Experimental Broadband OFDM System

Field Results for OFDM and OFDM with Frequency Domain Spreading
Vijay Nangia, Kevin L. Baum
Motorola Labs – Communication Systems Research Laboratory 1301 E. Algonquin Road, Schaumburg, IL 60196 USA
Abstract— This paper presents link performance results of a series of field experiments conducted for OFDM and spreadOFDM/MC-CDMA in the context of a fourth-generation cellular system (4G) downlink. The experimental system is a 20 MHz 2x2 MIMO-OFDM system operating at a carrier frequency of 3.676 GHz. Measurements were conducted in a moderate density urban and suburban setting outside Chicago, Illinois which exhibits varying degrees of delay spread and signal to noise ratio, with vehicle speeds ranging from 0 to 65 miles per hour. Coded performance (r=1/2 64-state convolutional coding) is compared for various link configurations: no receive diversity, 2-branch diversity, Alamouti transmit diversity, and 2x2 MIMO for QPSK, 16-QAM, and 64-QAM constellations. Results indicate that OFDM typically outperforms spread-OFDM, especially for higher order modulations. Moreover, the performance of dualstream MIMO is sensitive to the channel condition number. Keywords—OFDM, SOFDM, MC-CDMA, Alamouti, MIMO, Experimental 4G System

are directly applicable to a wideband OFDM system. It is well known that OFDM requires channel coding and interleaving across subcarriers to exploit the frequency diversity of the channel [11]. SOFDM is a transmission method that combines spreading and code division multiplexing concepts with OFDM to exploit the channel frequency diversity regardless of the channel coding scheme. In SOFDM, the original data stream is spread in the frequency domain over the different subcarriers using a spreading code, and the original data rate can be maintained by code-multiplexing N data symbols to N orthogonal codes. In the downlink, the use of orthogonal codes like Walsh-Hadamard or DFT codes guarantees no intra-cell interference (MAI) in a flat fading channel (i.e., the same channel gain on all occupied subcarriers). However, in a frequency selective delay-spread channel, the orthogonality between the codes is lost due to different channel gains on the different chips of a spreading code. Different linear equalizer/detection techniques have been investigated in the literature [8] to deal with the lost orthogonality, such as ORC (Orthogonality Restoring Combining), EGC (Equal Gain Combining), and MMSE equalization. The paper is organized as follows. Section II gives an overview of the experimental system setup, and Section III describes the OFDM and SOFDM system models. Section IV describes the procedure used to analyze the experimental data. Section V presents the performance results for OFDM and SOFDM for numerous transmit/receive configurations, including the baseline 1x1 configuration, 1x2 receive diversity, 2x1 transmit diversity, 2x2 transmit-receive diversity, and 2x2 dual-stream MIMO. Finally, Section VI summarizes the overall results of the experiments. II. EXPERIMENTAL SYSTEM This section presents an overview of the OFDM experimental system. The OFDM signal consists of 751 subcarriers spaced 25 kHz apart, resulting in a signal bandwidth of 18.775 MHz. Nulls are placed at each end of the spectrum to provide a guard band for D/A conversion and filtering. A 1024-point IFFT transforms this frequency domain data into the time-domain, and the resulting timedomain signal is sampled at 25.6 MHz. A 25% cyclic extension (256 samples) is added to the time-domain signal to combat multipath and inter-symbol interference. The system parameters are summarized in Table I. For SOFDM, the spreading factor is 748, and all 748 codes (with equal power per code) are used (multicode).

I. INTRODUCTION An experimental mobile broadband communication system has been developed [1][2] for evaluating a number of physical layer technologies for 4th generation mobile cellular systems, such as Orthogonal Frequency-Division Multiplexing (OFDM), advanced error-control coding techniques, adaptive array signal processing, and aggressive frequency re-use. The experimental system is a 20 MHz 2x2 MIMO-OFDM system operating at a carrier frequency of 3.676 GHz. The system consists of a base station with two transmit antennas and a two-antenna mobile receiver. This paper presents and compares link performance results of a series of field experiments for two promising broadband transmission technologies - OFDM and OFDM with frequency domain spreading. The latter is sometimes called spread-OFDM (SOFDM), OFDM-CDMA, or multicarrier CDMA (MC-CDM(A)) [5]-[9]. For convenience, we will use the term SOFDM in the sequel. OFDM is well known for its ability to maintain orthogonality between the transmitted symbols even in severe delay-spread channels, through the use of a cyclic extension. The orthogonality between symbols eliminates intra-cell interference [3], thus making it possible to boost data rate and system capacity with higher-order constellations when users are near the base station (i.e., adaptive modulation). Moreover, since each subcarrier experiences flat fading, numerous adaptive antenna and spacetime coding techniques that have been developed under a flat fading framework (such as Alamouti transmit diversity [4])

0-7803-7467-3/02/$17.00 ©2002 IEEE.

223

S/P add prefix IFFT P/S TABLE I SYSTEM PARAMETERS Soft decision Viterbi decoder channel Σ noise Channel Bandwidth Number of subcarriers Number of data subcarriers Subcarrier spacing Cyclic extension length FFT size Modulation Coding Interleaving 20 MHz 751 748 25 kHz 10 µs 1024 QPSK. mobile. SOFDM system model. The receiver. r[ k . For SOFDM.00 ©2002 IEEE. consists of two independent branches with 5 dBi gain omnidirectional antennas on the roof approximately 2. WHT is a Walsh-Hadamard Transform. To maintain suitable input levels at the A/Ds. Block diagrams of the overall OFDM and SOFDM system models for one transmit and one receive antenna (1x1) case are shown in Figure 2 and Figure 3. which comes at the expense of higher processing requirements at the transmitter/receiver due to non-binary complex-valued codes. and the different transmission techniques (OFDM and SOFDM). Two 80° beamwidth 18 dBi gain antennas mounted 160 ft above the average surrounding terrain radiate the RF signals at 66 dBm EIRP (effective isotropic radiated power). n] . (2) For the OFDM/SOFDM system. diversity.4 MHz IF signals and sampled at 51. As shown in Figure 2 and Figure 3. n] + N [k . n] = P[k ] X [k . which steps through the different constellations.Convolutional encoder Constellation mapping Information bits Intra-Block interleaver S/P add prefix IFFT P/S channel Σ noise de-interleaver Soft decision Viterbi decoder channel gain correction Symbol to bit mapping decoded bits P/S FFT S/P Figure 2. the cyclic prefix of Np samples is discarded and the remaining samples are DFT’d into the frequency domain. Let r[k. Thus.n] is the noise at sample k of symbol block n. The received RF signals are down-converted to 38. mounted in a test vehicle. n[k. remove prefix remove prefix (1) where pl is the sampled channel pulse response and pl is assumed to be zero for l<0 and l>Lp. The received signal is demodulated using conventional demodulation techniques for OFDM [10][11].2 MHz with 12 bits of resolution. III. within each symbol block. A mathematical description of the model is presented in [12]. The complex digitized (at 25. Convolutional encoder Constellation mapping Information bits WHT/ DFT Figure 1. n] . is stored in a programmable arbitrary waveform generator. 224 . to one or two receivers located in a test vehicle (shown in Figure 1). at a base station located on top of a building. which acts as a source to continuously transmit the encoded signals. IWHT is an Inverse Walsh-Hadamard Transform. OFDM system model. 16-QAM. SYSTEM MODEL This section describes the system models of the OFDM and SOFDM downlink transmission schemes. and MIMO). and SOFDM [7]-[9].7 m above the ground. Also. note that frequency interleaving of the coded bits is not needed in the case of SOFDM because the spreading/despreading process averages the channel quality across all subcarriers and spreading codes (SOFDM link simulations showed virtually identical performance with or without interleaving). each receiver gain is automatically adjusted based on the received signal strength and timing. The DFT of the sequence r[k. Base. and receiver for experimental system. Also. n] + n[ k . linear frequency-domain equalization is employed on each frequency bin according to 0-7803-7467-3/02/$17. 64-QAM R = 1/2 64-state convolutional Bit Interleaving for OFDM (over the subcarriers of one OFDM symbol period) No interleaving for SOFDM IWHT/ IDFT Symbol to bit mapping equalizer decoded bits P/S FFT S/P Figure 3.6 MHz) baseband time-domain samples are periodically captured for off-line analysis. the different antenna scenarios (baseline.n] can be written as R[k . The benefit of DFT spreading codes is a low transmit peak-to-average power ratio. n ] = ∑ p x[k − l + N l l =0 Lp p .n] denote the received samples corresponding to the nth symbol block after the cyclic prefix has been discarded. performance is evaluated with both Walsh-Hadamard and DFT-matrix orthogonal spreading codes. The system transmits from one or two antennas. The transmitted waveform.

2. only Walsh-Hadamard spreading results will be presented for SOFDM in the remainder of the paper. Figure 8 shows the results for the 2x2 transmit-receive diversity case. and for SOFDM. The results indicate that receive diversity is more effective than transmit diversity.6 MHz complex baseband time-domain samples captured by the experimental system are processed off-line. depending on location.5 dB). κ−1. The results show that as the diversity order increases. The above equations can be easily extended to the multiple receive antenna and MIMO cases [13][14]. For the dual-stream MIMO case. EXPERIMENTAL RESULTS Results are presented in terms of decoded BER as a function of Eb/No (Energy per information bit to noise power spectral density) for OFDM and SOFDM with several transmit/receive antenna combinations. 2x2 transmit-receive diversity. First. and snapshots of data were collected at approximately 2. results are presented for both Walsh-Hadamard and DFT spreading (SF=748). b] . (5). for SOFDM. 1x2 receive diversity. independent stream 2x2 MIMO with linear MMSE detection and channels with an average reciprocal condition number. 2x1 transmit 0-7803-7467-3/02/$17. Figure 10 also shows that OFDM outperforms SOFDM for dual-stream 16-QAM while the performance of SOFDM is comparable to OFDM for dualstream QPSK. EGC performance is comparable to MMSE for QPSK (within 0. The area surrounding the base is divided into sectors. w[k ] = w[k ] = P * [k ] | P[k ] | (EGC) . a linear MMSE approach is used [14] for joint equalization and stream separation. Also. while the lowest κ−1 are found when the delay spread is low. These results support the trends predicted by simulations [12]. (3) where for OFDM. V. The test area contains a mixture of single and multistory residential and commercial buildings as well as some undeveloped spaces. b] = w[k ]R[k . softdecision Viterbi decoding is performed.Z [k . For SOFDM. Frame and bit error statistics are gathered for each symbol interval in the snapshot. the gains are less pronounced compared to the 1x1 configuration. Figure 5 compares the performance of OFDM and SOFDM for the baseline 1x1 (single transmit and receive antenna) configuration. P[k ] (4) diversity. Figure 10 shows the results for open-loop. the performance difference between OFDM and SOFDM decreases. Therefore. w[k] will be chosen according to either an EGC (Equal-gain chip combing) or MMSE criterion. The higher the reciprocal condition number ( 0 ≤ κ −1 ≤ 1 ) the higher the theoretical capacity supported by the channel. Hence. IV. The results indicate that OFDM outperforms SOFDM and that the OFDM performance gain increases as the order of the constellation is increased. the impact of the receiver equalization/despreading algorithm for SOFDM is investigated. The received signal is then equalized using (4).n]. greater than 0. 225 . The performance gains of OFDM still increase with the constellation size. 16-QAM and 64-QAM constellations. 2x2 transmit-receive diversity. and these results are then averaged over all snapshots. Log-likelihood ratios are then generated and sent to the convolutional-decoder. Note that the OFDM implementation is computationally less complex than SOFDM. The poor performance of EGC for higher-order modulations is due to its inability to equalize the amplitude variations in the frequency domain channel. the long-code is removed from the equalized frequency-domain signal followed by despreading and gain correction. 2x1 transmit diversity. After computing (3). Figure 4 shows the performance of SOFDM with equal-gain chip combining (EGC) and linear MMSE detection for the single transmitsingle receive (1x1) configuration. This result was expected. of the 2x2 channel matrix. The reciprocal condition number. (6) 2 2 where σ n is the variance of n[k. In order to evaluate the performance at lower SNR. since the receiver collects 3 dB more energy when two receive antennas are used (results are plotted based on Eb/No per receive antenna). 1x2 receive diversity. EGC is not a viable receiver for SOFDM with high-order modulations and only the linear MMSE receiver will be considered for the remaining results of the paper. The reciprocal condition number. the performance of SOFDM is invariant to the type of orthogonal spreading codes used. However. Comparing MIMO QPSK to Alamouti 2x2 16-QAM in Figure 8 shows that for identical data rates.5 second intervals (a snapshot contains 500 OFDM symbol periods). Typically. 1 w[k ] = . MEASUREMENT AND ANALYSIS Outdoor measurements were performed on various drive routes around the base site. white noise of the appropriate power is added to each received signal. is computed for each 2x2 channel matrix and is averaged over time and frequency. Figure 9 compares the performance of 16-QAM OFDM for several transmit/receive antenna combinations: baseline 1x1. and each sector is covered by a pair of 80° beamwidth antennas (A1 and A2). including the baseline 1x1 configuration.n] and σ x is the variance of x[k.00 ©2002 IEEE. κ−1. The channel estimator uses a least-squares multiuser technique similar to that presented in [13]. (6) for the 1x1 case. Test vehicle speeds ranged from 0 to 65 mph. and is related to the degree of separability of the two streams. but it degrades severely (high BER floor) for the 16-QAM and 64QAM constellations. (5) P * [k ] 2 2 P * [k ]P[k ] + σ n / σ x (MMSE) . since OFDM uses a simple equalizer and has no spreading/despreading matrix operations. and 2x2 dual-stream MIMO for QPSK. The channel response is determined with a frequency-domain multi-transmit-antenna channel estimator designed using the fact that all of the transmitted data on both transmit antennas is known. Figure 6 and Figure 7 show the results for the 2-branch receive diversity (1x2) and Alamouti 2x1 transmit diversity case respectively. Finally. is equal to the ratio of the smallest to the largest singular values of the matrix. high delay spread channels exhibit high κ−1. κ−1. The 25.

VI. ACKNOWLEDGMENTS The authors would like to thank the Motorola Labs 4Gx team for developing the experimental system and for collecting the received waveforms over numerous drive routes (especially Mickael Batariere. “Coherent multicarrier/DS-CDMA and MC-CDMA for broadband packet wireless.00 ©2002 IEEE. the condition number of the channel has a significant impact on the performance of dual-stream MIMO. Fettweis.” IEEE PIMRC ’93. including the baseline 1x1 configuration. OFDM vs. 2000. N. T. T. Jeffrey Porter. “Performance of multi-carrier CDM and COFDM in fading channels. OFDM typically outperformed SOFDM. 2001. 2x2 transmit-receive diversity. Although this result highlights the benefits of 2x2 diversity. “A simple transmit diversity technique for wireless communications. Nangia. NJ: Prentice Hall. all channels with κ−1 > τ0 are included in the BER counting process (note that 75% of the data has κ−1 > 0. 16. κ−1. “Transmission techniques for digital terrestrial TV broadcasting. H.” IEEE Globecom ’99. S.” 2002 International Conference on 3G Wireless and Beyond. G. 226 . “Multicarrier modulation for data transmission: an idea whose time has come. Thus. H. May 2001. F. “A link performance study of DS-CDMA and OFDM for 4th generation cellular.” IEEE Globecom ’95. Mukthavaram. M. Abeta. K. Kepler. and for higher order modulations. Receive diversity was found to be more effective than transmit diversity. Baum. Atlantic City. non-linear receiver processing (which is not used here) can further improve the performance of MIMO. S. October 7-11. J. S. 2x1 transmit diversity. 1995. Alamouti. J. 19471951. Vook . Jeanclaude. S. pp. Krauss. Compton Jr. Englewood Cliff. “Experimental broadband mobile OFDM system: description and initial results. Kepler. Monticello. and by a substantial margin in many cases. September 1999. 1x1. The results show that the performance of dual-stream MIMO improves as τ0 is increased. r=1/2 64−state Conv 10 −1 16−QAM 64−QAM Decoded BER 10 −2 QPSK 10 −3 MMSE EGC 6 8 10 12 14 Eb/No (dB) 16 18 Figure 4.2). M. [6] [7] [8] [9] [10] [11] [12] [13] [14] S. τ. vol. 1993. C. For SOFDM. 2001. Sawahashi. 0-7803-7467-3/02/$17. “Forward link capacity of coherent DS-CDMA and MC-CDMA broadband packet wireless access in a multi-cell environment. Bingham. Results were obtained for several transmit/receive antenna combinations. Feb. Baum. Igor Lisica.” IEEE Journal on Select Areas in Communications. Thanks are also extended to Tim Thomas for providing the channel estimation. Abeta. Vol. and 50% has κ−1 > 0. V.” 2001 Mobile and Portable Research Group Wireless Symposium. Porter. Adachi. Vook. I. SOFDM: MMSE vs.. IL. and 2x2 dual-stream MIMO. no.” IEEE Communications Magazine. T. K. Baum. CONCLUSIONS This paper evaluated the performance of OFDM and spread-OFDM (SOFDM) using 20 MHz field experiments in a wide-area suburban system context. F. Karam. Vook. “Multi-carrier CDMA in indoor wireless radio networks. June 6-8. Krauss. 1918-1922. M. as it improves the likelihood of separating the two streams. pp. Atarashi. Thomas Krauss. G. 1995. 1x1. 28. S. J. R. r=1/2 64−state Conv 10 −2 OFDM SOFDM−WHT SOFDM−DFT 64−QAM Decoded BER 16−QAM [2] 10 −3 [3] QPSK [4] 10 −4 [5] 6 8 10 14 12 Eb/No (dB) 16 18 Figure 5. on the performance of 2x2 MIMO-OFDM for different condition number thresholds. 1998. J.. K. pp. Thomas. and Keith Blankenship). Kaiser.transmit diversity performed a couple dB better than MIMO for the field data. 109-113. James Kepler. pp. 1988. May 2000. F. 33. Oct. J. “Least-squares multi-user frequencydomain channel estimation for broadband wireless communication systems. I. V. τ0. A. 1999. This trend was most pronounced for the baseline single-antenna case. Porter. 847-851. In the results. May 1990. “On the performance of different detection techniques for OFDM-CDMA in fading channels. The improved performance and lower implementation complexity indicate that OFDM is very promising for 4G systems. Nangia. H.” IEEE VTC-2001/Fall. Kaiser. Sawahashi.” IEEE VTC 2000/Spring.” 2001 International Conference on 3G Wireless and Beyond. “OFDM and OFDM with frequency domain spreading: observations for higher-order modulations. J. Figure 11 shows the effect of the average reciprocal condition number. Atarashi. 100-109. F. REFERENCES [1] M. 1x2 receive diversity.” IEEE Communications Magazine.1.. Sept. both DFT spreading and Walsh-Hadamard spreading were found to have virtually identical performance. Sandeep Mukthavaram. pp. Vol. and the condition number of the channel had a significant impact on the performance of dual-stream MIMO. EGC receiver. Ramasubramanian. 8. May 2002. K.” IEEE VTC 2000/Fall. Blacksburg. 5-14. Linnartz. Adaptive Antennas: Concepts and Performance. Lisica. M. For a particular threshold. NJ. Batariere. VA. Batariere. Yee. Sari. pp. T. “An experimental OFDM system for broadband mobile communications.” 37th Allerton Conf. P. SOFDM: 1x1 (baseline).

antenna config: 1x1.2x2 Decoded BER 10 −2 Decoded BER 16−QAM 10 −3 QPSK 10 −3 10 −4 10 8 6 Eb/No (dB) Figure 6. r=1/2 64−state Conv OFDM SOFDM 10 −2 10 Decoded BER 64−QAM Decoded BER 16−QAM 16−QAM 10 −3 10 −3 QPSK QPSK 10 −4 2 10 8 12 Eb/No (dB) Figure 7. 16−QAM. OFDM vs. SOFDM: Alamouti 2x1. 2 4 6 0-7803-7467-3/02/$17. Al2x2. SOFDM: 1x2 (Rx diversity). r=1/2 64−state Conv OFDM Mimo 2x2.2. OFDM vs. channel condition number threshold. r=1/2 64−state Conv OFDM SOFDM −2 Mimo 2x2. OFDM 2x2 MIMO vs. r=1/2 64−state Conv 10 −1 OFDM. 0 2 4 12 12 10 8 6 Eb/No (dB) Figure 9. SOFDM: 2x2 MIMO. Thresh=0. OFDM vs. Number Thresh=0.2x1 1x2 Al. r=1/2 64−state Conv 10 −1 OFDM SOFDM 10 −1 QPSK 16−QAM Decoded BER Decoded BER 10 −2 10 −2 64−QAM 10 −3 16−QAM QPSK 10 −3 10 −4 Cond.1 Cond.2 −2 −1 7 6 5 4 3 Eb/No (dB) Figure 8. Cond. Thresh=0. 0 1 2 8 9 16 14 12 10 8 Eb/No (dB) Figure 11. 16-QAM OFDM vs. r=1/2 64−state Conv OFDM SOFDM 10 −2 64−QAM 1x1 Al. 4 6 14 Alamouti 2x2. Al2x1. SOFDM: Alamouti 2x2.00 ©2002 IEEE. 2 10 −4 4 Alamouti 2x1. 1x2. 4 6 14 10 −4 2 10 8 12 Eb/No (dB) Figure 10. 227 .1x2. OFDM vs. Thresh=0 Cond.