Hybrid Strategies for Link Adaptation exploiting several degrees of freedom in OFDM based Broadband Systems

Suvra Sekhar Das1,2, Muhhamad Imadur Rahman1,YuanyeWang1, Flemming B. Frederiksen1, Ramjee Prasad1

1Center for TeleInFrastruktur (CTiF), Aalborg University,Denmark. 2Tata Consultancy Services, India. e-mail: [email protected]

Abstract— In orthogonal frequency division multiplexing spread, average SNR condition, heavily influence the values (OFDM) systems, there are several degrees of freedom in time to be selected for the LA parameters. and frequency domain, such as, sub-band size, forward error con- The complexity increases significantly when LA schemes trol coding (FEC) rate, order, power level, modulation adaptation interval, coding rate adaptation interval and power attempt to achieve the highest possible spectral efficiency by adaptation interval, which can be exploited for link adaptation to trying to optimally adapt so many link parameters, which achieve high spectral efficiency. The variable channel parameters depend on another large set of varying channel conditions. such as delay spread, Doppler frequency spread, average signal Therefore hybrid strategies, i.e. to limit some degrees of to noise ratio and instantaneous channel gains influence the freedom by slowly varying a few parameters, while using link adaptation parameters. Optimal adaptation of the link parameters based on the channel conditions would lead to fast adaptation for other parameters are investigated in this highly complex systems with high overhead. Hybrid strategies work. The objective is to analyze the tradeoff between spectral to vary the adaptation rates to tradeoff achievable efficiency and efficiency loss, complexity and overhead reduction that can be complexity are presented in this work. obtained by the hybrid strategies. Based on the findings, strategies for simplified LA, which I. INTRODUCTION reduce processing complexity but does not compromise on the OFDM systems overcome the impairments of the frequency throughput significantly, are proposed in this paper. Results ob- selective wireless channel elegantly and provides high tained in this perspective are also very important for multi user spectral efficiency [1]. Link Adaptation (LA) techniques are resource allocation, since, the resource unit to be allocated to also known to maximize spectral efficiency [2] by exploit- one user must be such that there is maximum benefit in terms ing the channel variability. LA schemes adapt transmission of overall throughput considering the signaling overhead. This parameters according to the channel conditions so that the work serves as a first step to such systems. maximum data rate is transmitted while keeping the error rate The organization of the rest of the paper is as follows. below the target. Once the values of the adaptable parameters The system model is described in Section II. The hybrid LA are selected they are kept constant over a region in time and strategies are evaluated in Section III. Section IV discusses the frequency domain where the channel is relatively flat. OFDM suggestions on hybrid LA strategies that can be made out of with its fine granularity of the minimum allocation unit as a the investigation done in this work, which is followed by the sub carrier, which experiences flat fading, provides the inherent conclusions in Section V. support needed to exploit the advantage of LA techniques in multiple dimensions [3], [4]. The degrees of freedom that can II. SYSTEM MODEL be exploited by LA techniques increase when it is applied A. System Parameters in OFDM systems and this brings the benefit of improved The parameters used in this work closely match that of the spectral efficiency though it leads to increase in complexity WiMAX [5], [6] system and are given in Table I. of the system. LA involves adaptation of the modulation level (M), the FEC rate (C) and the power level (P) at the B. Frame Structure transmitter as per the the channel state information fed back The frame structure fundamental to LA being considered in from the receiver. When applied in the OFDM framework, this work is given in Fig. 1. The minimum unit over which LA additionally includes selection of adaptation interval for FEC and interleaving is applied is called a block. It is a set of M & C, adaptation interval for P, sub-band size, and choice consecutive sub carriers which spans a successive sequence of of bit–power loading algorithm. Other than the fast fading of OFDM symbols over a period of 0.5ms. The sub-band size is the channel gains, variation of the channel parameters such as defined by the number of consecutive sub carriers that make the root mean square (rms) delay spread, Doppler frequency the block. The modulation level, code rate and power level

1-4244-0264-6/07/$25.00 ©2007 IEEE 1807 TABLE I Several adaptation rates are considered in this work. It is SYSTEM PARAMETERS possible to fit in all values into a FDD framework however for Parameter Value TDD setup only some of values are applicable. The signal to Carrier Frequency 3.6GHz noise ratio (SNR) of the data block is fed back as the channel Bandwidth 5MHz Sampling Frequency 5.714MHz quality indicator. The framework is described in Fig. 2. Sub Carrier Spacing 11.16 MHz At the beginning of each adaptation window, the measured Rms delay spread 0.5µs–2µs CQI for each block is sent back to BS. The fed back CQI is FFT size 512 used to decide on the LA method. Useful Symbol period 89.6µs Cyclic Prefix 11.2µs Pilot Symbol Pilot Symbol Pilot Symbol 1 1 2 Code rates 3 , 2 , 3 Modulation 4-QAM, 16-QAM, 64-QAM Target Block error rate (BLER) 10−1

Link Adaptaed Data Symbols Link Adaptaed Data Symbols

CQI Feedback

Channel E stimate at R eceiver

Fig. 2. Link Adaptation Model

III. HYBRID LA STRATEGIES Hybrid LA strategies mean methods which do not exploit all degrees of freedom simultaneously. As mentioned, several parameters can be adapted, namely, modulation level (M), the FEC rate (C), power level (P), adaptation interval for M & C, adaptation interval for P, sub-band size and choice of bit–power loading algorithm. Different combination of slow and fast adaptation can be made between these parameters so that only few parameters are adapted instantaneously using immediate channel gains, while others are adapted statistically, i.e. using average information such as rms delay, Doppler, average SNR. These situations are analyzed here.

A. Bit and Power Loading Algorithms There can be some variation on the bit and power loading algorithm used for LA as discussed below. Fig. 1. Link adaptation frame structure. 1) APMC: will adapt power, modulation and coding rate all together [4], [7]. It uses iterative procedure to distribute power, and find bit loads, whose details are in [3]. is applied at the block level in this work. The block size can 2) AMCFP: This case considers fixed power, i.e the itera- be made to vary in the frequency domain, by changing the tions for bit and power loading are not used, which makes it sub band size. The modulation and coding level is adapted very simple and fast. once every adaptation interval, which can be multiple of the 3) AMC adapt P: In the AMCFP algorithm above, since block duration. Adaptation interval is the period in time over power is fixed, power is wasted as the SNR threshold required which the modulation and coding remains unchanged and for the M& C rate selected for a block is chosen to be less than power adaptation interval is the period between each power the available SNR. Therefore to save this power, the transmit adaptation. power level can be adjusted so that the received SNR at the block is maintained just at the threshold required to maintain C. Link Adaptation Set up the target error rate. The channel state information (CSI) is needed at the Base 4) Performance Comparison: The throughput comparison Station (BS) to use LA in the forward transmissions. In of the three algorithms mentioned above is shown in Fig. 3. Frequency Division Duplex (FDD) system the User Equipment In the figures in this paper, the legend ’u’ is used for ’micro (UE) can feedback this information over a different frequency seconds’ which refers to the rms delay spread of the channel. band. In Time Division Duplex (TDD) mode, either the UE In all cases the AMCP has the best throughput. It is followed feeds back the information or the BS can measure the CSI from by AMCFP, while AMC adapt P has the worst performance. the reverse link transmission considering channel reciprocity. AMCFP has almost similar performance to AMCP, when there

1808 7 x 10 0 2.5

APMC, rms d.s. 0.5 us, 20kmph −1 AMC fixP, rms d.s. 0.5, 20kmph AMC adaptP, rms d.s. 0.5, 20kmph 2 −2 APMC,rms d.s. 2 us, 70kmph AMC fixP,d.s. 2 us, 70kmph AMC adaptP,d.s. 2 us, 70kmph −3

1.5 −4

−5 APMC,2 us, 200 kmph,subN=8 AMC fixP,2 us, 200 kmph, subN=8 1 Saved power in dB −6 AMC adaptP,2 us, 200kmph, subN=8

Throughput (bits/sec) AMC fixP,0.5 us, 20 kmph, subN=8 AMC adaptP,0.5 us, 20 kmph, subN=8 −7 APMC,0.5 us, 20 kmph, subN=32 0.5 AMC fixP,0.5 us, 20 kmph, subN=32 −8 AMC adaptP,0.5 us, 20 kmph, subN=32

−9 5 10 15 20 25 30 0 5 10 15 20 25 30 SNR in dB SNR in dB

Fig. 3. Throughput comparison of different Link adaptation algorithms at Fig. 4. Power utilization comparison of different Link adaptation algorithms different rms delay spread and Doppler condition for sub-band size of 8 sub at different rms delay spread and Doppler condition for sub-band size of 8 carriers and 32 sub carriers

7 x 10 is high diversity in the channel condition, i.e. large Doppler 2.5 trms=half,velocity=20kmph,subN=8 and rms delay spread, but it has notable performance loss trms=half,velocity=20kmph,subN=32 in case of low Doppler and low rms delay spread condition. trms=half,velocity=20kmph,subN=128 trms=half,velocity=20kmph,subN=512 Therefore it can be suggested that for low Doppler and low 2 trms=2us,velocity=200kmph,subN=8 rms delay spread condition, ACMP be used, whereas when trms=2us,velocity=200kmph,subN=128 the diversity in the channel increases, it is better to use AMCFP. Finally it can be suggested to combine selection of 1.5 bit and power loading algorithm at a very slow rate (based on statistical measure) along with fast(instantaneous) adaptation of modulation and coding rate. Throughput 1 5) Power Utilization: Fig. 4 shows the power utilization of the different algorithms. A low power utilization means low power transmission. This in turn means low interference con- 0.5 dition in multi cellular scenario, where aggressive frequency re-use is followed. In such a scenario, the algorithm which 0 has the lowest power utilization, may be the best one use. In 5 10 15 20 25 30 this viewpoint the algorithm which brings down the transmit SNR in dB power to meet the threshold of the received SNR, but avoids Fig. 5. Throughput performance of different sub-band sizes for different rms iterative power distribution seems to have the best performance delay spread, Doppler velocity. under all channel conditions. In most of the analysis presented in this work, the fixed power algorithm will be used unless mentioned. by 512. At high velocity and high rms delay spread i.e. B. Sub band Size small coherence bandwidth and small coherence time, the sub- In this part the influence of rms delay spread and band size of 8 sub carriers has very similar performance as the Doppler frequency spread on different sub-band sizes that of sub-band with 128 sub carriers. Therefore it can be (8,32,128 and 512 carriers) are investigated. Fig. 5 shows the concluded that for very high velocity and high rms delay throughput for LA system when M&C are adapted every 2ms spread condition, it is better to use a large sub-band size since while keeping P fixed. It can be seen from the figure that, it will use significantly low overhead, while on the conditions when rms delay spread is small i.e. coherence bandwidth is of low velocity and small rms delay spread, it is better to use large, and velocity is low, i.e. coherence time is large, the small sub-band size. Finally the sub band size selection can sub-band size of 8 sub carriers has the highest throughput. be a statistical adaptation in combination with instantaneous It is followed by sub-band size of 32 which is followed adaption of modulation and coding rate.

1809 TABLE II C. Fixed Coding rate AVERAGE SNR THRESHOLDS (IN DB) FOR SWITCHING CODING RATE FOR In this part comparison is made on the performance of DIFFERENT RMS DELAY SPREAD AND DOPPLER CONDITION systems when power is fixed, while both modulation and coding is adapted for each frame, against the case when coding 0.5µs, 20kmph 2µs, 20 kmph 2µs, 200 kmph Code rate rate is kept fixed while only modulation is varied. Using more subN↓ 1 1 2 1 1 2 1 1 2 3 2 3 3 2 3 3 2 3 than one coding rate simultaneously for a user means that the 8 – 11.5 15.5 – 14.5 24 – 19.5 30 UE needs multiple FECs and decoders, which would increase 32 – 13.5 21.5 – 20.5 N.A. – 20 N.A. the complexity prohibitively. Therefore using only a single 128 – 21.5 N.A. – 21.5 N.A. – 20 N.A. 512 – 23.5 N.A. – 22 N.A. – 22.5 N.A. FEC coder / decoder (i.e only one FEC rate) for one user is highly desired. Then, selection of the FEC code rate becomes 7 x 10 very important which is discussed in this section. Fig. 6 shows 2.5 the throughput comparison for sub-band with 8 sub carriers. LA per 0.5ms, subN=8 In the figure ‘RateAdapt’ means adaptive modulation and LA per 2ms, subN=8 2 LA per 5ms, subN=8 coding simultaneously. The performance of fixed coding rate is LA per 0.5ms, subN=512 not far from the optimal adaptive modulation coding scheme. LA per 10ms, subN=512 LA per 0.5ms, subN=32 The average SNR along with rms delay spread and Doppler LA per 2ms, subN=32 1.5 velocity can be used to choose the threshold for switching LA per 5ms, subN=32 from one coding to another so that performance is close to being optimal. Similar behaviour has been found for different sub band size. The average SNR values for switching from 1 one coding rate to another is given in Table II. The mark ’-’ Throughput indicates that coding rate is the default coding rate to start with, while the SNR values indicate the starting average SNR 0.5 from where the particular coding rate can be used and ’NA’ indicates the corresponding coding rate not be used. It can be 0 concluded that if small sub-band size is used then all coding 5 10 15 20 25 30 SNR in dB rates are important, but when large sub-band size is selected then coding rates ’1/3’ and ’1/2’ are enough. Fig. 7. Throughput comparison for different adaptation rates, for rms delay 7 x 10 spread of 0.5µsat20kmph 2.5 RateAdapt,trms=half,velocity=20kmph,subN=8 BitAdapt,trms=half,velocity=20kmph,subN=8,rate=2by3 BitAdapt,trms=half,velocity=20kmph,subN=8,rate=1by3 RateAdapt,trms=2us,velocity=20kmph,subN=8 the earlier case for sub band size of 8 sub carriers and also 2 BitAdapt,trms=2us,velocity=20kmph,subN=8,rate=2by3 BitAdapt,trms=2us,velocity=20kmph,subN=8,rate=1by2 to some extent for subN=32. When sub band size is made BitAdapt,trms=2us,velocity=20kmph,subN=8,rate=1by3 large there is little impact on the adaptation time interval; i.e. RateAdapt,trms=2us,velocity=200kmph,subN=8 short term adaptation in time domain is not necessary when 1.5 BitAdapt,trms=2us,velocity=200kmph,subN=8,rate=1by2 BitAdapt,trms=2us,velocity=200kmph,subN=8,rate=1by3 the adaptation window is large in frequency domain. From results for large rms delay spread and velocity, (figure not

1 presented here due to space limitation) it is found that for low and moderate average SNR levels a large sub-band size can be good enough since efficiency loss is not quite high.

Throughput (bits/sec) 0.5 E. Power Adaptation Rate Now we consider that modulation and coding rate are

0 adapted together but at a different rate than the power adap- 5 10 15 20 25 30 SNR in dB tation rate. In this way a tradeoff can be achieved between performance and complexity. The power adaptation is done Fig. 6. Throughput performance comparison for fixed coding with adaptive by using a one bit feedback information on whether transmit modulation Vs adaptive modulation and coding for sub-band size of 8 sub carriers. power has to be increased or decreased. Fig. 8 shows the important effect of doing power adaptation at a different rate compared to the modulation and coding adaptation. The case D. M & C Adaptation Rate with LA per 10ms with fixed power is taken as the base line In this section, the power is kept constant, while modulation throughput curve. In the figure, LA per x ms implies, the and coding are adapted. Fig. 7 shows the impact of changing interval of adaptation of modulation and coding is x, while the adaptation rate for rms delay spread of 0.5µs and 20 kmph. P LA means power adaptation interval. It may be said that There is a large impact of the decreased adaptation rate as in LA interval can be reduced to 5ms with P-LA interval as 1ms

1810 10 LA per 10ms, P LA per 5ms coding rate selection, sub band size can be made at a very LA per 10ms, P LA per 2ms 9 slow rate using statistical channel parameters while modulation LA per 10ms, P LA per 0.5ms LA per 5ms, P LA per 1ms can be adapted at a medium rate using instantaneous channel 8 LA per 5ms, P LA per 0.5ms gains if fast power control is applied in time domain. The LA per 2ms, P LA per 1ms selection of the values for statistical adaptation depend heavily 7 LA per 2ms, P LA per 0.5ms LA per 1ms, P LA per 0.5ms on the rms delay spread, Doppler and average SNR values.

6 Some guidelines can be as follows. APMC should be used at low rms delay spread and Doppler condition, while AMCFP 5 is suggested for high rms delay spread and high Doppler frequency spread situations, but their performance with respect 4 Normalized throughput to AMC adapt P needs to be analyzed in cellular scenarios 3 using aggressive frequency reuse. Coding rate 1/3 and 1/2 are expected to be used most often while 2/3 is used only for 2 small sub band size. Coding rate selection can be based on

1 the average SNR condition for a particular channel condition. 5 10 15 20 25 30 SNR in dB Sub band size can be made small for low mobility and small delay spread while large sub band size is to be used for large variation of the channel. Power control is found to be very Fig. 8. Throughput comparison for different power adaptation rates, for rms effective, since modulation and coding can be made to vary at delay spread of 0.5µs at 20 kmph for sub band size of 8 sub carriers a medium rate, which simplifies LA implementation and the overhead needed. without much decrease in throughput. This kind of systems V. C ONCLUSION will be simpler than fast modulation and coding adaptation. In this work, several aspects of link adaptation in OFDM Therefore it can be suggested that fast power adaptation is systems have been presented. It is seen that there are a large a viable option for implementation of LA since it does not number of options to maximize throughput in these systems, sacrifice efficiency while minimizes complexity and overhead. but they are very complex since it involves several parameters. Similar nature of the curves has been found at other channel Using hybrid strategies, i.e. using a combination of slow and conditions. fast adaptation of different parameters can simplify the LA process while there is little impact on the spectral efficiency F. Overhead calculation performance. These results are expected to trigger a lot of For uplink (CQI feedback), 5 bits for one sub-band will be investigation in resource allocation algorithms which considers used. For power adaptation only 1 bit is needed to indicate LA. whether the power should go up or down. For downlink, i.e. to indicate the modulation and coding rate VI. ACKNOWLEDGEMENT to be used, 4 bits for each sub band will be needed to indicate The authors are grateful to TCS India for funding the up to 24 =16different rates. The overhead is presented in project, to the RATE section members of Aalborg University Table III. for delightful discussions especially to Mr. Sanjay Kumar for helping in preparing the manuscript. TABLE III OVERHEAD IN MBPS FOR ADAPT POWER LA REFERENCES [1] Prasad R., OFDM for Wireless Communications. Artech House Publish- TLA 1ms 2ms 5ms ers, 2004. TPA 0.5ms 0.5ms 1ms 0.5ms 1ms [2] Chung S. T., Goldsmith A. J., “Degrees of freedom in adaptive modula- subN=8 0.64 0.384 0.32 0.2302 0.1664 tion: a unified view,” IEEE Trans. Commun., vol. 49, no. 1, pp. 1561– subN=32 0.16 0.096 0.08 0.0575 0.0416 1571, Sep. 2001. subN=128 0.04 0.024 0.02 0.0144 0.0104 [3] Das S.S., M.I. Rahman et al., “Influence of PAPR on Link Adaptation subN=512 0.01 0.006 0.005 0.0036 0.0026 Algorithms in OFDM Systems,” in IEEE VTC Spring’07, Dublin, Ireland, 22-25 April 2007. [4] Toyserkani A. T. et al., “Sub-carrier based Adaptive Modulation in HIPERLAN/2 System,” in IEEE International Conference on Commu- IV. GUIDELINES DISCUSSION nications, 2004 , June. Some channel parameters such as average SNR, rms delay [5] “Mobile WiMAX Part I:A Technical Overview and Performance Evalu- ation,” WiMax Forum,” Techinal Report, June 2006. spread and Doppler frequency spread can be classified as sta- [6] , Tech. Rep. tistical measure, while channel gains for each sub band can be [7] Lei M., et al., “An Adaptive Power Distribution Algorithm for Improving considered as instantaneous measure. Adaptation of the system Spectral Efficiency in OFDM,” IEEE Trans. Broadcast., vol. 50, no. 3, pp. 347 – 351, Sep. 2004. parameters can be called statistical (slow) of instantaneous (fast) as per their dependence on the channel parameters. It can be said that bit and power loading algorithm selection,

1811