Improved propagation models for lte path loss prediction in urban & suburban Ghana

Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 35 I MPRO VED P ROP AGATION M ODELS F OR L TE P ATH L OSS P REDICTION I N U RBA N & S UBURBAN G HANA James D. Gadze , K wame A. Agy ekum, Steph en J. Nuagah and E. A. Affum Department of Telecommunication Engineering, Kwame Nkrumah University of Science and Technology, Ghana A BSTRACT To maximize the benefits of LTE cellular networks, careful and proper planning is needed. This requires the use of acc urate p ropagation models to quantify the path loss required for ba se statio n d eployment. Deployed LTE netwo rks in G hana can barely meet the desired 100 Mbps throughput leading to customer dissatisfaction. Network operators rely on transmission p lanning to ols d esigned for generalized environments that come with alread y embedd ed propagation mo dels suited to o ther environmen ts. A challenge therefore to Ghanaian transmission Network planners will be choosing a n accurate and p recise propagation model that best suits the Ghanaian environment . Given this, extensive LTE path loss measurements at 800MHz and 2600 MHz were taken in selected urban and suburban environments in Ghana and compared with 6 commonly u sed pro pagation models. Improved versions of the Ericson, SUI, and ECC-33 developed in this study p redict more precisely the path loss in Gh anaian environmen ts co mpared with co mmonly used propagation mod els. K EYWORDS Propagation Mo dels, Path loss Expon ent, Root Mean Square E rror, Signal Referen ce Received Power(RSRP). 1. I NTRODUCT ION Cisco's visual networking index, 2017-2022 predicts that the IP traffic recorded annually aroun d the globe is estimated at 4.8 Zb by 2022. This translates to a threefold increase ove r the next five years [1]. Mobile data subscription in Ghana as of July 2018 stood at twenty -nine million, one hundred and eighty-one thousand, eight hundred and sixty -three (29,181,863) [ 2]. For a country with an estimated total popula tion of thirty million [3], i t shows t he high demand for data and broadband servic es. With t his g rowing demand fo r ba ndwidth in mob ile co mmunication as use r numbe rs keep increasing significantly, mobile networks have evolved from 1G - 4G to meet the demand over the years. In Ghana, Blu telecom, Busy internet, Surfline, MTN, and recently Vod afone have commercially d eployed 4G LT E networks for high er throughpu ts and improv ed user experience . Ho wever, this hasn't been completely achieved since the expected throughput of 100Mbps is barely realized leading to dissatisfaction among customers. This h as resulted in a lot of comp laints and sanctioning from the Nati onal communicat ions authority [4]. Tr ansmitted signals from a base station su ffer seve re attenua tion as the y propag ate through space leading t o degrada tion in signal strength and qu ality [ 5]. This s evere attenua tion is introduced du e to reflection, diffraction, and scattering of the signal as it impinges on obstacles. For subscribers of a network who have va rying mobility, it is imperative to design a mobile network so that the y Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 36 have robust signal levels at all locations. To achieve this, Conditions f or radio propagation need to be well understood a nd predicted as accuratel y as possible. propagation models ar e instrumental in wireless network planning as they support i nterference estimates, frequency assignments, and ce ll coverage a ssessment and o ther parameters[6]. Empirical p ropagation models that are mostly used are however environment specific and are developed based on a spe cific propa gation environment of interest [7]. Any little deviation in characterizing the propag ation environm ent under investigatio n affects the ef fici ency of propagation m odels designed from the area[8],[9]. Therefore, the use of propagation models in settings other t han those in tended t o be used might lead to inaccurate prediction which affects system perfor mance[10],[1 1]. To investigate t hese clai ms, the approach adopted i n this p roject is to take Signal Referenc e Received Power (RSRP) values from deployed cell sites and compare with predictions from 4 propagation models at 800MHz and 6 models at 2600MHz. This approach will help us develop modified and improved versions of already existing propagation models suited for the Gh anaian environ ment. The rest of the paper is organized as follows : Section 2 reviews some relevan t work in this field. Section 3 presents the measurement procedure at invest igated e nvironment s and empirical propagation models und er consideration. Results a re pr esented in Sectio n 4 and Section 5 concludes the study. 2. R ELATED W ORKS Considering t he i ncreased demand placed on mobile communication, higher throughputs and seamless connectivity, designing LTE networks in compliance with the performance m etrics it promises is crucial. Nu merous studies have gone into finding propagation models that predict accurately the path loss i n the USA, Europe, Africa, and Asia to improve network performance for bo th voic e and data comm unication. How bes t current propagat ion models will perform when used i n wirele ss environmen ts other than tho se origin ally intended fo r fre quently dev iate from the ideal [6]. Numerous studie s aroun d the globe, however, show that many industry-standa rd path loss models perform effective ly when adjusted to measured data from t hese areas [12].In [12], Path loss measured data at 3.5GHz in Cambridge was compared with t he predictions of three empirical propagation models. Results indicated th at the SUI and COST-231 m odels over-estimated path l oss in this environ ment. The closest fit t o the measurement data was the ECC-33 model. It was therefore recom mended for use in urban environ ments. The least-square method was used in [13] to optimize the Hata empirical path loss m odel for accurate prediction suited to a suburban ar ea in Malaysia. Outdoor measurements were taken in Cyberjaya, Malaysia at a frequency range of 400MHz t o 1800 MHz. Measurements were then compared with the existing models f rom w hich the Hat a mo del showed the best fit. The optimized Hata model was used and validated i n the Putrajay a region to detect the relative e rror to evalua te its efficien cy. Smalle r mean relative e rror was recorde d hence s howing that the optimizat ion was done successful ly. Propagation m odels are presented in [14] for LTE Advanced Networks. Path l oss for varying environments ( rura l, suburban and dense urban)we re computed using the following propagat ion models, COST-231 Walfisch – Ikegami model, SUI, ECC-33, Okumura extended Model and COST-231 Hata Model using MATLAB. Three f requencies between 2.3GHz and 3.5 GHz were considered in this work. Results presen ted indicated the COST-231 Hata model agreed better, giving the least path loss in all the environments compa red with t he other models. This work, Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 37 however, did not compare the prediction of empirical models with measured data but only base d on the model with the least path l oss. The conclusion made favoring the cost 231 Hata model by simulation as ag reeing best in all environments might be m isleading. Extensive measurements in [10] taken i n Lagos at a frequency of 3.4GHz made a comparison with 6 standard propagation models. It was concluded that the COST 231 -Hata and Ericson models showed the best performance in urban and suburban areas. Recent works in [16] also compared the efficiencies of empir ical, heur istic, and geospatial m ethods use d f or signal path loss predictions using da ta collected in urban Nigerian cities to develop path l oss models. Th e developed models and empirica l models were compared with field measured da ta. All models gave acceptable RMSE values excluding the ECC -33 and Egli models. Empirical models wer e the simplest and most commonly applied of the three techniques submitted. Their work , therefore, emphasized the further improvement of empirical models for optimum prediction. A hybrid of heuristic a nd empirical mod els for predi ction was recommended t o decrease the errors asso ciated with empirical models. Works have also gone into comp aring path loss of urban and suburban areas and t o ascertain if a particular propagation model can be used for both settings.[17] showed that propagation models in urban areas experience higher losses compared with suburban areas. For all environments, no si ngle model c ould be proposed. On the background that deployed WiMAX networks, failed to meet the optimum service quality requirements for delivering continuous wireless connectivity requests in the sub -Sahara region needed for emerging mobile application s, [18] i nvestiga ted the throughput performance of a deployed 4G LTE Site to ascertain if LTE meets the bandwidth demand needed for data -centric broadband appli cations. Field d ata from a deployed 4G LTE BS in Ghana o perating at 2600 MHz recorded a maximum throughput of 29.9 Mbps per sector. A maximum throughput of 62.318 Mbps was recorded at th e dow nlink for customers within 2.5 k m of the cell range from the BS. I t was concluded t hat 4G LTE can meet the ever -increasing demand of Ghanaians for broadband. This conclusion was made afte r comparing these throug hputs with the desired throu ghput required to sustain datacen tric broadband ap plications. Works in the Ghanaian environment focusing on W iMAX networks in the 2500-2530 MHz ban d was presented in [ 11]. The m easurement from a deployed WiMAX site around the university of Ghana, Accra was compared with the prediction of four empirical models. The extended COST- 231 mod el w as s elected as the model that best fits the measured data because i t r ecorded the leas t RMSE and a higher corre lation coefficient. Th is model w as recommended ther efore for effic ient radio network planning in Ghana and the sub -region at large. It was also concluded that no particular propagation model can be used to f orecast coherent outcomes f or all pr opagatio n settings. The r eason for this was the variations in weather and geography. Recomme ndations were made to conside r varying terra in parameters. Intensive measureme nts in separate environments must be conducted to parameterize a model. The paramete rs of the channel model are then a djusted to suit the me asurement outcomes [19]. It is imperative therefore from the works reviewed to evaluate the performance of ind ustry-standard propagation m odels prop osed for 4G LTE networks by considering different Ghanaian environments. Wit h several path loss models performing diffe rently in different e nvironments, i t is, therefore, essential t o determine which of the most frequently used models is bes t suited for 4G LTE networks i n Ghana. Further improving the suited model for m ore accurate prediction pertinent to the Ghanaian and Sub-Saharan environment will facilitate effective deploymen t of LTE networks by operators, meeting the promises the Standard came with. This, in the end, will Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 38 afford subs cribers the chan ce to enjoy seamle ss conne ctivity leading to custo mer satisfaction and loyalty. 3. M EASUR EMENTS 3.1 Procedure Received signal refe rence power (RSR P) values in dBm were taken at 10 Ba se stations in seven selected areas in Ghana wi th varying environ mental conditions. A drive test was c onducted using phones connected vi a the USB port to a co mputer with LTE software (Genex probe) instal led on it. Genex probe serves as a data collection software interface. A GPS was attached for location finding and track ing distance cover ed. The frequ ency w as set to 800MHz for the first test case a t five base stations and 2600 MHz for the second t est case for the other five base stati ons. At the various sectors of each LTE site in these environments, RSRP values at a varying dist ance starting from a referenc e dis tance (d o ) of 50m to 500m w ith 50m i ntervals were re corded. The Trans mit - Receiver distance was limited to 500 m to reduce the impact of interference from neighboring cells and also to cater for obstructions in the way of the drive. A receiver antenna height of 1.5m was maintained throughout the measurement campaign. Measured data is sent via the phones to the computing device which stores t he data as recorded log fil es. These recorded log files are then interpreted and analy zed. F ield measurem ents w ere taken between F ebruary and May. The RSRP in dBm was taken along the LOS and NLOS of t he fixed base stations with heights ranging between 16m and 35m . The laptop having GENEX software installed on it, the phone and t he GPS were set up in the drive te st vehicle as shown in figure 1 Figure 1 Measu rement set u p 3.2 Description of Environments Drive test measur ements w ere taken in the follow ing environments in Ghana. 1. Adum: This is an urban are a located in the central hu b of Kumasi, in the Ash anti Regio n of G hana w ith coo rdinates 6.6919°N ,1.6287°W. It is h ighly populated and Characterize d by a lot of busine ss activity. Present in Adu m are a lot of high-rise building s. 2. Techiman: Th is i s an urban area th at se rves as the capital of the newly created B ono East Region of Ghana with coordinates 7.5909° N, 1.9344° W . It is characterized by qui te many high-rise bui ldings and a lot o f farming and business act ivities. 3. Agogo: This is a Suburban area i n the Asante Akyim North Municipa l District of the Ashanti Region of Ghan a with coordinates 6.7991° N, 1.0850° W. Agogo is approximately 80 kilometers east of Kumasi, with moderate populatio n and buildings. Buildings are mostly no t high rise and are a little isolated from eac h other. The terrain i s relatively flat. Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 39 4. Afrancho: It is a Populated suburban Community in the Bosomtwe District of the Ashanti region of Ghana with coordinates 6° 33' 0" N,1° 38' 0"W. I t is characterized by relatively hilly terrain with the presence of valleys. 5. New Dorma: Suburban area in the Brong Region of Ghana. It is Characterised by a mixture of flat and hilly terrains covered with a lot of vegetation. It lies on coordinates 7° 16' 39" N,2° 52' 42"W. 6. Berekum: This is a Municipal located in the Bono Region of Ghana. It lies on coordinates 7° 27'N,2° 35'W. 7. Sunyani: This is an Urban populated city serving as the capital of the Bono Region of Ghana. Sunyani is surrounded by the forested Southern Ashanti uplands. It lies on coordinates 7° 20'N,2° 20'W. Modeling Para meters Parameters used in generating the path loss for the different propagation models are given in Ta ble 1. Table 1 M odeling parameters 3.3 Propagation Models The following p ropagation models were considered in this work. I. Free Space Path Loss Model II. Hata Model III.COST - 231Model IV.ECC-33 Model V. Stanford University In terim (SU I) Model VI.Ericson Model 3.3.1 Free Space Pa th Loss Path loss estimati on by this model is as given in equa tion (1)       󰇛󰇜  󰇛󰇜 (1) f =frequency in MHz , d =distance in K m parameters values Operating freque ncy 800MHz &2600MHz Transmit power 46dBm Transmitter Anten na Height Techiman 35m Adum 24m Agogo 25m Afrancho 32m New Dorma 32m Berekum 32m Sunyani 25m Shadowing facto r urban 10.6 dB suburban 8.2 dB Distance 50 m – 500 m Referenc e distance (d o ) 50m Receiver antenn a height 1.5m Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 40 3.3.2 HATA Mod el Path loss for the Hat a model as give n in [5] and [2 1 ] is given in (2). Pl urban ( dB ) = 69.5 5 + 26. 16 log( f ) - 13.8 2 lo g( h t ) - a h r + [44.9 - 6.55 log ( h t ) ] log ( d ) (2) in suburban area s path loss is compu ted as in (3 ) Pl sub urb an ( db ) = Pl urb an - 2 [ lo g( f 28 ) ] 2 - 5.4 (3) f =frequency in MHz, d=distance i n Km, h r =mobile antenna height in meters and h t = base station antenna he ight in mete rs In small and med ium cities, a h r = ( 1 . 1l o g ( f ) - 0. 7) h r - ( 1.5 6 l og ( f ) - 0. 8 ) (4) For large cities, for    (5) (5) 3.3.3 COST-231 Model The path loss equa tion for thi s model expressed in dB as given in [24] is shown below  󰇛  󰇜        󰇛  󰇜      󰇛   󰇜  󰇛   󰇜  󰇛     󰇛   󰇜󰇜  󰇛  󰇜    (6) where i s the frequency specified in , i s the distance between the base station and mobile antennas given in km,   is the base station antenna height above ground level in meters.   is t he mobile antenna hei ght in meters, is def ined as 0 dB for suburban or open environments and 3 dB for urban env ironments. is defined for la rge areas as  󰇛   󰇜      󰇛  󰇛    󰇜󰇜        (7)  󰇛   󰇜    󰇛  󰇛      󰇜󰇜       (8) In medium or sma ll cities, (9) 3.3.4 ECC-33 Path L oss Model The path loss equa tion for this mod el is given in [1 5 ] (10) a h r = 3. 2 [ l o g ( 11 .7 5 h r ) ] 2 - 4.9 7 ( ) 2 8 .2 9 l o g 1 .5 4 1 .1 f o r f 3 0 0 M H z rr hh   = −   f M H z d C m a ( h a ) a ( h a ) dB = ( 1.1 lo g ( f ) - 0.7) h r - ( 1.5 6 lo g( f ) - 0.8 ) – – fs bm b r P L A A G G =+ Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 41 Where is the free space path loss, is the basic median path loss , i s the transmitter antenna height ga in factor a nd is the receiver antenna h eight gain fac tor. Each of these param eters is express ed fully as; (11) (12) (13) When considering medium city env ironments (14) For large cities, (15) where is frequency expressed i n is the distance between the transmitter and receiver in , is the transmitter antenna height i n meters and is the receiver antenna height in meters. 3.3.5 Stanford Un iversity Interi m (SUI Model) Path loss for this model is given in (16) as present ed in [2 2] PL SUI = A + 10 g lo g( d d o ) + s for f < 2 G H z (16) 8 .2 d B < s < 10 .6 d B d is the distance be tween the trans mitter and rece iver d o =50m f is the frequen cy in MHz A = 20 log( 4 p d o l ) (17) g = a - bh b + c h b (18) Where; h b is the base stat ion antenna heigh t 10 m < h b < 80 m l is the wavelength expressed in meters. a, b and c ar e terrain facto rs specified in T able 2 fs A bm A b G r G 10 10 92. 4 20 ( ) 20 ( ) Afs log d log f = + + 2 10 10 10 2 0 .4 1 9 .8 3 ( ) 7 .8 9 4 ( ) 9 .5 6 [ ( ) ] bm A lo g d lo g f lo g f = + + +   2 10 10 ( / 200) 13.958 5.8 [ ( )] b G log hb log d =+ ( ) ( ) 42.57 13.7 10 10 – 0.585 r G log f log hr   =  +     0.75 9 1.862 r G hr =− f GHz d Km hb hr Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 42 g ( f ) = 4 4 .4 9 l o g ( f ) - 4.7 8 ( lo g( f ) ) 2 a + b + c Table 2 Terrain Par ameters Parameter Category A Category B Category C a 4.6 4 3.6 b 0.0075 0.0065 0.005 c 12.6 17.1 20 3.3.6 Ericson Mode l The equatio n specifying path loss for this model as present ed by J. Milanovic et a l [2 6] is shown in equation (19). (19) (20) The parameters          , given in equation ( 1 9 ) are constants , that can be tuned to be st fit specified propagation condition s. The default values of          for diffe ren t environment categ ories are spec ified in Table 3 Table 3 Defau lt values of        Category of Are a     Urban 36.2 30.2 12.0 0.1 Suburban 43.20 68.93 12.0 0.1 3.4 Path Loss Exponent The pat h loss e xponent which shows the lossy nature of a par ticular propagation e nvironment was computed f rom the measurement data f or eac h of the areas con si dered. [23] pr esents an approa ch to finding the path loss exponent as s hown in (21)    󰇛      󰇜    󰇛    󰇜     󰇛      󰇛    󰇜󰇜    󰇛  󰇜 Where   is t he received power at the reference distance   ,    is t he path loss at the reference distance and  is the path lo ss exponent. Pl Er ics on = a 0 + a 1lo g( d ) + a 2 log ( h b ) + a 3 log ( h b ) log ( d ) - 3.2 ( l og( 11.7 5 h m ) ) 2 + g ( f ) Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 43 3.5 Root Mean Square Error(RM SE) The RMSE which measures the difference between the signal power predicted by a m odel and the actual measured sign al was impleme nted in MATLAB . It served as a measure of accuracy t o compare forecasting er rors of the different propag ation models given the driv e test measurement data. It is defined m athematically by equation (22 ) (22) Where represents the m easured power value at a specified distance, is the predicted power value at a spec ified distance,  represen ts the number of measured samples. 4. R ESULTS The results prese nted are two-fold. The first is at an oper ating frequency of 800MHz and the second at an opera ting frequency of 260 0MHz.Results are Validated a t the end of th is section 4.1 Results at 800MHz The average received power w as co mputed for e ach of the measurement environments by averaging the readings taken at the three di fferent sector antenna s of the ba se stations. The mea n received power for the different environments was compared and analyzed by plots against varying distances from the base station using MAT LAB. This is show n in figure 2 . As can be observed from the graph, the Rec eived power decrea ses as distance awa y fr om the Bas e station is increased. Deviations from this trend, however, occurred on a few occasions. Th is was partly as a resul t of obstacles and a contributi on from the terrain of those environments. Figure 2 Receiv ed power of all Sites 4.1.1 Path Loss of Measu red data The experienced path loss at each measurement location at a distance 󰇛󰇜 was computed as follows;  󰇛  󰇜   󰇛  󰇜   󰇛󰇜 (23) 2 1 k ii k pp RM S E k  =  −   =  i p i p  50 100 150 200 250 300 350 400 450 500 Distance -100 -95 -90 -85 -80 -75 -70 -65 -60 -55 Received Power (dBm ) Techiman Adum Agogo Afrancho Dorma Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 44 where   = Mean received pow er in  ,  = Effective isotropic radia ted power in  .  is given in (24) and (25)      (24) Wher e  stands for G a ins an d  for l o sses Typical g ains con sidered are t he antenna ga ins both at the transmit ter and receiver end Typical losses are connector, body and combiner loss Expanding this yield s ,            (25)  = Transmit power (dBm),  = Gain of Transmit Antenna (dBi),  = Gain of Receive antenna (dBi),  = Connector loss (dB),  = Body loss (dB),  = Combiner lo ss(dB). The Values of t he stated parameters commonly applied in LTE Networks are given by S. A. Mawjoud [24] as;        ,        ,     ,    ,     ,     These paramete rs are substituted into equation (25)     The path l oss is obtained by substituting the calculated value of EIRP (dBm) and the m ean received powe r Pr (dBm) in to equation (23). The effect of varying distance on Path loss for each m easurement environment was investigated by plots of path loss versus distanc e and the graph sh own below in figure 3 illustrates this. Figure 3 Path loss of a ll environments It can be observed from the gr aph given in figure 3 t hat Path loss increases as the di stance from the Base station increases. Comparing the path l oss experienced for all the m easure ment environments, the P ath loss of Adum and Afrancho is relatively higher compa red to the other areas. The hilly nature of Afrancho and the presence of many high-rise buildings in Adum are good reasons to supp ort the high p ath loss in th ese areas. 50 100 150 200 250 300 350 400 450 500 DISTANCE 110 115 120 125 130 135 140 145 150 155 Path Loss (dB) Techiman Adum Agogo D. Afrancho New Dorma Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 45 4.1.2 Comparison of Pa th Loss Measure ment Results with Propagation Models The path loss of each measurement environ ment was compared with the path loss estimations of the understudied propagation models at 800MHz for both urban(Adum and Techiman) and suburban scenarios (Ag ogo, Afrancho & Dorma). Figure 4 Path Loss of Adum Compared With Path Loss of Prop agation Models Figure 5 Path loss of Techiman compared with path loss of propagation models Figure 6 Path loss of Afrancho comp ared with path loss o f propagation models 50 100 150 200 250 300 350 400 450 500 Distance 60 70 80 90 100 110 120 130 140 150 Path Loss (dB) Adum HATA SUI ERICSON fspl 50 100 150 200 250 300 350 400 450 500 Distance 60 70 80 90 100 110 120 130 140 150 Path Loss (dB) Techiman HATA SUI ERICSON fspl 50 100 150 200 250 300 350 400 450 500 Distance 60 80 100 120 140 160 180 200 220 240 Path Loss (dB) Afrancho HATA SUI ERICSON fspl Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 46 Figure 7 Path loss of Agog o compared with path loss of pr opagation mod els Figure 8 Path loss of New Dorma com pared with p ath loss of propag ation models 4.1.3 Choice of Propagat ion Model that best fits Mea surement dat a Root Mean squa re er ror w as used as a quantitativ e mea sure of accura cy for choos ing th e propagation model that best fits the measured data in the Ghanaian environment. The best propagation model was the model that had the l east Root Mean squared errors (Lea st RMSE). The RMSE computed for the measurement areas together with the various propa gation models are given in Tabl e 4. Table 4 RMSE V alues Root mean square e rror(urban) Environments Hata model SUI model Ericson model Techiman 30.66 32.96 44.41 17.98 Adum 40.64 40.39 52. 31 25.17 Root mean square e rror(suburban ) Agogo 50.22 45.88 173.99 Afrancho 52.39 48.48 171.18 New Dorma 44.03 39.21 182.86 50 100 150 200 250 300 350 400 450 500 Distance 60 80 100 120 140 160 180 200 220 240 Path Loss (dB) Agogo HATA SUI ERICSON fspl 50 100 150 200 250 300 350 400 450 500 Distance 60 80 100 120 140 160 180 200 220 240 Path Loss (dB) Dorma HATA SUI ERICSON fspl Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 47 The Ericson model had the l owest RMSE values in Urban environment s as shown in Table 4.Thi s model is therefore chosen as the model that predicts be st in Urban a reas in Ghana and it is f urther modified and improved for more accurate predictions. In the suburban environments, the SUI model had the lowest RMSE values and hence was chosen as the best m odel for path loss prediction in subur ban cities in Ghana. I t is also fu rther modified for a more accura te prediction. 4.1.4 Modification o f Ericson Model The Ericson model which best fit measurement in the urban environments was chosen and modified to fit the measured data in urban environments. To m odify and further improve t he Ericson model the mean square e rror betw een the urba n environments and the Ericson mo del wa s added to the stan dardized Eric son path loss equ ation.     󰇛  󰇜     󰇛   󰇜     󰇛   󰇜   󰇛  󰇜   󰇛󰇛      󰇜󰇜       (2 6)  for Adum =17.98 Adding the RMSE y ields;      󰇛  󰇜     󰇛   󰇜     󰇛   󰇜   󰇛  󰇜   󰇛󰇛    󰇜󰇜          (27) This new equa tion w ith the  add ed was plo tted with me asurement data from A dum together with the ini tial standardi zed Ericson m odel equation and the g raph is show n in figure 9. It can be observed that adding the RMSE to the initial equat ion improves the accu racy of prediction as th e modified Ericson e quation fi ts best with the m easured data. Figure 9 Com parison of th e modified mod el and original Er icson model for Adum The values of various parameters                in the Ericson model suited for an urban area were substitute d into the modified equation and approximated to make the Ericso n equation si mple and less tedious to use yet no t comp romising accuracy. The res ulting eq uation is as in equation (28)        󰇛 󰇜  󰇛 󰇜 (28) A similar a nalysis was carri ed out for T echiman and f igure 10 shows the modified Eri cson model fitting closely to measurement data from Techim an 50 100 150 200 250 300 350 400 450 500 Distance 100 120 140 160 180 200 220 Path Loss (dB) ADUM ERICSON ORIG ERICSON MODIFIED Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 48 Figure 10 Comparison of the modified model and original Ericson mod el for Techiman 4.1.5 Modification o f SUI Mod el R esults o f RMSE for suburban areas favor ed the SUI model which had the lowest RMSE valu es. On this basis, the SUI model was c hosen as the best-fit pr opagation model for path loss estimation in suburban areas in Ghana. It was further modified for m ore ac curate predictions in Ghanaian suburban environ ments. The RMSE each of Ag ogo, A franch o and New D orma were added to the original SUI equa tions and fur ther simplif ied in equations ( 30 ) - (32)        󰇡    󰇢     (29) 1) modified mode l for Agogo  󰇛󰇜        󰇛    󰇜   (30) 2)modified mode l for Afrancho  󰇛󰇜        󰇛    󰇜   (31) 3)modified mode l for New D orma  󰇛󰇜        󰇛    󰇜   (32) A graph comparing t he performance of the modified model for Agogo with the original SUI model is shown in figure 11 Figure 11 Comparison of modified SUI mod el and original SUI m odel for Agogo 50 100 150 200 250 300 350 400 450 500 Distance 100 120 140 160 180 200 220 Path Loss (dB) Techiman ERICSON ORIG ERICSON MODIFIED 50 100 150 200 250 300 350 400 450 500 Distance 70 80 90 100 110 120 130 140 150 Path Loss (dB) AGOGO SUI ORIG SUI MOD Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 49 4.2 Results at 2600 MHz Comparing the path loss predictions of propagation models at 2600MHz with drive t est measurements of five dif ferent propaga tion environ ments, th e EC C-33 m odel pred icted close to the measurement data. T his m odel was further modified to predict more accurat ely the path loss in these envi ronments. This is show n in figures 12 & 13 Figure 12 Compar ison of modified ECC- 33 mo del and original ECC- 33 mo del for urban enviro nments Figure 13 Compar ison of modified ECC- 33 mo del and original ECC- 33 mo del for suburban environments 4.3 V ALIDATION The dev eloped mod els in this study wer e va lidated by calculatin g th e e rror betwe en the measure d and estimated path loss for the various measurement environm ents using the modified equations presented. This is achieved by using equa tion (28) programmed in MATLAB . The values of RMSE closer to zero indicate a better fit [2 5],[11]. Thus, the developed models are described as valid and suitable for t he test ed environments since the RMSE between t he measured and the predicted path loss value s are clo ser to zero than th e initial RMSE v alues. Tables 7 & 8 show the RMSE generated using the developed models in thi s thesis at 800MHz and 2600MHz respectively. 50 100 150 200 250 300 350 400 450 500 Distance 105 110 115 120 125 130 135 140 145 150 155 Path Loss (dB) Urban ECC33orig ECC33modified 50 100 150 200 250 300 350 400 450 500 Distance 115 120 125 130 135 140 145 Path Loss (dB) Suburban ECC33orig ECC33modified Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 50 Table 7 RMSE Values using developed mo dels at 800MHz Root Mean Squar e Error Values o f Developed Mod els at 800M Hz Measurement Env ironment Root Mean Squar e Error Adum 11.8494 Techiman 9.6717 Agogo 5.2510 Afrancho 7.1129 New Dorma 29.8491 Table 8 RMSE values using developed mod els at 2600MHz Root Mean Squar e Error Values o f Developed Mod els at 2600M Hz Measurement Env ironment Root Mean Squar e Error Site 1 9.1408 Site 2 13.3313 Site 3 16.8445 Site 4 15.2780 Site 5 11.9498 The improved m odels developed were further compared with the path loss simulated by the use of the NYUSI M simulator. This is a simulation tool developed by the New Yor k Universi ty (NYU) wireless team and relies on h uge amount s of true measured data at mm-wave frequencies in New York[27]. The simulator incorpora tes the CI propagat ion model [28]. Developed models s how consistent pred iction behav ior compared with th e NYU simula tor’s path l oss and hence can be considered valid m odels for use i n t he Ghanaian environment with similar environmental features as the m easuremen t environments. Figure 14 Comparison of Performance of Developed Models against NYUSIM at 2600MH z 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Distance 110 120 130 140 150 160 170 180 Path Loss (dB) techiman 1 techiman 2 techiman zongo sunyani brekum Nyusim Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 51 Figure 15 Compar ison of Performance of Dev eloped Models against NYUSIM at 800MHz 5. CONCLUS ION This s tudy w as f ocused on developing improv ed versions of industry-standa rd pr opagation models suited for LTE path loss prediction in the Ghanaian environmen t. Path loss of four propagation models was compared with Path loss of propagation measurements taken from five LTE 800MHz base stations located i n the urban and suburban areas of Ghana using MATLAB. Results confirmed t he i nitial assumption of the st udy, that propagation models predict far from the ideal. The Ericson m odel showed satisfac tory performance in t he urban environme nts at 800MHz. This model however over predicted the path loss in the suburba n environment s. The SUI model outperformed the othe r models in predi cting close to th e propaga tion measur ement in suburban areas at 800MH z.The Ericson and SUI models were further improved for a more ac curate predict ion of LTE path loss in urban and suburban Ghanaian en vironments at 8 00MHz. For similar studies at an LTE frequency of 2600MHz, t he Pat h loss of five propagation models was compa red w ith the Path loss of propaga tion measuremen ts taken at f ive base stations located in th e u rban and suburban a reas of Gha na. Th e ECC-33 model best fit propagat ion measurement s both in the urban and suburban environments and hence it was developed further for use in LTE path loss estima tion at 2600MHz From the resul ts presented, measurement data ascer tained the f act that propaga tion models pr edict far fr om the ideal. The modified equations presented in this paper can be use d i n Ghanaian settings having similar charac teristics with the Areas consider ed in this pap er by network operators for accu rate and s implified path loss predict ion. R EFERENC ES [1] Cisco, V. N. I. (2018) . Cisco visual network ing index: Forecast and trends, 201 7 – 2022. White Paper, 1. [2] National Commu nication Authority Accessed date:8 th July 2 018. https://www.nca.o rg.gh/ ind ustry - data-2/mar ket-share-statistics-2 /data- 3/ [3] World Pop ulation Review. ( 2019). Ghana population . Retrieved 10 th May 2019 fromhttp://wo rldpopulationreview.com/cou ntries/ghana -population 50 100 150 200 250 300 350 400 450 500 Distance 80 100 120 140 160 180 200 220 240 260 Path Loss (dB) Adum Techiman Agogo Afrancho Nyusim Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 52 [4] Nimako, S. G. , & Mensah, A. F. (201 2). Motiv ation for customer com plaining and non -comp laining behavior toward s m obile telecommunication ser vices . Asian Journal o f Business Managemen t, 4(3), 310 - 320. [5] Rappapor t, T. S. (1996). Wireless communicatio ns: prin ciples and prac tice (Vol. 2). New Jersey: prentice hall PTR. [6] Chebil, J., Lawas, A . K., & I slam, M. D. (2 013). Compariso n between measur ed and predicted path loss for mob ile communication in Ma laysia. World App lied Sciences Journal, 21, 123 -1 28. [7] Mollel, M. S., & Kisangiri, M. (20 14). Comp arison of empirical propag ation path loss mod els for mobile com munication. Computer Eng ineering and Intell igent Systems, 5 (9), 1- 10. [8] Ibhaze, A. E., Imo ize, A. L., Ajose, S. O., John, S. N., Ndujiuba, C. U. , & Idachaba, F. E. (2017). An empirical propagation mod el for path loss pred iction at 2100M Hz in a d ense urban environmen t. Indian Journal of Science and Technology, 10(5), 1 - 9. [9] Ajose, S. O., & Imoize, A. L. (20 13). Propag ation measuremen ts and modeling at 1800 MHz in Lagos Nigeria. International Journal of Wireless an d Mobile Computin g, 6 (2), 165 -174. [10] Imoize, A. L., Ibhaze, A. E., Nwosu , P. O., & Ajose, S . O. (2019). Dete rmination of Best -fit Propagation Mod els for Pathloss Prediction of a 4G LTE Network in Suburban and Urban Areas of Lagos, Nigeria. West Indian Journal of E ngineering , 41(2). [11] Halifa, A., Tchao, E. T., & Kponyo, J. J. (2017). I nvestigating the Best Radio Propagation Model for 4G -WiMAX Netwo rks Deployment in 2 530MHz Band in Sub -Sahar an Afr ica. arXiv preprint arXiv:1711.0806 5. [12] Abhayawardhana, V. S., Wassell, I. J., Crosby , D., Sellars, M. P., & Brown , M. G. (2005, May). Comparison of empirical propagation path loss mod els f or fixed wir eless ac cess system s. In 2005 IEEE 61st Veh icular Technology Conference(Vol. 1, pp. 73 -77). [13] Roslee, M. B., & Kwan, K. F. (2010). Optimization o f Hata propagation prediction model in sub urban area in Malaysia. Progress in Electromag netics Research, 13, 9 1 -106. [14] Kale, S., & Jadhav, A. N. (2013). An Emp irically Based Path Loss Models for LTE Ad vanced Network and Modeling for 4G Wirele ss Systems at 2.4 GHz, 2 .6 GHz, an d 3.5 GHz. International Journal of App lication or Innovation in Engineering & Management ( IJAIEM), 2(9), 252 -257. [15] Ekpenyong, M., I sabona, J., & Ekong , E. (2010). On Propagation Path Loss Models For 3 -G Based Wireless Networks: A Com parative An alysis. Computer Scien ce & Telecom munication s, 25(2). [16] Faruk, N., Popoo la, S. I., Surajudeen -Bakin de, N. T., Olo yede, A. A., Abdulkarim , A., Olawoyin, L. A., ... & Atay ero, A. A. (2 019). Path Loss Predictions in the V HF and UHF Bands Within Urb an Environments: Ex perimental Investigatio n of Empirical, Heuristics and Geospatial Mod els. IEEE Access, 7, 772 93-77307. [17] Khan, I., Eng, T. C., & Kamboh, S. A. (2012) . Performance Analy sis of Var ious Path Loss Models for Wireless Network in Different Environmen ts. International Journal of Engine ering and Advanced Technology (IJEAT) ISSN, 2249 -8958. [18] Tchao, E. T., Gadze, J. D., & Agyapong, J. O. (2018). Performance evaluation of a deployed 4G LTE network. ar Xiv preprint arX iv:1804.05771. [19] Unterhuber, P., Pfletsching er, S., Sand, S., Soliman, M., Jo st, T., Arriola, A., ... & Rodríguez, C. (2016). A survey of channel measurements and models for current and future railway communication systems. Mobile In formation Systems, 201 6. Internation al Journal of Wireless & Mobile Networks (IJWMN) Vo l. 11, No. 6, Decemb er 2019 53 [20] Hata, M. (1980). Empirical form ula for propagation loss in land mob ile radio ser vices. IEEE transactions on Vehicular Technology, 29(3), 317 - 325. [21] Neskovic, A., Neskovic, N., & Paunovic, G. (2000). Modern approaches in modeling of mobile radio systems propag ation environment. IEEE Communications Survey s & Tutorial s, 3(3), 2 - 12. [22] Sulyman, A. I., Nassar, A. T., Samimi, M. K., MacCartn ey, G. R., Rappaport, T. S., & Alsanie, A. (2014). Radio propag ation path loss models for 5G cellular networks in the 28 GHz and 38 GHz millimeter-wav e bands. IEEE Commun ications Mag azin e, 52(9), 78- 86. [23] Zhou, T., Sharif, H., Hempel, M., Mahasukhon, P., Wang, W., & Ma , T. (2009, October). A deterministic approach to e valuate path loss ex ponents in large -scale outdoor 802 .11 WLANs. In 2009 IEEE 34th Confer ence on Local Computer Net works (pp. 348 - 35 1). [24] Mawjoud, S. A. (2013). Path loss propagation model prediction for GSM netwo rk p lanning. Internation al Journal of Computer App lications, 84(7). [25] Wu, J., & Yuan, D. (1998, September). Propagation m easurements and modeling in Ji nan city. In Ninth I EEE Internatio nal Symposium on Personal, In door and Mobile Radio Comm unications (Cat. No. 98TH836 1) (Vol. 3, pp. 1157 -1160). [26] Milanovic, J., Rimac-Drlje, S., & Bejuk, K. (200 7, December) . Compar ison of propagation mo dels accuracy for WiMAX on 3 .5 GHz. In 20 07 14th IEEE International Con ference on Electronics, Circuits, and Systems (p p. 111- 114). [27] Rappaport, T. S., Sun, S., & S hafi, M. (2017, September). Investigation and comparison of 3GPP and NYUSIM channel models for 5G wireless communications. In 2017 IEEE 86th Vehicular Technology Conference (VTC-Fall) (pp. 1-5). [28] Sun, S., MacCartney, G. R., & Rappap ort, T. S. (2017, May). A novel millimeter -wave channel simulator an d applications for 5G wireless communication s. In 2017 IEEE International Confer ence on Communicatio ns (ICC) (pp. 1 -7).