You are on page 1of 13

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO.

7, SEPTEMBER 2009

3661

A Performance Summary of the Evolved 3G (E-UTRA) for Voice Over Internet and Best Effort Trafc
Mika Rinne, Markku Kuusela, Esa Tuomaala, Pasi Kinnunen, Istvn Kovcs, Kari Pajukoski, and Jussi Ojala
AbstractEvolved Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access (E-UTRA) is a new technology that targets the long-term evolution (LTE) of third generation. The E-UTRA technology has recently been standardized by the Third-Generation Partnership Project (3GPP) as a set of (series_36) specications. The standardization process is almost complete; this paper does not focus on comparing specic proposed schemes but instead aims to outline the system performance more in general. E-UTRA operates in the packet-switched domain; hence, all trafc is handled by the packet protocols. The analysis is shown for Voice over Internet Protocol (VoIP) and Best Effort (BE) data, as these are common examples of real-time and non-real-time trafc, respectively. The performance is given in the VoIP study as the number of satised voice users per cell and in the BE study as the cumulative probability distribution of user (or cell) throughput, which was scaled to the spectral efciency values, given both at the mean and at the cell edge. The VoIP capacity extends beyond 350 users in downlink and 230 users in uplink, which is a considerable increase compared with the high-speed packet-access (HSPA) reference. The BE spectral efciency meets the 3GPP targets and yields impressive gains of order 2.53 times the well-performing HSPA reference. This clearly exceeds the mean of 1.5 b/s/Hz/cell in downlink and 0.8 b/s/Hz/cell in uplink for a baseline antenna conguration with penetration loss of 20 dB. In some scenarios, these mean gains exceed four times the reference. Index TermsCellular performance, discrete Fourier transform-spread-OFDMA (DFT-s-OFDMA), Evolved Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access (E-UTRA), multiple access, orthogonal frequency-division multiple access (OFDMA), single-carrier frequency-division multiple access (SC-FDMA), third-generation long-term evolution (3G LTE).

method and the use of packet-switched protocols, which enable fast scheduling of resources. The selected multiple-access method enables bandwidth scaling, improvement of intracell orthogonality despite the increased symbol rates resulting from wider bandwidth, and a reduction of intersymbol interference. Furthermore, the opportunity of frequency-domain processing keeps the complexity reasonable and allows the exploitation of frequency-selective channel properties. In this paper, the performance evaluation of E-UTRA is shown for Voice over Internet Protocol (VoIP) and Best Effort (BE) data according to the guidance and assumptions in [4][6]. The evaluation concentrates on the frequency-division duplex. The structure of this paper is as follows. The multiple-access technology and the transmitterreceiver structures are briey described in Section II. The channels and resource allocation including frame structure, transport formats, and scheduling are addressed in Section III. The trafc scenario and simulation methodology are described in Section IV. The simulation results are presented and analyzed in Section V, and the summary is given in Section VI. II. M ULTIPLE A CCESS AND T RANSMITTER R ECEIVER S TRUCTURES A. Downlink Technology The downlink technology [2] is based on orthogonal frequency-division multiple access (OFDMA), which allows fast allocation of radio resources and orthogonal multiuser multiplexing in the frequency domain. The wideband signal is generated from modulated frequency-domain subcarrier signals using a scalable fast Fourier transform (FFT) transform. The resolution of resource allocation is a set of adjacent subcarriers that form a physical resource block (PRB). The PRBs are scheduled and exibly allocated for user equipment (UE) in the frequency domain [exible-frequency-domain multiplexing (FlexibleFDM)] so that any set of PRBs may be allocated to any UE as convenient. A special embodiment of the Flexible-FDM is a PRB level distributed transmission, which seeks to maximize frequency diversity, when frequency domain scheduling is not expected to gain. For a small payload (as a VoIP packet), it is further feasible to share the same set of distributed PRBs among multiple (two to three) users. The localized frequency allocations are preferred whenever frequency scheduling is expected to gain and whenever the needed payload is large. Distributed allocations are valuable for small limited payload sizes and if frequency diversity is appreciated by other means than

I. I NTRODUCTION VOLVED Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access (E-UTRA) technology [1], [2] aims to improve third generation (3G) by meeting a new set of signicantly increased target requirements in [3]. The improvements include, e.g., a new multiple-access
Manuscript received March 2, 2008; revised October 27, 2008 and January 7, 2009. First published February 3, 2009; current version published August 14, 2009. The review of this paper was coordinated by Prof. W. Zhuang. M. Rinne, M. Kuusela, and J. Ojala are with Nokia, Itmerenkatu 11-13, FI-00180 Helsinki, Finland (e-mail: mika.p.rinne@nokia.com; markku.kuusela@nokia.com). E. Tuomaala is with Nokia, Economic and Technological Development Area, Beijing 100176, China. P. Kinnunen and K. Pajukoski are with Nokia Siemens Networks, FI-90630 Oulu, Finland. I. Kovcs is with Nokia Siemens Networks, 9220 Aalborg, Denmark. Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TVT.2009.2014457

0018-9545/$26.00 2009 IEEE

3662

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 7, SEPTEMBER 2009

Fig. 1. OFDMA transmitterreceiver (TXRX) structure for E-UTRA downlink. Signal processing blocks are shown including the CP extension in the transmitter and CP removal in the receivers. Front-end (FE) processing includes ltering and high-power/low-noise ampliers and RF/antenna sections.

frequency-adaptive scheduling. The granularity of distributed allocations is coarser than that of the localized allocations. The signal processing blocks of orthogonal frequencydivision multiplexing (OFDM) transmitter and receivers for a multiuser downlink transmission are shown in Fig. 1. The timecontinuous signal waveform of OFDM is given in [2, eq. (1)]. The layer mapping and the precoding stages add the possibility of multiantenna transmission. The numerology is given in [2] and has been designed to allow signal generation by a 2kpoint FFT with a subcarrier spacing of 15 kHz and a PRB of 12 subcarriers. This scheme allows scaling of the system bandwidth to the needs of an operator. At least, the bandwidths of {6, 15, 25, 50, 75, 100} PRBs in the range of 1.2520 MHz are available in the specications. The bandwidth efciency is about 0.9 with ltering. The subcarrier spacing was selected to allow for Doppler tolerance over a large range of UE velocities. This will yield a sufciently high SNR to modulate either by quadrature phase shift keying, 16QAM, or 64QAM and up to two independent symbol streams in the presence of imperfect power control, phase noise, and RF imperfections [7], while keeping the computational complexity at a reasonable level. Further denitions for precoding up to four symbol streams have been included in the specications. (p) The time-continuous signal Sl (t) on antenna port p in OFDM symbol l in a downlink slot is dened in [2, eq. (1)] as Sl (t) =
DL k= NBW /2 DL NBW /2

DL for 0 t < (NCP,l + N ) Ts , where k () = k + NBW /2 , (+) DL DL DL PRB = k + NBW /2 1, N = 2048, NBW = NPRB NSC , k (p) the subcarrier spacing is f = 15 kHz, and ak,l is the complex value content of resource element (k, l) on antenna port p.

B. Uplink Technology The uplink technology [2] is based on single-carrier frequency-division multiple access (SC-FDMA), which provides excellent performance due to intracell orthogonality in the frequency domain comparable to that of OFDMA multicarrier transmission, while still preserving a single-carrier waveform and, hence, the benets of a lower peak-to-average power ratio (PAPR) compared with OFDMA [8]. As the linear range of the UE transmission power for a practical power amplier is constrained, a lower PAPR will increase coverage, not only at the cell edge, but for each of the supported modulation alphabets and code rate sets as well. Further, due to the lower PAPR, the power amplier backoff can be decreased, which increases the battery active time. The signal-processing blocks of SC-FDMA transmitters and a receiver for multiuser uplink transmission are shown in Fig. 2. The time-continuous signal waveform of SC-FDMA is given in [2, eq. (2)]. The uplink signal may be created by a cascade of discrete Fourier transform (DFT) and inverse FFT, which is also known as DFT-spread-OFDMA (DFT-s-OFDMA). The uplink numerology is comparable to the downlink numerology. The bandwidth adaptation, in the uplink, allows emitting the UE power in a very narrow bandwidth, which extends the coverage, e.g., compared with a spread-spectrum transmission of 5 MHz (wideband code division multiple access (WCDMA) [9], [10]). However, the bandwidth adaptation allows shortterm bandwidth sharing between multiple users with resource

(p)

ak() ,l ej2kf (tNCP,l Ts )


(p)

(p)

+
k=1

ak(+) ,l ej2kf (tNCP,l Ts )

(1)

RINNE et al.: PERFORMANCE OF THE EVOLVED 3G (E-UTRA) FOR VOICE OVER INTERNET AND BE TRAFFIC

3663

Fig. 2. SC-FDMA transmitterreceiver (TXRX) structure for E-UTRA uplink. The signal processing blocks are shown including the CP extension in the transmitters and the CP removal in the receiver. FE processing includes ltering and high-power/low-noise ampliers and RF/antenna sections.

Fig. 3. E-UTRA frame structure. The symbol blocks (control/data/pilot) are time multiplexed, and the resource allocations are frequency multiplexed. The subframe is shown with 14 symbol blocks due to the normal CP length.
PRB size of NSC and can be expressed as radix {2, 3, 5} so that UL NBW = 2r1 3r2 5r3 NSC , where r1 0, r2 0, and r3 > 0.

allocations ranging from the minimum bandwidth of a PRB to the maximum available bandwidth. Thus, adaptive transmission bandwidth (ATB) is valuable in terms of coverage extension, load sharing, and maximizing the bandwidth usage. The time-continuous signal Sl (t) in SC-FDMA symbol l in an uplink slot is dened in [2, eq. (2)] as
UL NBW /2 1

III. C HANNELS AND R ESOURCE A LLOCATION A. Frame Structure The E-UTRA frame structure is shown in Fig. 3 and consists of ten subframes, each of which contains 12 or 14 symbol blocks, depending on the choice of cyclic signal extension (prex) between the symbol blocks. The principle of keeping the signal orthogonal in OFDMA is to let the cyclic prex (CP) be at least as long as the maximum delay dispersion of the channel so that interference from the previous symbol block is attenuated for the following symbol block.

Sl (t) =
UL k= NBW /2

ak() ,l ej2(k+1/2)f (tNCP,l Ts ) (2)

UL for 0 t < (NCP,l + N ) Ts , where N = 2048, NBW = UL PRB NPRB NSC determines the ATB, the subcarrier spacing is f = 15 kHz, and ak,l is the complex value content of resource UL element (k, l). NBW takes values that are multiples of the PRB

3664

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 7, SEPTEMBER 2009

To support a wide range of delay spreads without sacricing the spectral efciency, two alternatives of CP lengths have been dened. The normal prex (on the order of 5 s) is for channels with a short delay spread, which is typical for small urban cells and microcellular environments. The extended prex (on the order of 17 s) is for channels with a longer delay spread, which is typical for large cells and macrocellular environments. The prex isolation of the symbol blocks maintains the cyclic convolution properties of the signal in a dispersed channel and allows frequency-domain processing in the receiver. In particular, frequency domain equalization (FDE) is known to be much less complex than time-domain equalization when the bandwidth extends beyond 5 MHz. In Fig. 3, sync marks the OFDMA subcarrier resources in the central portion of the bandwidth that carry the primary and secondary synchronization sequences in the last two symbol positions of the rst slot in subframes #0 and #5. System information is present, respectively, in subframe #0. In the downlink, time multiplexing is applied between the control resources and the shared data resources in every subframe. The rst symbols of every subframe contain a number of physical downlink control channels (PDCCHs). Each PDCCH contains scheduling and allocation information for either the downlink or the uplink, which is separately encoded. The remaining symbols in the subframe carry the physical downlink shared channel (PDSCH) for data. PDSCH contains the physical resource allocations in the resolution of PRBs with frequency multiplexing between users. In the uplink, frequency multiplexing is applied between the control resources and the shared data resources in every subframe. The physical uplink shared channel (PUSCH) is frequency multiplexed between the UE transmissions in the resolution of PRBs. To maintain single-carrier properties, only an adjacent set of PRBs can be allocated to one UE. However, the allocated PRBs may have a frequency shift (a hop) between the slots of a subframe. In case UE does not have data allocation, the non-datadependent control signaling is transmitted in the physical uplink control channel (PUCCH). A PUCCH may carry timemultiplexed channel quality indication (CQI) reports, acknowledgements of the downlink packets, or scheduling requests. The number of PUCCHs per cell is congured by a higher-layer protocol, and each PUCCH reserves two PRBs. Inside each PUCCH, the resources are code multiplexed, e.g., an acknowledgement bit of a downlink packet from one UE is mapped to a zero autocorrelation (ZAC) sequence in the frequency domain and is block-wise spread (Hadamard) in the time domain. The signaling bits of UE inside one PUCCH are modulated on orthogonal sequences. When UE has data allocation, non-datadependent signaling is time multiplexed with the uplink data into the PRBs of the PUSCH to maintain the single-carrier property. Every symbol block of the PUSCH subframe carries demodulation reference symbols (DMRS) for channel estimation, equalization, and demodulation. For scheduling decisions, in addition to DMRS, a sounding reference symbol (SRS) may be transmitted in a subframe common symbol position with a UE-specic periodicity of {2, 5, 10, 20,. . ., 360} ms. SRS

transmission in the frequency domain is also congurable per UE, and it may be transmitted over a given narrow band of {4, 8, 12, 16, . . .} PRBs or over the full bandwidth. B. Resource Allocation Similarly, in downlink and uplink, each set of PRBs allocated to one UE carries a single transport block (TB) of a selected transport format. The transport format denes a given modulation alphabet and a channel code rate from a set of modulation and coding schemes (MCSs). In case of a dualcodeword multistream transmission [multiple inputmultiple output (MIMO)], two transport blocks of independent transport formats are modulated to the same physical resources in that subframe. The packet scheduler allocates physical (PRB) resources at a subframe resolution, and the link adaptation algorithm selects a proper transport format. Resource allocations for downlink and uplink are decided in E-UTRAN NodeB (eNB) that is the base station of the E-UTRA network. Adaptive modulation and coding, as well as Hybrid Automatic Repeat reQuest (HARQ) retransmissions, provide short-term channel-dependent gains and self-adaptation of the operation points. Again, the frequency scheduling and the multiuser channel sharing are fundamental performance gain mechanisms for E-UTRA. The fast HARQ process cycles are depicted in [11]. Resource allocation in a PDCCH refers to the current downlink subframe and to the fourth uplink subframe. This allows encoding of non-data-dependent downlink acknowledgements and CQI reports to the same PUSCH as the actual uplink data. The new resource allocation for the next data (new data or a retransmission) will appear in the eighth subframe. This denes the number of HARQ processes to eight and leaves 3-ms processing time for eNB (decoding, scheduling, Layer 2 protocol processing, encoding, and multiplexing) and 3 ms minus timing advance time for UE processing (decoding, Layer 2 protocol processing, encoding, and multiplexing). For uplink transmission, the UE transmit power is separately controlled for PUSCH, PUCCH, and SRS. For a PUSCH transmission in subframe i, the transmit power (in dBm) is set by an MCS-dependent open-loop value adjusted by a closed-loop correction [11] given as PPUSCH (i) = min {Pmax , 10 lg (MPUSCH (i)) + P0_PUSCH + P L + MCS (MCS(i)) + f (i)} (3)

where Pmax is the maximum allowed transmit power for that UE power class, MPUSCH (i) is the number of allocated PRBs, P0_PUSCH is the UE specic component set by a higherlayer protocol, is the cell-specic parameter for fractional path loss compensation, and P L is the path loss estimate. MCS (MCS(i)) is the MCS-dependent power offset set by a higher-layer protocol for the serving cell, and it is valid for the selected transport format MCS(i) in subframe i. f (i) is the closed-loop power control adjustment based on the power control command received from PDCCH with a delay of about 3 ms. The uplink power control is analyzed in [12].

RINNE et al.: PERFORMANCE OF THE EVOLVED 3G (E-UTRA) FOR VOICE OVER INTERNET AND BE TRAFFIC

3665

C. Multiantenna Transmission and Reception E-UTRA transmission in a single cell is orthogonal in both time and frequency, resulting in a large probability of high signal-to-interference-plus-noise ratio (SINR) at the receiver. Thus, environments with high channel rank are more probable (several large eigenvalues may exist above the SINR oor), and transmission of multiple information streams from several antennas can be exploited. The multiantenna transmission is provided as a precoded layered transmission, where the number of layers is bounded by the number of antenna ports. Transmit diversity is created as a single codeword transmission, where precoding applies over the layered mapping of symbols. Spatial multiplexing is created as a dual-codeword transmission, where precoding applies per codeword over the layered mapping of symbols. A multistream MIMO transmission has the potential of increasing the instantaneous peak data rates. In case the channel rank or SINR suddenly changes, the multistream transmission may no longer be feasible, and the diversity order has to be increased instead. This is achieved by channel-dependent MIMO adaptation, a.k.a. rank adaptation, which is based on CQI feedback from the UE [13]. Both for the increase of diversity order and as an enabler for multistream transmission, a dual-antenna receiver in the UE is essential. IV. T RAFFIC S CENARIO AND S IMULATION M ETHODOLOGY A. VoIP Scenario Packet transmission of voice is expected to gradually replace the circuit-switched voice connections due to the exible and cost-efcient delivery of packets over the Internet. The voice frames that are output from the codec are delivered by the real-time transmission protocol (RTP) [14], which includes format descriptions and time stamps of the coded voice frame, typically carrying a phoneme. For RTP, the transport protocol is the User Datagram Protocol (UDP) [15], which includes the application port numbers, but does not provide retransmissions [like Transmission Control Protocol (TCP)]. For RTP/UDP, the Internet protocol [16], [17] is applied for end-to-end routing. Thus, every packet of voice over Internet service contains large networking headers compared to the relative size of the voice payload. The RTP/UDP/IP protocol overhead decreases the efciency and impacts the capacity and coverage of the radio access. However, in a wideband transmission as E-UTRA, this is not a major problem, particularly if the network load is low, as the offered capacity is large, and the communication ows of different trafc types can efciently be multiplexed. However, when the load increases, the large overhead is not tolerable, and header compression is recommended. Robust header compression (ROHC) [18] can actually reduce the size of networking headers to the minimum and achieve maximal efciency. Furthermore, the scheduler may impact efciency, because it may decide to bundle individual VoIP packets within the delay constraints to the same transport block, and it may optimize the physical resource allocations for that transport block size. As voice frames have a fairly deterministic interarrival process, which is moderately jittered by the Internet, the resource al-

location scheme may use persistent knowledge of scheduling events as an alternative of fully dynamic (FD) scheduling [19], [20]. Among such persistent allocation schemes, talk-spurt persistency seems to be the most appropriate choice [21]. B. BE Data Scenario Packet transmission of BE data over the Internet includes TCP and IP. The TCP/IP networking headers are large, but they can be compressed to the minimum at the radio access without any complexity. As the E-UTRA system targets at serving packet trafc, the system performance largely depends on the packet-scheduling algorithms [22][24]. In this paper, two types of schedulers were chosen to be analyzed, namely, a round robin (RR) scheduler and a proportional fair (PF) channel-dependent scheduler. The RR scheduler equally allocates resources in the timefrequency sequential order. The PF scheduler allocates resources in the frequency resolution of a PRB and in the time resolution of a subframe, so that each UE gets resource allocations according to a dened fairness criterion. Such a criterion may be tuned to gain in the average cell throughput, to boost the cell edge throughput, to balance the throughput experienced by the served UEs, or to unequally weight the throughput of logical channel ows. In this paper, the RR results are mainly presented for calibration purposes, and the PF results show the expected E-UTRA performance. C. BE Data Scenario With MIMO The downlink performance of E-UTRA may further be improved by MIMO transmission technology. This is particularly viable for BE trafc, which enables a scheduling choice of large transport block sizes. Unlike the BE trafc, VoIP with small transport block sizes and lengthy interarrival times does not largely gain from MIMO. In this paper, the MIMO schemes include a well-known perantenna rate control (PARC) [25] and a precoded MIMO. A single codeword rank 1 precoding scheme [2] provides a fallback mode whenever multistream transmission is occasionally not feasible and diversity transmission is needed. For diversity transmission, maximal ratio combining (MRC) receiver or interference rejection combining (IRC) receiver is applied. For PARC and precoded MIMO, linear minimum mean squared error (LMMSE) receivers are separately used for the independent streams. The CQI feedback per fractional bandwidth allows both rank adaptation and independent selection of MCS per codeword [13]. This paper primarily studies large cell sizes, where rank adaptation is an essential feature. In a microcell scenario, where SINR is higher, MIMO could provide even larger gains. D. Simulation Methodology This paper aims to show the performance numbers of E-UTRA for the most typical trafc proles, i.e., VoIP and BE data. The simulation scenario and the assumptions are summarized in Tables IIII. The assumptions for trafc and protocol

3666

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 7, SEPTEMBER 2009

TABLE I SUMMARY OF THE MOST IMPORTANT ASSUMPTIONS FOR THE S YSTEM S IMULATION S CENARIO

TABLE II SUMMARY OF THE MOST IMPORTANT ASSUMPTIONS FOR THE D OWNLINK S IMULATIONS

models are given in Table IV. More detailed descriptions can be found in [4][6]. For VoIP trafc, a simple two-state voice activity model was used. During the voice-active state, a talk-spurt is generated with a sequence of voice packets having a constant interarrival time and a constant size after header compression. For the voice-inactive state, silence descriptor (SID) packets are generated. For BE trafc, innite buffer model was applied with a log-normal truncated le-size distribution. The simulation methodology uses a quasi-static approach with intercell interference generated by full load packet trafc in a simulated area. The statistics is collected from a set of center cells of the simulation area or alternatively from all cells with wraparound to ensure rich interference characteristics. The frequency reuse factor is 1, which makes scheduling and resource allocation a challenging task. The quasi-static approach allows a large number of snapshot simulations, yet each snapshot has a continuous-time model to include a sufcient amount of channel state changes for link adaptation, scheduling, and transport format selection. These channel state changes include not only frequency-selective fast fading but changes of correlation and phase information of the complex channel as well. In addition, interference changes during a snapshot. Link adaptation and MCS selection algorithms target the largest instantaneous throughput in the allocated resources as a function of SINR and expected block error rate (BLER). The outer loop of link adaptation may locally tune the MCS se-

lection thresholds and also the expected BLER target of the rst transmission of a transport block to increase the cell throughput. Control channel resource utilization is analyzed relative to the resource allocations by calculating the PDCCH formats [2] and their aggregation on average. For VoIP, this is taken into account as control channel limitations as the number of PDCCHs available in a subframe. For BE, the worst-case overhead is assumed during the simulations, and scaling for the real expected overhead is done by postprocessing. The overhead scaling is in the resolution of OFDM symbols, and its shortterm variations are not taken into account. Uplink signaling is calculated as the overhead that is present in PUSCH, or it appears as resource reservation for PUCCH so that CQI reports, downlink acknowledgements, and scheduling requests can be transmitted for all the served UEs. The short-term variations

RINNE et al.: PERFORMANCE OF THE EVOLVED 3G (E-UTRA) FOR VOICE OVER INTERNET AND BE TRAFFIC

3667

TABLE III SUMMARY OF THE MOST IMPORTANT ASSUMPTIONS FOR THE U PLINK S IMULATIONS

TABLE IV SUMMARY OF THE MOST IMPORTANT ASSUMPTIONS FOR THE TRAFFIC MODELS AND PROTOCOLS

and error performance of uplink signaling are not taken into account. For power control in downlink simulations, the transmit power is equally shared between the PRBs. In uplink simulations, power control is an ideal open-loop algorithm, which assumes path loss measurements with inaccuracy and reporting delay (both given in Tables II and III). The power setting is accurate, however, having fractional path-loss compensation. The power setting at a transmit time is a function of the selected MCS; it controls the power spectral density, and it scales with the allocated bandwidth. The power factor P0 is varied as a parameter, and it is shown to have an impact on the cell edge

to the cell mean efciency ratio (see Section V). In practice, open-loop power control is very inaccurate at discontinuity, and closed-loop power control correction may additionally be given at every instance of uplink allocation. This closed-loop power correction may be an absolute or an accumulated control, and it will result in accurate-enough power setting, as HARQ provides further adaptation to the channel dynamics. The propagation characteristics of a physical link, such as distance-dependent path loss, shadowing, and frequencyselective fast fading, are modeled in the system simulator, whereas the link performance is fed to the system simulator either through the exponential effective SNR mapping (EESM) interface [26] or through the actual value interface (AVI). The Third-Generation Partnership Project Typical Urban channel model (3GPP TU) with 20 taps was applied for training the EESM interface in subcarrier symbol resolution. Further, the spatial channel model (SCM-C) [4], [5] was applied, including spatial antenna correlations that were calculated beforehand to the channel covariance matrix. The number of served VoIP users per cell is analyzed to give comparable results to the circuit-switched voice capacity in conventional systems. The VoIP capacity is a measure of the number of users per cell according to the satised user criterion (SUC), i.e., when 98% (95%) of the users are satised. A VoIP user is dened satised if 98% of the packet latency distribution of that user is at or below 50 ms. This denition lets us assume an end-to-end delay below 200 ms for seamless mobile-to-mobile communications [6]. For BE data, the statistics are created as cumulative probability density functions (cdfs). Both the mean cell throughput and the cell edge (5 percentile) throughput were calculated from the statistics and scaled to the spectral efciency values. For scheduling, a fairness criterion was set. The fairness bound to exceed was given at points {0.1, 0.2, 0.5} of the normalized throughput for probability of {0.1, 0.2, 0.5}, respectively.

3668

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 7, SEPTEMBER 2009

TABLE V VoIP CAPACITY AS THE NUMBER OF SATISFIED USERS PER CELL. EVALUATION WITH FD AND SP SCHEDULER (UPLINK MEAN IoT WAS ABOUT 1314 dB.)

As the velocity of the mobile UE is expected to impact the scheduling and feedback, a study was conducted to draft the performance degradation in case, where the mixed-velocity links are present during simulation. In here, a set of modied ITU models (PedB and VehA) [5] were used. The modied model refers to the wideband properties of the channel providing a large number of taps, frequency selectivity, and correlation. The actual scenario had a random choice of link property for each radio link so that the velocity was 3 km/h at 60%, 30 km/h at 30%, and 120 km/h at 10% probability. The scenario had to further be extended to a large intersite distance (ISD), where high velocity may appear in practice. V. E-UTRA S YSTEM P ERFORMANCE FOR VoIP AND BE T RAFFIC For VoIP, the E-UTRA capacity is summarized in Table V. The results were obtained both for an FD scheme and for a semipersistent (SP) scheme. The FD scheme relies on CQI feedback and efcient scheduling of the frequency resources. The tolerable time constraints of scheduling are bound by the SUC (50 ms). The SP scheme has nonscheduled blind constant allocations for every rst transmission of a packet. The transport format of the rst transmissions is set constant for the duration of a talk-spurt [19][21], but packet retransmissions are scheduled, and their transport format is selected to be similar to the FD allocations. This obviously saves control channel resources. In addition to the scheduling and allocation schemes, the VoIP capacity largely depends on E-UTRA features and properties such as the control channel overhead and signaling efciency. The capacity can instantaneously be limited either by the shared data channel or by the control channel resources (power or symbol resources). The control channel capacity is exible up to the upper bound of three full symbols. The consumption of control channel resources varies as a function of the scheduler. Packet bundling was optionally applied here for the downlink, and, therefore, the control signaling need was slightly unbalanced, having about four to six signaled allocations in a downlink subframe and about four to eight signaled allocations in an uplink subframe.

The VoIP capacity of E-UTRA extends beyond 350 users in downlink and beyond 230 users in uplink for ISD up to 1000 m. This is a considerable increase compared to a reference system reported in [27]. Increasing the ISD from 500 to 1000 m shows that the downlink performance remains the same, whereas the uplink capacity may decrease over 10%, depending on the scheduler. Comparing the schedulers, we note that the performance of the FD scheduler gets limited by the control channel resources. The results indicate that the VoIP capacity with FD scheduler increases almost linearly as a function of the increasing number of PDCCH channels. Control channel limitation to the VoIP capacity with FD scheduler is also visible in Fig. 4(a), which contains the distribution (cdf) of the scheduled PRBs per subframe with six PDCCH channels. Due to control channel limitations, the average PDSCH utilization with FD scheduler is only 40%. However, the control channel limitation can partly be avoided by packet bundling, which increases the average PDSCH utilization to 70%, because larger transport blocks can be allocated to small physical resources during good channel conditions. This is enabled by the fast CQI-based link adaptation. On the other hand, the average PDSCH utilization for the SP scheduler gets even higher (above 90%), which shows that the VoIP capacity with SP scheduler tends to mainly be limited by the PDSCH resources. The packet delay distribution (cdf) using different schedulers is shown in Fig. 4(b). Due to severe control channel limitations, the amount of packets having delays close to the delay bound (50 ms) is clearly higher for the FD scheduler. With packet bundling, control channel limitations are relaxed, and, therefore, delay critical packets can be scheduled earlier. Bundling itself adds delays in cumulation of 20 ms. For the SP scheduler, the control channel limitations mainly vanish, as the rst packets are directly scheduled into the reserved resources, and the packet delay distribution is smooth. In all these schemes, retransmissions are scheduled at the earliest convenience, i.e., after 8 ms of HARQ cycle. For BE, the results are created with a full set of radio resource management (RRM) features. The mean cell throughput and the cell edge throughput are calculated from the full statistics, and they were scaled to the spectral efciency values. Earlier studies of this kind appear in [28][31], and a numerical summary is available in [32]. The specication and simulation models are updated since, and the new results are given in Figs. 5 and 6 and Table VI for downlink and in Fig. 7 and Table VII for uplink. In these tabular results, the downlink control channel overhead was taken according to the realistic expectations, and the uplink SRS overhead was moderated to practical values. For downlink, the baseline results are shown for a twoantenna receiver and for the selected MIMO transmission schemes. The fairness bound is shown with the normalized throughput curves as a probability line that needs to be exceeded. (The cdf curves are on the right side of the fairness bound.) Looking at the throughput statistics, we see that the IRC receiver algorithms particularly gain in the lowSINR regime, which is critical for cell edge throughput. As regards the channel models, the received subcarrier-powerto-interference-power ratio is slightly lower over the full

RINNE et al.: PERFORMANCE OF THE EVOLVED 3G (E-UTRA) FOR VOICE OVER INTERNET AND BE TRAFFIC

3669

Fig. 4. VoIP distributions with FD and SP schedulers. (a) Number of scheduled PRBs per subframe and (b) the packet delay. The results with the fully dynamic scheduler are shown without (w/o) and with (w) packet bundling.

Fig. 5. Downlink 1 2 results with PF channel-dependent scheduler with MRC and IRC receivers. (a) User goodput with the lowest 20 percentile zoomed. (b) User goodput normalized to the geometric mean. (c) Modulation and coding statistics.

range for the SCM model compared to the TU model, which converts to slightly lower reported throughput values. The full distributions show high-throughput observations for all users, and the fairness bound is clearly exceeded. High

throughput is primarily achieved by the receiver algorithms, schedulers, and adaptation algorithms. According to Fig. 5, link adaptation seems to provide MCS selection over a large set of transport formats.

3670

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 7, SEPTEMBER 2009

Fig. 6. Downlink 2 2 results. (a) User goodput normalized to the geometric mean with the lowest 20 percentile zoomed. (b) Effective received SINR for the scheduled PRBs with the selected MCS shown for the highest SINR regime (above 80 percentile) when the dual-codeword transmission was actually applied. TABLE VI DOWNLINK SPECTRAL EFFICIENCY AT THE MEAN AND AT THE CELL EDGE FOR VARIOUS MULTIANTENNA CONFIGURATIONS, MRC/IRC RECEIVERS, AND RR/PF SCHEDULERS

When comparing the E-UTRA baseline performance with the reference, gains of order 250%300% were obtained. As stated, the performance may further be improved by MIMO. However, in the selected large-cell scenario, where the amount of users who have any benet from the multistream transmission was modestly around 30%, MIMO only provided marginal average gains. This is visible in Fig. 6(b), where the effective received SINR of the scheduled PRBs indicates the probability of the dual codeword being actually in use for the selected transport format. In these studies, the 4 4 MIMO scheme had more ideal assumptions than the other schemes, and it can be seen as the upper bound of performance in Table VI. Anyway, in a microcell scenario, where a larger proportion of users will benet from multistream transmissions due to higher SINR distribution, the average gains would be much higher. For uplink, the baseline results are shown for a singleantenna transmitter and for a two- or four-antenna receiver (see Fig. 7). The operation regime was selected following the guidelines of loading in [33], that is, not to exceed the Interference over Thermal noise (IoT) level 20 dB by a probability larger than 0.05. For this criterion, we observed the mean IoT of about 14 dB with less than 3-dB deviation. The statistics of MCS selected per transport block are shown in Fig. 7(c) for the applicability of adaptive modulation and coding. It was observed

that by tuning the algorithms and MCS selection thresholds, it is feasible to increase the cell edge performance with a minor impact on the mean. For a small ISD (we had P0 close to 60 dB), this is still not as dramatic as it gets for the large ISD shown later [see the mixed velocity case in Fig. 8(c)]. There, the cell edge performance may easily be sacriced by inappropriate parameterization. Typically, the uplink cell edge efciency is one of the most critical measures of system performance, and gains are particularly favored in that regime. Throughout all the cell throughput results, PF channel-dependent scheduling gains are signicant. Further, the receiver diversity gains achievable by the increase of the number of antenna elements in the eNB are notably large. For BE in the mixed-velocity case, a subset of transmission schemes and receivers are reported to show the relative performance loss. As the target here was not to reach the highest of all mixed-velocity performance but rather to study the relative performance loss, multistream MIMO schemes were omitted. The results are shown in Fig. 8 for both downlink and uplink. Performance at 3 km/h still dominates these results, because the scheduling gains at that velocity are signicant, as discussed earlier. The large intersite distance set for the mixed velocity analysis will as such imply severe limitations to the observed SINR and, hence, to the throughput. Performance at about 30 km/h is critical, as the channel-dependent scheduling may still gain over a blind diversity transmission. However, scheduling gains can only be obtained by fairly accurate and fast CQI feedback (for downlink) or accurate channel sounding measurements (for uplink). Thus, the increased overhead (in uplink) of feedback (for downlink), as well as the inaccuracy of reports, will partly mitigate the achievable gains. For 120 km/h, the channel coherence time is so short that it is not reasonable try to do channel-dependent scheduling at all. On the contrary, it is better to save in feedback and blindly schedule for maximal frequency diversity. This is doable by allocating PRBs with large frequency spacing according to the Flexible-FDM scheme or to frequency shift (hop) PRB allocations between consequent transmission instances. The latter mechanism applies for the uplink as well. Further in the uplink, the power control setpoint clearly impacts the cell edge to cell mean ratio, as shown in Fig. 8(c) for the large intersite

RINNE et al.: PERFORMANCE OF THE EVOLVED 3G (E-UTRA) FOR VOICE OVER INTERNET AND BE TRAFFIC

3671

Fig. 7. Uplink 1 2 and 1 4 results with RR and PF channel-dependent schedulers. (a) Cell throughput. (b) Cell throughput normalized to the geometric mean. (c) Modulation and coding statistics. TABLE VII UPLINK SPECTRAL EFFICIENCY AT THE MEAN AND AT THE CELL EDGE FOR VARIOUS MULTIANTENNA CONFIGURATIONS, FDE/MRC RECEIVERS, AND RR/PF SCHEDULERS

these studies, the performance-versus-overhead tradeoff was not yet optimized for different velocities, and neither was the ATB exploited to the full extent. These means have further potential to improve the results, particularly for the mixedvelocity scenario. VI. S UMMARY In this paper, an update on the performance evaluation of E-UTRA transmission technology has been given for the VoIP and BE trafc scenarios. The results for E-UTRA were shown to exceed the targets, which were set in [3] relative to the highspeed packet-access (HSPA) reference in [32]. Since the standard specication work for E-UTRA is near completion for the rst approved version of the standard, it was necessary to verify that the high performance expected in [3], [28][30], and [32] has not been compromised along the process but, on the contrary has been maintained or even improved. The high performance is evidently shown by the extensive simulation studies reported in this paper. However, further algorithmic improvements are foreseen, but they need not have an impact to the standard specication. The signaling overhead remains a critical issue, and that shall be kept minimized. The efciency, coverage, and capacity of the signaling channels and their dynamic formats can be studied in further detail.

distance. Setting P0 larger will cause a larger IoT (the mean IoT was about 6 dB), which allows an increase of the cell mean efciency with the expense of the cell edge. However, in the case of larger cells, decreasing the power control setpoint is well motivated to generate lower IoT (the mean IoT was about 2.2 dB), which signicantly increased the cell edge efciency with only a moderate impact to the cell mean. In the overall mixed-velocity results, the dominance of 3-km/h links kept the loss of spectral efciency small (from 2.6% to 9%) in downlink. In the uplink, the loss was larger (around 30%) due to the nonoptimal sounding feedback. During

3672

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 7, SEPTEMBER 2009

Fig. 8. Mixed-velocity results with a large intersite distance. (a) User goodput in downlink with 1 2 IRC receiver. (b) Cell throughput in uplink with 1 2 and 1 4 receivers. (c) Uplink spectral efciency with the cell edge to the cell mean ratio varied as a function of the power control setpoint.

ACKNOWLEDGMENT The authors at Nokia and Nokia Siemens Networks would like to acknowledge 3GPP partners for creating technology proposals to the standard specications. R EFERENCES
[1] Evolved Universal Terrestrial Radio Access (E-UTRA), LTE Physical LayerGeneral Description, Dec. 2008. 3GPP TS 36.201, TSG-RAN, Rel. 8. [2] Evolved Universal Terrestrial Radio Access (E-UTRA), Physical Channels and Modulation, Dec. 2008. 3GPP TS 36.211, TSG-RAN, Rel. 8. [3] Requirements for Evolved UTRA (E-UTRA) and Evolved UTRAN (E-UTRAN), Mar. 2006. 3GPP TR 25.913, TSG-RAN, Rel. 7. [4] R1-070674 LTE Physical Layer Framework for Performance Verication, Orange, China Mobile, KPN, NTT DoCoMo, Sprint, T-Mobile, Vodafone, Telecom Italia, Feb. 1216, 2007. 3GPP TSG-RAN WG1 Meeting #48. [5] IEEE802.16 Broadband Wireless Access Working Group, IEEE 802.16 m Evaluation Methodology Document, Oct. 2, 2008. [6] R. Irmer, G. Liu, S. Xiadong, J. Krmer, S. Abeta, T. Slzer, E. Jacks, A. Buldorini, and G. Wannemacher, Next Generation Mobile Networks (NGMN) radio access performance evaluation methodology, A White Paper by the NGMN Alliance, Jan. 31, 2008. [7] B. E. Priyanto, C. Rom, C. Navarro, T. B. Soerensen, P. Mogensen, and O. K. Jensen, Effect of phase noise on spectral efciency for UTRA long term evolution, in Proc. IEEE Int. Symp. PIMRC, Helsinki, Finland, Sep. 1114, 2006, pp. 15.

[8] T. Lunttila, J. Lindholm, K. Pajukoski, E. Tiirola, and A. Toskala, EUTRAN uplink performance, in Proc. 2nd Int. Symp. Wireless Pervasive Comput., San Juan, Puerto Rico, Feb. 57, 2007. [9] Physical Layer Aspects of UTRA High Speed Downlink Packet Access, Mar. 2001. 3GPP TR25.848, TSG-RAN, Rel. 4. [10] Feasibility Study for Enhanced Uplink for UTRA FDD, Mar. 2004. 3GPP TR 25.896, TSG-RAN, Rel. 6. [11] Evolved Universal Terrestrial Radio Access (E-UTRA); Physical Layer Procedures, Dec. 2008. 3GPP TS 36.213, TSG-RAN, Rel. 8. [12] X. Weimin, R. Ratasuk, A. Ghosh, R. Love, Y. Sun, and R. Nory, Uplink power control, interference coordination and resource allocation for 3GPP E-UTRA, in Proc. IEEE VTCFall, Montreal, QC, Canada, Sep. 2528, 2006, pp. 15. [13] I. Kovacs, K. I. Pedersen, T. E. Kolding, A. Pokhariyal, and M. Kuusela, Effects of non-ideal channel feedback on dual-stream MIMO-OFDMA system performance, in Proc. IEEE VTCFall, Baltimore, MD, Sep. 30Oct. 30, 2007, pp. 18521856. [14] H. Schulzrinne, S. Casner, R. Frederick, and V. Jacobson, A Transport Protocol for Real-Time Applications (RTP), Jan. 1996, IETF. RFC 1889, Proposed Standard. [15] J. Postel, User Datagram Protocol (UDP), Aug. 1980, IETF. RFC 0768, USC/Inf. Sci. Inst. [16] J. Postel, Internet Protocol (IP), Sep. 1981, IETF. RFC 0791, USC/Inf. Sci. Inst. [17] S. Deering and R. Hinden, Internet Protocol, Version 6 (IPv6) Specication, Dec. 1998, IETF. RFC 2460, USC/Inf. Sci. Inst. [18] C. Burmeister, M. Degermark, H. Fukushima, H. Hannu, L.-E. Jonsson, R. Hakenberg, T. Koren, K. Le, Z. Liu, A. Martensson, A. Miyazaki, K. Svanbro, T. Wiebke, T. Yoshimura, and H. Zheng, Robust Header Compression (ROHC), C. Bormann, Ed., Jul. 2001, IETF. RFC3095.

RINNE et al.: PERFORMANCE OF THE EVOLVED 3G (E-UTRA) FOR VOICE OVER INTERNET AND BE TRAFFIC

3673

[19] Scheduling of LTE DL VoIP, Jan. 1519, 2007. R2-070006, Nokia, 3GPP TSG-RAN WG2 Meeting #56bis. [20] Uplink Scheduling for VoIP, Feb. 1216, 2007. R2-070476, Nokia, 3GPP TSG-RAN WG2 Meeting #57. [21] D. Jiang, H. Wang, E. Malkamki, and E. Tuomaala, Principle and performance of semi-persistent scheduling for VoIP in LTE system, in Proc. Int. Conf. Wireless Commun., Netw. Mobile Comput. WiCom, Shanghai, China, Sep. 2125, 2007, pp. 28612864. [22] C. Wengerter, J. Ohlhorst, and A. G. E. von Albwart, Fairness and throughput analysis for generalized proportional fair frequency scheduling in OFDMA, in Proc. IEEE VTCSpring, Stockholm, Sweden, May 30Jun. 1, 2005, vol. 3, pp. 19031907. [23] J. Lim, H. G. Myung, and D. J. Goodman, Proportional fair scheduling of uplink single carrier FDMA systems, in Proc. IEEE Int. Symp. PIMRC, Helsinki, Finland, Sep. 2006, pp. 16. [24] K. I. Pedersen, G. Monghal, I. Kovacs, T. E. Kolding, A. Pokharial, F. Frederiksen, and P. Mogensen, Frequency domain scheduling for OFDMA with limited and noisy channel feedback, in Proc. IEEE VTCFall, Baltimore, MD, Sep. 30Oct. 3, 2007, pp. 17921796. [25] R. W. Heath, Jr., S. Sandhu, and A. Paulraj, Antenna selection for spatial multiplexing systems with linear receivers, IEEE Commun. Lett., vol. 5, no. 4, pp. 142144, Apr. 2001. [26] K. Brueninghaus, D. Astley, T. Slzer, S. Visuri, A. Alexiou, S. Karger, and G.-A. Seraji, Link performance models for system level simulations of broadband radio access systems, in Proc. IEEE Int. Symp. PIMRC, Berlin, Germany, Sep. 1114, 2005, vol. 4, pp. 23062311. [27] H. Holma, M. Kuusela, E. Malkamki, K. Ranta-aho, and C. Tao, VoIP over HSPA with 3GPP release 7, in Proc. IEEE Int. Symp. PIMRC, Helsinki, Finland, Sep. 1114, 2006, pp. 15. [28] E. Dahlman, H. Ekstrom, A. Furuskar, Y. Jading, J. Karlsson, M. Lundevall, and S. Parkvall, The 3G long-term evolutionRadio interface concepts and performance evaluation, in Proc. IEEE VTCSpring, May 710, 2006, vol. 1, pp. 137141. [29] N. Wei, A. Pokhariyal, C. Rom, B. E. Priyanto, F. Frederiksen, C. Rosa, T. B. Soerensen, T. E. Kolding, and P. E. Mogensen, Baseline E-UTRA downlink spectral efciency evaluation, in Proc. IEEE VTCFall, Montreal, QC, Canada, Sep. 2528, 2006, pp. 15. [30] Y. Ofuji, T. Kawamura, T. Y. Kishiyama, K. Higuchi, and M. Sawahashi, System-level throughput evaluations in evolved UTRA, in Proc. IEEE ICCS, Singapore, Oct. 2006, pp. 16. [31] LTE Performance Benchmarking, Apr. 2324, 2007. R1-071992, Nokia and Nokia Siemens Networks, 3GPP TSG-RAN WG1. [32] LS on LTE Performance Verication Work, May 711, 2007, Kobe, Japan. R1-072580, GPP TSG-RAN WG1 Meeting #49. [33] LTE BS Dynamic Range, Nov. 59, 2007. R4-071870, 3GPP TSG-RAN4, NTT DoCoMo.

Esa Tuomaala received the M.Sc.(Tech) degree (with honors) from Helsinki University of Technology, Espoo, Finland, in 2002. He joined Nokia in 2000 and is currently a Project Manager with Nokia Devices R&D, Beijing, China. His research focuses on the performance analysis of wireless communication systems.

Pasi Kinnunen received the M.Sc. degree from the University of Oulu in 1993. He is currently with Nokia Siemens Networks, Oulu, Finland, where he is working on thirdgeneration radio standards and their evolution. His current research interests include both radio link and system performance topics.

Istvn Kovcs received the B.Sc. degree from Politehnica Technical University of Timisoara, Timisoara, Romania, in 1989, the M.Sc.E.E. degree from The Franco-Polish School of New Information and Communication Technologies, Poznan, Poland, and cole Nationale Suprieure des Tlcommunications de Bretagne, Brest, France, in 1996, and the Ph.D.E.E. degree in wireless communications from Aalborg University, Aalborg, Denmark, in 2002. From 2002 to August 2005, he was an Assistant Research Professor with the Antennas and Propagation Division, Center For TeleInFrastruktur, Department of Communication Technology, Aalborg University. He is currently a Wireless Networks Specialist with Nokia Siemens Networks, Aalborg, where he conducts system-level studies and multiple-inputmultiple-output radio resource management research for UTRAN long term evolution (LTE), LTE-A, and ITU IMT-A systems. He has contributed to a number of publications as well as to three book chapters on ultrawideband propagation topics.

Mika Rinne received the M.Sc. degree in signal processing and computer science from Tampere University of Technology, Tampere, Finland, in 1989. He is currently a Principal Scientist with the Wireless Systems and Services Laboratory, Nokia Research Center, Helsinki, Finland. His background is in the research of multiple-access methods, radioresource management, and implementation of packet decoders for radio communications. His current interests are in the systems research of protocols and algorithms for wireless communications including third-generation (3G) WCDMA/HSPA, beyond 3G, long-term evolution (LTE) (E-UTRAN), and LTE-Advanced/IMT-Advanced.

Kari Pajukoski received the B.S.E.E. degree from the Oulu University of Applied Sciences in 1992. He is currently with Nokia Siemens Networks, Oulu, Finland, where he is working on various topics related to third-generation radio standards and their evolution. His current research interests include radio performance and transceiver algorithms.

Markku Kuusela received the M.Sc. degree in mathematics from the University of Turku, Turku, Finland, in 1998. He currently is a Senior Research Specialist with the Nokia Devices Research and Development Unit, Helsinki, Finland. He has contributed to three book chapters on voice over internet protocol on 3GPP high-speed packet access (HSPA)/long-term evolution (LTE). He has worked on the physicallayer performance of different transmit diversity and mulitple-inputmultiple-output concepts for 3GPP wideband code division multiple access and LTE. During the past few years, he has focused on physical-layer optimization mechanisms for VoIP on HSPA and third-generation LTE.

Jussi Ojala received M.Sc. degree in mathematics from the University of Helsinki, Helsinki, Finland, in 1999. From 1999 to 2001, he was with the University of Helsinki, mainly in cooperation projects with Nokia Research Center. He is currently with the Nokia Devices Research and Development Unit, Helsinki. He has worked on channel modeling, spectrum requirements, physical- and network-layer performance analysis, and various L1 and L2 research areas of variable radio access techniques. Over the past few years, he has focused on L1 and L2 research/standardization of third-generation long term evolution.

You might also like