Malawi Television White Spaces (TVWS) Pilot Network Performance Analysis

Journal of Wireless Networking an Communications 2014, 4(1): 26-32 DOI: 10.5923/j.jwnc.20140401.04 Malawi Television White Spaces (TVWS) Pilot Network Performance Analysis C. Mikeka 1,*, M. Thoi 1, J. S. P. Mlatho 1, J. Pinifolo 2, D. Konwani 2, L. Momba 2, M. Zennaro 3, A. Moret 3 1 Physics Department, Chancellor College, University of Malawi 2 Malawi Communications Regulatory Authority (MACRA) 3 T/ICT4D Marconi Wireless Lab, Abus Salam, International Center for Theoretical Physics (ICTP) Abstract Performance status of the Malawi TVWS pilot as of December 2013 is presente. Basic performance metrics like throughput, latency, SNR have been analyze using known path loss empirical moels like Hata, Asset an Friis. Aitionally, this paper presents new ata mining tools evelope using Python an Perl which were eploye at each TVWS station s ALIX boar running on Voyage Linux to abstract useful ata for computation of network latency an throughput. Typically, for the longest teste link at 7.5 km, an average SNR = 24.7 B, ata-rate of 420 kbps an latency of 118 ms were observe using the collecte ata. Empirically, Hata an Asset resolve an average path loss value equal to 155 B which was 51 B higher than the ieal Friis Free Space Path Loss (FSPL) at UHF CH31 (554 MHz), BS antenna height of 23 m, receiver antenna height of 2m, an farthest station istance of 7.5 km. These results prove a 2.6 times superior propagation performance of the TVWS network over the commercial fixe broaban wireless network (assuming 2 Mbps backhaul, an all other parameters controlle). These measurements were one in ry season. Results for rainy season will be stuie from December 2013 to June 2014. Keywors Latency, Python, SNR, Throughput, TVWS 1. Introuction While the expectations from using TVWS for wireless broaban service are high, the commercial an technical viability of TVWS operations is still largely unknown [1]. Several researchers have stuie the use of TV white spaces for other applications [2-5]. In 2011, [3] carrie out a technical feasibility on the use of TV White Spaces for broaban access in ifferent sectors namely; rural an urban with the potential for traffic offloa from congeste mobile broaban networks, Smart city applications an location-base services. The primary aim of the stuy in [3] was to assist Ofcom in eveloping the regulatory framework which was to facilitate the use of TV white spaces an help improve spectrum efficiency an provie universal broaban access. One of the key finings of the stuy was that TV white spaces spectrum can be use for a range of applications, from improving rural broaban connections to machine-to-machine (M2M) applications. Similar stuies have also been carrie out in USA an Singapore. Most of the stuies on TV white spaces have either focuse on the * Corresponing author: chomora@gmail.com (C. Mikeka) Publishe online at http://journal.sapub.org/jwnc Copyright 2014 Scientific & Acaemic Publishing. All Rights Reserve availability of the TV white spaces, avances in TV white space raios or the evelopment of TV white spaces regulatory framework an white space evice specifications [3-8]. Very few stuies, to the knowlege of the authors, have focuse on the performance of a eploye TV white space network. Of the few TV white space network eployments in Africa, the network performance analysis has not been publishe. Thus, this paper presents a performance of the TV white spaces network uner harsh African environment for example intermittent power an constraine Internet banwith resource. This work also presents low cost an innovative tools for monitoring an analysing TV white space networks. An evaluation of the performance of well-efine seconary systems in realistic scenarios will eventually help to gauge the market prospects for TVWS-riven technologies an potentially guie subsequent regulatory rule-making. In Malawi, the partnership team between the regulator (referre to as the Authority in the TVWS regulations) an the University of Malawi, Chancellor College have in the mi of December, 2013 evelope the TVWS rules an regulations, investor sie business moel an a performance analysis of the network since eployment. The goals for the Malawi TVWS Pilot are similar to that of the Cambrige Trial [9] which intens to help inustry unerstan the capability of TV white spaces to serve a wie

Journal of Wireless Networking an Communications 2014, 4(1): 26-32 27 range of applications, through key factors such as the coverage an performance that can be achieve. The rest of this paper is organize as follows. Section II presents the network setup an the propose metho for the efinition of the performance inicators or metrics an their analyses. Results are presente in Section III while Section IV iscusses the results. Conclusions are rawn in Section V. Motivation for the TVWS Deployment in Rural Malawi A number of rural communities in Malawi have complaine about the unavailability or poor broaban performance from the currently available commercial ISP services. Their experience is not unusual in Africa an other rural areas of the eveloping worl. The key attraction of TV white spaces in this application is the enhance coverage istances which lower frequencies enable (compare to the higher frequency bans traitionally use for wireless broaban access). The extene range translates into fewer base stations being require to cover a given area an, hence, lower CAPEX an OPEX costs. An aitional avantage of subsiize license-scheme (ISM ban-like) propose in the December 2013 Malawi TVWS regulation raft, is that rural communities woul have the capacity to provie their own wireless networks in the iconic fashion calle citizen-science. In the Malawi pilot, measurements were mae to establish the coverage an performance that coul be achieve aroun pilot base station (BS) in the network setup as epicte in Section II. Measurements were taken with a spectrum analyzer at street level height of 2m connecte to a TVWS Yagi antenna pointing to the BS, with checks on both receive signal level (RSSI) an ata throughput (uplink an ownlink). The coverage achieve matche reasonably well with preictions mae using Friis, Okumura-Hata, an Asset propagation moeling tools in a quasi-line of sight conition [6, 10] an script ata mining tools using Python an Perl. The path loss estimations escribe in this paper will provie guielines for researchers an practicing engineers in choosing appropriate path loss moel(s) for coverage optimization an interference analysis for wireless evices operating in the TV ban in our environment an also, to preict TV coverage an keep-out istances for potential seconary users operation in the TV white spaces [10]. Another challenge for TVWS raio transceiver is the choice of the best igital harware to meet the requirements. Features like flexibility, performance an power consumption are key factors for cognitive raios [11]. In the Malawi TVWS network the harware can mitigate intensive interference by invoking aaptive moulation functionality in the base station communication subsystem. The manufacturer of our raio equipment is working on auto channeling for future proucts. To exten the coverage range further than the current 20km teste range, a stuy is in progress to use UHF ban power amplifiers with fast switching capability to increase transmit power above the current 4W towars 10W using 50B gain UHF amplifiers. The authors herein have previously publishe on the finings of a TVWS spectrum measurement initiative in Malawi an Zambia [12]. In [12] they introuce an open harware evice that geo-tags spectrum measurements an saves the results on a micro SD car. The evice can also be use to recor the use of spectrum over long perios of time. An assessment stuy on TV white spaces in Malawi using afforable tools was presente in [13]. In this paper, however, the scope is on the performance of the network since it is manatory that appropriate performance inicators of cognitive raio systems (White Space Devices, WSDs) are ientifie [14]. 2. Malawi TVWS Network Setup an Propose Methos for Capturing Network Data The Malawi TVWS network topology assumes a star configuration. It has a single base station an three client stations as shown in Table 1. Table 1. Station ientification in the online Operation an Management Center (OMC) Terminal Description Station Name CST00162 ICTP CPE 162 St. Mary s Girls Sec School CST00163 ICTP CPE 163 Malawi Defense Force Air-wing CST00164 ICTP CPE 164 GPS (Seismology Dept.) CSB00490 ICTP Base 490 ZA TVWS Base Station (BS) 2.1. Station Design an Description Each station comprises client premise equipment (CPE) an a Yagi-Ua type of antenna mounte outoors an powere by a UTP cable that terminates into an inoor Power-over-Ethernet (PoE) aapter. Aitional station evices inclue a LAN switch an an ALIX boar (functionality iscusse in later sections). The TVWS base station is an inoor evice, an has ultra-low power consumption compare with cellular base stations. It transmits using a huge monopole antenna (case of Carlson raios) mounte outoors at a height base on rigorous computations or simulation per esign coverage an link quality of the star topology. Internet supply in the Malawi TVWS pilot network is provisione through a eicate 2Mbps wireless backhaul as shown in Fig. 1. The actual network eployment terrain an coverage scope is shown is Fig. 2.

28 C. Mikeka et al.: Malawi Television White Spaces (TVWS) Pilot Network Performance Analysis Figure 1. Malawi TVWS pilot network: Base station locate at the GPS coorinates: -15.376415S, 35.318349E Figure 2. Malawi TVWS terrain coverage scope 2.2. Low Cost Monitoring Platform The network was eploye using Broaban Rural- Connect equipment from Carlson Wireless Technologies. The Carlson equipment comes with a clou base Operation an Management Center (OMC) which provies a clou level view of the evices that you register uner a given name. This online tool is use for the initial configuration as well as management of the network. The OMC has graphical isplays of SNR (uplink an ownlink), application ata throughput (uplink an ownlink) as well as application packet throughput (uplink an ownlink). The ata presente in OMC has one major shortfall in that it cannot be save or exporte in any format for later viewing or analysis. Therefore the performance ata in the OMC is only useful when one only wants to see the current performance an not the trens over a perio of time beyon 24 hours. This prompte the authors to ientify some low cost innovative tools of capturing an saving performance ata using Python an Perl. To enable the real time collection of performance ata, ALIX boars running Voyage Linux were fitte at each client station. i. Installation of Voyage Linux an Configuration of the Network Interfaces on ALIX boars Voyage Linux was installe on the ALIX boar using the proceure escribe in [15]. After successful installation, Voyage Linux was mounte

Journal of Wireless Networking an Communications 2014, 4(1): 26-32 29 (with the compact flash (CF) car still connecte to the Linux machine use for the installation) on /mnt/cf (the mounting point set in the installation configuration). Then, one network interface (eth0) was enable an given an IP aress (192.168.x.y/24). The configuration was save; the CF car was un-mounte an slotte it into the ALIX boar which was then boote. The interface eth0 connects to the client raio (the CPE) an another interface (eth1) was configure to be proviing ynamic IP aresses in the range 192.168.0.5/ to /240 to computers on the client sie using DHCP. The DHCP server use was nsmasq whose configuration file is /etc/ nsmasq. conf. The interface itself was configure with a static IP aress (192.168.0.1). ii. Python Scripts Given the limitations of the OMC, a nee arose to have custom scripts that woul capture an save network throughput ata. For this purpose, Python programming language was chosen because it has numerous existing open source libraries an moules that can be use as-is as well as moifie to suit ifferent nees. The evelope scripts were use to collect network an packet throughput ata. The ata was collecte for each ay on a one secon interval. The collecte ata was then save using comma separate values (CSV) files; chosen mainly for their portability. The files use the ate stamp for the name for easy sorting an ientification uring analysis. Fig. 3 below shows a snippet of the ata. The columns in each row contain the information; interface, bytes sent, bytes receive, packets sent, packets receive, bytes sent per secon, bytes receive per secon, packets sent per secon, packets receive per secon an time in that specific orer. In aition to the scripts that capture throughput ata, we also use some Python scripts to plot the ata obtaine into graphs. We mainly exploite the plotting functionalities of the matplotlib moule for this purpose. iii. Perl Scripts The Python scripts that were evelope i not capture network latency ata. For this purpose, other scripts were evelope in Perl programming language to capture an save network latency from the base station to all the clients sites. Again, the same Python language coul have been use, but Perl was opte for ue to the relative ease foun by the researchers. Figure 3. A snippet of the ata store in csv file

30 C. Mikeka et al.: Malawi Television White Spaces (TVWS) Pilot Network Performance Analysis 3. Results The important results in this paper are on the throughput, latency an path loss. Throughput an latency are compute averages from the measure ata over one month at station premises. The path loss is compute using known empirical moels with all equation variables replace by the actual eployment parameters an conitions. Any assumptions mae an limitations observe have been iscusse in the subsequent sections. case is propagation elay. Propagation elay, in our case, is a function of how long it takes the information to travel at the spee of light in the wireless channel at 554 MHz from source to estination. These results are comparable to TENET stuy, the South African TVWS Trial [16]. 3.1. Performance Graphs The ownstream throughput (blue line) an upstream throughput (black line) are shown in Fig. 4 below. Figure 6. One hour latency measurement at GPS Station (1.7 km from BS) Figure 4. Average ownstream throughput at the farthest TVWS station, Airwing (7.5 km from BS) The average latency in millisecons (ms) for 3 r December in an hour stuy is shown in Fig. 5 below. Figure 5. One hour latency measurement at AirWing station Latency an overall throughput is ominate by two factors namely; the length of the route that the packets have to take between sener an receiver an the interaction between the TCP reliability an congestion control protocols. To confirm this point, a plot for latency at GPS station (closest station from TVWS BS) is shown in Fig. 6 below, where it is evient that the latency is lower than for the farthest station. The reuce latency however is not proportional to the scale factor of the station istance ratios because latency is a function of several factors. These inclue; propagation elay, serialization, ata protocols, routing cum switching, an queuing cum buffing. Of the aforementione factors, the eterminant for the reuce latency between the shortest an farthest stations in this 3.2. Equations In this subsection, path loss will be estimate base on Friis Free Space Path Loss (FSPL), Hata an Asset moels.. Path loss estimation is the basis for the erivation or computation of the receive power at a given station from the known base transceiver station. 1) Estimating Friis Path Loss The funamental aim of a raio link is to eliver sufficient signal power to the receiver at the far en of the link to achieve some performance objective. For a ata transmission system, this objective is usually specifie as a minimum bit error rate (BER). In the receiver emoulator, the BER is a function of the signal to noise ratio (SNR) measure in ecibels (B). In esigning a spectrally efficient wireless communication system like TVWS, it is important to unerstan the raio propagation channel. The characteristics of the raio channel will change mainly ue to the operating frequency an the propagation environment. Typical examples may inclue line of sight (LoS) compare to non-line of sight (NLoS). In this work, all the stations are eploye in a quasi-nlos environment. The major LoS obstructions are ue to builings, hills an trees which impose a slow faing conition over the channel. For shorter istances to the base station, it is still viable to employ Friis moel given receiver antenna height of 2 meters above street level groun. The actual receive power P r at each TVWS station was calculate using (1) which has taken into account; system losses an antenna gain for the transmitter an the receiver. 2 4π Pr = Pt + Gt + Gr Lsys 10log 20log λ Where P r = Receive power in B P t = Transmit power in B G = Antenna gain for transmitter t ( ) (1)

Journal of Wireless Networking an Communications 2014, 4(1): 26-32 31 G r = Antenna gain for receiver L =System losses sys = Distance c λ = f Where c = spee of light, f = 554MHz The compute receive power P r ( Bm) is shown in Table 2 below. Pt Table 2. The compute receive power P r ( Bm) Gt G r 4 6 11 12 L sys λ GPS 0.5 4 StMarys AirWing 1700 2400 7500 P r ( Bm) -80.9-83.9-93.8 The compute receive power took into account the following path losses: FSPL = 91.9 B for GPS, 94.9 B for St. Mary s an 104.8 B for AirWing. These results compare very closely to the actual spectrum measurements for GPS an St Mary s (closest stations) at 554 MHz (given 20 MHz observation winow) using a hanhel spectrum analyzer terminate to the TVWS Yagi antenna (given same cabling losses). The measurement results for Malawi Defense Force AirWing were worse than the value presente in Table 2. A typical measurement was unsteay aroun -103 Bm on average at 2 m height an only improve to -93.8 Bm or better when the antenna height position was raise to 8 m above street level groun. These observations attracte the authors to estimate the path losses using other moels like Okumura-Hata an Asset. Unlike Friis Free Space Path Loss (FSPL), Hata an Asset are realistic moels because they take into account real environment factors such as reflection, scattering, iffraction, refraction an absorption of signals by many features such as vegetation, builings an people hence ieal for moelling a network. Asset moel closely resembles Hata moel. Nevertheless, Asset moel iffers from Hata moels because of the aitional iffraction an clutter losses. The two formulae are presente in (2) an (3) below. 2) Asset Propagation Moel PL( B) = K1+ k2log ( ) + k3( Hm) + k4log ( Hm) (2) + K5log Heff + k6log Heff log + k7 iffn + CL ( ) ( ) ( ) ( ) 3) Hata Propagation Moel L ( ) = 69.55 + 26.16log( ) 13.82log( eff ) ( ) 44.9 6.55log( ) log( ) H B f H A Hm + Heff Where, f = Transmitting frequency (MHz) (3) H eff H m = Effective antenna height of base station (m) = Antenna height of the CPE station (m) A( H m ) = CPE height correction factor = Distance between CPE an base station (km) iff = Diffraction loss K1 an K 2 = Intercept an slope corresponing to a constant offset in Bm an a multiplying factor for the logarithm value of istance [ ( km) ] K3 K 4 K5 = CPE antenna height correction factor = Multiplying factor for H m = Effective antenna height gain 6 = Multiplying factor for log log K ( ) ( ) = Multiplying factor for iffraction loss = Clutter loss The compute path loss comparison is shown in Fig. 7 where Hata compares closely to Asset as expecte. However, Asset exhibits a 1.04% increment factor on the path loss compare to Hata which is ue to the inclusion of the iffraction an clutter losses. Future work will inclue a careful stuy on the comparison of the measure receive power to simulations base on these empirical moels. In this work, the authors have simply confirme the superiority of the Asset moel to Okumura-Hata in path loss estimation. However, the authors think that the unity values assume by the multiplying factor for the iffraction loss (K7) an the clutter loss are too ieal. Path Loss [B] K7 C L 160 140 120 100 80 Figure 7. Compute path loss (a comparison between Friis Free Space, Hata an Asset Moels) 4. Discussion H eff Friis FSPL Moel Hata Moel Asset Moel 0 3 6 9 Distance [km] The results presente in the preceing sections have emonstrate the superiority of Hata an Asset moels over

32 C. Mikeka et al.: Malawi Television White Spaces (TVWS) Pilot Network Performance Analysis Friis FSPL in terms of the precision of the calculate path loss. At AirWing TVWS station, the measure SNR was equal to 24.7 B implying a noise level of -118.5 Bm. There is however, a small correction on the effective receive-sie antenna height uring the computation of Hata an Asset path losses. The smallest antenna height of 2 m as eploye at St. Mary s TVWS station was use to provie the worst case scenario for the compute path loss (eliberately one to simulate an NLoS environment). In reality the antenna height for GPS is 3 m while for AirWing is 8 m (to approximate LoS conition). In this paper, the ownstream throughput coul be efine by DT avg, then for AirWing, reaing from Fig. 4 250 DT avg 420kbps. This ata rate is 2.6 times better than the typical ata rates (at 176kbps) provie by commercial operators in rural areas, given the same backhaul internet banwith of 2 Mbps. The commercial operators use 5.8 GHz ISM ban Ubiquiti raios in Wi-Fi hotspot fashion. 5. Conclusions In this paper, performance analysis of the TV white spaces in the UHF ban in Malawi was performe. It is observe that unlike other fixe broaban services, TVWS services emonstrate 2.6 times better ata rates given the same operating conitions. The longest teste operational range for the Malawi TVWS network is 18.56 km with the lowest SNR i.e. below 10 B (Pirimiti station). However, the teste functional range at the moment is 7.5 km (AirWing) which measures an SNR of 24.7 B, average latency of 118 ms an maximum throughput of 420 kbps given simultaneous usage of three client stations from a TVWS BS backhaule to a 2 Mbps internet banwith. These results are unpreceente; for example, the longest known TVWS link to the authors is 6 miles (approx. 7.2 km) from the South African Trial. Also, the Malawi TVWS eployment merits others, in that; it is one that has been one with the most constraine resources of banwith (meager line spee of 2 Mbps cf. 1 Gbps in South Africa). The collaboration arrangement between the University an the Regulator is worth learning from the eployment in Malawi, where the regulator unerstoo the significance of ynamic spectrum stuy, re-use an re-farming through soli science research an supporte the project using its Universal Access Fun. In the future, the authors plan to obtain an inclue the missing actual measurements comparison with selecte fixe wireless services like Wi-Fi an WIMAX. A thorough stuy on interference mitigation by the white space evices is also reserve for further stuy. REFERENCES [1] Achtzehn, A., Petrova, M., Mähönen P., On the Performance of Cellular Network Deployments in TV Whitespaces, in Proc. IEEE ICC, 2012, pp.1789-1794. [2] Gomez C., TV White Spaces: Managing space or better managing spaces; Discussion paper, Raio Communication Bureau, ITU. [3] Peter Flynn., White Space Potentials an Realities; Texas instrument, Texas, USA, Available: http://www.ntia.oc.gov /files/ntia/publications/spectrum_wall_chart_aug2011.pf (Accesse: December 14, 2013). [4] Info-communications evelopment authority (IDA) of Singapore, Propose Regulatory Framework for TV White Space Operations in the VHF/UHF Bans, IDA Singapore, 2013. [5] Keltz I., State of TV White Spaces in the in the Unite States, Paper presente at the TV White Space an Dynamic spectrum African Forum, 2013. [6] Noguet D., Gautier, M., an Berg V., Avances in opportunistic raio technologies for TVWS, EURASIP Journal on Wireless Communications an Networking 2011, 2011:170. [7] Nekovee M., A Survey of Cognitive Raio Access to TV White Spaces, International Journal of Digital Multimeia Broacasting Volume 2010. oi:10.1155/2010/236568. [8] Stirling A., White Spaces the New Wi-Fi? International Journal of Digital Television, Vol. 1. DOI: 10.1386/jtv.1.1. 69/1. [9] Cambrige White Spaces Consortium, Recommenations for Implementing the Use of White Spaces: Conclusions from the Cambrige TV White Spaces Trial, 2012. [10] Faruk, N., Ayeni, A. A., an Aeiran, A. Y., On The Stuy Of Empirical Path Loss Moels for Accurate Preiction of TV Signal for Seconary Users, Progress In Electromagnetics Research B, Vol. 49, 155 {176, 2013}. [11] Charalambous, E., Stavrou, S., Raspopoulos, M., Marques, P., Dionisio, R., Alves, F., Gonçalves, Roriguez, J., Balz, C., Lauterjung, J., Schuberth, G., Schramm, R., an Fore, T., Cognitive raio systems for efficient sharing of TV white spaces in European context, report elivere to CEC, 2011. [12] Zennaro, M., Pietrosemoli E., Arcia-Moret, A., Mikeka C., Pinifolo, J., Wang, C., an Song, S., 2013, TV White Spaces, I presume? in Proc. Sixth International Conf. on ICTD, Cape Town, South Africa. [13] Zennaro M., Pietrosemoli E., Mlatho J. S. P., Thoi M., Mikeka C., An Assessment Stuy on White Spaces in Malawi Using Afforable Tools, in Proc. IEEE GHTC, 2013. [14] Gomez, C., TV White Spaces: Managing Spaces or Better Managing Inefficiencies? 13 th GSR 2013, iscussion paper. Available:http://www.itu.int/en/ITU-D/Conferences/GSR/D ocuments/gsr_paper_whitespaces_gomez.pf. [15] http://linux.voyage.hk/content/getting-starte-v08x. [16] http://www.tenet.ac.za/tvws/latency-sept-25-2013 (Accesse: January 8, 2014).