LINK ADAPTATION IN GENERAL PACKET RADIO SERVICES (GPRS) Juan Li

Size: px

Start display at page:

Download "LINK ADAPTATION IN GENERAL PACKET RADIO SERVICES (GPRS) Juan Li"

Lorin Tucker
5 years ago
Views:

1 LINK ADAPTATION IN GENERAL PACKET RADIO SERVICES (GPRS) Part II Juan Li, HUT 28th.Jan, 3 Link Adaptation in GPRS 1/27 Simulation Model ---One Example Figure 1 One example of link level model 28th.Jan, 3 Link Adaptation in GPRS 2/27 1

2 Selective ARQ Transmission of radio blocks over the radio interface is controlled by a selective ARQ (Automatic Repeat Request) mechanism, [1] whose function consists of error detection and retransmission. In GPRS the transmitter side has a sender window of 64 blocks and the receiver side has a same size sender window accordingly. When the transmitter sends blocks from the sender window, the receiver sends back Temporary Packet ACK/NACKS to indicate whether the received block is correct or erroneous. The period between two consecutive Temporary Packet ACK/NACK is here referred as polling period and could be of length 1-64 blocks. Too long polling period leads to stalling while too short polling period decreases the efficiency of utilizing radio resources. According to [2], for an ACK/NACK message delay of 8 blocks, a BSS processing delay of 3 blocks and BLER = % as worst case, the polling period should be no shorter than blocks. Hence here we choose polling period as blocks. 28th.Jan, 3 Link Adaptation in GPRS 3/27 Mode of Operation For MAC and RLC The Medium Access Control (MAC) and Radio Link Control (RLC) layer operates above the Physical Link layer in the reference architecture. [3] The RCL/MAC function provides two modes of operation: - acknowledged operation; and - unacknowledged operation Acknowledged mode for RLC/MAC operation The transfer of RLC Data Blocks in the acknowledged RLC/MAC mode is controlled by a selective ARQ mechanism coupled with the numbering of the RLC Data Blocks within one Temporary Block Flow. Unacknowledged mode for RLC/MAC operation The transfer of RLC Data Blocks in the unacknowledged RLC/MAC mode is controlled by the numbering of the RLC Data Blocks within one Temporary Block Flow and does not include any retransmissions. 28th.Jan, 3 Link Adaptation in GPRS 4/27 2

3 Initial Coding Scheme There are several different ways to determine the initial coding scheme to be used in the system. A general point of view is that if the system has a relatively low load (typically below %) the simplest choice is to always start with the weakest coding, i.e., coding scheme 4. Actually most of the time the radio channels will have sufficient radio resource to allow this. In the algorithms presented in this presentation the common approach has been used, i.e., is always chosen as a starting coding scheme. 28th.Jan, 3 Link Adaptation in GPRS /27 Performance of Four Coding Schemes (1) Before discussing the link adaptation algorithm itself, the performance of four coding schemes in certain channels is desired to be investigated first. For this purpose, the throughput of four coding schemes will be analysed The comparison is undertaken in the multipath fading channel, where Gaussian noise is added at the receiver. Each time a webpage download of 6 bits was assumed in the transmitter side. With different coding schemes, data block carries different information bits. At the receiver side, the received block was compared with the transmitted block. If it is totally correct, this block would be counted into throughput and otherwise into BLER. After repeating this test three times, taking the average values of throughput and BLER, the resulted curves Throughput. vs. CIR (Figure 2- Figure ) were generated. 28th.Jan, 3 Link Adaptation in GPRS 6/27 3

4 Performance of Four Coding Schemes (2) Figure 2 Throughput as function of CIR, without FH, MS speed =3.6km/h (1m/s) Figure 3 Throughput as function of CIR, without FH, MS speed =4km/h (1m/s) Figure 4 Throughput as function of CIR, with FH, MS speed =3.6km/h (1m/s) Figure Throughput as function of CIR, with FH, MS speed =4km/h (1m/s) 28th.Jan, 3 Link Adaptation in GPRS 7/27 BLER-based Link Adaptation (1) The implementation of BLER-based algorithm in the presentation can be described as: (1) Starting with coding scheme 4. (2) The RLC blocks are continuously transmitted until the end of one reporting window ( blocks was selected in the simulation). (3) Then the BLER is averaged over the previous reporting window. If the result locates outside the range of the current coding scheme defined by specified threshold, the coding scheme is modified in the following data blocks and keeps unchanged during the following averaging window. (4) If there are more data to transmit go back to step 2. 28th.Jan, 3 Link Adaptation in GPRS 8/27 4

5 BLER-based Link Adaptation (2) Figure 6 Throughput as function of CIR, without FH, MS speed =3.6km/h (1m/s) Figure 7 Throughput as function of CIR, without FH, MS speed =4km/h (1m/s) Figure 8 Throughput as function of CIR, with FH, MS speed =3.6km/h (1m/s) Figure 9 Throughput as function of CIR, with FH, MS speed =4km/h (1m/s) 28th.Jan, 3 Link Adaptation in GPRS 9/27 BLER-based Link Adaptation (3) The thresholds are obtained from the intersection points (similarly as A, B, C in Figure 11 in part I) by using the results from Figure 6 to Figure 9, which is similar to those figures (Figure 2 to Figure ) presented in the slide #7 but with considering the attenuation. For example, if the transmitter is now coding with, after the transmission of blocks, if the calculated BLER is lower than the threshold determined by point B, the coding scheme will be switched to weaker in the next RLC/MAC data block; if the BLER is higher than the threshold determined by point A, the coding scheme will be switched to stronger ; if the BLER is between the thresholds determined by of point A and B, the coding scheme will be kept unchanged. 28th.Jan, 3 Link Adaptation in GPRS /27

6 BLER-based Link Adaptation (4) One example of throughput vs. CIR for four coding schemes 28th.Jan, 3 Link Adaptation in GPRS 11/27 CIR-based Link Adaptation (1) The implementation of the CIR-based algorithm in the thesis can be described as: (1) Starting with coding scheme 4. (2) The RLC blocks are continuously transmitted until the end of one reporting window ( blocks was selected in the simulation). (3) Then the CIR is estimated by subspace-based estimator [4]. If the result exceeds specified threshold, the coding scheme is modified in the following data blocks and keeps unchanged during the following averaging window. (4) If there are more data to transmit go back to step 2. 28th.Jan, 3 Link Adaptation in GPRS 12/27 6

7 CIR-based Link Adaptation (2) The thresholds for different situations can be directly taken from Figure 6 to Figure 9 based on the throughput intersections of different coding schemes. Further the intersections for different user speed was measured and plotted in Figure, which illustrates that the throughput intersection of different coding schemes is a non-linear function of user speed. Throughput intersection (db) & & & 4 1m/s 3m/s m/s 7m/s 9m/s 11m/s 13m/s 1m/s MS Speed (m/s) Figure Threshold of CIR vs. MS speed, without frequency hopping 28th.Jan, 3 Link Adaptation in GPRS 13/27 Under Ideal Situation (1) Ideal link adaptation here refers to applying link adaptation algorithm by assuming the speed of users be fixed and known so that the selection of thresholds can be identical to the real situation. Typical urban user speed, either 1m/s or 1m/s, is involved. In this presentation, comparison of two link adaptation algorithms, BLER-based and CIR-based, under ideal link adaptation is carried out in terms of throughput. Three typical channel qualities CIR = 13, 19 and 2dB have been chosen to stand for bad, normal and good channel quality. The throughput of each link adaptation algorithm under three types of channel in a simulation run is shown in Figure th.Jan, 3 Link Adaptation in GPRS 14/27 7

8 Under Ideal Situation (2) 23, 23, 22, 22, 21, 21,, BLER CIR 22, 21,, 19, 18, 17, BLER CIR, 16, 19, without FH, MS speed = 1m/s 1, without FH, MS speed = 1m/s 23, 23, 22, 21,, 19, BLER CIR 22, 21,, 19, 18, 17, BLER CIR 18, 16, 17, , with FH, MS speed = 1m/s with FH, MS speed = 1m/s Figure 11 Comparison of throughput for different link adaptation algorithms under ideal situation 28th.Jan, 3 Link Adaptation in GPRS 1/27 Under Ideal Situation (3) Conclusion: -- The resulted throughput indicates that the estimated CIR-based algorithm is superior to the BLER-based algorithm. Under extreme situations, with either very bad or very good channel qualities, the same results might be obtained for both algorithms. As a matter of fact, one single coding scheme is continuously adopted under such circumstances no matter which link adaptation algorithm is employed. 28th.Jan, 3 Link Adaptation in GPRS 16/27 8

9 Under Practical Situation (1) In a real system, a crucial problem arises because of the unknown user speed. The curves of throughput vs. CIR presented in Figure 6 to Figure 9 are only available with knowing the user speed. As a result, the selection of threshold in link adaptation scheme experiences an uncertainty. Once a link adaptation scheme is chosen, there is a risk that it is unfair to certain users. Assuming there are two user groups in one urban environment with speed of 1m/s and 1m/s, the link adaptation algorithm is determined with thresholds for 1m/s user. This link adaptation scheme is suitable for users with speed of 1m/s but adverse to users with speed of 1m/s. 28th.Jan, 3 Link Adaptation in GPRS 17/27 Under Practical Situation (2) A statistic evaluation method is proposed. Under the urban environment, the typical user speeds are 1m/s and 1m/s so two sets of thresholds can be determined with respect to these two speeds. It is also a natural thinking that averaging these two sets of thresholds offers another possibility. Hence, according to the combinations of thresholds and link adaptation algorithms, six schemes were proposed and they were denoted as BLER1, BLER1, BLERm, CIR1, CIR1, and CIRm. The leading capital letters stands for the algorithm and the following numbers specifies how the thresholds are obtained (m means by averaging two sets of thresholds). For example, BLER1 stands for the BLER-based link adaptation algorithm whose threshold corresponds to user speed of 1m/s while CIRm standing for the CIRbased link adaptation algorithm whose threshold is an average of the thresholds of user speed 1m/s and 1m/s. 28th.Jan, 3 Link Adaptation in GPRS 18/27 9

10 Under Practical Situation (3) Six different typical situations are of interest in the practical situation under consideration. They are combinations of three typical channel qualities (CIR=13, 19, 2dB represents bad, normal and good channel respectively) and two typical user speeds (1m/s, 1 m/s). The combinations are named as: 13dB_1m/s, 19dB_1m/s, 2dB_1m/s, 13dB_1m/s, 19dB_1m/s, 2dB_1m/s. Simulations are thus performed under each situation for each scheme and the resulted throughputs are recorded. For each scheme, the throughputs under six typical situations are averaged and the standard deviations are calculated. One result is listed in Table 1. The final mean values and standard deviations are plotted in histograms as shown in Figure 12 to Figure th.Jan, 3 Link Adaptation in GPRS 19/27 Under Practical Situation (4) 28th.Jan, 3 Link Adaptation in GPRS /27

11 Under Practical Situation () BLER1 BLER1 BLERm CIR1 CIR1 CIRm BLER1 BLER1 BLERm CIR1 CIR1 CIRm Link Adaptation Schemes mean_thr. Std. Link Adaptation Schemes mean_thr. Std. Figure 12 Comparisons of six link adaptation schemes, without frequency hopping Figure 13 Comparisons of six link adaptation schemes, with frequency hopping 28th.Jan, 3 Link Adaptation in GPRS 21/27 Under Practical Situation (6) Two criteria can be then introduced to evaluate the performance of different schemes. One is that the highest throughput is expected, which is measured by the mean value of throughput; the other is that the throughput should be as stable as possible, which is measured by the standard deviation. As indicated in Figure 12, under the condition without frequency hopping, BLERm owns the highest average throughput and BLER1 appears to be the most stable scheme. However, CIR1 bears the most promising feature combining high average throughput and low standard deviation. As indicated in Figure 13, under the condition with frequency hopping, BLER1 appears to be the most stable scheme. However, CIR1 bears the most promising feature combining high average throughput and low standard deviation. 28th.Jan, 3 Link Adaptation in GPRS 22/27 11

12 Comparison of Link Adaptation and Fixed Coding Scheme(1) In the same way, a comparison was made among CIR1 and two fixed coding schemes, fixed and fixed. See Figure 14 and Figure 1. In these two figures, fixed though has the lowest standard deviation (too low to be shown in the figures) but also has the poorest throughput, fixed has a poor throughput and the highest standard deviation. It is thus apparent that fixed coding scheme will lead to considerable throughput loss and probable large variation in throughput. Therefore link adaptation that presents much better performance than fixed coding schemes is indispensable in GPRS system. 28th.Jan, 3 Link Adaptation in GPRS 23/27 Comparison of Link Adaptation and Fixed Coding Scheme(2) mean_thr. Std. CIR1 fixed fixed mean_thr. Std. CIR1 fixed fixed Figure 14 CIR1 vs. fixed coding scheme, without frequency hopping Figure 1 CIR1 vs. fixed coding scheme, with frequency hopping 28th.Jan, 3 Link Adaptation in GPRS 24/27 12

Preliminary Explanation of Superior Performance of CIR1 The throughput difference between CS-3 and CS-4 is shown in Figure 16 as a function of CIR.

13 Preliminary Explanation of Superior Performance of CIR1 The throughput difference between CS-3 and CS-4 is shown in Figure 16 as a function of CIR. The curve for 1m/s is inversed so as to simplify the discussion below. Assuming the current channel quality expressed in CIR is at point D, CIR1 makes a misjudgment for 1m/s users and its throughput loss is represented by DF. Similarly, the throughput loss for 1m/s user using CIR1 is represented by DE. The possibility of such misjudgment is the same for CIR1 and CIR1. It is then not surprising that the performance of CIR1 is superior to CIR1. The reason why CIRm is inferior to CIR1 is the similar. Figure 16 Throughput difference between and, without frequency hopping, MS speed = 1m/s 28th.Jan, 3 Link Adaptation in GPRS 2/27 Conclusion The above discussions proved that the proposed statistic evaluation method offered a powerful tool in evaluating and optimizing link adptation scheme. It is interesting to note that the method is flexible. A more accurate model can be obtained by increasing the number of typical situations if necessary. The method is also open to other link adaptation schemes of interest. 28th.Jan, 3 Link Adaptation in GPRS 26/27 13

14 Reference 1. William Stallings, Data and Computer Communications, fourth edition, Macmillan publishing company, Gutierrez, P.J.A.; Wigard, J.; Andersen, P.N.; Damgaard, H.C.; Mogensen, P., Performance of link adaptation in GPRS networks, Vehicular Technology Conference,. IEEE-VTS Fall VTC. 2nd, Volume: 2, Page(s): vol.2 3. ETSI TS 1 3 V8.8. (1-4), Digital cellular telecommunications system (Phase 2+); General Packet Radio Service (GPRS); Overall description of the GPRS radio interface; Stage 2 (3GPP 3.64 version 8.8. Release 1999) 4. Queseth, O.; Gessler, F.; Frodigh, M. Algorithms for link adaptation in GPRS., Vehicular Technology Conference, 1999 IEEE 49th, Volume: 2, 1999 Page(s): vol.2 28th.Jan, 3 Link Adaptation in GPRS 27/27 14

General Packet Radio Service Performance Evaluation Based on CIR Calculation, Considering Different Radio Propagation Models

C38 1 General Packet Radio Service Performance Evaluation Based on CIR Calculation, Considering Different Radio Propagation Models SAMI A. EL-DOLIL and AMIR S. EL-SAFRAWY Dept. of Electronics and Electrical