Performance Evaluation of VoLTE Based on Field Measurement Data

www.u5gig.ae Performance Evaluation of V oL TE Based on Field Measurement Data Aym a n E ln as h ar, M o ha m ed A. El - Saidny , and Mohamed Y eh ia Abstract — Vo i c e o v e r Long - Te r m E v o l u t i o n (V oL TE) has been witnessing a rapid deployment by network carriers worldwide. Duri ng t he phases of V oL TE deployments, carriers would typicall y face challenges in understanding the factors affecting the V oL TE performance and then optimizing it to meet or exceed the performance of the legacy circuit switche d (CS) network (i.e., 2G/3G) . The main challenge of Vo LT E s e r v i c e q u a l i t y i s t h e LT E network optimization and the pe rformance aspects of the service in different L TE deploy ment scenarios. In this paper , we present a detailed practica l perform a nce analysis of V oL TE based on comm ercially deployed 3GPP Release - 10 L TE network s . The analys is evaluates Vo LT E pe r f o r m a n c e i n t e r m s o f r e a l - time tran sport pro tocol (RTP) error rate, RTP jitter and delays, block error rate (BLER) in dif ferent rad io conditions and V oL TE voice quality in terms of mean opinion score (MOS). In addition, the paper evaluates key V oL TE features such as RO bust Header Compression ( ROHC ) an d transmission time interval (TTI) bundling . This paper provides guidel ines for best practices of V oL TE deployment as well as pra ctical performance evaluation based on field measurement data from commercial LT E networks . Index T erms — LT E , V o LT E , RLC, RTP , ROHC , TTI Bundling, BLER , MOS 1 I NTRODUCTION Vo i c e o v e r Long - Te r m E v o l u t i o n ( Vo LT E ) is an IP multimed ia sys tem (IMS ) - based voice service over the L TE network [1]. The IMS supports various access and multimed ia s ervices and has re ce nt ly become the standard architectur e of evolved packet core (EPC) [1], [2]. 3GP P h as adopted GSMA IR.92 IMS profile for voice and SMS [3] and GSMA IR.94 IMS pr ofile for conversational video [4] t o provide high quality IMS - based telephony services over L TE radio access ne twork. The pr ofiles define optim al sets of existing 3GPP - specifi ed functionalities that network infra - vendors, service providers a nd handset manufacturers can use to offer compatible L TE voic e/video solutions. Therefor e, the commer cial deployment of V oL T E mandates extensive testing b etween term inals and networks includ ing L TE ra dio access n etwork ( i.e., eNodeB (eNB ) ), L TE EPC , and IM S. In addition to these challenges , the Vo L T E o p t i m i z a t i o n f o r different radio/loading condi tions to f ind an accept able tradeof f between the user ’s experience and netw ork deployment complexi ties has led to a substantial delay in widely adopting Vo L T E service . In this paper , we addr ess these challenges by provid ing the best practice s for V oL TE rel at ed fe at ure s a nd pr ac ti c al p er f or man ce evaluation based on field - testing results from commercial LT E 3 G P P R e l 1 0 networks. The deployment of V oL TE brings variety of benefits to telecom operators as v oice is still the main source of r evenue. Hence, t elecom carriers need voi ce e voluti on to effectively compete with o ver - the - to p (OTT) voice over IP ( Vo I P ) applications that create a significant load on the mo bile broadband networks and accordingly affecting other services. V oL TE also imp roves the spectral ef ficiency and red uc es n et wo rk c os ts compar ed to legacy circuit switched (CS) networks . Moreover , the spectral efficiency of L TE networks is higher than the traditional GSM/UM TS networks, which makes V oL TE a really suitable voice solution in 4G networks [5], [ 6]. For the same channel bandwidth, an L TE cell offers twofold cell - edge throughp ut of a UMTS cell [5], [6]. Vo LT E e n h a n c e s t h e e n d - user ex perience by providing a better quality of e xperience (QoE) . Vo LT E i s e q u i p p e d w i t h a numerous set of integrated fe atures that improves QoE aspects such as better call setup time, higher efficiency i n deep cove rage conditions and lower battery consumption. Moreover , Vo L T E s u p p o r t s d i f f e r ent range of speech codec rates i.e., adaptive multi - rate (AMR) with both wide - band (AMR - WB) and narrow - band ( AMR - NB) and e nhanced voice services (EVS) codec . While high d efiniti on (HD) voice is being rolled out by telecom operators using AMR - WB with audio bandwidth of up to 7 kHz, 3GPP Rel - 12 introduced EVS codec to offer up to 20 kHz audio bandwidth [7] . EVS at 13.2kbps provides super - w ideband ( SWB) voice quality at comparable bit - rate to AMR and AMR - WB and of fers high rob us tn es s to ji tt er a nd p ac ke t l os se s. The feasi ble solutions for providing voice call and service continuity over L TE - based net works , a comparison between various aspects of these s olutions , and a possible roadmap that mobile operators can adopt to provide seamless v oice over L TE are provided in [8 ]. A comprehensive evaluation and validation of V oL TE quality of service ( QoS ) is pro vided in [9]. The r esults in [9 ] give clear evidence th at the V oL TE service fulfills the ITU - R and 3GPP standard r equirements in terms of end - to - end delay , jitter and packet loss rate. The Vo LT E p e r f o r m a n c e i n t e r m s of Quality of Service (QoS) is evaluated and validated using OPNET modeler wireless suite in [9]. T he r ailway voice communication based on Vo LT E is proposed in [ 10 ]. The simulation results in [10 ] indicate that V oL TE is a viable option for provid ing railway voice com munication i.e., GSM - R. Therefor e, L TE is proven to be a strong candidate for becoming the future commun ication network for railways [1 1 ]. The feasibility of semi - persistent schedulin g (SPS) for V oIP is an alyzed in [12 ], which evaluate s its performance in terms of through put of random access dela ys and traffic channels . A m e t h o d o l o g y to evaluate the V oice - over - IP (V oIP) capacity and performance of orthogonal frequency - division multiple - access (OFDMA) - based systems is pr ovided in [13 ]. This methodology can also be applied to V oL TE. A m ethod for estimating cell capacity www.u5gig.ae from network me asurements in a multiservice l ong - term evolution (L TE) system i s presented in [14 ]. The performance of circuit switched fallback (CSFB) , Vo LT E a n d enhanced single radio voice call continuity (eSR VCC) are key elements to guarantee seamless voice experience wit hin the L TE/UMTS/2G networks [15] - [20], especially t hat a mix of V oL TE and CSFB devices can coexist in the same L TE network. The three features were analyzed in [6], [20] to evaluate the call setup delays and the details behind the interruption ti me wi thin L TE network or over e SR VCC . It is demonstrated in [20 ] that V oL TE provides a better end - user experience i n terms of call set up delay . However , the eSR VCC voice inter ruption is hi gher than the LT E i n t r a - freq uency interr uption by ~ 200 m sec, but still w ithin the acceptable audio quality range of 300 mse c. A comprehens ive performance evaluation of RO bust Header compression ( ROHC ) for V oL TE by means of a testbed implementatio n is presented [21]. In [22], it is envisaged that qu ality of call measured through mean opinion s core (MOS) is always better in L TE compar ed with UMTS network. The Vo LT E M O S c o n d i t i o n o n l y b e c o m e s u n s t a b l e u p o n r e a c h i n g very poor RF condi tion – where re fe ren ce sig na l rec ei ve d power (RSRP) < - 11 7 d B m a n d refe ren ce si gn al re ce i ve d q uality (RSRQ) < - 12dB. The implementation of transm ission time interval ( TTI ) bundli ng feature helps to improve the uplink ( UL ) coverage an d minimizes the BLER . It is pr oposed in [22] that serving cell SR VCC thr eshold RSRP = - 108dBm for networks withou t TT I bundling and RSRP = - 11 7 d B m f o r networks with TTI bundling deployment. Ho wever , this paper demonstrates that RSRP = - 11 0 dBm is the optimum thresh old to maintain voice quality even with TTI bundling deployment, as real - time tra nsport pr otocol (R TP) performance can be adver sely impacted beyond t his level . The SR VCC requires UE to support the abi lity to transmit and receive , sim ultaneously , on bot h net works ( Packe t Switch (PS) such as L TE and Circuit Switch (CS) such as 3G ) . SR VCC went through differ ent stages in the standard in order to reduce the voice int erruption ti me that impacts the user ’s experience as well as improving the call s etup success rate at different stages of the V oL TE call. 3GPP has started with the support of SR VCC in Release 8/9 a nd then enhanced the mechanism to support enhanced SR VCC (eSR VCC) in Release 10 [20] . The main target of the eSR VCC is to reduce the voice interruptio n during the inter - technology handover . eSR VCC targets an interruption of < 300 msec. In this p aper , we p rese nt compr ehensive practical perfo rmance analysis of V oL TE performance based on commer cially deployed 3GPP Release - 10 L TE network s . The analysis demonstrates V oL TE performance evaluation in terms of RT P err or rat e, RTP jitt er and delays , BLER and Vo LT E v oice quality in terms of MOS . In additio n, this pape r evaluates key V oL TE featur es such as RO HC , T TI bundling and SPS . The rem aining of the paper are organized as follows . Vo LT E p r i n c i p l e s a r e s u m m a r i z e d i n s e c t i o n I I . M a i n Vo LT E f e a t u r e s a l o n g w i t h d e p l o y ment best practi ces are outlined in section III. T esting environm ent is explained in section IV . Practical performance analysis is provided in section V . The conclusions and future work are summarized in section VI. 2 V O LT E P RINCIPLES The IMS server carri es all sources of V oL TE traf fic and provides f unctions including subscribers’ re gi s tr at io n, authentication, control, r outing, switching, media negotiation and conversion . The voice co dec type and r ate of Vo LT E s e r v i c e a r e n e g o t i a t e d a t c a l l s e t u p b y t he UE (User Equipment) and IMS. The eNB pr ovides only the media - plane bearer channel and the IMS signaling is transpar ent to the air interface. After the V oL TE ca ll is establ ished and the media pa ckets start flowing, the eNB performs dynamic scheduling an d uses power c ontrol policies that are suitable for such scheduling. The eNB selects a modulation and coding scheme (MCS) index and physical r esour ce b locks (P RBs) for voice user s similar to the mechanism in scheduling the PS data. The main purpose of the Vo L T E s c h e d u l i n g t e c h n i q u e is to maintain continuous transmission on uplink and downlink in a way that minimizes the packet delays. The data packets for voice se rvices have a relatively small and fixed size, and therefor e, scheduling RT P packets on the rad io and core network requires stringent delay budget to control the inter - arrival time that minimizes the j itter . There are several pr otocol interfaces used between the UE’s IMS client and network ’s I MS se rver . They are t he language between the IMS client/ server to exchange either signaling information or actual media packets (i.e. voice packets). The s ummary of these protocols is as follows: a) SIP: delivers IMS signaling to negotiate a media se ssion between two users. b) RT P : d e l i v e r s I M S m e d i a p a c k e t s . c) RT P c o ntrol protocol (R TCP): used to synchr onize streams by providing feedback on QoS information. d) IP Security (IPSec): used to carry the authentication and key agreement (AKA) as a secured tunnel for IMS clients. Due to the delays in V oL T E net work from differen t network elements (i.e., eNB, EPC, IMS, and transport network) , the R TP packets inter - arrival time can vary in time. The R TP packets during ta l k spurts are generated every 20 ms. However , the pac kets do not arri ve p recisely at that exact interval. This means that the V oL TE packets cannot be played out as they arrive due to variance in packet arrivals. Jitter i s defined as a statistical variance of the R TP data packet inter - arri val time. In PS networks, the occurrence of variable delay is much higher than the values in CS networks. Fi gure 1 illustrates the concept of j itter and the delay requir ements to keep the inter - arrival time within an acceptable range for better j itter buff er management. Jitter buffers are used to change asynchronous packet arrivals into a synchronous stream by tuning variable network del ays int o consta nt delays at the destination . The role of the jitter buffer is to set a trade - off between delay and the pro bability of interrupted playout because of late packets. Late or out - of - order pa ckets are discarded. IMS clients shall implement adaptive jitter buffers t o overcome thes e issues by d ynamically tuni ng the jitter buffer size to the lowe st acceptable value. I ncreasing the buffer size can increase latency . Even if the RTP packets rem ai n i n t he c or rec t s eq ue nc e a nd t he re i s z ero pa ck et l o ss , large variations in the end - to - end transmission time for the packets may cause degr adation of aud io quality that can only be fixed thr ough the u se of the jitter buf fer . T ypi cal R TP error www.u5gig.ae r ate is < 1% for audio and < 0.1% for v ideo [23]. On the other hand, the radio network scheduler has a key role by securing sufficient scheduling re so urc es in a ll ra di o co nd it io ns in order to keep the R TP error rate, jitter and end - to - end delays within a range t hat can be efficient ly h andled by the jitter buffer . The V oL TE call incl udes an end - to - end voice/media flow transmitted on a dedicated guaranteed bit rate (GBR) bearer with QoS class i dentifier (QCI) = 1 through R TP protocol, user datagram protocol (UDP) or IP pr otocol. Another default bear er for session initiati on protocol (SIP) signaling is established beforehand using QCI = 5, through UDP protocol, t ransmission control protocol (TCP) or IP protocol. The IMS network and evolved packet syst em (EPS) tra nsfer the IMS access point name (APN ) as well as the IMS PDN connection that host the tw o IMS bearers. The packet - switched (PS) services continue to use the default packet data network (PD N) connecti on with a non - GBR bearer (e.g. QCI = 9). Radio bearers with QCIs of 1, 2, and 5 are esta blished between two V oL TE UEs to carry conversational voic e, signaling, and video, respectively . The eNB performs admission and congestion control for conve rsational voice (QCI = 1), signaling (QCI = 5), and video (QCI = 2). Mor eover , the eNB performs dynamic or SPS scheduling and uses power contr ol policies that ar e suitable for dynamic or SPS scheduling. Fig. 1. The concept of jitter and j itter buffer management 3 Main VoLTE F EA TURES The eNB scheduler is typically designed to efficiently schedule the UEs during the DL and UL transmissions for either sm all or large packet sizes. However , the V oL TE media stream has small voice packets with fi xed inter - arrival intervals. Therefor e, the eNB scheduler work s with various special features specific for V oL TE in order to enhance the c overage, c apacity a nd quality of the voice calls. The se eNB feature s are ROHC , TTI Bundling, and SPS , wh ich are all designed to assist the V oL TE in - call performance . These feature s will be presente d in theses section. 3.1 Ro bust Head er Compression ROHC is a header compr ession pro tocol originally designed for real - time audio/video communica tion in wireless environment. As V oL TE media p ackets are transported as IP traffic, the generated headers of the IP protocols can be massively large. ROHC compresses t he R TP , UDP , and I P heade rs to reduce th e size of the entire voic e packets. This decreases the radio resou rce utiliza tion on the cell - edge and therefor e impr oves the overall cell cov erage on both UL and DL. In addition, it reduces the number of voice packe t segments, which impr oves the BLER associated with these smaller - size tr ansmissions. This improve s the V oL TE end - to - end delays and jitter . The ROHC oper ation is ill ustrated in fi gure 2 . The ROHC function in L TE is part of Layer - 2 at the user plane of the UE and eNB. Both UE and eNB behave as a compressor and de - compressor for the user - plane packet s on UL and DL. The compression efficiency depends on the ROHC operating mode and t he variat ions in t he dynamic part of t he packet headers at the application laye r. A h e a d e r c a n b e c o m p r e s s e d to one byte with ROHC, which efficiently reduces the voice packet size. ROHC in L TE oper ates in three modes: U - Mode, O- Mode, and R - Mode (Uni directiona l, Bidirect ional Optimisti c and Bidirectiona l Reliable , respectively) . The r eliability of these modes and overheads used for transmitting feedback are differen t. In U - Mode, pac kets c an only be sent from the compressor to the de - compr essor , with no mandatory feedback channel. Compared with O - Mode and R - Mode, U - Mode ha s t he lowest r eliability but requires the minimum overhead for feedback. In O - Mode, the de - compressor can send feedback to indicate failed decompression or successful context update. Therefor e, it provides higher reliability than U - Mode but it gener ates less feedback co mpared to R - Mode . I n R - Mod e, context synchronization between the compressor and de - compr essor are ensured only by the feedback. That is, the compr essor sends the context updating packets r epeatedly until acknowledgment is received from the de - compressor . T herefor e, R - Mode provi des the highe st reliab ility but generates the maximum overhead due to the mandatory acknowledgment. Fig. 2. ROHC Operation Mech anism 3.2 Tr ansmission Time Interval B undling TTI bundling enables a data block to be transmitt ed in consecutive TTIs, which are packed together and treated as the same resour ce during the scheduling process. TTI bundling makes full use of hybrid automati c repeat request (HARQ) gains and therefor e red uces the number o f www.u5gig.ae ret ra ns mi ss io ns an d rou nd t r ip t ime (RTT). When the user ’s channel quality is degraded or the transmit power is limited, the TTI bundling is trig gered to impr ove the up link co verage at cell - edge , r educe the number of differ ent transmissio n segments at the radio link control ( RLC ) layer and the ret ra ns mi ss io n o ve rhe ad . T he ma in advantage of TTI bundling is enhancing Vo L T E u p l i n k c o v e r a g e whe n the UE has limited uplink transmit power . Thus, it guarantees Vo LT E Q o S f o r cell - edge us ers. In a conventional scheduling mechanis m, i f the U E is not ab le t o a ccumulat e su fficient power to transmit a small amount of data, like a V oI P packet, the data packets can be segmented into smaller size packets that fit within the UE transmit power . Eac h segment will b e transmitted with a separate HARQ proce ss. Ho wever , this segmentation mechanism increases t he amount of control information that needs to be transmitted. Therefor e, the control channel load increases with the amount of segments since every segment requir es new transmission resour ces on these channels. Additionally , the probability for HARQ feedback erro rs increases with the number of segments causing higher BLER at cell - edge . T herefor e, the ne ed to utilize b etter segmentati on method li ke TT I Bundling is important. For UEs at cell - edge RLC segmentation is done fir st, then TTI bundling transmit same packets four times in one scheduling period to extend the coverage by increasing uplink transmission reliabilit y . The eNB decides when to activat e or deactivate the TTI bundling f or certai n users based on the measured signal - to - interfer ence noise ratio ( SINR ) and PRB usage on uplink. T he data block in a bundle of TTIs, where the chunk of each bundle (up to 4 chunks) , is modulate d with different redundancy version s with in the same H ARQ identity . In the case of un successful deco ding of the HARQ identity , the eNB sends negative acknowledgement the UE, which requir es retransmission. The resource allocation during this ope ration is restricted to a certain number of PRB and transport block size (TBS) in order to improve the probability of decoding at lower data rate s . The mech anism to transmit same packets four times in one scheduling instance expands the c overage by increas ing the uplink transmission reliability with a better success rate gain. In addition, it guarantees that V oL TE packets are transmitted at cell - edge when resources are limi ted to impr ove the latencies in bad radio conditions . TTI bundli ng is estimated to pr ovide 2 - 4 dB uplink coverage impr ovement for V oL TE servi ces which extend the cell radius for V oL TE service [ 25 ]. Figur e 3 describes t he TTI bundling mechanis m and provide comparison with a scheduling mechanis m t hat depend s only on RLC se gmentati on proced ur es [ 26 ]. 3.3 Se mi - persistent scheduling SPS reduces the control signaling ove rhead on the air interface while increasing the overall system capacity by means of s chedulin g the UE to re ceive th e regular P RB resources in a fixed period with no scheduling grant on the physical downlink control channel (PDCCH) . The feature is needed in scenarios where voice service UEs and other data UEs c oexist in same cell . Thu s, t he i ncrease in the number of PRBs allocated to voice service U Es w ill cause a d ecrease in the number of PRBs available to other data UEs, and consequently the cell throughput (capaci ty) will decrease. In SPS, th e allocated traf fic channel i s released when a certain number of empty transmission sl ot s are detected on the allocated data traffic channel. This is effectively achi eved without ch anging the amount of resources or the packet size (i.e. MCS), at the beginning of SPS allocation period (e.g. 10, 20, 32…640 subframes). With this reduction mechan ism in air inter face, SPS scheduling offers up to 2.5x capacity improvemen t over the conventional dynamic scheduling in limited PDCCH scenario i.e., 5 PDCCH [27] . However, the dynamic scheduling mechanism is adopte d in most deployed networks unless VoLTE l oad reaches the required SPS activation threshold (i.e. significant increase in load w ith sub - optimal control channel capaci ty). Therefore, this paper does not present practical results for SPS since it is not widely deployed . 4 T ESTING Environment and VoLTE M ain KPIs In this paper, the VoLTE performance is assessed in terms of ROHC efficiency , RTP error rate, jitter, DL/UL BLER, handover delays and voice quality in terms of MOS . The data is p rocessed from field measurement with a large sample s ize (i.e., long VoLTE ca lls during mobility ) and results were averaged over two different LTE access networks (i.e., two different clusters) from two different suppliers and using two different smartphones to capture the main trend and mi tigate any network or handset impact . The Two di fferent network infra - vendors are tested where both a re deployed with 1800MHz (20MHz channel) and collocated with UMTS at 2100MHz and GSM900/1800 MHz . The K ey P erformance I ndicators ( KPIs ) are estimated from the device side through post processing script s t o the coll ect ed logs from the device modem. Tabl e I summarizes the LTE network parameters, average RF conditions in mobility and VoLTE related features . The testing is con ducted i n live commerc ial LTE networks wi th normal loa ding i.e. , ~ 50%. The testing is conducted with and without PS data sessi on to evaluate the impact of concurrent services (voice and data) on the VoLTE performance. The testing methodology is illustrated in fig ure 4. Th e eNB st rategy is based on proportional fair scheduler with d ynamic a ssignment based on load and channel condit ions. QCI for VoLTE is always 1 and for data i t can be 6, 8 or 9 (9 is used in the tested network). The scheduler differentiat e s voice and data base d on QCI=1 or QC I = 9 giving priority to QCI=1 when a conflict occurs (e. g. channel condition i s not suitable to schedule data from both services at same time , or in case of netw ork congestion ). T ABLE I N ETWORK AND V O LT E P ARAMETERS Configurati on DL/UL LTE System Bandwidth 20 MHz UE Category 6 MIMO Confi guration MIMO 2x2, TM3 Mobilit y Speed 80 km/h RF Condition s in Mobility Average Valu es Serving Cell RSRP [dBm] - 83.8 Serving Cell RSRQ [dB] - 8.7 Serving Cell RSSI [dBm] - 54.9 Serving Cell S IN R [dB] 20.2 VoLTE Relevant Parameter Value ROHC ON TTI Bundling ON Dynamic Sche duling ON SPS OFF C- DRX for VoLTE co nfiguration ON (LongDrxCycle = 40 ms) www.u5gig.ae Fig. 3. TTI bundling and RLC s egmentation procedur es Fig. 4. Testing methodology summary The focus of this paper is assessing the end - to - end performance of VoLTE call s. Therefore, we have conducted long VoLTE calls in the each cluster with a duration of ~2 hour. The main VoLTE KPIs demonstrated in this paper are summarized as follows: Relative Jitter: inter - arrival time of subsequent RTP payloads calculated with reference to the previous RTP packet received in - sequence during a talk spurt. The ji tter is calculate d per RTP stre am during talk spur ts as follows : 𝐽𝑖𝑡𝑡𝑒𝑟 & = & ( 𝐼𝑀𝑆 & 𝑇𝑖𝑚𝑒 & 𝑜𝑓 & 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 & 𝑅𝑇𝑃 & − & 𝐼𝑀𝑆 & 𝑇𝑖𝑚𝑒 & 𝑜𝑓 & 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 & 𝑅𝑇𝑃 ) − & ( 𝑇𝑖𝑚𝑒 & 𝑜𝑓 & 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 & 𝑅𝑇𝑃 − & 𝑇𝑖𝑚𝑒 & 𝑜𝑓 & 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 & 𝑅𝑇𝑃 ) (1) Then, the relati ve jitter 𝑗(𝑡 ) at ti me 𝑡 can be calculated as follows : 𝑗(𝑡 ) & = & 𝑎𝑏𝑠 ((𝑠 (𝑡 )&– &𝑠 (𝑡– 1 ))&– & (𝑟(𝑡 )&– &𝑟 (𝑡– 1))) , (2) W here abs (.) is the absolu te value of a n umber and s(t)& is the RTP timestamp embedded insid e the recent received RTP packet which is the actual timestamp of the RTP packet , s(t– 1) is the R TP timestamp embedded inside the previously received RTP packet which is the actual timestamp of the RTP pac ket , r(t) is the timestamp of the recent received RTP packet i.e., the timestamp of arrival current RTP packet and r(t– 1) : the timestam p of the previous ly received RTP pa cket i.e., the timestamp of the previous RT P packet. RTP DL E rror Rate : The percent age of the RTP packets that are not recei ved by the UE based on RTP sequence number. The num ber of lost packets “ 𝐸 ” is calc ulated per RTP flow by adding the number of RTP packets lost (i.e. the gaps in RTP s equence number) . Similarly, the number of RTP packets successfully received “ 𝑁 ” is calculated per RTP flow by counting the number of RTP sequence number and payloa d received in order . Then, t he RTP downli nk e rror rate is calculated as follows 𝑅𝑇𝑃 & 𝐷𝐿 & 𝐸𝑟𝑟𝑜𝑟 & 𝑅𝑎𝑡𝑒 = & 𝐸 /(𝐸 + 𝑁 ) (3) Both j itter and RTP error r ate are calculated with reference to all RTP packe ts received over certain interv al (e.g. 1 sec). 5 V O LT E Performance Evaluation In this section, the VoLTE perfo rmance is evaluated in terms of ROHC , TTI, RTP error rate , jitter, BLER, handover delays (C - Plane and U - Plane) and voice quality in terms of MOS. All results are obtained from commercial networks as explained in the previous s ection. 5.1 ROHC Efficiency and Perf ormance Evaluation To evaluate t he maximum capabil ity o f the ROHC , we have tested two scenarios; the first scenario for concurrent VoLTE and data connections and second scenario for a VoLTE standalone call. In both scenarios, the voice activity was continuous with minimum silent periods. In the first scenario, the packets at the radio side are t ypically multipl exing betwe en both PS data and IMS wit hin the same transmission time interval. However, because the compression takes place at the upper layers, then the impact on the a ctual size of the radio packet ( efficiency ) is not too much co mpared t o the s econd s cenario (IMS c all on ly). Table II pro vides pract ical results for ROHC hea der compression efficiency from real network deployment as described in section IV based on IPv4 . The table illustrate s DL / UL header size and average compression efficiency in mobilit y scenario for long VoLTE - to - VoLTE ca ll cov ering 100 eNBs. As observed, the ROHC in both scenarios (i.e., with and without PS data) is capable of offering significant gain to the radio resources by reducing the packet size and compressing the headers with an average efficiency of 81% to 92 % a nd ove rall avera ge effici ency of 86.7% . Therefore, ROHC is very beneficial for VoLTE traff ic transmitted a lone or alongside other data traffic, w hich is a typical case in smartphone (i.e. background data is ongoing whil e the user is on a voice call). In terms of channel rate saving, and using the practical values in t able II , it is obvious that ROHC can reduce the transmission data rate on radio interface from : 𝑃ℎ 𝑦𝑠𝑖𝑐𝑎𝑙 &𝑐ℎ 𝑎𝑛𝑛𝑒𝑙 & 𝑑𝑎𝑡𝑎 & 𝑟𝑎𝑡𝑒 & = & (𝑃 + 𝐻 + 𝑂) ∗ 8/𝐼 & = && 32 .4& 𝑘𝑏𝑝𝑠 , (4 ) w here P = AMR p ayload, H = average original header s ize, and O = other protocol h eaders , and I = RTP packets i nter val ( i.e., 20ms), t o 𝑃ℎ 𝑦𝑠𝑖𝑐𝑎𝑙 &𝑐ℎ 𝑎𝑛𝑛𝑒𝑙 & 𝑑𝑎𝑡𝑎 & 𝑟𝑎𝑡𝑒 & = & (𝑃 + 𝑅𝐻 + 𝑂 ) ∗ 8/𝐼 & = & 18 .5& 𝑘𝑏𝑝𝑠 , (5) IMS$ Performance Ass es s $ RTP$ and$ VoI P$pa cket$ performance $ IMS$Cl ient Modem $ IMS$ Serve r eNB &$ E PC LTE$ Prot ocol $Laye rs Ass es s $ impact$ of$ LTE$ P rot ocol$Laye r $on$ VoL TE Dat a$ P o st$ Proces s ing UE$ Loggin g$T oo l Netwo rk$T races$ for $UE$ un d er$ test Applicati on$Layer Access$ Stratum Vo L T E to $ Vo L T E (with o u t$D at a) Vo L T E to $ Vo L T E (with $ D at a) Multip le$T est$ rou nds $ wit h$ sa me$ &$d iff erent$ chipset/SW $v ersion Multip le$T est$ rou nds $ wit h$ sa me$ &$d iff erent$ chipset/SW $v ersion www.u5gig.ae w here RH is the average of compressed ROH C header s in all scenarios listed in t able II which is 5.3 bytes . This indicat es that ROHC boosts the air interface resources by almost twofold . This air interface saving provi ded by ROHC contributes directly to enhanced capacity and coverage. The exact capacity gain due to ROHC may vary depend ing on the deployment scenari o (e.g. RO HC depl oyed with or with out TTI Bundling features, RO HC mode of operati ons configured as explained in section II) . It is expec ted that ROHC will even offer higher gain with IPv6 since the header is 60 bytes [24]. As evi dent f rom th ese re sults, the original voic e packe t siz es on the UL and DL are same; however, the compression rate is different. It is observed that th e UL has sli ghtly better compression efficiency than DL in both scenarios, which is attributed to different compression methods used by U E and by eNB. In addition, the volume of information carried by the compressed data packets varies with the state in which the d ata packets are compressed. The decision about compression state transit ions (from sending uncompressed data i nto compres sing dat a with maximum compression efficiency) are made by the compressor based on many factors like the variations in the static part or dynamic part of packet headers and the acknowledgment f eedback status from the de - compressor. T ABLE II ROHC COMPRESSION EFFICIENCY FROM FIELD M EASUREMENTS AMR - NB with 12.65 Kbps and IPv 4 Concurrent VoLTE and Data sessio ns VoLTE session Only Grand Average UL DL UL DL AMR Payload ( Bytes) 33 33 33 33 33 Average Orig inal Header Size (Bytes) 40 40 40 40 40 Other Prot ocol Overhead ( L1/MAC/ RLC/ PDCP *) (Bytes) 8 8 8 8 8 Average Compre ssed Header Size (Bytes) 3.9 7.5 3.2 6.5 5.3 Average Compres sion Efficiency (%) 90.1 81.2 91.9 83.6 86.7 Required Channel Data Rate after ROHC (kbps) 18.0 19.4 17.68 19.0 18.5 * L1: p hysical layer , MAC: m edium access control layer , PDCP: p acket data conversion protocol . 5.2 TTI Bundling TTI bundling in general can achieve very good coverage and reliability [ 22 ] . TTI bundling opti mization mainly depends on cel l edge user RB utilization and SI NR which are key factors to trigger TTI bundling, in fact the triggering criteria differ from one infra - vendor to another. However, careful optimization is required especially in the choice of t he codec rate of VoLTE calls. For example, a VoLTE cal l with WB - AMR codec rate of 23.85 Kbps may face voice quality challenges even when TTI bundling is applied. In t his scenario, every RTP packet sent on the UL wi ll requir e an extra 8 ms to be transmitted. Assuming the maxim um TBS of 63 byt es is granted to the UE during the TTI bundli ng operation ( typical s ize d uring this operation) and when the ROHC is not used (either not configured or not applied due to ROHC state transition), the PDCP protocol data uni t ( PDU ) size (including A MR payload and IPv4 RTP/UDP/IP * Mobile - originated UE is the UE which ori ginate s the voice call while mobile - terminated UE is the U E which terminat es the voice call . headers) will be 102 bytes. In thi s scenario, one AMR payload cannot fit within one TTI and hence segmentation is needed . If ROHC is applied with the best compression efficiency (as demonstrated previ ously ), the header size can be reduced to 3 bytes making the PDU arrives at the MAC l ayer with si ze of 65 b ytes. In both cases, TTI bundling grant , to transmit a complete AMR payload , will not be suffic ient ov er a si ngle 4 ms ec bundle. Therefore, it require s more than one bun dle to transmit one AMR payload (i.e. one RTP packet) ; hence, it can take ≥ 8 ms , whi ch i ncreases the del ays, and in ret urn det eriorate the voice quality at cell - edge . In all cases, it is obvious that the usage of ROHC and TTI bundling together at cell - edge engend ers the optimum performance in terms of coverage and voice quality [ 21 ], [ 24 ] , [27]. The lower codec rates provide larger coverage and better radio robustness because fewer v oice information bits need to be sent over t he air . Since media packets are gene rated in 20ms interv als then, the RTP T otal P acket S ize (RT PS) ( i.e., one voice frame) can be estimated as foll ows: 𝑅𝑇𝑃𝑆 = 𝑃𝑎𝑦𝑙𝑜𝑎𝑑 + 𝐴𝑀𝑅 &ℎ 𝑒𝑎𝑑𝑒𝑟 + 𝑅𝑇𝑃 &ℎ 𝑒𝑎𝑑𝑒𝑟 + 𝑝𝑟𝑜𝑡𝑜𝑐𝑜𝑙 &ℎ 𝑒𝑎𝑑𝑒𝑟 (6) Then the t otal RTP packe t si zes f or 12. 65kbps and 23,8 5kbps codec rates can be estimated, respectively, as follows: 𝑅𝑇𝑃𝑆 = 12 . 65 𝑥 20 𝑚𝑠 + 11 & 𝑏𝑖𝑡𝑠 + 96 & 𝑏𝑖𝑡𝑠 + 64 & 𝑏𝑖𝑡𝑠 = 424 & 𝑏𝑖𝑡𝑠 (7) 𝑅𝑇𝑃𝑆 = 23 . 85 𝑥 20 𝑚𝑠 + 11 & 𝑏𝑖𝑡𝑠 + 96 & 𝑏𝑖𝑡𝑠 + 64 & 𝑏𝑖𝑡𝑠 = 648 & 𝑏𝑖𝑡𝑠 &&&& (8) & When TTI bund ling is enabled, the resource allocation size is restricted to a maxim um of three PRBs and the modulation scheme must be Q PSK [ 28 ] . T herefore, the selected MCS index cannot be greater than 10. After TTI bundling is enabled, the maximum available TBS is as large as 504 bits that can be bundled in 4 TTIs (i .e. , 504 LTE bits sent every 4ms). Therefore , one RTP packet of 424 bits can fit within the bundled TTI every 4ms for 12.65kbps whil e we ne ed to send one RTP packet bundled every 8ms in case of 23.85kbps (i.e. two RTP pack ets bundled every 4 ms with the size of 648 bits) . Since, VoLTE service is delay - sensitive , i f higher - layer data is not transmitted within the specified delay budget, the voice quali ty deteriora tes. This implies that RTP throughput is cut in half i mpact ing the jitter and voice quali ty at c ell edge , in addition to the negative impact on the upl ink coverage. On the other hand, the TTI bundling does not apply to the DL, , instead methods like RLC segmentation concept are typically used to accommodate larger size RTP packets within smaller protocol layer packets at cell edge . Therefore, for codec rate like 23.85 kbps , pack ets will be segmented into more smaller size RLC l ayer PDUs t han 12.65 kbps and hence higher voi ce d elays can be observed on the downlink as well . B ased on the above discussion, it is recommended to reduce the codec rate to 12.65 Kbps to gain more uplin k coverage with AMR payload that can fit within a bundled packet. This is another reason why the high d efinition codec rate of 23.85 k bps provides better voice qualit y in cell center whi le lower rates like 12.65 kbps provide s better voice quality at cell - edge as demonstrated late r in this section. The adaptive switching between different codec rates based on radio conditions or network load is still not widely applied. Duri ng an on - going call, the mobile - originated UE and mobile - terminated UE * have the ability to modify the codec mode and ra te. The radi o www.u5gig.ae condition and user experience can be taken into accou nt in this procedure. For the network loa d, the network congestion can be indicated by explicit congestion notifi cation (ECN) mechanis m vi a t he RTP protoc ol using the RTCP as a feedback m echanism. H owever, ECN usage is very limited as it can only indicat e the occurrence of congestion at the E - UTRA side without further info rmation on its le vel. As for radio conditions indication betw een UE and eNB that can be used for codec ra te adaptation, thi s mechanism is not cl early standardized at this point in 3GPP. Th erefore, developing prope r optimization processes is recommended to provide consis tent user experience at all radio condi tions. More details on the voice quali ty comparison between di fferent codec rates are discussed later in this section. Als o, the impact of TTI bundling on MOS and cell radius ra nge are analyzed. 5.3 RTP and Jit ter Evaluation We conduc ted a VoLTE to Vo LTE long cal l in mobili ty conditions as summarized in t able I and f igure 4 and evaluated the perfor mance o f the KP Is menti oned in section IV . W e have tested three different scenarios: concur rent VoLTE and Data connectio ns wit h ROHC , VoLTE s tandalone call with ROHC and VoLTE standalo ne call wit h ROHC turned off from the eNB side. In all scenarios, the voice activity was continuous with minim um silent periods and we have pe rformed ful l download using an FTP server when data session is present in first scenario . The distri butions (probability density function (PDF) and cumulative distribution function (CDF)) of the RTP error rate for all tested scena rios are illustrated in figure 5 . With ROHC activated, the average downlink RTP error is withi n t he ac cepted range of ≤ 1% . However, due to the presence of data packets in pa rallel with RTP voice packets, the RTP errors observed are slightly higher. This can happen in cases especially when both data and vo ice packets are multiplexed within the same TTI as explained in subsection (a) of this section . In this case, the HARQ scheduled with higher number of bits can jeopardize the VoLTE per formanc e causing more air interface BLER . However, as will later be shown, the impact of higher BLER i s more obvious on the jitter than the RTP error rate. On the other hand, when ROHC is dis abled, the RTP error rate significantly degraded . This is because the Vo LTE packets are trans mitted with uncompressed headers leading to hig h packet sizes on the radio side. In return , this leads to more RT P errors and holes in IMS transmission (i.e. out of sequence) that impacts the overall voice quality. It is evident tha t enabling ROHC has positive im pact on the performance alongside improvements to the capacity and coverage aspects as discussed previously . From RTP jitter perspective, figure 6 illustrates the PD F and CDF of the downlink relati ve jit ter in the same three scenarios. The presence of the concurrent PS data session significantly degraded the average relative jitter by 4 0% even with while ROHC is enabled . Additiona lly, the RT P error increased by 50% as shown in figure 5 however it is still within the accepted range. In t he case of ROH C is disabled during VoLTE standal one c all, the re lative jitter is 2 0% highe r compared to the same scenario with ROHC enabled. This stresses on the i mportance of ROHC feature to the VoLTE overall performance. Test Scenario for RTP Error rate Avrg Median Min Max Concurrent VoLTE and Data sessions ( ROHC ON ) 0.74% 0% 0% 51.2% VoLTE sessio n Only ( ROHC ON) 0.34% 0% 0% 54.2% VoLTE sessio n Only ( ROHC OFF) 7.77% 7.84% 0% 34.7% Fig. 5. Distribution of the RTP error rate in mobility condi tions Test Scenario for Relative Jitter Avrg Median Min Max Concurrent VoLTE and Data sessions ( ROHC ON) (ms) 3.14 1.0 0 34.7 VoLTE sessio n Only ( ROHC ON) (ms) 1.87 1.5 0 25.6 VoLTE sessio n Only ( ROHC OFF) (ms) 2.32 1 0 220 Fig. 6. RTP DL jitter with ROHC enabled and without and with PS data 5.4 Scheduler Eval uation for VoLTE The presence of PS data alongside VoLTE call has obvious degradation on the overall RTP performance as shown in figure 5 and Figure 6 . It is therefore important to consider applying techniques to mitigate this negative impact. One of the options is to handle both data and VoLTE sessions in parallel at the eNB scheduler level. In this context, eNB scheduler can consider sending VoLTE pac kets in a given TTI without any data packets m ulti plexed especially in bad radio conditions. When the downlink scheduler tends to utiliz e the same physical layer transmission to send both voice and PS Data in the same TTI, then t he TBS can increase and hen ce higher BLER. Ano ther technique is to use the 2x2 MIMO codewords (or higher order MIMO) for splitting the VoLTE and PS data into two different streams with Rank - 2 spatial multipl exing , i.e. segm entation of packets at MIMO codeword level . This can improve the spectral efficiency and minimize the j itter and ove rall B LER. Table III provi des scheduler comparison between the two infra - vendors used in this testing , for the case of concurrent 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0 5 10 15 20 25 30 35 CDF PDF RTP , Erro r,Ra t e, [ %] VoL TE, sessio n,Only,(RoHC ,O N ) VoL TE, and,Dat a,sessi o ns, (RoHC , ON) VoL TE, sessio n,Only,(RoHC ,O F F) CDF - VoL TE,sessio n,Onl y, (RoHC ,O N) CDF - VoL TE,and,Dat a, sessi on s,(RoHC , ON) CDF - VoL TE,sessio n,Onl y, (RoHC ,O FF) 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% ! " #! #" $! $" %! %" &! &" "! '() *() +,- ./ 01, 230 / / ,42 567 8 VoL TE& se ssi o n&Onl y& (R o HC &ON) VoL TE& an d &Dat a&s essi ons &(Ro H C&ON ) VoL TE& se ssi o n&Onl y& (R o HC &OFF) CDF - VoL TE& se ssi o n&Onl y& (R o HC & ON) CDF - VoL TE& an d &Data &s essi o ns &(R oH C &O N) CDF - VoL TE& se ssi o n&Onl y& (R o HC & OFF ) www.u5gig.ae VoLTE and data sessions . The first scheduler does not tend to utilize the same Physical Downlink S hared Channel ( PDSCH ) transmission to send both IMS Media and PS Da ta in the same TTI . This m eans that more VoLTE m edia packets need to be transmitted alone which can increase waiting time in buffer and hence the ove rall jitter . S ince m ulti plexing i s not used, the TBS scheduled is low with high padded bits that waste the PRB/MCS resources and the overall capacity . Also, this scheduler does not seem to util ize MIMO Rank2 for IMS media packets and then there is no RLC segme ntation over the tw o MIMO code - words . The second scheduler tend s to utilize the same PDSCH transmissi on to send both IMS Media a nd PS Dat a in the s ame TTI . Since m ultiplexi ng is used in this deploymen t , the TBS schedule d is higher and the s chedul er tends to minim ize the padding and hence improve the PRB/M CS utilization and th e overall capacity . Unl ike the first schedu ler , this one sometimes utilizes MIMO Rank 2 for IMS media packets and in this case use RLC segme ntation over MIMO two code - words. This can add an additional improve ment to the DL spectral efficiency. T ABLE III S CHEDULER I MPLEMENT A TION C OMP ARISON Scheduler Behavior Scheduler 1 Scheduler 2 IMS Media Packets sent witho ut Data (i.e. , packets are not multiplexed) 88% 53% IMS Med ia + PS Data multip lexed in same TTI 12% 47% IMS packets transmitted using MIMO two code - words 0% 5.42% Inter - TTI scheduling delay ( i.e., delay between two MAC layer packets scheduled by eNB carrying IMS media) (ms) 34.3 20.9 Average TBS Scheduled (Bytes) 585 1291 Average paddi ng per TBS at MAC (Bytes) 302 41 So far , we have analyzed the relative jitter in different scenarios with concurrent PS data , duri ng VoLTE standalone call and also with and without ROHC . In this section, w e will analyze the relative jitter versus BLER at the radio interface . Figure 7 demonstrates the relative jitter versus DL BLER with and without data session. As evident from f igure 7 , there is a clear relation bet ween relative downlink jitter and downlink BLER at physical layer as there is trend of jitter increase with the increas e of re - transmission at lower layers. It is also obvious that there is high variati on in jitter wh en PS dat a i s present. More specifica lly , the jitter reached an average of 3.5 ms at BL ER of 10% wi th PS data while the j itter reached an average of 2 ms at the same BLER level and with VoLTE standalone call . This implies that DL BLER and the BLER target at eNB can affect the end - to - end RTP delays and therefore careful implementation of BLER convergence algorithms at the eNB scheduler is mandatory . Due to delays at di fferent int erfaces, the RTP packets transfer with diffe rent inter - arrival intervals. From eNB side, the convergence into the targeted B LER requires a stable flow of packets in order to maintain a suitable tracki ng t ime t o achieve the target. Typically, the eNB scheduler selects the downlink TBS based o n the reported CQI whi ch controls the MCS se lecti on. Th e amount of data to b e sche duled in a TTI determines the number of PRBs to be scheduled for UEs. Based on the feedback mechanism between UE and eNB, the scheduler keeps tracking the BLER measured versus the target BLER an d hence starts the adjustment of the M CS and PRB to this UE. This me chanism enables the scheduler to allocate r esources in a manner to maximize the resources utilization eff iciency. However, in so me cas es where the packets flow irregularly, the track ing mechanis m cannot converge which can cause fluctuations of BLER as obse rved in fig ure 7. To ma intain a good tr adeoff b etween DL BLER and DL resources, the scheduler should be designed to minimize the TBS (and the padd ing bi ts i n each TBS) while maintai ning a spectral e fficie ncy in terms of good throughput. Fig. 7. RTP relative jitter vs. DL BLER T ABLE IV DL S CHEDULING AND T HROUGHPUT C OMP ARISON BETWEEN V O LT E AND OTT Near Cell Conditions (VoLTE scenarios operating at 12.65 Kbps and IPv4) OTT VoLTE ( ROHC OFF) VoLTE ( ROHC ON) Avera ge Time Domain Scheduling Rate ( %) 6% 2% 2% Average MAC La yer TBS (b ytes ) 360.8 344.4 163.1 Average Sch eduled Resourc e Blocks 4.2 5.4 4.1 Average Bit Rate (kbps) 155.1 54.7 25.4 To evaluate t he scheduler behavior , we compare the downlink t hroughput of three cases: VoLTE with and without ROHC and Over - the - Top (OTT) VoIP using a commercial application. Typical OTT applications include Skype or Facetime, where t he IP packet transmission is handled w ithout the network m anagement of an operator. In this case, the OTT utilize QCI=9 as a normal PS data sessio n carrying VoIP packets . Table IV provides a comparison between VoLTE (with and without ROHC ) and OTT in terms of radio resource utilization at different layers based on scheduling attempts . As shown in t able IV , OTT consumes the most schedul ing resources in the time domain which means i t would require more TTI utiliza tion co mpared t o VoLTE in general , while the PRB utili zation in the frequency domain is almost similar for VoLTE and OTT . As a result, OTT wi ll r equire higher throughpu t to mainta in th e voic e quality and that consumes more scheduling resources from the network, affecting the overall cell capacity. As shown, OTT requires six times more data throughput compared to VoLTE with ROHC , w hich means more resources and mainly consumed in time - domain. It is worth mentioning that this result is for the tested ne tworks and with one OTT 0 1 2 3 4 5 6 7 8 9 10 0 1.852 3.252 4.8 6.122 7.609 9.278 10.843 12.329 14.05 16.162 18.333 20.755 23.188 26.829 32.231 49.686 Relative Jitter [ms] DL BLER [%] VoLTE session Only (ROHC ON) 0 1 2 3 4 5 6 7 8 9 10 0 1.572 2.813 3.746 4.469 5.28 5.907 6.593 7.317 7.907 8.388 9 9.589 10.267 10.97 11.972 13.511 15.717 18.689 49.121 Relative Jitter [ms] DL BLER [%] Concurrent VoLTE and Data sessions (ROHC ON) www.u5gig.ae application . On the ot her hand, VoLTE with ROHC enabled provides the most eff icient scheduling and the lowest data rate. This indicates that when VoLTE deployed with optimized feat ure will outperform OTT i n cell capacity aspects. However, it is to be noted that the TBS in all cases fairly exceeds the actual R TP payload (i.e. 72 b ytes w ithout ROHC and ~ 37 b ytes with ROHC ). This indicate s that th e scheduler still strives to meet proper scheduling size when the IMS packets transfer irregularly . This is also an observation that may require further investigation in the scheduler behavior and in future work in Vo LTE evolution . The indication of the TBS an d the padding bits are highlighted in table III. With this data, we o bserve that network schedulers have a r oom of improvements to map PRB to MCS in a way to uti lize sufficient TBS and minimize the padding bits and consequently the LTE spectrum efficiently . 5.5 RT P and Jitter Evaluation ver sus Radio Conditions In this section , we evaluate the RTP performance versus radio conditions. Figures 8 to 10 demonstrate the average RTP error rate and j itter as a function of RSRP , RSRQ and SINR, respective ly. The figures illustrate that the jitter and RTP e rror rate t end to increase when th e RF conditions degrade . However, the most impacting factor is the RSRQ (loading and interference indica tor) where the j itter and RTP error rate exhibit significant degradation (i .e., jitter > 10 m s and RTP error rate > 10% ) wh en the RSRQ i s degr aded . T his indicates that high interfe rence or high loading at the eNB will dir ectly af fect th e scheduling of the RTP packets and thus increases the RTP delays . It is therefore important to optimize the handover parameters based on RSRQ as well as RSRP to improve the overall radio conditions of VoLTE performance. RSRQ is a typical quality measure of t he loading and interference in connected mode. In a multiple bands sc enario, the RSRQ can be used as a trigger to distribute the load uniformly across the two bands. The RSRP reflects only the channel condition, e.g. if UE is n ear or far from cell center, however the RSRQ can ind icate wheth er the cell on that carrie r is loaded or not. Hence, the RSRQ can be used to trigger an inter - frequency handover from loa ded carrier into an unloaded one. As the voice service is more sensitive to the rad io and l oading conditions, then improving the handover trigger b ased on RSRQ can be beneficial to VoLTE calls. Moreo ver, The eNB s can be configured with different handover parameters for VoLTE call compared to other data sessions. This can be configured since the eNB is aware of the VoLTE and data bearers and then it c an define the han dover para meters differe ntly for each service. Mo re importantly, the jitter starts to increase significantly when th e RSRP < - 110 dBm . This can be a decisive factor to set the threshold for triggering eSRVCC to 3G or 2G to maintain a good voice qual ity. Wit h thi s rec ommended sett ing , the coverage of VoLTE service will reach up to RSRP of - 110 dBm, beyond which, the UE wi ll be handed over to 3G/2G using eSRVCC in order to enhance voice quality. 5.6 Ha ndover Impact on VoLTE Call Performan ce It is expected that users during VoLTE call will be in mobilit y conditi ons where hand overs betw een differ ent eNBs will occur frequentl y. Therefore, evaluating the delays of RTP packets during the VoLTE call is important to ensur e consistent voice quali ty in mob ility conditions. The authors in [20] evaluated the performance of VoLTE call in mobility conditions for both i ntra and inter - systems handovers. The results showed that inter - system handovers incurred higher delays than intra - system handovers. The y evaluated both the user and control plane delays associated with these handover types. In this paper, we will evaluate the impact of intra - frequency handover on the characteristics of VoIP call represented by the jit ter and RTP errors that could occur du ring mobility conditions. Figure 11 demonstrates the intra - frequency handover procedures and the associated delays affecting the RTP interruption and ji tter. Fig. 8. average RTP and Jitter vs. RSRP Fig. 9. average RTP and Jitter vs. RSRQ Fig. 10 . average RTP and Jitter vs. SINR * r efer to [20] for more details Fig. 11. LTE Intra - Frequency Handov er Procedures As explai ned in fig ures 8 to 10, the radio conditions have direct impact on the VoLTE per formance. We will go one - step further by evaluating the jitter and interruption during the handover procedure. Handover is essential during mobilit y to move the user from one eNB to another. If the parameters are set to delay the han dover, then higher jitter and errors can be observed as discusse d before in t erms of www.u5gig.ae RSRP a nd RSRQ. The analysis in [20 ] presents the details of the voice interruption time where the a ctual RTP pac kets are suspended during the handover execution. We here evaluate the distribution of the RTP interruption and jitter dur ing the handover execution proce dure as shown in figur e 12 , with ~100 handover attempts in the testing route. As observ ed from figure 12 , the average jitter during handover is h igher than the normal average observed in figure 6. As estimated in [20 ], the average interruption time is ~ 75 ms, which means that the jitter i s dir ectly impacted by the delay in transferri ng the VoLTE context from one eNB to another over X2 interface. Assuming that the RTP pa ckets are not lost as evident from the RTP error rate i n figure 12, then the t ime difference between the j itter a nd RTP i nterruption is ~ 40 ms. This means that there i s ~ 4 0 ms del ay i n downl ink scheduling of two consecutive RTP packets during handover. T he r eduction of such interrupti on time is very critica l especially for certain a pplications that utilize VoIP services. This topic requires further research for enhancements and it can be proposed as contribution in LTE - Advanced Pr o (3GPP R el - 13 and beyond) . I t is important to know that the IMS server and client maintai n timers to detec t any timeout in RTP packet transfer , after which the VoLTE call can be released. In this context, UE IMS client detects that the DL RTP packets are not detected wi thin a certain time windo w , e.g. for 10 sec, then the IMS c li ent will terminate the voice call to prevent the call from being active without audio. Therefore, optimizing the handover performance is essenti al to keep the established VoLTE with good ret ain ability and consistent performance. KPIS Avrg Min Max Number of RTP Packet lost duri ng handover 0.33 0 21 Relative DL Ji tter during han dover (ms) 35.6 2 922 Fig. 12. RTP perf ormance during I ntra - Frequency Handover execution 5.7 VoLTE Voice Quality Evaluat ion In this section, we will evaluate the VoLTE voice quality in terms of POLQA (Perceptual Objectiv e Listening Quality Analysis) MOS scor e [29 ] and compare it with OTT application (as described earl ier in section), 3G with AMR - WB an d 2G with AMR - NB. POLQA wa s adopt ed in 2011 a s ITU - T Rec ommendation P.863 [ 30 ]. In this testing, we ha ve used same clusters as des cribed in section IV. The vo ice quality testi ng is conducted bas ed on mobility wit h average speed o f 80km/h. We used S pirent Nomad voice quality tool with 4 ch annels for testing. I n ea ch t est scenario, w e used four devices as t wo for UL and two for DL. In the first time, we conducted 2G and 3G and in the second time we used two VoLTE call s with different coding r ate i.e ., 23.8 5kbps and 12.65kbps. The test devices are locked on the tested network i.e., 2G, 3G or LTE with no handover or SRVCC. The OTT application is tested separately and the device was locked on LTE. The calls are long calls i .e, around 16 min which is around 50 cycles of the MOS testing device. We averaged the DL and UL and averaged the res ults over 10 t imes for each round. Figure 13 provides the average MOS values for VoLTE with codec of 23.85, VoLTE with codec 12.65, OT T, 3G with AMR - WB o f 1 2.65 and 2G AMR - NB. The figur e i ndicates tha t VoLTE engenders the best voi ce qua lity compared to OTT and CS voice calls. Also, the average MOS of VoLTE with 12.65 kbps codec rate is better than VoLTE with 23.85 kbps codec rate. This is because the 12.65 kbps is more r obust at cell - edge. Therefore, in mobility scenario and since MOS values are averaged , the overall measured MOS of VoLTE with 12.65 kbps is better than VoLTE with 23.85kbps . It i s a lso noted that, LTE with 12.65 Kbps f alls within the range of “good quality” which is speci fied in ITU - T P.863 as 3.6 ≤ MOS ≤ 4.0. On the other sid e, 3G and OTT fall in the range of “acceptable quality” specified as 3.1 ≤ MOS ≤ 3.6, while the 2G falls into the “poor quality” of 2.6 ≤ MOS ≤ 3.1, as it is n ot high definition. Figure 14 provides comparison bet ween Vo LTE with the two codec rates at near cell and fa r cell scenarios. The RSRP range for near cell scenario is - 60 to - 65 dBm and the far cell scenario is - 102 to - 105 dBm. It is evident that the VoLTE with 23.85 kbps codec offers bad voice quality at cell - edge as explained earlier in TTI bundling section . Theref ore, it is recommended to use the speech rate of 12.65kbps to guarantee a consistent user experience . Also, 23.85 k bps would consume higher resources compared to 12.65 k bps (due to higher payload size) however with a minimal improvemen t to the voice quality in near cell an d with highly degraded quality at the cell - edge . EVS codec was not evaluated in this paper but it is anticipated that EVS with SWB can outperform AMR - WB even at mi xed and music content and off ers 1.2 MOS improv ement at comparable bitrates [31 ]. Fig. 13 . Voice Quality Meas urements for Differe nt Technologies (Mobility) Finally, in or der to evaluate the TTI bundling gain at cell - edge, we have conducted a field test to measure the average MOS at cell - edge with and without TTI bundling. Fi gure 15 provides t he average MOS versus RSRP for the two scenarios. As depicted form the figure, the TTI bundling improves MOS when RSRP £ - 118dBm. This is the threshold that can be set to trigger the TTI bundling. Forcing the TTI bundling at RS RP > - 118dB unnecessarily repeats the packet four time s and h ence increasing the jitter and reducing the MOS. The T TI bundl ing ex tends t he cell - edge to RSRP = - 129dBm while without TTI bundling, the cell radius is limited to RSRP =~ - 123dBm (i. e., call drop ped at this l evel). 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0% 5% 10% 15% 20% 25% 30% ! " #! #" $! $" %! %" &! &" "! "" '! '" (! (" )! )" *! *" #!! #!" +,- .,- /0 123 450 67 43 30 8 69:8 4; <6=;38 2 > ?80 @ 6A 2; 9 B5 0 8 6C D EF PDF CDF www.u5gig.ae Therefore, the TTI bundl ing has extended the coverage by 6dB. This gain can be beneficial for new technologies such as 3GPP Narrow Band Internet of Things (NB - IoT) that relies on the extended cove rage concept and requir es additional 20d B increase in LTE coverage to reach undergrou nd and deep indoor sensor s [32] , [ 33 ] . Fig. 14 . Voice q ua lity for diffe rent VoLT E codec r ates at different RF conditions Fig. 15. TTI Bundling impact on MOS 8 C ONCLUSIONS In this paper, w e have a nalyzed the practical pe rformance of VoLTE based on commercially deployed 3GPP Release - 10 LTE net work s . The evaluation demonstrates VoLTE performance in terms of ROHC , RTP error rate, RTP jitter and delays, TTI b undling, BLER and VoLTE voice quality in terms of MOS. This paper provided best deployment practices for VoLTE deployment and VoLTE related featur es. We have demonstr ated that ROHC is capable of offering significant gain to the rad io resources by reducing the packet siz e and compressing the headers with an average effi ciency of 81% to 92 % and ov erall av erage eff iciency of 86.7% based on evaluated networks . Accordingly, ROHC can boost the air interface resources by almost twofold. Also, ROH C offer s ~7% improve ment in RTP error rat e and ~24% reduction in jitter. The use of ROHC and TTI bundling together at cell - edge will be t he best in terms of improved coverage and best voice quality. H owever, i t is recommen ded to reduce the codec rate to 12.65 Kbps to gain more uplink cov erage with AMR payload that ca n fit wi thin a b undled pac ket. The concurren t data and VoLTE session causes 40% degradation in jitter and 50% increase in the RTP error however they are still wit hin the accepte d ranges. Also, the jitter reached an average of 3.5 ms at BLER o f 1 0% with P S data while the jitter reache d an average of 2 ms at the same BLER level with VoLTE sta ndalone. The pre sence of PS data alongside VoLTE cal l has obvious i mpact to the overall RTP performance. We have provided techniques to m itigate these drawbacks. The degradation in the RSRQ leads to significant degradation in the jitter an d RTP e rror rate . More importantly, the jitter starts to increase significantly when the RSRP < - 110 dBm. Therefore, it i s recommended to l imit the coverage of VoLTE service to RSRP = - 110 dBm by trigging SRVCC . However, TTI bundling can extend VoLTE coverage to RS RP = - 129 dBm . There i s ~ 40 ms del ay in downlink scheduling of two consecutive RTP packets during handover. The reduction of such interr uption time is very critical especially for certain applications that utilize VoIP services. we have evalua ted VoLTE in terms of voice quality usin g POLQA MOS. It is demonstrated that VoLTE engenders the best voice quality compared to CS voice calls. Als o, t he average MOS of VoLTE with 12.65kbps is better than VoLTE with 23. 85kbps. I t is hi ghlight ed that 2 3.85 kbps offers v ery bad voice quality at cell - edge . Therefore, it is recommended to fix the codec rate at 12.65kbps to gu arantee consistent user experience. Future work may include proposing techniques to reduce the RTP data interrupt ion of 40msec during handover. Also, eva luating th e a daptive code c s election t o guarantee best voice quality and minimi ze radio resources utilization . R EFERENCES [1] 3GPP TS 22.228 V12.9.0, Service requir ements for the Internet Protocol (IP) Multimedia cor e network Subsystem (IMS). [2] 3GPP TS 22.173 V9.5.0 (2010 - 03): “IP Mu ltimedia Core Network Subsystem (IMS) Multimedia T elephony Service and supplementary services; Stage 1 (Release 9)” [3] GSMA, IR.92 IMS Profile for V oice and SMS, V ersion 7.0, 03 March 2013. [4] GSMA, IR.94 IMS Profi le for Convers ational V ideo Service V ersion 5.0 04 March 2013. [5] A. Elnashar , M. A. El - Saidny , “Looking at L TE in Practice: A Performance Analysis of the L TE System based on Field T est Results,” IEEE V ehicular T echnology Magazine, V ol. 8, Issue 3, pp. 81:92, Sept. 2013. [6] Ay m a n E l n a s h a r, M o h a m e d A l - saidny , and Mahmoud Sherif “Design, Deploym ent, and Performance of 4G - LT E N e t w o r k s : A practical Appr oach,” W iley , May 2014. [7] 3GPP TS 26.441, “Codec for Enhanced V oice Services (EVS); General ove rview” , version 12.0.0 Relea se 1 2 [8] Myasar R. T abany and Ch ris G. Gu y , “Per formance Analysis and Deployment of V oL TE Mechanisms over 3GPP L TE - based Networks,” International Journal of Computer Science and Te l e c o m m u n i c a t i o n s , V o l u m e 4 , I s s u e 1 0 , O c t o b e r 2 0 1 3 . [9] Myasar R. T abany , Chri s G. Guy , An End - to - End QoS Performance Evaluation of V oL TE in 4G E - UTRAN - based W ire less Networks, ICWMC 2014, pp . 90:97. [10] J. Calle - Sánchez, M. Molina - García , J. I. Alonso, and A. FernándezDurán, “Long term evolution in high speed railway environment s: feasibility and challenges,” Bell Labs T ech. J., vol. 18, no. 2, pp. 237 – 253, 2013. [11] Sniady , A. ; Sonderskov , M . ; Soler , J. , V oL T E Performance in Railway Scenarios: Investigating Vo L T E as a Viable Replacement for G SM - R, IEEE V ehicular T echno logy M agazine, 2015, Vo l u m e : 10, Issue: 3 , Pages: 60 - 70 [12] J. Seo and V . Le ung , "Performance modeling and stability of semi - persistent scheduli ng wit h initial random a ccess in L T E" , IEEE Tr a n s . Wi r e l e s s C o m m u n . , vol. 1 1 , no. 12 , pp.4446 - 4456 , 2012 [13] Q. Bi, S. V itebsky , Y . Y ang, Y . Y uan, and Q. Zhang, “Per formance ! !" # $ $" # % %" # & &" # ' - 104 - 115 - 118 ( $%$ - 123 - 129 )*+ ,+,- . /0 123 TTI# Bu n d l in g #O N TTI# Bu n d l in g #O F F www.u5gig.ae and capacity of cellular OF DMA systems with voice - over - IP traffic,” IEEE T rans. V eh. T echnol., vol. 57, no. 6, pp. 3641 – 3652, Nov . 2008. [14] J. A. F ernández - Segovia, S. Luna - Ramírez, M. T oril, and Juan J. Sánchez - Sánchez, IEEE Comm unications Letters, VOL. 19, NO. 3, MARCH 2015 431 Estimating Cell Capacity From Network Measur ements i n a Multi - Service L TE System [15] A. Sanchez - Esguevillas, B. Carro, G. Camarillo, Y . - B. Lin, M. A. Garci a - Martin and L. Hanzo , "IMS: the new generation of Internet - protocol - based multimedia services " , Proc. IEEE , vol. 101 , no. 8 , pp.1860 - 1881 [16] J. E. V argas Bautista et. all, “Performance of C S Fallback from L TE to U MTS” IEEE C ommunications Magazine, pp. 136:143, Sept. 2013. [17] R. - H. Liou et. all, “Pe rformance of CS Fallback for Long T erm Evolution Mobile Network,” IEEE T ransactions On V ehicular Te c h n o l o g y, V O L . 6 3 , N O . 8 , p p . 3 9 7 7 : 3 9 8 4 , O C T O B E R 2 014. [18] Yi - Bing Lin, “Performance Evaluation of L TE eSR V CC w ith Limited Access T ransfers,” IEEE T ransaction s On W ir eless Communications, VOL. 13, NO. 5, MA Y 20 14. [19] 3GPP TS 23.272 V10.3.1 (201 1 - 04), Circuit Switched (CS) fallback in Evolved Packet System (EPS ). [20] Ay m a n E l n a s h a r, M o h a m e d A . E l - Saidny , M ohamed M ahmoud, “Practical Performance Analyses of Circuit Switched Fallback (CS FB) and V oice over L TE (V oL TE),” IE EE T ran sactions on Ve h i c u l a r Te c h n o l o g y , Vo l . 6 6 , I s s u e 2 , p p . 1748 - 1759 [21] Andreas M aeder and Arne Felber , “Perfor mance Ev aluation o f ROHC Reliable and Optimistic Mode for V oice over L TE,” In Proc. Ve h i c u l a r Te c h n o l o g y C o n f e r e n c e ( V T C S p r i n g), 2013 IEEE 77th . [22] M. Vi l l a l u z e t . a l l , V o LT E S RV C C O p t i m i z a t i o n a s I n t e r i m S o l u t i o n for L TE Networks with Coverage Discontin uity , International Confere nce on Infor mation and Communication T echnology Convergen ce (ICTC), 2015 , pp. 212:21 6. [23] 3GPP TS 26.1 14, IP Multimedia Subsystem (IMS); Multimedia telephony; Media handling a nd interaction [24] Daniel Philip VENMANI et. all, “Impacts of IPv6 on Robust Header Compressi on in L TE Mobile Networks,” ICNS 2012: The E ighth International Conference on Networking and Services, pp. 175:180. [25] NSN white paper , “From voice over IP to voice over L TE” Nov . 2013. [26] 3GPP TS 36.321 , Medium Access Control (MAC) protocol specification, version 12.5.0 Release 12 [27] O. Ozturk and M. V ajapeyam, "Performance of V oL TE and data traffic in L TE heterogeneous networks", IEE E GLOBECOM, 201 3 . [28] 3GPP TS 36.213, Evolved Universal T er re st ri al Radi o Ac ce ss (E - UTRA); Physi cal layer pr ocedur es, versio n 12.4.0 Rel ease 12 [29] ETSI CTI Plug tests Report, V oL TE QoS Assessment, 0.1.0 (2013 - 12) [30] ITU - T POLQA Recommendation P .863 , webs ite http://www .itu.int/r ec/T - REC - P. 8 6 3 / e n [31] Fraunhofer I nstitute for Integrated C irc uits IIS, Enhanced V oice Services (EVS) Codec, T echnical paper: (online) http://www .iis.fraunhofer .de/content/dam/iis/de/doc/ame/w p/FraunhoferIIS_T echnical - Paper_EVS.pdf [32] TR 45.820 v13.1.0, “Cellular system support for ultra low complexity and low throughput internet of things,” Nov . 2015. [33] A. Elnashar , Mohamed A. El - Saidny , “ Practical Guide to L TE - A, Vo LT E a n d I o T: P a v i n g t h e w a y t o w a r d s 5 G ,” Wil e y , J u l y 2 0 1 8 . A yman Elnashar received the B.S. degree in electrical engineering from Alexandria University, Egypt, in 1995 and t he M.Sc. and Ph.D. degrees in electr ical comm unications engineering from Mansoura University, Egypt, in 1999 and 2005, respectively. He has 20+ years of practical experience and leadership positions in ICT industry including planning, deployment and operation . He was part of three ma jor start - up service providers (Orange/Egypt, Mobily/KSA, and du/UAE). Currently, he is VP and Head of Inf rastructure P lanning – ICT & Cloud w ith the Emirates Integrated Telecommunicat ions Co. “du”, UAE. He is the founder of the Terminal Innovation Lab and UAE 5 G innovation Gate (www.u5gig.ae) to derive 5G ado ption in UAE. H e is leading the mobile/f ixed network tra nsformation tow ards SDN/N FV and dig ital transformation of EITC Infrastructure to becom e a digital ICT servic e provider. Prior to this, he was S r. Director – Wireless Networks, Terminals and IoT where he managed and directed the evolution, evaluation, and introduction of du wireless networks, terminals and IoT including LTE/LTE - A, HSPA+, WiFi, NB - Io T and currently working towards deploying 5G network in UAE. Prior to this, he was with Mobily, Saudi Arabia, from June 2005 to Jan 2008 as Head of Projects. He played key role in contributing to the success of the mobile broadband network of Mobily/KSA. From March 2000 to June 2005, he was with orange Egypt. He pub lished 3 0+ papers in wireless communica tions an d wirel ess networks in highly ranked journals and international conferenc es. H e is the author three b ooks published by W iley: • Design, Deployment, and Performance of 4G - LTE Networks: A Practical Approa ch, published in May 2014, • Practical Guide to LTE - A, VoLTE and IoT: Paving the Way Towards 5G, publishe d in Jul y 2018, • Simplified Robust Adaptive Detection and Beamforming for Wireles s Communicat ions, pu blished in July 2018. His r esearch in terests include pr actical performance analysis, planning and optimization of wireless networks (3G/4G/WiFi/IoT/5G), d igital signal processing for wireless communications, multiuser detection, smart antennas, massive MIMO, and robust adaptive de tection and beamforming. He is more focused now on network transfor mation t owards SDN/NFV , Cloud transformat ion, digital transfo rmation, ICT application s, IoT evolution and 5G use cases. Mohamed A. El - saidny rec eived the B.Sc. degree in Computer Engineering and the M.Sc. degree in Electrical Engineering from the University of Alabama in Huntsville, USA in 2002 and 2004, respectively. He i s currently Director - Technology at MediaTek. He is a leading technical expert in wireless comm unication systems for modem chipsets and network design . He is driving a team responsible for the technology evolution and the alignment of the network operators to the device and chipset roadmaps/products in 3G, 4G and 5G. His main focus is on expanding MediaTek technologi es and technical expe rtise with the mobile ne twork operators worldwide. Prior to Media Tek, he worked a t Qualcomm CDMA Technolog y, Inc. (QCT), San Di ego, Cali fornia, USA. He later moved to mobil e network design in Qualcomm’s Corporate Engineering Services divisi on in Dubai, UAE. A t Qualcomm, he was responsible for performance evaluation and analysis of the UMTS and LTE system solutions for user equipment. He developed and implemented system studies to optimize the performance of various UMTS and LTE algorithms including Cell re - selection, Hando ver, Cell Search and Paging, CSFB, C - DRX, Inter - RAT, VoLTE/IMS, Carrier Aggregation and multi - band load balancing techniques. His current research interest is on the 5G evolution and gap analysis in 5G requirements compared t o the 4G deployment challenges in the a reas of p hysical layer, High Reliable/Low Laten cy s ystems, and waveforms design concepts . He is the inventor of n umerous patents in CDMA and FDMA systems, the co - author of “Design, Deployment and Performance of 4G - LTE Networks: A Practical Approach” book by Wiley & Sons, in addition to contributions to 3GPP algorithms. He published several in ternational research papers in IEEE Communication s Magazine, IEEE Vehicular Technology Magazine, and other IEEE Transactions. www.u5gig.ae Mohamed Yehia received the B.S. degree in electrical communications engi neering fr om Cairo University, Egypt, in 2006. Current ly, he is currently Manager of Terminal s and Performance with the Emirates Int egrated Telecommunicat ions Co. “du“ , UAE. He is respon sible for testing and validation of new terminals/chipsets and new 2G/3G/LTE features. H e has over all 9 years of experience in multi - culture environments, focusing on GSM, WCDMA and LTE protocols , troubleshooti ng, features testing and deployment, capacity planning, network dimensioning, network optimization and network performance monitoring . And Ha ving management and technical experience within different projects for network implementations, managed services in Middle East and Africa, he has proven experience in Wireless 2G/3G/HSPA/HSPA+/LTE/LTE - A.