Throughput Enhancement for MIMO OFDM using Frequency Domain Channel Length Indicator and Guard Interval Adaptation

Transcription

1 Throughput Enhancement for MIMO using Frequency Domain Channel Length Indicator and Guard Interval Adaptation Marco Krondorf Technische Universität Dresden Vodafone Chair Mobile Communication Systems Dresden, Germany. Gerhard Fettweis Technische Universität Dresden Vodafone Chair Mobile Communication Systems Dresden, Germany. Abstract In this paper a novel scheme for guard interval adaptation is presented, that uses the Frequency Domain Channel Length Indicator (FCLI) algorithm for estimating the length of the channel impulse response (CIR). In contrast to other schemes, our method does not require any time averaging to estimate the CIR length and is applicable in SISO and MIMO, with and without spectrum mask. By estimating the CIR length in an ad hoc manner our results show, that it is possible to instantaneously react on changing propagation environments and to significantly increase throughput without decreasing detection performance. I. INTRODUCTION In todays wireless LANs and future mobile communication systems, orthogonal frequency division multiplex () is used as physical layer concept. In the system bandwidth is divided into a number of orthogonal narrow band subchannels. The simple frequency domain equalization using an FFT operation gives rise to low cost terminals, highly suitable for consumer electronics and mobile applications. One drawback of current systems is the loss of spectral efficiency due to guard interval (GI) insertion. The GI contains the cyclic extension of the transmit data to preserve subcarrier orthogonality and to prevent inter symbol interference (ISI). There are some approaches to increase the spectral efficiency of. In [9] a guard interval free system is presented, that uses guard intervals only at the beginning and the end of a block of symbols. In addition, perfect channel knowledge is assumed. Without any complex decision directed interference cancellation algorithm at the receiver the throughput of such systems is poor due to SNR loss and decreased packet error rate. With respect to the instantaneous CIR length (L), using an adaptive GI preserves the orthogonality of the subcarriers and reduces the idle time of the system to a minimum. In order to instantaneously adapt the GI to a changing CIR length, a suitable CIR length estimation scheme is required. In [] a CIR length estimation method called NCLE (Noise Variance and CIR Length Estimation) is presented, that performs well in continuous streaming systems by long time averaging of the estimated CIRs. Especially in WLAN systems, the long acquisition time of this method makes it impracticable when considering the bursty nature of packet switched traffic. Furthermore, the NCLE method is not applicable to systems using a spectrum mask. The spectrum mask defines a set of null-subcarriers to reduce leakage at the spectral edges. In order to avoid these problems, we choose an alternative joint CIR length and channel estimation method called FCLI [7]. FCLI finds the CIR length that minimizes the mean squared channel estimation error (MSE), with respect to current noise power. Therefore FCLI intentionally does not even try to estimate the correct CIR length, but the effective length with respect to noise power by omitting low power channel taps. When adapting the GI length based on the FCLI CIR length estimation, residual ISI and ICI would degrade the system due to GI mismatch. The results of this paper show, that this effect is compensated due to much better MSE performance of FCLI by even increased BER/PER performance compared to standard channel estimation and fix GI. Furthermore we compare our approach with two alternative methods for GI adaptation given in [5] and []. The paper is organized as follows. In Sec.II the MIMO system model is presented. Moreover, we give a short revision of MIMO channel estimation and introduce the main idea of FCLI. Our GI adaptation scheme together with the mentioned alternative approaches can be found in Sec.III. Simulation results are presented in Sec.IV and finally we conclude in Sec.V II. SYSTEM MODEL The used MIMO- System Model is shown in Fig.1. It uses N T transmit and N R receive antennas, represented by indexes t and r. Generally, the system divides the system bandwidth into N C narrow band subcarriers. In order to perform size N FFT FFT/IFFT operations for modulation and demodulation, an oversampling factor of N FFT /N C is used. Hence, there are N FFT N C carriers at the spectral edges, forming some kind of spectrum mask mainly to reduce outer band radiation. When perfect time and frequency synchronisation is assumed at receive antenna r, the i-th

2 received frequency domain symbol vector 1 Y i,r of size [N C 1] can be written as follows N T Y i,r = X i,t H r,t + N i,r (1) t=1 where X i,t denotes the i-th [N C N C ] diagonal matrix containing the frequency domain data symbols of transmit antenna t. Furthermore, N i,r denotes the i-th AWGN vector at receive antenna r while H r,t represents the [N C 1] vector of frequency domain channel coefficients between antennas r and t. The channel coefficient at subcarrier position n is given by H r,t (n) = h r,t (k)e jπ nk N FFT () k=0 where L represents the length of the corresponding channel impulse response h r,t. Applying inverse FFT operation to Eq.(1) and using convolution operator, the i-th time domain symbol at receive antenna r takes the form N T y i,r (k) = x i,t (k) h r,t (k)+n i,r (k) (3) t=1 with N G k N FFT 1, sample index k and x i,t (k) = 1 N FFT N C/ 1 n= N C/ X i,t (n)e jπ nk N FFT () The parameter N G determines the length of cyclic prefix, chosen such, that L > N G to prevent ISI and preserve subcarrier orthogonality. For further details on please see []. Coding Modulation Antenna Mapping... Fig N T H N R MIMO System Model 1 I I... I Decoding Demodulation Antenna Demapping A. Channel Estimation Scheme In this section a short review of MIMO channel estimation in given, mainly to set the scene for FCLI joint channel and CIR length estimation. Therefore we only focus on the most common pilot aided channel estimation scheme: Time Domain Least Squares (TDLS) Channel Estimation using frequency orthogonal (FO) pilot data due to its low computational complexity [7]. In FO preamble data, each Txantenna uses a different subset of pilot carriers simply to avoid inter pilot distortion at receiver side. Using such a FO pilot 1 Notation: Upper (lower) letters will be generally used for frequencydomain (time-domain) signals; boldface letters represent matrices and column vectors; letters with both boldface and underline represent block matrices or vectors in MIMO. grid consisting of N PC pilot carriers requires interpolation of the N C N PC missing coefficients after initial frequency domain channel estimation. This can be done via time domain FFT based interpolation. The initially N PC estimated channel coefficients are obtained by omitting symbol index i and ˆḢ r,t = 1 σx Ẋ H t Ẏ r (5) In Eq.(5), the dotted notation shows the dimensional reduction of the related vectors/matrices from N C to N PC. Simultaneously we observe, that N T N R MIMO channel estimation can be subdivided into a set of N T N R channel estimation problems due to FO pilot design. Therefore we omit parameters r and t in the next sections. For FFT based interpolation, CIR ĥ is calculated from ˆḢ using dimensional reduced [N PC L] Fourier matrix Ḟ NPC,L having elements ( [Ḟ] n,k = exp jπ nk ) () N FFT with 0 k L 1 and n P t where P t denotes the index set of the N PC pilot carriers of Tx-antenna t. After CIR calculation, zero padding is performed in the interval [L...N FFT 1] to obtain full channel vector Ĥ via FFT. Thus, using [N C L] Fourier matrix F NC,L according to Eq.() and pseudo inverse matrix (.) we have Ĥ L = F NC,Lĥ with ĥ = Ḟ N PC,L ˆḢ (7) Substituting Eq.(5) into Eq.(7) and assuming σx =1we get Ĥ L = F NC,LḞ N PC,LẊH Ẏ () From these equations it is easy to observe that knowledge of CIR length L is necessary for channel estimation. Since L is a generally unknown parameter, it is typically approximated by using the fixed GI length of the system. B. FCLI joint channel and CIR length estimation Channel estimation performance in terms of mean squared channel estimation error (MSE) is highly related to CIR length L [7]. The basic idea of FCLI is to estimate MSE e (ˆL g ) using some ˆL g out of a given set of CIR length hypotheses with g [1...G], where MSE e (ˆL g ) denotes the estimated MSE when using a certain CIR length hypothesis ˆL g for channel estimation. Therefore, a priory knowledge of noise power σn is used as shown in pseudo code table below. The whole procedure can be summarized as follows: ˆL = arg min ˆL g {MSE e (ˆL g )} (9) Using matrix A L = F NC,LḞ N PC,L and exploiting a set of G CIR length hypotheses [ˆL 1 <... < ˆL G ], FCLI can be described in pseudo code notation as follows

3 (1) init: σ N, Y, [ˆL 1 <...< ˆL G], [A ˆL1,...,A ˆLG ] () for i=g...1 Ĥ i= Channel Estimation(Y,ˆL i) EĤ(ˆL i) = sum( Ĥi ) MSE n(ˆl i)=σ N trace{a ˆLi A HˆLi } end () ˆL = arg min ˆL g MSE e (ˆL i)=eĥ(ˆl G) EĤ(ˆL i) MSE n(ˆl G) +MSE n(ˆl i) {MSE e (ˆL g)} (5) Ĥ = Ĥ g where ˆL = ˆL g It should be noted that the estimated CIR length is not necessarily the maximum length of the actual CIR, but the length of the CIR minimizing the MSE of the channel estimate. For further details on FCLI and a detailed mathematical description please see [7]. III. GUARD INTERVAL ADAPTATION Guard interval adaptation is an option for throughput enhancement by limiting the GI to the minimal necessary amount, without performing any complex signal processing at receiver side. Basically, GI Adaptation is done by FCLI channel and CIR length estimation using preamble P at the beginning of each block as shown in Fig.. In order to transfer back the estimated CIR length to the transmitter, an additional feedback channel has to be used. When receiving the CIR length estimation, the transmitter adapts the GI in the next block, directly incorporating the CIR length estimate. In order to react on channel variations, the following criterion should be fulfilled: N B (T Sym + T GI ) < T C, where T C represents the channel coherence time, T GI the GI duration, T Sym the symbol time and block size N B. blocks are formed of N B serial concatenated data symbols headed by a group of preamble symbols for Rxsynchronization and channel estimation. Especially in low SNR, FCLI intentionally underestimates the CIR length to minimize MSE. Hence, adapting the GI accordingly gives rise to some residual ICI and ISI because of GI mismatch that potentially reduces detection performance quite contrary to the performance increase provided by FCLI MSE reduction. Time P Block P Block P CIR Length Estimation and GI Adaptation Fig.. CIR Length Estimation and GI Adaptation GI adaptation procedure A. GI Adaptation in MIMO In MIMO systems, the same GI adaption procedure can be applied in order to increase spectral efficiency and to save transmit power. In this paper we assume a single user MIMO link according Fig.1. As a result, all N T Tx-antennas transmit their symbols in a coherent manner, all using the same GI length, mainly to simplify Rx-synchronization. Distributed MIMO multi-user multi-base station systems are not addressed here. The single user MIMO link gives rise to the assumption that all N T N R channels have the same power delay profile but are independed in terms of individual channel realisations. Joint channel and CIR length estimation yields a set of N T N R GI adaptation propositions, one for each individual channel. In order to coherently transmit all N T symbols, we have to calculate a resulting GI length out of all proposals. Therefore, we choose the simple averaging method L GI = mean{ˆl 1,1,..., ˆL NT,N R } () where ˆL t,r are the FCLI-based CIR length estimates used for GI adaptation between antennas r and t. In order to study the potential of GI adaptation techniques, a perfect signalling is assumed to propagate L GI to the mobile terminals. Nevertheless, superposed signalling as studied in [] seems to be an attracktive candidate for this task. B. Alternative Approaches for GI Adaptation In order to compare our FCLI-based GI adaptation approach with other methods, we shortly introduce two alternative schemes given in [5] and []. In [5] a criterion in terms of required SINR γ req is used as follows T GI ˆτ log (A/B) (11) where A = { ˆτ γ req + 1 } { ( ) ( ) } ˆτ ˆτ + (1) N FFT N FFT N FFT and using instantaneous SNR γ we have B =1 γ req with γ req <γ (13) γ The parameter ˆτ represents the estimated maximum excess delay of the channel in terms of time domain samples and is given by l=0 ˆτ = ĥ(l) l (1) l=0 ĥ(l) In [5] it is not expressed how to estimate ĥ(l) and its length L. Therefore we assume a standard TDLS channel estimation using FO pilots and fixed GI length L-approximation. The approach in [] stated T GI =ˆτ RMS (15) where ˆτ RMS denotes the estimate of the channels RMS spread as follows ˆτ RMS = ( ) l=0 ĥ(l) l l=0 ĥ(l) l (1) l=0 ĥ(l) l=0 ĥ(l)

4 As seen before, we assume conventional TDLS channel estimation and fixed GI length L-approximation. IV. SIMULATION RESULTS FCLI joint channel and CIR length estimation together with GI adaptation provides a significant increase of throughput especially in situations, when the CIR length is smaller than a certain fixed GI. Simulation results show, that residual ISI and ICI do not cause any performance (BER/PER) loss in comparison to standard using fixed GI and standard channel estimation techniques. In this paper two standard WLAN systems are investigated: IEEE 0.11a (0 MHz, SISO) and WIGWAM (0 MHz, MIMO, Home Office scenario) [3]. The PHY parameters used in simulation are given in Tab.I. Parameter 0.11a WIGWAM HO Bandwidth 0MHz 0 MHz Data Carriers 1 FFT Size Sampling Time 50 ns.5 ns Guard Interval 00 ns 00 ns N B 50 Modulation QPSK QPSK Conv. Coding r=0.5 r=0.5 Packet Size 1 sym. 1 sym. ARQ none none Long Training Sym. N T Terminal Velocity static static TABLE I PHY PARAMETERS USED IN SIMULATION We choose the V-BLAST scheme for MIMO transmission, performing linear MMSE detection [1] at receiver side due to its good tradeoff between performance and computational complexity. Besides system throughput (in Bit/Sec) we use averaged idle time t idle% in percent to compare GI adaptation schemes and standard conform using fixed GI. Averaged idle time is given by { } T GI t idle% = E 0 (17) T Sym + T GI In Simulation three frequency selective indoor fading channels are used [] : IEEE 0.11n Channel Model D (L = 00 ns), IEEE 0.11n Channel Model E (L = 00 ns) and IEEE 0.11n Channel Model C (L = 300 ns). A static channel is assumed during one block. For FCLI algorithm and GI adaptation procedure we use a set of CIR length hypotheses in terms of time domain samples as follows: for IEEE 0.11a [1, 1, 1,,,,, ]; for WIGWAM HO [1, 1, 90, 0, 70, 0, 50, 0, 30, 0, ]. In Fig.5, Fig. and Fig.7 throughput performance of GI adaption schemes and standard WIGWAM is presented. For maximum excess delay GI adaptation mathod given in [5] we choose γ req =5.5dB in order to further reduce idle time. In Fig. it is shown that there is no BER performance loss due to FCLI GI adaptation. In Fig. and Fig.9 we choose γ req =db Fig. 3. Comparison of standard 0.11a with FCLI GI adaptation method, 11n Channel Model D BER a fixed GI, TDLS, 11nD Channel,QPSK adaptive mode, 11nD Channel,QPSK 0.11a fixed GI, TDLS, 11nE Channel,1 QAM adaptive mode, 11nE Channel,1 QAM Eb/N0 in db Fig.. BER of standard 0.11a and FCLI GI adaptation method, 11n Channel Model D/E with RMS spread GI Adaptation with Max Excess Delay GI Adaptation Fig. 5. Comparison of standard WIGWAM V-BLAST MIMO with FCLI GI adaptation method, 11n Channel Model C

5 Idle Time FCLI GI Adaptation Idle Time RMS Spread GI Adaptation Idle Time Max Excess Delay Spread GI Adaptation with RMS spread GI Adaptation with Max Excess Delay GI Adaptation Idle Time in % Fig.. Comparison of standard WIGWAM V-BLAST MIMO with FCLI GI adaptation method, 11n Channel Model D with RMS spread GI Adaptation with Max Excess Delay GI Adaptation Fig. 7. Comparison of standard WIGWAM V-BLAST MIMO with FCLI GI adaptation method, 11n Channel Model E Idle Time in % 1 1 Idle Time FCLI GI Adaptation Idle Time RMS Spread GI Adaptation Idle Time Max Excess Delay Spread GI Adaptation SNR in db Fig.. Comparison of idle times of GI Adaptation Schemes V-BLAST MIMO, 11n Channel Model D SNR in db Fig. 9. Comparison of idle times of GI Adaptation Schemes V-BLAST MIMO, 11n Channel Model E V. CONCLUSIONS We show that FCLI-based guard interval adaptation is superior to other GI adaptation methods and conventional fixed GI in terms of system throughput. This holds especially in low SNR and in situations when there is a big difference between mobile channel maximum excess delay and a certain fixed GI length. Our results show that our approach enables conventional systems to significantly increase throughput and spectral efficiency without utilizing computational complex signal processing techniques. REFERENCES [1] R. Böhnke V. Kühn K.D. Kammeyer D. Wübben, J. Rinas. Efficient Algorithm for Detecting Layered Space-Time Codes. In Proc. th International ITG conference on Source and Channel Coding, Berlin, January 00. [] IEEE /90r. IEEE P0.11 Wireless LANs, TGn Channel Models. January 00. [3] Ralf Irmer and Gerhard Fettweis. WIGWAM: System Concept Development for 1 Gbit/s Air Interface. In Proc. WWRF 05, San Diego, USA, July 005. [] G. Fock M. Speth, S.A. Fechtel and H. Meyr. Optimum Receiver Design for Wireless Broad-Band Systems Using -Part I. In Proc. IEEE Transactions on Communications, volume 7, November [5] E.D. Carvalho R. Prasad S.S. Das, F.H.P. Fitzek. Variable Guard Interval Orthogonal Frequency Division Multiplexing in presence of Carrier Frequency Offset. In Proc. GLOBECOM, 005. [] W. Rave T. Deckert and G. Fettweis. Superposed Signaling Option for Bandwidth Efficient Wireless LANs. In Proc. WPMC, Abano Terme, Italy, September 00. [7] M. Krondorf T.-J. Liang and G. Fettweis. Improved Channel Estimation for complexity-reduced MIMO- Receiver by Estimation of Channel Impulse Response Length. In Proc. European Wireless Conference, April 00. [] Hans-Peter Kuchenbecker Van Duc Nguyen and Matthias Pätzold. Estimation of the Channel Impulse Response Length and the Noise Variance for Systems. In Proc. VTC Spring, 30.May-01.Jun 005. [9] Y. Wu X. Wang, P. Ho. Robust Channel Estimation and ISI Cancellation for Systems With Suppressed Features. In IEEE Journal on selected Areas in Communication, volume 3, May 005. [] L. Lai Z. Zhang. A novel transmission scheme with lengthadaptive Cyclic Prefix. In Journal of Zhejiang University SCIENCE, November 003.