Performance of Multiflow Aggregation Scheme for HSDPA with Joint Intra-Site Scheduling and in Presence of CQI Imperfections Dmitry Petrov, Ilmari Repo and Marko Lampinen 1 Magister Solutions Ltd., Jyvaskyla, Finland, 2 Renesas Mobile Europe, Oulu, Finland {dmitry.petrov,ilmari.repo}@magister.fi {marko.lampinen}@renesasmobile.com Abstract. Lately several Single Frequency Dual Cell (SF-DC) techniques for High Speed Downlink Packet Access (HSDPA) were considered in the Third Generation Partnership Project (3GPP), which were aimed to enhance network performance for users at the cell edges. In this paper we are concentrating on intra-site multiflow aggregation scheme, where User Equipment (UE) can receive two independent transmissions from separate sectors. Furthermore, this paper focuses on the impact of Channel Quality Indicator (CQI) reporting interval and measurement imperfections on the system performance. Moreover, we present a joint scheduling algorithm, which selects users site-wide instead of independently in each sector. Simulation results for scenarios mentioned above are presented and analysed in the paper. This study shows that joint scheduling gives some extra gain compared to independent scheduling. In addition, impact of CQI errors and CQI update period are studied. Keywords: HSDPA, SF-DC, 3GPP, multiflow, scheduling, CQI 1 Introduction Currently High Speed Packet Access (HSPA) is the leading mobile broadband technology globally with 451 networks commercially launched in 174 countries. Even though Long Term Evolution (LTE) networks are rapidly developing (already 49 commercial networks worldwide) the Wideband Code Division Multiple Access (WCDMA) networks still show constant subscriptions growth about 40% per year. Over 41% of commercial HSPA networks have already launched Evolved HSPA, also called HSPA+ [1]. According to the latest issued 3GPP Release 10, achievable peak data rates in downlink direction (HSDPA) are up to 168 Mbps. Deployment of LTE networks will still require time and considerable capital expenses from operators. At the same time already existing HSPA networks will need to accommodate huge mobile traffic growth. For that reason new techniques that can enhance 3G networks performance are constantly examined in the 3GPP.
2 Multiflow HSDPA with Joint Scheduling and CQI Imperfections Good examples of such enhancements is Dual-Cell HSDPA (DC-HSDPA) feature first defined in 3GPP Release 8 [2]. It provided considerably better user experience in two geographically overlapping cells in deployments when several 5 MHz paired spectrum allocations were available to an operator. Performance was improved especially at the cell edges where channel conditions were not favourable and existing techniques such as Multiple Input and Multiple Output (MIMO) could not be used. At the same time, it is rather usual situation that only one 5 MHz frequency band is assigned to an operator in the specified geographical region. Thus multicarrier multi-cell approach cannot be utilized. Nevertheless reception quality at the cell edges may degrade considerably. Multipoint HSDPA transmission, i.e. transmission from two different cells to a UE situated in Soft (SHO) or Softer Handover (SofterHO) region can be used to evolve system performance in Single Frequency (SF) deployments. Similar approach was firstly introduced for LTE Release 8 where it was called Coordinated Multipoint (CoMP) [3]. Adaptation of this technology into the HSDPA system was proposed as a study item at the 3GPP Radio Access Network (RAN) Technical Specification Group (TSG) meeting in December 2010 [4]. In this paper our own research is focused on so called intra-site aggregation scheme where UEs have two independent receiving chains and are capable to receive data blocks simultaneously and independently from two cells of one site. It is quite probable that this scheme will be included into the coming 3GPP Release 11 planned for the third quarter of 2012. Our aim is to go further than standard 3GPP simulation scenario [5] by evaluating the performance of the HSDPA system under two assumptions: firstly, when the standard Proportional Fair (PF) scheduling algorithm acting independently in every sector is substituted by a joint site-wise PF scheduler; secondly, when CQIs are reported less often and with errors. The study is done by means of quasi-static system-level simulator, in which the most essential physical parameters and radio resource management functions like channel fading, UE scheduling, etc. are modelled in details. The rest of the paper is organized as follows: Section II contains the brief overview of SF multipoint concepts for HSDPA. In the following section the most important parameters and assumptions are introduced together with more detailed description of joint scheduling scheme and CQI imperfections. In Section IV simulation results are presented and analysed. Finally Section V concludes the article. 2 Multipoint Concepts for HSDPA The idea of multipoint transmission is to perform spatial processing over several transmitting antennas from different cells. Two or more independent communication channels with different characteristics can be used to either improve the reliability of the transmission or to increase the capacity. Development of multipoint HSDPA techniques are motivated, firstly, by the problems with the
Multiflow HSDPA with Joint Scheduling and CQI Imperfections 3 Fig. 1. Intra-site multipoint transmission scheme. reception quality at the cell edges, when users are far from the transmitter and being interfered strongly by the neighbouring cell. Secondly, by the load balancing function between the sectors, when UEs can utilize resources from less loaded cells. The possibility of multipoint transmission is determined individually for every UE according to the difference in the received power between the two strongest NodeBs. This procedure is similar to one used to define the HandOver (HO) state ofthe UE. If the receivedpowersofthe two cells arecloseenough, the UE is informed to monitor the High Speed Shared Control Channel (HS-SCCH) on both of them. Transport blocks can be scheduled on High Speed Downlink Shared Channel (HS-DSCH) from secondary or even both sectors (Fig. 1). Two scenarios can be considered: intra-site case when UE falls into SofterHO coverage region and is served by only one NodeB; and inter-site case referred to SHO state when primary and secondary cells belong to different NodeBs. In this publication we study only intra-site scenario for two main reasons. Firstly, inter-site case has more impact of the network and require certain enhancements to Radio Link Control (RLC) and Iub interface flow control [6]. Secondly, practical realisation of proposed joint scheduling algorithm is more complicated when separate NodeBs are involved. Now we will briefly introduce multipoint techniques-candidates, which were considered in the 3GPP for the SF deployments (for more detailed information see [5] and [7]):
4 Multiflow HSDPA with Joint Scheduling and CQI Imperfections High Speed Data-Discontinuous Transmission (HS-DDTx) and Fast cell switching techniques are not really multipoint transmission schemes because they do not assume that UEs can receive data from several cells simultaneously. This is similar to single point data transmission and can be used with UEs having only one receiving antenna chain. UEs can get data blocks from only one cell in each Transmission Time Interval (TTI) whereas cells can be changed dynamically from TTI to TTI. Thus sectors are chosen in compliance with reported CQI values. The main difference between DDTx and Fast cell switching is that in the first scheme it is not required to switch off HS-SCCH reception from the secondary sector. High Speed Single Frequency Network (HS-SFN) technique allows UEs, which are in SofterHO state, to receive data blocks from two sectors simultaneously. Bothofthem transmitexactlythe samedatablockstooneue.in suchaway HS-SFN-capable UEs benefit from extended spatial diversity. Transmission from assisting cell adds additional power and also removes the strongest source of interference. Single Frequency Dual Cell HSDPA (SFDC-HSDPA) also called Multiflow aggregation technique originates from DC-HSDPA feature and gives use to two receiving chains of DC-HSDPA capable UEs in SF scenario. It can be understood as a spatial multiplexing MIMO scheme when transmitting antennas are situated in different sectors. Achievable gains in this scenario are the highest among all multipoint schemes presented above. It is quite probable that it will be adopted in 3GPP Release 11. For that reason we have selected it for evaluation in this paper. 3 Simulation Assumptions and Methodology This study has been performed by means of proprietary comprehensive quasistatic system simulator, which models HSDPA with a slot resolution. It is widely used to support 3GPP standardization work [8, 9]. The simulation tool enables detailed simulation of users in multiple cells with realistic traffic generation, propagation and fading. Quasi-static simulations mean that UEs are stationary but both slow and fast fading are explicitly modeled. Fast fading is modeled as a function of time for each UE according to the International Telecommunication Union (ITU) channel profiles which are modified to chip level sampling [10]. Actual simulation area consists of 19 NodeBs which results into 57 hexagonal cells. Statistics are collected from all cells. Statistical confidence is obtained through running multiple drops, i.e. independent simulation iterations. In each iteration UE locations, fading, imbalance and other random variables are varied. The statistics are gathered and averaged over all drops. Actual Value Interface (AVI) mapping is used for mapping link level Eb/N0 values to frame error rates [11].
Multiflow HSDPA with Joint Scheduling and CQI Imperfections 5 3.1 Scheduling Algorithms Independent Scheduling Basically, SFDC-HSDPA scheduling is considered to be independent. Practically it means that even multiflow capable UE is considered like a combination of two separate receivers connected to the different sectors and placed to one physical point. Nevertheless such UEs have one file buffer divided in two flows. Scheduling decisions are made depending on the whole load profile in the sector. Specifically, every TTI PF scheduler selects user with the highest priority SP = max i (SP i ) = max i ( Ri M ) where R i is the rate i achievable by user i associated to this sector and M i is its average rate [12]. The achievable rate R i is defined directly from the channel quality measurements, i.e. from the CQI. The average rate refers to the PF metric and is calculated as moving average over a one TTI time window: M i (t+1) = (1 α)m i (t)+αr i where α is the forgetting factor. Thus PF scheduling achieves high throughput while maintaining proportional fairness among all users in the sector. Multiflow capable UEs are particular in that sense that there are two Scheduling Priorities (SP) associated with them: SP 10 = R10 M 10 at the primary sector and SP 02 = R02 M 02 at the assisting sector. Moreover, In the multiflow case special prioritization is used. For each cell, two classes of UEs are defined during scheduling [13]: Class A UEs have this cell as serving, i.e. are connected via strongest link; Class B UEs do not have this cell as serving, i.e. are connected via the weak link. Class A UE prioritization means that serving HS-DSCH cell UEs are always scheduled before multiflow UEs connected to this cell as to a secondary one. This rule is important in high load scenarios. In such a way loss in performance of regular UEs belonging to this serving cell is avoided. Retransmissions are also taken into account in the scheduling algorithm. They always have highest priority over regular transmissions. It is achieved by simple adding of big enough positive number to the scheduling priority. Such approach always forces retransmissions to be scheduled first and at the same time keeps the orderof UEs. This rule is fulfilled even in the case when retransmission is needed for Class B UE, while active Class A UE exists in the same sector. Joint Intra-Site Scheduling In this paper we propose and evaluate a different scheduling algorithm based on site-wide user selection. Taking into account that in intra-site scenario scheduling decision are made inside one NodeB we can use PF algorithm to select users with maximum scheduling priorities from the list of all active users of the site. In the case when multiflow is switched off there will be almost no difference with regular independent scheduling algorithm because, again, UEs with the highest SP will be selected in every cell. The situation will change if there are multiflow capable users which can be scheduled simultaneously from both sectors. To take that into account we are considering three SPs for these kind of UEs:
6 Multiflow HSDPA with Joint Scheduling and CQI Imperfections Fig. 2. Scheme of joint intra-site scheduling algorithm. SP 10 = R10 M and SP 02 = R02 M, where average rate M = M 10+M 02 takes into account whole traffic received by the UE; SP 12 = R12 M, where R 12 = R 10 + R 02 is the achievable rate in the case of aggregated transmission. The conceptual scheme of joint scheduling can be seen on Fig. 2. First, all active users from three sectors are collected to one list, which is then sorted according to the scheduling priority value. After that we go through the sorted list and select one UE for every sector taking into account Class A UE prioritization. In joint scheduling the additional gain is expected from the better organized user selection mechanism. It allows to choose users in an optimal way out from three possible modes available for multiflow capable UEs: aggregated transmission, transmission only from primary sector, transmission only from the secondary sector. Simulation results can be found in the next section.
Multiflow HSDPA with Joint Scheduling and CQI Imperfections 7 Feature / Parameters Cell Layout Inter-site distance Carrier Frequency Table 1. Main simulation parameters Description / Value Hexagonal grid, 19 NodeBs, 3 sectors per NodeB with wrap-around. 1000 m 2000 MHz Number of UEs/cell 1, 2, 4, 8, 16, 32 (distributed uniformly across the system). Receiver Type Channel model CQI Traffic Scheduling MP-HSDPA UE capabilities Type 3i PA3; Fading across all pairs of antennas is completely uncorrelated. CQI measurements are 3 slot delayed; CQI update period is 3, 6, 9, 12 TTIs; CQI estimation is ideal and with Gaussian errors of variance 3 and 6 db; CQI decoding at NodeB is ideal. Bursty traffic model: File size - truncated lognormal distribution (µ = 11.736, σ = 0.0); Inter-arrival time - exponential distribution (Mean = 5 sec.). Proportional Fair: Independent; Intra-site joint; Forgetting factor 0.001. All MP-HSDPA UEs are capable of 15 SF 16 codes and 64QAM for each cell; Percentage of SFDC-HSDPA capable UEs is 100%. 3.2 CQI Imperfections CQI reports are sent from the UE to the NodeB every CQI reporting period. These values are used by the scheduling algorithm to select appropriate scheduling grants and to observe changes in the UE channel conditions. However CQI might be reported incorrectly and with varied period length. Thus it should be studied how these variations affect the performance of considered multipoint scheme. For this purpose, errors of 3 and 6 db were simulated in addition to ideal case, and CQI reporting period from 3 TTIs to 12 TTIs is analyzed. 3.3 Simulation Parameters The main parameters used in the system simulation are summarized in TABLE 1. A hexagonal wrap-around multi-cell layout is utilized. Wrap-around is used to model the interference correctly also for outer cells. This is achieved by limiting UE placement inside the actual simulation area but replicating the cell transmissions around the whole simulation area to offer more realistic interference situation throughout the scenario. UE is also able to connect to the replicated
8 Multiflow HSDPA with Joint Scheduling and CQI Imperfections cells, for example as a part of SHO active set. UEs are created according to a uniform spatial distribution over the whole area. This results into some cells being more heavily loaded while others can be even empty. 100% of the UEs are SFDC capable but multiflow is available only for user in SofterHO state. As it was shown in [7] considerable gain from multiflow feature can be achieved only with interference aware Type 3i receivers. Thus in our study all UEs use only this kind of receiver. A bursty traffic generation model is assumed, which means that the UEs do not constantly have data in the transmission buffers. File inter-arrival time is modeled with exponential distribution. File size is also variable and follows log-normal distribution (see TABLE I). The data available in the UE buffer is transmitted as fast as it is allowed by the NodeB. Those decisions are made according to the scheduling algorithms discussed above. Only pedestrian A channel with 3 km/h is studied in this paper. 4 Results Analysis Simulation results are presented in this section. The performance is evaluated through mean user throughputs for both all and SofterHO UEs. Gain is calculated as the difference in throughputs divided by the throughput of baseline scenario. On Fig. 3 independent PF scheduling plays the role of baseline scenario. Joint scheduling is benchmarked against it. Actual values of throughput can be found in the TABLE 2. As Fig. 3 and TABLE 2 indicate there is a positive gain from joint scheduling in higher load scenarios. The gain increases and reaches the maximum value of about 6% in 32 UE/sectorcase. We see the explanation of this relatively small gain in several facts. Firstly, the ratio of multiflow capable UEs, i.e. percentage of UEs which are in SofterHO region is about 9% so it is difficult to expect their high influence on the overall system performance. Secondly, precise assessment of SPs requires more detailed knowledge of CQI. Currently it is impossible to distinguish situations when it is better, for example, to use aggregated instated of only one-flow transmissions removing the main source of interference for the neighbouring sector. Nevertheless the first step is the study of the influence of CQI imperfection on the system performance, which is also done in this paper. Fig. 4 and 5 illustrate the degradation of system performance in the case of CQI imperfections. Legends in the figures refer to different cases so that: Mflow off means normal HSDPA operation with data transmission only from serving HS-DSCH cell; Intra equals to the case where UEs in SofterHO state can utilize HS-DSCH transmission from multiple cells (depending on scheduling decisions). Results are expected for both scenarios. With higher errors in CQI for all users (Fig.4(a)) average throughput goes down for baseline scenario without multiflow feature and for intra-site as well. As it can be noticed from Fig.4(b) for SofterHO UEs the gain from multiflow also becomes less significant. The
Multiflow HSDPA with Joint Scheduling and CQI Imperfections 9 8 6 All UEs SofterHO UEs 4 Gain, % 2 0 2 4 0 5 10 15 20 25 30 35 UEs/sector Fig. 3. Gain from joint intra-site scheduling for all and SofterHO UEs. UEs / Sector Independent scheduling, Mbit/sec Table 2. Mean user throughputs Joint scheduling, Mbit/sec Independent scheduling, Mbit/sec SofterHO UEs All UEs 1 7.92 7.93 10.86 10.58 2 7.69 7.57 9.74 9.97 4 7.00 6.90 9.07 8.87 8 5.61 5.69 7.06 7.03 16 2.83 2.89 3.30 3.29 32 0.34 0.36 2.85 3.06 Joint scheduling, Mbit/sec influence of CQI reporting period is not so strong. However, from Fig.5(b) it follows that in high load scenarios delayed information on channel quality causes loss of the gain for SofterHO users. 5 Conclusion This paper continues our study of multipoint HSDPA transmissions. It presents the results of our own simulations for one of the most practically interesting intra-site aggregation scheme. The effect of intra-site joint scheduling on system performance was studied and benchmarked against regular independent algorithm. In intra-site case moderate gain can be achieved by changing the scheduling algorithm. It is also shown how the precision of the channel quality feedback impacts on the efficiency of the multiflow feature.
10 Multiflow HSDPA with Joint Scheduling and CQI Imperfections 8000 Average throughput, kbps 7000 6000 5000 4000 3000 2000 Mflow Off, 0dB Intra, 0dB Mflow Off, 3dB Intra, 3dB Mflow Off, 6dB Intra, 6dB Average throughput, kbps 12000 10000 8000 6000 4000 Mflow off, 0dB Intra, 0dB Mflow off, 3dB Intra, 3dB Mflow off, 6dB Intra, 6dB 1000 2000 0 0 5 10 15 20 25 30 UEs/sector (a) All UEs 0 0 5 10 15 20 25 30 UEs/sector (b) SofterHO UEs Fig. 4. Influence of CQI errors on average user throughput Average throughput, kbps 8000 7000 6000 5000 4000 3000 2000 Mflow off, 3TTIs Intra, 3TTIs Mflow off, 6TTIs Intra, 6TTIs Mflow off, 9TTIs Intra, 9TTIs Mflow off, 12TTIs Intra, 12TTIs Average thoughput, kbps 12000 10000 8000 6000 4000 Mflow off, 3TTIs Intra, 3TTIs Mflow off, 6TTIs Intra, 6TTIs Mflow off, 9TTIs Intra, 9TTIs Mflow off, 12TTIs Intra, 12TTIs 1000 2000 0 0 5 10 15 20 25 30 35 40 UEs/sector (a) All UEs 0 0 5 10 15 20 25 30 35 40 UEs/sector (b) SofterHO UEs Fig. 5. Influence of CQI reporting period on average user throughput Acknowledgement This study is done in cooperation with Renesas Mobile Corporation. The authors would like to thank all of the co workers and colleagues for their comments and support. References 1. Global mobile Suppliers Association (GSA), GSM/3G Market/Technology Update, Feb. 7, 2012, http://www.gsacom.com/downloads/pdf/mobile_broadband_fact_ sheet_070212.php4. 2. Tapia P., Lui J., Karimli Y., and Feuerstein M.J.: HSPA Perforamnce and Evolution: A Practical Persective, Wiley (2009).
Multiflow HSDPA with Joint Scheduling and CQI Imperfections 11 3. Taoka H., and others: MIMO and CoMP in LTE-Advanced, NTT DOCOMO Technical Journal, Vol. 12, No. 2, pp. 10-28 (2010). 4. HSDPA multipoint transmission, 3GPP Work Item Description RP-101439, Dec. (2010). 5. HSDPA Multipoint Transmission, 3GPP Technical Report (TR) 25.872, Release 11, Sept. (2011). 6. DL Scheduling, RLC and Flow Control assumption for Inter-NodeB Multi-Point Transmissions, 3GPP Contribution R1-110126, Jan. (2011). 7. Petrov D., Repo I., Lampinen M.: Overview of Single Frequency Multipoint Transmission Concepts for HSDPA and Performance Evaluation of Intra-site Multiflow Aggregation Scheme, in Proc. of IEEE VTC, Yokohama, Japan, May (2012). 8. System Performance Evaluation of SF-DC Inter-NodeB aggregation with Type 3 and Type 3i receivers in Pedestrian A channel, 3GPP contribution R1-112303, Aug. (2011). 9. System Performance Evaluation of SF-DC Inter-NodeB aggregation with Type 3 and Type 3i receivers in Vehicular A channel, 3GPP contribution R1-112304, Aug. (2011). 10. Hamalainen S., Slanina P., Hartman M., Lappetelainen A., Holma H., and Salonaho O.: A Novel Interface Between Link and System Level Simulations, in Proc. of ACTS97, Aalborg, Denmark, Oct. (1997). 11. Guidelines for Evaluation of Radio Transmission Technologies for IMT-2000, ITU- R M.1225 Recommendation (1997). 12. Bu T., Li L., and Ramjee R.: Generalized Proportional Fair Scheduling in Third Generation Wireless Data Networks, IEEE INFOCOM (2006). 13. Simulation Framework for System Evaluation of Muti-point HSDPA, 3GPP Contribution R1-110563, Jan. (2011).