ final channel models. – Tech. Rep. D5.3, Jun. 2010.5. A. Osseiran, M. Boldi, J. Monserrat, J. Vihriala, A. Tolli, and A. Gouraud. Radio enabling techniques for IMT-advanced (4G) and

beyond: WINNER+ project// In: Wireless Technology Conference (EuWIT) 2010European, Sep. 2010, pp. 17–20.6. J. G. Andrews, W. Choi, and R. W. Heath. Overcoming interference in spatial multiplexing MIMO cellular networks. – Wireless

Communications, IEEE 14(6) (2007) 95–104.7. M. Chiani, M. Z. Win, and H. Shin. MIMO networks: The effects of interference. – IEEE Transactions on Information Theory 56(1)

(2010) 336–349. – <http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=5361482>.8. M. Di Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo. Spatial modulation for generalized MIMO: challenges, opportunities, and

implementation // In: Proceedings of the IEEE 102(1) (2014) 56–103.9. A. Goldsmith, Wireless communications. – Cambridge; New York: Cambridge University Press, 2005.10. L. Hanzo, J. Akhtman, L. Wang, and M. Jiang. MIMO-OFDM for LTE, Wi-Fi, and WiMAX: coherent versus non-coherent and

cooperative turbo-transceivers. – Chichester, West Sussex, U.K. ; Hoboken, N.J: Wiley, 2011.11. S. Catreux, P. Driessen, and L. Greenstein. Simulation results for an interference-limited multiple-input multiple-output cellular sys-

tem. – IEEE Communications Letters 4(11) 334–336, Nov. 2000. – <http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm? arnum-ber=892193>

12. R. Blum. MIMO capacity with interference. – IEEE Journal on Selected Areas in Communications 21(5) (2003) 793–801. –<http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=1203164>.

13. J. Krivochiza, R. Aleksiejunas, K. Svirskas, B. Dzindzeleta, A. Pantelejev, J. Zverelo, M. Zilinskas, and J. Aleksandravicius. Mo-deling of radar interference impact on VoLTE service. – IEEE Nov. (2013) 25–27. – <http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=6716162>.

14. F. J. Rivas, A. Dı́az, and P. Merino. Obtaining more realistic cross-layer QoS measurements: A VoIP over LTE use case. – Journal ofComputer Networks and Communications (2013) 1–10. – <http://www.hindawi.com/journals/jcnc/2013/405858/>.

15. R. Aleksiejunas, J. Aleksandravicius, J. Krivochiza, and K. Svirskas. Analysis of radar interference effects on MIMO LTE downlinkperformance. // Submitted to EMC Europe 2014.

16. GNU Octave, <http://www.gnu.org/software/octave/>, accessed 2013.03.10.17. NumPy, <http://www.numpy.org/>, accessed 2013.03.10.18. HTCondor <http://research.cs.wisc.edu/htcondor/>, accessed 2013.03.10.19. IPython Notebook <http://ipython.org/notebook>, accessed 2013.03.10.


Documents

Simulation Framework for MIMO LTE Network Performance Analysisjournal.kolegija.lt/iitsbe/2013/Aleksiejunas... · the spectrum overlap between LTE and radar signals and can be estimated