You are on page 1of 5

Performance of Frequency Selective Scheduling and Fractional Frequency Reuse schemes for WIMAX

Chandrasekar Sankaran, Fan Wang, and Amitava Ghosh


Network Advanced Technology (NAT) Motorola Inc, Arlington Heights, IL 60004, USA csankar1,fwang2,qa0047@motorola.com
Abstract WiMaX is the commercial name for the 4G

OFDM based wireless broadband data transmission standard 802.16, developed by IEEE ([1] and [2]). It is currently being deployed in several countries around the world. Sub-channels (which are the basic unit of resource allocation) are formed by grouping together sub-carriers in one of two formats - PUSC or AMC. In PUSC the set of sub-carriers in a sub-channel are distributed throughout the bandwidth while in AMC the sub-carriers in a subchannel are contiguous within a portion of the bandwidth. In this paper we study two candidates that offer the potential to improve the system performance of PUSC and identify scenarios when this improvement is possible: (i) compare the performance of PUSC with two flavors of AMC; and (ii) study how the PUSC performance can be improved by a Fractional Frequency Reuse (FFR) scheme that tries to perform interference mitigation/co-ordination and improve the throughput of users at the edge of the cell.
Keywords WiMaX, PUSC, AMC, Frequency Selective Scheduling,

1 0.9 0.8 0.7 Probability 0.6 0.5 0.4 0.3 0.2 0.1 0 -10 -5 0 5 10 15 20 25 SNR (dB)
1x3x1 1x3x3 1x4x2

Fractional Frequency Reuse, Interference co-ordination, 1 cell reuse. I. INTRODUCTION WiMaX (Worldwide Interoperability for Microwave Access) is a 4th generation (4G) broadband wireless solution that enables convergence of mobile and fixed broadband networks through a common wide area broadband radio access technology and flexible network architecture [4]. The WiMaX air interface is based on the IEEE 802.16-2004 standard [1] and the IEEE 802.16e Mobile Amendment to the standard [2]. WiMaX air interface adopts Orthogonal Frequency Division Multiple Access (OFDMA) for improved multi-path performance in non-line-of-sight environments. The downlink (DL) and uplink (UL) transmissions are time division duplexed within a single frame. There are two modes of sub-carrier assignments in WiMaX PUSC (partial usage of sub-channels) where the set of subcarriers that constitute a sub-channel assigned to a user are distributed throughout the bandwidth and band AMC (adaptive modulation and coding) where the set of sub-carriers in a subchannel occupy a contiguous portion of the bandwidth (i.e., localized allocation). In the first part of this paper we compare the performance of the PUSC scheme with a band AMC scheme and provide insights into when AMC performs better than PUSC for different traffic models.

Figure 1: CINR Distribution for various Reuse Schemes Frequency reuse refers to the distribution of the available spectrum (bandwidth) across the cells and sectors in the system. The typical nomenclature used is [number of cell sites x number of sectors per site x total number of RF carriers required]. Standard frequency reuse types possible in WiMaX are [1x3x1] (1 cell site, 3 sectors, 1 RF carrier), [1x3x3], and [1x4x2]. The 1x3x1 system is also referred to as the one cell reuse configuration. In the 1x3x1 configuration all sectors are using the same RF carrier resulting in users experiencing interference from all their neighboring sectors in the system. This is reflected in the CINR (Carrier to Interference plus Noise Ratio) CDF plot shown in Figure 1. This figure plots a typical CINR CDF for the three different frequency reuse types: 1x3x1 (site-site spacing of 1.23km), 1x3x3 (site-site spacing of 1.23km) and 1x4x2 (site-site spacing of 1.62km), all using a 2W transmission power per sector. As can be seen from Figure 1 users in the 1x3x1 system experience the most interference. This results in the throughput for the cell edge users in the 1x3x1 system being low since they have a low CINR. In the second part of this paper we consider the PUSC scheme (for 1x3x1) and see how the throughput for cell edge users can be improved by different flavors of Fractional Frequency Reuse (FFR) schemes that attempt to perform some form of interference co-ordination across the sectors. II. PUSC AND BAND AMC In this section we describe how the band AMC scheme operates in the context of WiMaX and study its performance relative to PUSC for a mobile WiMaX system with users moving at slow speeds (3 km/hr).

978-1-4244-2517-4/09/$20.00 2009 IEEE

A. AMC SCHEDULING AND SWITCHING METHOD 1) Frequency Diverse and Frequency Selective Scheduling Since the sub-carriers allocated in band AMC are contiguous, it provides the scheduler a chance to take advantage of the constructive channel fading in specific bands of the spectrum as shown in Figure 2. Frequency selective (FS) scheduling combined with band AMC can allocate a higher MCS (modulation coding scheme) than whats possible with PUSC (FD).
SINR

rate - in this case 64 Kbps). There is a performance gap between AMC-I and AMC-P - as Figure 5 and Figure 6 show this gap narrows if the mobiles update the best 5 bands often (i.e., if the base station always has information on the 5 best bands for the band AMC mobiles). Note that UL capacity is consumed based on the frequency of these updates.
Parameter Number of Sectors Carrier Freq & BW Reuse Pattern No. of DL data symbols Propagation model Channel Model & Speed Link to System Mapping Penetration Loss Value 3 (center cell only of 19 cell system) 2.5 GHz, 10MHz 1x3x3 24 (HTTP), 12 (G711 VOIP, 50 users per sector) Erceg B with freq correction: L = [80.4 + 43.75 log (d/100]. d in meters 100% PedB, 3km/hr Exponential Effective SINR Mapping (EESM) 10dB 8dB 2W 1 Tx/2 Rx 90o with 14dBi gain 5.5 dBi 5 dB SIMO (1 data stream) Table 1: Simulation Parameters

FS
MCS 2 Avg. SINR

FD

MCS 1

MCS 0 Frequency

Lognormal shadowing BTS Transmit Power BTS Tx/MS Rx BTS Antenna model MS Antenna Gain MS Noise Figure Technology

Figure 2: Band AMC vs. PUSC MCS Allocation

2) Switching from PUSC mode to band AMC mode We focus on performance of a TDD system with a 10 MHz BW. This BW is divided into 12 bands and a mobile operating in PUSC mode can request a transition into band AMC mode if the following two conditions are satisfied for a configured number of frames (N): (i) mean CINR for the entire BW is above a threshold (CINR-entry); and (ii) the standard deviation of the mean CINR over time for each of the 12 bands is below a threshold (MAX). A mobile requesting transition into band AMC reports the 5 best bands and the associated CINR of these bands to the base. If the base grants the request to transition to band AMC (in which case the mobile would be scheduled only in 1 or more of these 5 bands) then the mobile provides a 1 bit 1 dB differential CINR feedback for these 5 bands at configured intervals. If the set of 5 best bands changes over time then the mobile can potentially report the new best 5 bands to the base station using UL messaging. At configured intervals the mobile can refresh the complete CINR information for the 5 best bands. We refer to the band AMC mode following the above procedure as AMC-P (Practical) as opposed to AMC-I (Ideal) in which the complete CINR is available for all 12 bands (at configured intervals) and the mobile can be scheduled in 1 or more of all the 12 bands. B. SIMULATION RESULTS The system simulation methodology follows the guidelines in [3]. The simulation parameters are listed in Table 1. Figure 3 and Figure 4 show the performance of PUSC and different AMC flavors for the web browsing/HTTP traffic model. We find that AMC-I provides ~ 10% improvement in sector throughput and significant improvement in outage performance compared to PUSC (outage is defined as the percentage of users who are not able to achieve a target data

For G711 VOIP model as the outage plot in Figure 7 shows PUSC performs better than all AMC flavors (an outage criteria of x% with a delay bound of y msec with an FER of z% implies that x% of the users are able to receive (100-z)% of their VOIP packets within the delay bound of y msec). The main reason for this is the fact that with 12 DL data symbols the minimum allocation size in AMC is 192 sub-carriers whereas for PUSC it is only 48 sub-carriers (thus there would be less unused/wasted resources in PUSC).
175 750 users 800 users User Packet Call Tput (Kb ps) 150 750 users
75 0u se rs

PUSC AMC-Ideal AMC-P (10dB CINR entry, N=100) AMC-P (0dB CINR entry, N=10) AMC-Ideal - Ave CINR for 12 bands available every 8 frames - Scheduling on all 12 bands AMC-P (Practical) - CINR entry = MIN - MAX set to high value - Ave CINR and best 5 bands updated every 255 frames
13.5 14 14.5

850 users

125

800 users
80

75 0u

se rs 0u se rs

850 users 850 users 850 users

100

800 users

75 12 12.5 13 Sector Tput (Mbps)

Figure 3: HTTP Sector & User Throughput

30

PUSC
25

AMC-Ideal AMC-P (10dB CINR entry, N=100) AMC-P (0dB CINR entry, N=10) AMC-Ideal - Ave CINR for 12 bands available every 8 frames - Scheduling on all 12 bands AMC-P (Practical) - CINR entry = MIN - MAX set to high value - Ave CINR and best 5 bands updated every 255 frames
800 Number of Users per Sector 850

20 Outage (%)

15

10

0 750

Figure 4: HTTP Outage for 64 Kbps


140 AMC-I 130 For AMC-P

User Packet Call Tput (Kbps)

120

AM

CP

- CINR entry = 0 dB - N = 10 frames - Best 5 bands and average band CINR updated every k frames (k value in plot) - Differential CINR feedback every 8 frames

(k =

110

M C -P

AMC-P (k=50) AMC-P (k=100)

100

90 PUSC 80 12.5 12.75 13 13.25 13.5 13.75 14 14.25

Sector Tput (Mbps)

Figure 5: Effect of Faster Update of 5 best bands on Throughput


26 24 22 20 18 Outage (%) 16 14 12 10 8 6 4 PUSC AMC-P AMC-P (every 255 (every 100 frames) frames) AMC-P (every 50 frames) AMC-P (every 8 frames) AMC-Ideal For AMC-P - CINR entry = 0 dB - N = 10 frames - Best 5 bands and average band CINR updated every k frames (k value in plot) - Differential CINR feedback every 8 frames

Figure 7: VOIP Results A. PUSC 1/3 Configuration In this configuration each sector is assigned 1/3rd of the carrier bandwidth thus effectively creating a 1x3x3 kind of interference scenario and improving the CINR for the cell edge users (without actually using 3 carriers). Even though each sector is assigned only 1/3rd of the available sub-carriers, the set of sub-carriers assigned are still distributed throughout the entire band (for obtaining enough frequency diversity in each sector). When comparing this scheme with a 1x3x1 scheme which uses all the sub-carriers (referred to as a PUSC 1/1 configuration) two possibilities exist for how the transmit power is distributed in the PUSC 1/3 mode - (i) with pwr boost: the transmit power per sub-carrier (power spectral density - PSD) in the PUSC 1/3 mode is three times the PSD in the PUSC 1/1 mode; (ii) no pwr boost: no increase in the PSD per sub-carrier - i.e., sector operating in the PUSC 1/3 mode would use 1/3rd of the transmit power being used by a sector operating in the PUSC 1/1 mode.

25

5)

(k =

8)

Figure 8: Combo DL sub-frame B. Combo-Frame Configuration In this scheme we partition the DL sub-frame into a PUSC 1/1 zone and a PUSC 1/3 zone as shown in Figure 8. The idea is that users with good CINR would be scheduled in the PUSC 1/1 region and users with bad CINR (cell edge users) would be scheduled in the PUSC 1/3 region. This effectively creates a fractional frequency reuse scenario in each sector wherein for some users in the sector (users closer to cell site) the full

Run Configuration

Figure 6: Effect of Faster Update of 5 best bands on Outage

III. FRACTIONAL FREQUENCY REUSE (FFR) In this section we propose two flavors of FFR and study their performance relative to a non-FFR scheme in the context of the PUSC allocation mechanism for a 1x3x1 configuration.

bandwidth (F) is available for scheduling and for other users (users in cell edge) only a limited portion of the bandwidth (Fj) is available as shown in Figure 9. Note that in the PUSC 1/3 region of the combo frame the sector is loosing 2/3rd of the bandwidth (compared to the PUSC 1/1 region) so any potential throughput improvement for cell edge users (achieved because of an improved CINR) would have to weighed in relation to this loss in bandwidth.

C. SIMULATION RESULTS The system simulation methodology follows the guidelines in [3]. The performance results are provided for full buffer/UDP traffic model and web browsing/HTTP traffic model for a mobile WiMaX system with users at different speeds. The simulation parameters are listed in Table 1.
Parameter Number of Sectors Carrier Freq & BW Frequency Reuse Cell Size (site-site) Number of DL OFDM symbols for bearer data Propagation model Channel Model & Speed Value 3 (center cell only of 19 cell system) 2.5 GHz, 10MHz PUSC 1/1. PUSC 1/3 and FFR 1 km 20 Path Loss (dB): L= 126.2 + 36 log (d), d in km 50% PedB @ 3km/hr 30% VehA @ 30 km/hr 10 % VehA @ 60 km/hr 10% VehA @ 120 km/hr EESM 11.5dB 8dB 20W 2 Tx/2 Rx 2x2 open loop MIMO with MMSE receiver and rank adaptation (adapting between 2 stream MIMO and 1 stream Alamouti based on spectral efficiency estimation) Proportional Fair 90o with 15dBi gain -2 dBi 7 dB HTTP and UDP

Figure 9: FFR Illustration of Combo Frame 1) Scheduler for Combo-Frame In the combo frame we need to decide whether a mobile should be scheduled in the PUSC 1/1 zone or the PUSC 1/3 zone. This decision is made every frame based on the computed Spectral Efficiency (SE) for the mobile. The SE is defined as the expected data rate (capacity) achievable per modulation symbol. This is computed with a modified Shannon capacity formula as: SE[x] = A*log2 (1+B*x). In this formula, x stands for the instantaneous CINR (which is fed back by the mobile), A and B are constants used to capture the deviation from the Shannon capacity formula (to account for non-idealities in the coding and decoding operations). A and B are typically set based on a target HARQ error rate performance. A mobile is scheduled in the PUSC 1/3 zone if SE [CINRr3] > 3*SE [CINRr1] where CINRr3 is the CINR achievable if the mobile is scheduled in the PUSC 1/3 region and CINRr1 is the CINR achievable if the mobile is scheduled in the PUSC 1/1 region. The scaling by 3 accounts for the fact that in the PUSC 1/3 zone only 1/3rd of the sub-carriers is used by the sector (compared to PUSC 1/1). Thus we decide to schedule the mobile in the PUSC 1/3 zone if the resulting data rate (with the reduced set of sub-carriers) is higher than what is achievable in the PUSC 1/1 zone. In this study we make a few ideal assumptions to obtain an upper bound on the expected performance improvement with FFR. Firstly we assume that the base station has access to both CINRr1 and CINRr3 from all the mobiles. In a real system the base can have access to only one of these two measurements. Secondly when we are dealing with the RF computation for a mobile in the PUSC 1/3 zone in one sector we assume that all the other sectors are also in the PUSC 1/3 zone during that OFDM symbol time (i.e., mobiles in PUSC 1/3 zone assume perfect co-ordination of the zones in other sectors so that the appropriate r3 SNR metrics can be computed).

Link to System Mapping Penetration Loss Lognormal shadowing BTS Transmit Power BTS Tx/MS Rx Antennas Technology

Scheduler BTS Antenna model MS Antenna Gain MS Noise Figure Traffic Models

Table 2: Simulation Parameters The simulation results are shown in Figure 10, Figure 11, and Figure 12 (the cell edge throughput is defined as the minimum data rate achieved by 95% of users in the sector). For web browsing we see that the sector and user average throughput are not impacted much across the different configurations but the cell edge throughput improves as we go from PUSC 1/3 to PUSC 1/1 to FFR-combo. The reason for PUSC 1/1 performing better than PUSC 1/3 is because of the availability of extra BW (the full 10 MHz vs. 1/3rd of 10 MHz even though the CINR is lower in PUSC 1/1). This becomes important as we add more users in the system. The reason for the FFR-combo improvement is because we are optimally placing the mobiles in the appropriate zone making sure that a mobile is placed in the PUSC 1/3 zone only if the spectral efficiency improvement (because of better CINR) can compensate for the loss of BW in the 1/3 zone.

100

90 Cell Edge Throughput (Kbps)

unlimited data in the buffer (buffer is limited by the offered load in the download/ON periods of the web browsing model and the frequency of the ON periods, [3]). IV. CONCLUSIONS In this paper we presented the design and associated performance results for two potential candidates for improving the performance of a WiMaX system operating under the PUSC sub-channelization mechanism. The conclusions are: For band AMC:

PUSC 1/3 PUSC 1/3 (with pwr boost) PUSC 1/1 FFR (combo) FFR (combo with pwr boost)

80

70

HTTP UDP

60

50

40

30

If the right settings are in place, at slow speeds band AMC performs better than PUSC for the web browsing model (similar results for the UDP model - results not included here due to lack of space). For VOIP PUSC performs better than the AMC flavors studied (AMC might perform better if we reduce the number of DL symbols in the AMC zone of the TDD sub-frame). When compared to PUSC 1/1, we see that the ideal version of FFR-combo offers a cell edge throughput improvement of 20% for the web browsing model and an improvement of 14% for the UDP model. There is no noticeable improvement in the sector throughput and the user throughput. When compared to PUSC 1/3 we see that the ideal version of FFR-combo offers an improvement in all three metrics, with a cell edge throughput improvement of 81% for web browsing and 16% for UDP (the sector throughput improves by 39% for UDP). For good cell edge reliability for the DL control channel in a 1x3x1 deployment (i.e., where the full BW is used in all sectors) it is preferable to send the control message in a PUSC 1/3 zone (instead of a PUSC 1/1 zone). The FFR-combo would be a good choice in this kind of deployment since it can combine both the PUSC 1/3 zone and a PUSC 1/1 zone in the same DL sub-frame to provide a good cell edge coverage and cell edge throughput performance while at the same time not compromising on the overall sector throughput and user average throughput. REFERENCES

Figure 10: FFR Cell Edge Throughput comparison


5.5

5 Sector Throughput (Mbps)

4.5 4

For Fractional Frequency Reuse (FFR):

HTTP UDP

3.5

2.5 PUSC 1/3 PUSC 1/3 (with pwr boost) PUSC 1/1 FFR (combo) FFR (combo with pwr boost)

Figure 11: FFR Sector Throughput comparison


550 500 Average User Throughput (Kbps) 450 400 350 300 250 200 150 100 PUSC 1/3 PUSC 1/3 (with pwr boost) PUSC 1/1 FFR (combo) FFR (combo with pwr boost)

HTTP UDP

Figure 12: FFR User Throughput comparison For the UDP traffic model we see that all three metrics improve as we go from PUSC 1/3 to PUSC 1/1 to FFRcombo. The reason for the sector throughput improvement is because in the UDP model users practically have infinite data to transmit, hence a user with a very good CINR (in PUSC 1/1) can benefit from availability of more bandwidth (since the user can send more data through). This is not the case with the web browsing traffic model since a user doesnt have

[1]

[2]

[3] [4]

IEEE 802.16-2004, IEEE Standard for Local and Metropolitan Area Networks - Part 16: Air Interface for Fixed Broadband Wireless Access Systems, October 2004. IEEE 802.16-2005, Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems Amendment 2: Physical and Medium Access Control Layers for Combined Fixed and Mobile Operation in Licensed Bands and Corrigendum 1 February 2006. 3GPP2, TSG-C.C30-DOAH-20031002-004, 1XEV-DO Broadcast Multicast Service Evaluation Methodology Proposal, June 16, 2004

F. Wang, A. Ghosh, C. Sankaran, P. Fleming, F. Hsieh, and S. J. Benes.. Mobile WiMAX Systems: Performance and Evolution, IEEE Communications Magazine, Oct 2008.

You might also like