Performance Evaluation on Dual-Cell HSDPA Operation

Danielle Morais de Andrade, Axel Klein, Harri Holma

Research, Technology and Platforms Department Nokia Siemens Networks Munich, Germany and Espoo, Finland danielle.andrade, axel.klein, harri.holma@nsn.com
Abstract This paper presents a quantitative performance analysis of the Dual-Cell High-Speed Downlink Packet Access (DC-HSDPA) operation focused on the gain achieved by pooling the radio resources of two carriers in the same base station and enabling a collaborative operation (joint-scheduling) on the lower radio layers. Keywords - Dual-Cell; HSDPA; joint-scheduler; multi-user diversity; frequency selectivity
U E1 1 x 5 M Hz U E2 2 x 5 MH z

Ingo Viering, Gnther Liebl

Nomor Research GmbH Munich, Germany viering, liebl@nomor.de

U plink 1 x 5 MH z

Dow nlink 2 x 5 MH z

I.

INTRODUCTION

One main target for the evolution of 3G mobile communication is to provide the possibility of significantly higher end-user data rates compared to what is achieved with the first releases of the 3G standards. This also refers to higher data rates over the entire cell area including users at the cell edge. 3GPP standards body has significantly enhanced the peak user throughput as part of Release-7 with features as MIMO and Higher Order Modulation (HOM) [1] and this has helped to improve the average user throughput to some extent. However, the rapid increase in mobile data usage calls for additional enablers to provide enhanced user experience throughout the cell, especially in the outer area of the cell coverage. One approach to increase the typical user experience consists in pooling the radio resources of two or more carriers in the same base station and enabling a collaborative operation on the lower radio layers (i.e. L2 layer) for a better resource utilization efficiency by dynamic radio resource management over multiple carriers. With corresponding enhancements of the terminal capabilities this would also allow increased data rates for users in all coverage conditions by receiving transmissions on multiple carriers simultaneously. Within 3GPP such operation has been investigated under the work item Dual-Cell HSDPA operation on adjacent carriers (hereafter DC-HSDPA) [2]. As the name suggests, the work item restricts the pooling of radio resources to two carriers and only for the downlink direction remaining the UL with the possibility of only one carrier utilization. The goal is a feasibility study for the new feature, including the quantification of enhancements to user throughput throughout the cell and the system impacts of introducing such feature to the existing UTRA system.

Fig. 1: 3GPP Work Item DC-HSPDA operation principle

With its potential to increase peak data rates, the feature DC-HSDPA is regarded as an alternative to MIMO without the cost and complexity of multi-antennas deployment (provided that the spectrum is available), and is claimed to achieve better performance gains particularly in unfavorable channel conditions, e.g. at the cell edge. This paper presents a quantitative performance analysis of the DC-HSDPA operation (based on the above described 3GPP assumptions), demonstrating the gains of implementing a collaborative scheduling mechanism (joint scheduler) for the two carriers. In section II the basic concepts and expected benefits of the feature are described and discussed. Section III introduces the simulation model used in the investigations, with its major assumptions, parameter settings and in particular, the implemented joint scheduling scheme. Section IV presents and reviews various selected performance results highlighting the advantages of DC-HSDPA in different conditions, including a comparison with the MIMO feature. Finally, section V summarizes the findings and concludes with an outlook on remaining open issues. II. DC-HSDPA CONCEPT

DC-HSDPA operation has the purpose of enhancing the user experience throughout the whole cell range, in particular in outer area of the cell coverage (at the cell edge where MIMO can not be operated with dual stream transmission), by pooling the radio resources of two carriers in the same base station and enabling a collaborative operation (joint-scheduler) on the lower radio layers. It aims to achieve double peak rates and more than doubled sustained data rates (compared to a single

978-1-4244-2515-0/09/$25.00 2009 IEEE

carrier deployment) without the cost of dual antenna deployment. In terms of system performance, DC-HSDPA operation enables efficient and flexible spectrum asset utilization offering efficient inherent load balancing across the carriers. Furthermore, the feature is a natural and smooth evolution of HSPA [3] in terms of radio equipment and spectrum. The carrier aggregation is backward compatible with legacy devices and it requires simple network upgrade. If a second carrier is available to the operator, it is highly probable that it would first be utilized for capacity enhancement by static load distribution (two independent carriers or two times single carrier, hereafter 2xSC) before introducing enhancement feature like collaborative dual carrier operation that requires new terminal capabilities. Therefore, the simulations carried out in this paper are regarded as a fair performance comparison of dual carrier (DC) over 2xSC cell configuration. The main sources for performance gains from the collaborative dual carrier over the 2xSC approach are: Frequency selectivity which depends on carrier allocation. It is most pronounced for carriers in different bands, giving the possibility to select the better of the two carriers at every transmission opportunity. Double peak rate by assigning all resources on both carriers to the same user. Note that increased peak rates require carrier aggregation and therefore are mutually exclusive with frequency selectivity Multi-user diversity gain [4], through more users to select from in time scheduling. Statistical multiplexing (or trunking gain), meaning better resource utilization efficiency of fast dynamic over semi-static load balancing (not treated in this presented study).

For dual carrier-capable terminals, it shall utilize the available resources of both carriers in every Transmission Time Interval (TTI), prefer the carrier with the higher Channel Quality Information (CQI) in case of single carrier allocation, and allow for dual carrier allocation whenever this brings a benefit. For legacy terminals (non-dual carrier capable), there shall be an operation mode with independent (i.e. noncooperative) scheduling on the two carriers, applying the already available single carrier scheduler schemes on statically allocated subsets of users for each carrier. Mixed configurations of legacy and dual carrier terminals shall be handled in an integrated way. III. SIMULATION ASSUMPTIONS

A. System-Level Simulation Model AMoRE (Advanced Mobile Radio Real-time Experience) [6] is a flexible and accurate simulation environment that can be used for testing and demonstrating the performance of IPbased applications over emulated radio networks (like WCDMA) providing data services. Being mainly targeted for the demonstration and observation of radio behavior at runtime through a powerful and flexible graphical user interface, the semi-static system-level simulator in the core of the tool also presents an efficient means for the systematic generation of performance statistics. The system-level model simulates a hexagonal 3-sectorized cell layout with multiple users in one target cell and synthetically generated interference from multiple rings of surrounding cells (which in the HSDPA model are assumed to transmit continuously with full power). Data channel transmissions are simulated in full detail with fast link-adaptive scheduling and resource allocation, error model based on linklevel lookup tables and Hybrid Automatic Repeat Request (HARQ) retransmission protocol. The control channel model is simplified with fixed power allocation and error-free transmission, yet realistic constraints in terms of signaling accuracy and delays. Statistical performance evaluations are taken without user mobility over a multitude of snapshots where users are randomly dropped into the target cell and the measured performance values are averaged over the duration of each drop. Channel fading and intracell interference is modeled according to [7]; the MMSE and MIMO extension is described in [8]. Throughput and delay is measured on PDCP layer (i.e. without L2 protocol overheads and after HARQ retransmissions and RLC reassembly). B. Simulation Setup The simulations presented in this paper were performed under the following assumptions: The inter-site distance in the cellular topology was fixed at 1500 m, the total transmission power per cell was 20W, of which 16W were available for HSDPA transmissions. Pedestrian channel models of type PedA and PedB were used, both for terminal velocities of 3km/h. The carrier bandwidth is 5MHz with both carriers

These are directly related to the resource allocation and coordination of the two carriers making the scheduling the most important part of the DC-HSDPA feature implementation, which determines the gain that can be achieved by collaborative over independent (2xSC) dual carrier configurations (joint and disjoint scheduler, respectively). The joint scheduler can be regarded as similar to dual stream scheduling in MIMO operation, yet is simpler (due to the orthogonal channels) and less constrained: the carriers can be allocated to the same or different users, which would correspond to multi-user (MU) MIMO scheduling [5]. The selection of specific scheduling schemes is up to the vendors, and its precise implementation is still for further study. However, the following requirements are reasonable: The joint scheduler shall be backward compatible to the already implemented scheduler for single carrier operation while fulfilling the requirements of joint scheduling for DC-HSDPA operation terminals.

located in the 2GHz band. The terminals are modeled with advanced receivers of type 3 (i.e. with 2Rx antennas and equalizing receiver). Terminals are assumed to have either 64QAM or MIMO capability (i.e HSDPA category 14 or 16). Dual Carrier capable terminals do not support MIMO, but 64QAM. The base stations always transmit with one antenna, except for single carrier transmissions to MIMO terminals. The fading correlation between the two carriers in the frequency domain was not modeled in detail instead independent fading processes were applied providing an upper bound for the frequency selectivity gain. It should be noted that, at least for the more dispersive PedB channel, independent fading is a realistic assumption. The evaluation section typically compares Dual Carrier configurations (where all terminals are DC-capable and jointly scheduled) with Single Carrier configurations of two independently operated carriers (2xSC) where half of the total number of terminals is statically assigned on each of the carriers, corresponding to a perfect load balance. The default traffic model for all terminals is full buffer, i.e. all users can at any time be assigned any data rate subject only to scheduler decision. For the delay statistics a streaming-like traffic model was used with a constant bit rate (cbr) of 256kbps. Statistical multiplexing gain was not treated in this study. C. Joint-Scheduling Scheme The joint scheduling model for DC-capable terminals assumes common Tx buffers and common scheduler and HARQ entities for both carriers in the transmitting base station. That means each packet provided from higher layers into the Tx buffer can be transmitted on any carrier, and also HARQ retransmissions can be allocated on the same or a different carrier as the original transmission. The common scheduler entity gets, for each user, periodic channel quality information (from the so-called CQI reports) on both carriers, computes a metric function independently for each carrier according to the selected single carrier scheduling scheme and assigns the resources on each carrier to the user with the highest metric value. This can be the same or different users on the two carriers. In the simulations for this paper only the Proportional Fair (PF) metric was applied, but our generic joint scheduler scheme also works with any other single carrier metric such as Round Robin, Maximum C/I or even QoS-aware metrics taking buffer status and waiting time into account [9]. The joint PF scheduler used in the investigations applies a metric function, in which the instantaneous throughput in the enumerator depends on the carrier-specific CQI whereas the average past scheduled throughput term in the denominator is common for both carriers. The modeled scheme is a straight-forward extension of already existing single carrier schedulers, which leads to a flexible utilization of the available resources on both carriers, exploits frequency selectivity (as far as CQI-based metrics are applied) and allows for the allocation of both carriers to the same user (carrier aggregation) whenever this brings a benefit.

IV.

PERFORMANCE EVALUATION

A. Capacity and Throughput Increase Simply doubling the bandwidth, i.e. going from single carrier to 2xSC configuration, will also double the average total cell throughput. Taking advantage of the joint scheduling, however, the cell capacity will additionally increase over systems where the carriers are used independently. This can be observed in Fig. 2 where the DC and 2xSC configurations are compared for different channel models.
Capacity Gain from Joint-Scheduler (8 users)
1 0.9 0.8 0.7 0.6

cdf

0.5 0.4 0.3 0.2 0.1 0 PedA, PedA, PedB, PedB, 0 5 10 15 DC; mean = 12.71 2xSC; mean = 10.42 DC; mean = 9.48 2xSC; mean = 8.09 20 25

DL cell throughput (Mbps)

Fig. 2: Cell Throughput: Capacity Gain from Joint-Scheduler (10MHz)

The benefit of joint scheduling comes from pooling all users on both carriers, leading to increased multi-user diversity, and from giving each user scheduling opportunities on either of two carriers, thus exploiting frequency diversity. In case of burst traffic with the load varying over time (not considered here), the pooling of users will also provide a statistical multiplexing gain by balancing the load between the two carriers at any time. In contrast to the diversity gains, this effect does not depend on channel properties.
Frequency Selectivity and Multi-User Diversity Gain for Full Buffer Traffic

12

mean DL cell throughput (Mbps)

10

PedA, PedA, PedB, PedB, PedA, PedB, 2 4 6 8 10 12 14

DC 2xSC DC 2xSC DC identical fading DC identical fading 18 20

16

number of users

Fig. 3: Frequency selectivity and Multi-user diversity Gain

95th percentile of DL packet delay (s)

Fig. 3 shows the improvement of the mean cell throughput, over different number of users, due to frequency selectivity and multi-user diversity. In order to separate the two effects, a reference curve for the DC case has been added that was generated with identical fading processes on both carriers, thus eliminating the frequency diversity effect. The observed multiuser diversity gain is more pronounced at low loads it vanishes with increasing number of users, while the gain from frequency selectivity remains and presents the majority of the overall DC gain in this scenario. With 2 users per sector in a PedA channel, the gain is around 24%, for 20 users per sector, it is around 19%. For PedB channel the gain is about 20% and 16% for 2 and 20 users per sector, respectively. It must be noted that for the less dispersive PedA channel our assumption of independent fading processes may be too optimistic (for adjacent carriers) and thus somewhat exaggerates the frequency diversity gain. The benefits of HOM (64QAM modulation) and MIMO are mainly seen in high geometry (good coverage) and high SNR. In contrast, DC-HSDPA transmissions can achieve significantly higher data rates for most users, including those near the cell edge experiencing only low or moderate SNR. This increases the number of users having access to improved data rates, as can be viewed in the table on Fig. 4.
Throughput Gain over entire Cell Range (8 users)
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

as the decreasing amount of assigned resources (by the resource-fair PF scheduler) is no more sufficient to maintain the fixed nominal bit rate.
Outage Capacity for Streaming Traffic (cbr) at 256 kbps
5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 PedA, PedA, PedB, PedB, DC 2xSC DC 2xSC

10

12

14

16

18

20

number of users

Fig. 5: DL Packet Delay: Outage Capacity for Streaming Traffic at 256 kbps

Fig. 5 shows this effect on the 95th percentile of the measured packet delays. Assuming a tolerable delay of 1s for the streaming service, the intersection of the delay curves with the 1s threshold line marks the outage limit where 5% of the simulated cases exceed that threshold, thus defining the delaybased outage capacity. In the example of the PedB channel, this outage capacity increases from slightly over 11 to 14 users corresponding to a gain of about 25%. C. DC vs. MIMO Comparison MIMO technique can be introduced in a HSPA network to increase the DL peak data rate up to 28Mbps; combined with HOM it can reach up to 42 Mbps [10]. Implementing DCHSDPA (64-QAM capable), through carrier aggregation the same 42Mbps can be reached with twice the bandwidth, yet without the extra cost of 2 Tx chains in the NodeB.

Percentile 10th 50th 90th

PedA, 2xSC (Mbps) 0,374 0,950 3,114

PedA, DC (Mbps) 0,480 1,219 3,672

PedA, PedA, PedB, PedB,

DC Gain (%) 28 28 18

cdf

DC; mean = 1.64 2xSC; mean = 1.35 DC; mean = 1.23 2xSC; mean = 1.04 4 4.5 5

0.5

1.5

2.5

3.5

DL user throughput (Mbps)

Fig. 4: DL User Throughput: Throughput Gain over entire Cell Range

Compared to 2xSC, DC has a throughput gain of 28%, also at cell edge (represented by the 10th percentile). At extremely good coverage, i.e. 90th percentile, the gain is somewhat lower only at 18%. This is because the throughput is here limited by the modulation and coding scheme and the DC configuration is meeting its hard upper bound. B. Delay Impacts As delay measurements are not meaningful with full buffer traffic, this investigation has been performed with constant bit rate (cbr) traffic of 256kbps, which can be regarded as an example for a video streaming service. In this scenario the system load increases with the number of users, and at some point the cell edge users start suffering from increasing delays,

In practice the achieved net data rates are much lower, considering the protocol overheads, HARQ retransmissions and the fading variations that often prevent the selection of the highest Modulation and Coding Scheme (MCS) necessary for the peak rate. Fig. 6 shows a snapshot view from the AMoRE simulation, where 4 users with different capabilities and in different locations are successively scheduled (from full buffers). The red and blue users close to the base station achieve sustained maximum rates of about 25 and 15 Mbps for the DC and MIMO users, respectively. The observed higher gain (compared with the theoretical 50% peak rate improvement for DC with 64QAM over MIMO without 64QAM) is due to the better separation between carriers than between spatial MIMO streams. About the same relative DC gain is also achieved for users near the cell edge, but for different reasons: Here, neither dual stream transmission nor 64QAM modulation can be applied. Although the carrier aggregation for the DC user can not always be applied due to occasional deep fading dips, better cell edge performance from DC over MIMO is observed.

V.

CONCLUSION AND OUTLOOK

In this paper a quantitative performance analysis of DCHSDPA operation was presented, giving an overview of the feature operation with its benefits and expected performance impacts in dependence of channel, load and traffic conditions. The results identified and distinguished DC-HSDPA gains from different sources: carrier aggregation, pooling gains (multi-user diversity) and frequency selectivity, which apply in and depend on different operating conditions (such as load level, traffic type and channel conditions). Carrier aggregation allows obtaining twice the peak rate for a single user than on a single carrier. In contrast to dual stream MIMO, a doubling of the user rate can be achieved in all channel conditions (i.e. even at the cell edge) and without the expenses of dual Tx antennas and power amplifiers. Joint scheduling over two carriers provides a pooling gain (over independently scheduled carriers) through multi-user diversity when link-adaptive scheduling is applied. Even more significant (particularly in cases of higher system load) are the frequency selectivity gains obtained from DC operation, when the channel fading on the two carriers is sufficiently decorrelated (as is the case in dispersive channels or for larger carrier distance. The reported investigations did not cover another important benefit from joint DC scheduling: the trunking gain from statistical multiplexing that applies for limited buffer or random burst traffic by improving the resource utilization efficiency through fast dynamic instead of (semi)-static load balancing. This will be an issue for further investigations. REFERENCES
[1] H. Holma, A. Toskala, K. Ranta-aho, J. Pirskanen, High-Speed Packet Access Evolution in 3GPP Release 7, in IEEE Communications Magazine, vol. 45, Dez. 2007, pp. 2935. [2] 3GPP, Feasibility study on Dual-Cell HSDPA operation - Work item description, 3GPP, RAN Plenary RP-080228, Mar. 2008. [3] H. Holma, A. Toskala, WCDMA for UMTS. England: Wiley, 2004, pp. 307345. [4] H. Holma, A. Toskala, HSDPA/HSUPA for UMTS. England: Wiley, 2006, pp. 137140. [5] P. Ting, J. Chen, C. Wen, J. Chen, Efficient multiuser MIMO scheduling strategies, in Proc. 60th IEEE Vehicular Technology Conference, Los Angeles, 2004, pp. 11391142. [6] I. Viering, C. Buchner, E. Seidel, A. Klein, Real-time network simulation of 3GPP Long Term Evolution, in Proc. IEEE Symposium on a World of Wireless, Mobile and Multimedia Networks 2007, Helsinki, Finland , 2007, pp. 13 [7] A. Seeger, M. Sikora, A. Klein, Variable orthogonality factor: a simple interface between link and system level simulation for High Speed Downlink Packet Access, IEEE Vehicular Technology Conference 2003 (VTC-Fall'03), Orlando, USA, 2003. [8] M. Wrulich, S. Eder, I. Viering, M. Rupp, "System level modeling of D-TxAA MIMO HSDPA, IEEE Globecom 2008 Wireless Communications Symposium, New Orleans, LA, USA, November, 2008. [9] E. Dahlman, S. Parkvall, J. Skold, 3G Evolution: HSPA and LTE for Mobile Broadband, UK: Elsevier, 2007, pp. 109-120. [10] J. Bergman, M. Ericson, D. Gerstenberger, B. Gransson, J. Peisa, S. Wager. (2008). HSPA Evolution - Boosting the performance of mobile broadband access. Ericsson Review. No. 1, pp. 32-37.

Fig. 6: AMoRE snapshot of comparative DC vs. MIMO Performance

Fig. 7 compares the cdf of the total cell throughput for a MIMO scenario with 8 users on two single carriers with nonMIMO scenarios in 2xSC or DC configuration. It can be seen that the DC gain (from joint scheduling) is slightly better than the MIMO gain (from Tx Diversity and occasional dual stream transmission) in all situations. Even though the assumption of independent fading may be too optimistic for adjacent carriers and the PedA channel, it can be concluded that the DC feature improves the cell capacity in the same order as the MIMO feature, and additionally allows much higher throughput for single users in all channel conditions.
Comparative DC vs. MIMO Performance (PedA, 8 users)
1 0.9 0.8 0.7 0.6

cdf

0.5 0.4 0.3 0.2 0.1 0 DC; mean = 12.71 2xSC; mean = 10.42 2xSC MIMO; mean = 12.27 0 5 10 15 20 25

DL cell throughput (Mbps)

Fig. 7: DL Cell Throughput: Comparative DC vs. MIMO Performance

This benefit comes at the expense of double bandwidth, so the DC gain in spectral efficiency is in fact smaller than with MIMO. As, however, a second carrier and the capability for 2*SC operation is already present in most deployed 3G networks, this makes upgrading to DC much less costly than a MIMO deployment.