You are on page 1of 6

RESOURCE AND SCHEDULING OPTIMIZATION IN HSDPA BASED UMTS NETWORKS Anis Masmoudi1,2,3, Djamal Zeghlache2, Sami Tabbane3

Institut National des Sciences Appliques et de Technologie (INSAT), Tunis, Tunisia, Anis.Masmoudi@insat.rnu.tn 2 Institut National des Tlcommunications dEvry (INT), Evry, France, Djamal.Zeghlache@int-evry.fr 3 Unit de recherche en Rseaux Radio Mobile Multimdia MEDIATRON, (SupCom), Tunis, Tunisia Sami.Tabbane@supcom.rnu.tn
Abstract HSDPA-based UMTS networks should offer high bit rate connections in the next generation wireless networks. However, radio resources available to achieve them are limited. This paper presents suggestions to optimize some HSDPA techniques such as scheduling in single and multiservice cases while distinguishing terminal capabilities. Keywords HSDPA, Scheduling Techniques, HARQ Retransmissions, Resource Optimization, Terminal Category, Single and multi-service networks, CQI, Transport Block Size. I. INTRODUCTION The success of 3G networks is mainly due to an efficient provisioning of radio resources required to satisfy different services QoS with respect to data rate, delay and error rate. In order to improve support of high data rate packet services, 3GPP has developed High Speed Downlink Packet Access (HSDPA) in Release 5 specifications. HSDPA is considered as 3.5 G offering peak data rates up to 10-14 Mb/s [1], and expected to be implemented within the few next years. In this context, HSDPA resources should be optimized and efficiently used. In this paper, we evaluate some scheduling techniques, HARQ retransmissions and optimize them to be suitable for terminals categories. We also suggest new scheduling techniques for multiple services. Next section gives an overview on the main techniques used in HSDPA. Then, Section III presents HSDPA simulation results by distinguishing single and multi-service cases. Finally, our conclusions are given in the last section. II. DIFFERENT TECHNIQUES USED IN HSDPA HSDPA is based on the three following techniques: A. Adaptive Modulation and Coding The most important technique enabling data rates up to 10 Mb/s for HSDPA is the fast link adaptation provided by the use of Adaptive Modulation and Coding (AMC) consisting in a constant power node B transmit while the Modulation and Coding Scheme is altered to adapt to channel variations. This results in higher average throughput because higher order MCS levels are assigned to users with favorable conditions. B. Hybrid Automatic Repeat Request (HARQ) Although the MCS level is selected to ensure a reasonable probability of successful transmission, errors could occur. This is a result of highly variable channel conditions caused by interference from other users and base stations. Under normal circumstances, approximately 10 - 30% of first transmissions must be resent to be successful. The 3GPP selected HARQ for retransmissions due to its ability to quickly retransmit packets. HARQ functionality is implemented at the MAC layer to decrease delays. When a retransmission occurs, the data is combined with data stored in the soft buffer to effectively increase the coding gain, average throughput and enable the retransmission process to require fewer transmissions. Two fundamental schemes have been proposed to combine packets with those stored in the soft buffer: Chase Combining (CC) [2] and Incremental Redundancy (IR) [3]. C. Fast Scheduling The fast scheduling entity is also very important in the operation of HSDPA. One primary change from the previous implementations is that the scheduler is located at the Node B enabling the scheduler to quickly respond quickly the changes in the channel conditions. Examples of schedulers include Round Robin (RR) and Maximum C/I. III. SIMULATION RESULTS A. Single service case 1) One simulation 16 HSDPA mobile users uniformly distributed in the cell have been simulated (located respectively at distances 800 m, 750 m, 700 m, 650 m, 600 m, 550 m, 500 m, 450 m, 400 m, 350 m, 300 m, 250 m, 200 m, 150 m, 100 m and 50
1

m to base station) all using HTTP web browsing service simultaneously (with 384 Kb/s maximum bit rate) according to Best Effort policy. HTTP traffic is modeled by a Pareto On/Off distribution [4] with a mean burst time equal to 2 s and a mean idle time of 5 s. One simulated HTTP session lasts 190 seconds per user. Propagation environment is Pedestrian A. EURANE (Enhanced UMTS Radio Access for NS-2 Extensions) open source simulator including HSDPA functionalities [5]-[6] has been used. Four sample users have been analyzed by measuring their mean session throughput, packets delivery delay and number of HARQ retransmissions. Measured users are those distant to 700, 500, 300 and 100 meters from node B. Some statistics have been extracted using Awk and Matlab. Category 10 terminals offering the highest bit rates have been used in the simulations of this subsection. Measured instants are not regular but according to jitter variation. Mobiles are considered as vehicles moving at equal distances from the base station (circular movement). Figures 1 to 3 show RLC packet delivery delay CDF plots for each of the four users analyzed using respectively Max C/I, Fair throughput, and Fair time scheduling techniques [7]-[8]. The two last methods have been implemented and added to the original Eurane NS-2 extensions. Table 1 summarizes those curves into statistics such as average session delay and its 80-ile (packets delay value than which delivery delay is higher in 80% of session time). In Fair throughput scheduling method, more time resources, in average, are allocated to mobiles farther from node B, so packets intended for mobiles near base station have a longer waiting time in the MAC queue, that' s why the nearer the mobiles are from node B, the less time resources they are allocated, and thus the higher the endto-end packet delivery delay is when Fair throughput scheduling is used (See Figure 1 and Table 1). In contrast, Max C/I scheduling technique gives in average more priority to nearer mobiles, which makes their packet delivery delay shorter (less duration spent at the MAC buffer): See Figure 2 and Table 1. By assigning the same priority for mobiles to be scheduled by applying Fair Time technique, we note according to Figure 3 and Table 1 that the further from node B the mobile is, the higher the packet delivery delay. This is due to the fact that Fair time scheduling handles different flows block by block (Transport Block TB from each flow in a Round Robin manner), and since far mobiles buffers include more packets waiting because their associated Transport Block Size TBS is smaller due to a lower average CQI indicator, and thus data from further mobiles should include more TBs.

Figure 1: End-to-end delivery delay CDF plots using Fair Throughput scheduling. Mean Delay delay 80-ile 2.100 1st user (700 m) 2.647 4.034 Fair 2nd user (500 m) 3.424 Throughput 6.702 3rd user (300 m) 3.846 10.378 4th user (100 m) 6.768 117.112 1st user (700 m) 80.466 0.818 2nd user (500 m) 0.452 Max C/I 0.072 3rd user (300 m) 0.200 0.012 4th user (100 m) 0.008 3.162 1st user (700 m) 3.342 2.096 2nd user (500 m) 1.594 Fair Time 0.762 3rd user (300 m) 0.504 0.074 4th user (100 m) 0.046 Table 1: Numerical delay statistics for different scheduling techniques. 16 users

Figure 2: End-to-end delivery delay CDF plots using Max C/I scheduling. Table 2 shows statistics on session throughput and jitter delay. Note that Fair throughput ensures higher throughput percentile (guaranteed bit rate) except for mobiles very far from node B. Max C/I scheduling couldn' t be convenie nt

especially for remote mobiles (the 1st Max C/I user has about 85% relative throughput loss compared to its throughput by applying Fair Throughput method due to excessive jitter delay). Fair Time scheduling technique seems to be the best method appropriate to Web browsing interactive service. In fact, it doesn' t present excessive jitter delay as Max C/I and Fair Throughput, and offers better total cell throughput than Fair Throughput (more than 6% gain over Fair Throughput).

Mean ThrouMean session 16 users ghput throujitter 20%-ile ghput 0.036 26.6 1st user (700 m) 75.6 5.2 34.3 Fair 2nd user (500 m) 84.2 Throughput 3rd user (300 m) 102.6 27 42.3 32.7 111.7 4th user (100 m) 81.7 0.145 430.5 1st user (700 m) 11.3 0.273 46.5 2nd user (500 m) 67.8 Max C/I 0.313 22.6 3rd user (300 m) 121.6 67 0.674 59.4 4th user (100 m) 0.493 35 1st user (700 m) 59.8 0.078 33.9 2nd user (500 m) 81.1 Fair Time 0.564 30.7 3rd user (300 m) 153.2 0.233 84.7 4th user (100 m) 71.8 Table 2: Numerical throughput and jitter statistics for different scheduling techniques.

Figure 3: End-to-end delivery delay CDF plots in Fair Time / Resource scheduling. 2) Several simulated HTTP sessions In this paragraph, 20 simulations have been addressed for each of HSDPA terminal categories 1 (offring up to 3.5 Mb/s theoretical bit rate) and 10 having a the greatest capability (offering a 12.7 Mb/s maximum theoretic bit rate) owing to the use of up to 15 codes and a TBS of 25558 bits for a CQI of 30 in category 10 HSDPA terminals [9]. Fair Throughput is used in next simulations. Figure 4 to 6 show different retransmissions probabilities (in %) CDF plots for Fair Throughput scheduling and category 10 terminals by simulating 20 sessions of 190 seconds, and Table 3 presents their average values. In addition to the evident aspect that the nearer to node B the mobile is, the higher the probability of successful reception after the first HARQ transmission. The first retransmission probability presents the opposite tendency among the four users to compensate the amount of data not received correctly especially by remote mobiles. The second HARQ retransmission has the same tendency as the first one except for the remote mobile distant to 700 m from node B, proving the lack of efficiency of the only two possible retransmissions for such mobiles far from base station (more rejected packets). So, to keep the same efficiency of the two possible standardized HARQ retransmissions, cell coverage range should not exceed 500 to 600 m. This is confirmed by retransmissions results in

Figure 4: Probability of transmission success without any retransmission.

Figure 5: Probability of successful first retransmission. tables 4 and 5 referring to the use of Max C/I and Fair Time scheduling techniques respectively. Consequently, the high packet delivery delay and the important resource

Distance to node B 700 500 300 100 Success probability without 99.03 99.30 98.67 98.49 any retransmission (%) st Probability of the 1 0.56 0.51 1.19 1.29 retransmission (%) nd Probability of the 2 0.41 0.19 0.14 0.22 retransmission (%) Table5: HARQ probability retransmission probability statistics (Fair Time / Resource). Distance to node B Category 1 terminals 700 [53.3, 72] (62.7) 500 [64.6, 71.6] (68.1) 300 [75.3, 92.1] (83.7) [70, 86] (78) 100 [68.8, 84] (76.4)

Figure 6: Probability of successful second retransmission. consumption of mobiles distant more than 600 m from node B in addition to the inefficiency of the only two HARQ retransmissions to serve remote mobiles make more convenient for operators to limit HSDPA deployment to concentric cells with a core not exceeding 600 m range; only a basic pure UMTS layer can be used in the ring. Tables 3, 4 and 5 show also that Fair Time scheduling is the most suitable for interactive services such as web browsing due to its high transmissions success probability. Table 6 compares both terminal categories 1 and 10 in terms of mean session throughput (in Kb/s) for mobiles in different distances from node B with indication of 95% confidence intervals and average values. We note clearly that all category 1 mobile terminals have a higher throughput (from 2 to 5 Kb/s additional throughput per user). This is due to the higher consumption of the limited bandwidth by category 10 terminals since they allow a more important instantaneous throughput (greater TB) Distance to node B 700 500 300 100 Success probability without 97.51 98.32 98.57 99.52 any retransmission (%) Probability of the 1st 2.12 1.30 1.18 0.40 retransmission (%) nd Probability of the 2 0.36 0.37 0.26 0.09 retransmission (%) Table 3: HARQ probability retransmission probability statistics (Fair Throughput). Distance to node B 700 500 300 100 Success probability without 92.13 79.33 84.99 96.87 any retransmission (%) Probability of the 1st 7.77 19.27 14.44 2.93 retransmission (%) nd Probability of the 2 0.10 1.41 0.57 0.21 retransmission (%) Table 4: HARQ probability retransmission probability statistics (Max C/I).

[66.2, 81.2] (73.7) Table 6: Average session throughput for different terminal categories. Category 10 [52, 68.8] [56, 72.9] (60.4) (64.5) terminals when propagation conditions (and thus CQI) are good. In fact, the higher average TB for nearer mobiles makes the time allocation of further mobiles more important in average in order to ensure a fair throughput for all cell users, thus decreasing throughput due to the greater use of smaller TBS for remote mobiles, and to the higher delay and jitter for mobiles next to node B. Tables 7 and 8 compare respectively categories 1 and 10 in terms of HARQ retransmissions probability by applying Fair Throughput scheduling (95% confidence intervals and average probability values are shown for each (re)transmission). Here, we also note that categ. 1 is more efficient than categ. 10 (in Fair Throughput scheduling). Hence, Fair Throughput is efficient only for lower terminal capabilities such as categ. 1. Figure 7 shows instantaneous RLC throughput CDF plots for categ. 1 terminals with several simulations. Note the comparable throughputs. In conclusion, Fair throughput is not recommended for higher capability terminals. Instead, a terminal capability detection component can be implemented in the node B to differentiate connected terminals categories and prevent Fair Throughput scheduling for higher capabilities. B. Multi-service case By using multiple services having different characteristics (some requiring a guaranteed bit rate such as streaming, voice, at the opposite of interactive services), new scheduling techniques should be used. The following scheduling methods have been introduced and implemented in NS-2 Eurane simulator and tested with multiple services.

Distance to node B Success probability without any retransmission (%) Probability of the 1st retransmission (%) Probability of the 2nd retransmission (%) Table 7: Probability category 1 terminals.

700 500 300 100 [98.21, [98.43, [98.64, [99.58, 98.63] 98.75] 99.03] 99.96] (98.42) (98.59) (98.84) (99.77) [1.08, [0.90, [0.77, [0.03, 1.46] 1.19] 1.11] 0.38] (1.27) (1.05) (0.94) (0.20) [0.26, [0.31, [0.18, [0.01, 0.36] 0.41] 0.27] 0.05] (0.31) (0.36) (0.22) (0.03) retransmissions probabilities for

2) Prioritized Rayleigh Peak scheduling This method attempts to schedule guaranteed bit rate services first only in Rayleigh Peak instances, i.e. if their quality is good enough (or, in terms of implementation, provided that their CQI is greater or equal than a specified CQI threshold value). In this case Fair Throughput scheduling is applied to guaranteed bit rate flows having an acceptable CQI value. Otherwise, if CQI is less than this CQI threshold, flows are treated at Fair Time scheduling. 3) Weighted Differentiated Services scheduling It is similar to the Prioritized Differentiated Services scheduling except that instead of taking an absolute priority for the guaranteed bit rate services, and in order to increase the resources reserved to non-guaranteed bit rate services, this method assigns weights both to guaranteed and non-guaranteed bit rate services to balance the bandwidth dedicated for each of them being scheduled respectively in Fair Throughput and Fair Time manners. 4) Weighted Rayleigh Peak scheduling It has the same principle as the previous method but adds the condition for CQI to be above a pre-determined threshold (at Rayleigh peaks) for guaranteed bit rate flows before deciding to serve them according to Fair Throughput scheduling. The aim of this method is avoiding to waste resources and dedicating them to bad TTI links (especially at fair throughput). If CQI is less than the threshold, non-guaranteed bit rate flows are scheduled at Fair Time fashion. Figure 8 and 9 show the different mobile users CDF plots for some of the previously described scheduling techniques. Four services are simulated simultaneously: 64 Kb/s Constant Bit Rate (CBR 64), 128 Kb/s Constant Bit Rate (CBR 128), FTP and web browsing, at the rate of 3 users per service (located respectively at 300 m, 500 m and 700 m from the node B) involving 12 users in total. Tables 9 and 10 present average and 20%-ile throughput statistics per user and per service. By increasing CQI threshold of Rayleigh peaks for guaranteed bit rate services (CBR) from 12 to 17 (Figure 8 and 9), more time resources are allocated to non real time services and thus their throughputs are enhanced without much degrading other services throughputs. By introducing a CQI threshold (8) to weighted differentiated services scheduling (Figure 10) and after assigning a small weight to the FTP time resources (1/11), the two remote web browsing users are satisfied and other services are well maintained. The small FTP weight is because FTP is a great consumer of bandwidth.

Distance to node B 700 500 300 100 Success probability [96.86, [98.08, [98.30, [98.96, 98.16] 98.57] 98.83] 100] without any retransmission (%) (97.51) (98.32) (98.57) (99.52) [1.50, [1.08, [0.94, [0, Probability of the 2.75] 1.52] 1.42] 0.87] 1st retransmission (2.12) (1.30) (1.18) (0.40) (%) [0.31, [0.32, [0.22, [0, Probability of the 0.41] 0.43] 0.29] 0.18] 2nd retransmission (0.36) (0.37) (0.26) (0.09) (%) Table 8: Probability retransmissions probabilities for category 10 terminals. 1) Prioritized Differentiated Services scheduling This scheduling method differentiates between services requiring guaranteed bit rate (such as CBR services, video streaming, voice, ) and non-guaranteed bit rate ones (such as interactive Web browsing, FTP, Email,). It introduces a parameter specifying the maximum priority value for which service is considered as guaranteed bit rate and are scheduled at Fair Throughput manner, while the remaining others (non-guaranteed bit rate flows) are scheduled at Fair Time fashion. Priorities among all services is taken into account.

Figure 7: Instantaneous throughput CDF plots (Category 1 terminals).

Service CBR 64 Kb/s CBR 128 Kb/s Distance 300 500 700 300 500 700 (m) Moyenne 63.5 63.4 63.4 127 5.1 15.9 (Kb/s) 20%-ile 41.8 34 32 60.6 0.029 3.4 (Kb/s) Service www FTP Distance 300 500 700 300 500 700 (m) Moyenne 0.033 0 0 7.4 2.1 0.346 (Kb/s) 20%-ile 0.012 0 0 0.889 0.222 0.015 (Kb/s) Table 9: Throughput statistics of different services (Prioritized Rayleigh Peak, CQI threshold = 12).

Figure 9: CDF plots for each user for Weighted Differentiated Services scheduling technique (a weight of 1/3 for non-guaranteed bit rate services and 2/3 for guaranteed bit rate ones). Service CBR 64 Kb/s CBR 128 Kb/s Distance 300 500 700 300 500 700 (m) Moyenne 63.7 63.7 63.8 127.6 117.1 93.8 (Kb/s) 20%-ile 53.3 47.5 42.7 76.2 37.5 3.1 (Kb/s) Service www FTP Distance 300 500 700 300 500 700 (m) Moyenne 135.3 46.3 8.4 203.9 3.3 0 (Kb/s) 20%-ile 55.2 0.288 0.147 114.9 0.274 0 (Kb/s) Table 10: Throughput statistics of different services (Weighted Differentiated Services, CQI threshold = 8, 1/11for FTP). [3] Amitava Ghosh et al., Incremental Redundancy (IR) Schemes For W-CDMA HS-DSCH, IEEE Pers., Ind. and Mob. Radio Com., Lisbon, September 2002, [4] 3G TR 25.892, Feasibility Study for OFDM for UTRAN enhancement, [5] Neill Whillans (Editor), End-to-end network model for Enhanced UMTS, IST SEACORN Project Deliverable D3.2v2, [6] Eurane user Guide, User manual for Eurane v 1.3, [7] T. E. Kolding, F. Frederiksen, and P.E. Mogensen, Performance Aspects of WCDMA Systems with High Speed Downlink Packet Access (HSDPA), VTC, , vol. 1, pp. 477-481, Sept. 2002, [8] T. E. Kolding, et al., High Speed Downlink Packet Access: WCDMA Evolution, In IEEE Vehicular Technology Society News, IEEE, Feb. 2003, [9] 3G TS 25.214, Physical layer procedures (FDD).

Figure 8: CDF plots for each user for Prioritized Rayleigh Peak scheduling (CQI threshold of 17). IV. CONCLUSION In this paper, we have evaluated some scheduling techniques and HARQ retransmission both in single and multi-service cases. In the first case, we have addressed a comparison study between higher and lower terminal capabilities and, have recommended the most efficient scheduling technique for each category. We have also tested and suggested new scheduling methods for multiple services by differentiating real time and non-real time services. If those techniques are implemented in HSDPA, operators can benefit of increased cell throughput and user QoS satisfaction according to service requirements. References [1] H. Holma and A. Toskala, WCDMA for UMTS. Radio Acc. for 3 G Mobile Comm., Second Edition, Chapter 11, John Wiley and Sons, England, 2002, [2] D. Chase, Code combining: A Maximum-Likeli. Decoding App. for Comb. an Arbitrary Number of Noisy Packets, IEEE Trans. on Comm., May 1985,

You might also like