IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 26, NO. 6, AUGUST 2008 1 Transmitter Noise Effect on the Performance of a MIMO-OFDM Hardware Implementation Achieving Improved Coverage Hajime Suzuki, Member, IEEE, Thi Van Anh Tran, Iain B. Collings, Member, IEEE, Graham Daniels, and Mark Hedley, Member, IEEE Abstract— This paper presents analysis of performance mea- surements from a MIMO-OFDM IEEE 802.11n hardware imple- mentation at 5.2 GHz using four transmitters and four receivers. Two spatial multiplexing systems are compared; one which uses a zero-forcing (ZF) detector and the other a list sphere detector (LSD). We show that the measured results do not align with standard prediction based on simulation assuming uncorrelated receiver noise. We show that the discrepancy can be explained by the inclusion of transmitter noise into the channel model. This effect is not included in existing MIMO-OFDM channel models. The measured results from our hardware implementation show successful packet transmission at 600 Mbit/s with 15 bit/s/Hz spectral efﬁciency at 73% coverage for ZF and 84% coverage for LSD with an average receiver signal to noise ratio (SNR) of 26 dB. Index Terms— MIMO-OFDM, LDPC, ZF, LSD, SNR estima- tion. I. I NTRODUCTION M IMO-OFDM (multiple-input multiple-output orthogo- nal frequency division multiplexing) is currently be- ing considered as a strong candidate for the physical layer transmission schemes of next generation wireless communica- tion systems [1], [2]. Commercial products utilizing MIMO- OFDM with two transmitters and three receivers, denoted as 2 × 3 in this paper, are currently available for wireless local area networks (WLAN), achieving up to 300 Mbit/s physical layer (PHY) data rate with 7.5 bits/s/Hz spectral efﬁ- ciency. The WLAN standardization group is aiming to achieve 15 bits/s/Hz bandwidth efﬁciency using four transmitters [3]. This paper reports on a hardware implementation of high spectral efﬁciency MIMO-OFDM without the knowledge of the channel at the transmitter. The implementation is based on the draft IEEE 802.11n standard 1 with the optional low density parity check (LDPC) codes [4] at 5.2 GHz using four transmitters and four receivers. Two spatial multiplexing systems are compared; one which uses a zero-forcing (ZF) Manuscript received July 1, 2007; revised December 1, 2007 and January 21, 2008. H. Suzuki, M. Hedley, G. Daniels, and I. B. Collings are and T. V. A. Tran was with the ICT Centre, Commonwealth Scientiﬁc and Industrial Research Organisation (CSIRO), PO Box 76, Epping, NSW 1710 Australia (e-mail: Hajime.Suzuki@csiro.au). Digital Object Identiﬁer 10.1109/JSAC.2008.0808xx. 1 The draft standard has been continually updated to this date. While the exact construction of the transmitted signals by the latest draft differs from the one used in this paper, very similar performance is expected between the two speciﬁcations. detector [5], [6] and the other a list sphere detector (LSD) [7]. The main contributions of our paper are summarized as follows: • We report on successful (no error) packet transmission of MIMO-OFDM using LDPC codes achieving PHY data rate of 600 Mbit/s and 15 bit/s/Hz spectral efﬁciency. This was achieved at 73% of the measured locations for the ZF system, with an average measured receiver signal to noise ratio (SNR) of 26 dB; and the coverage was 84% for the LSD system. • We demonstrate that noise generated at the transmitter which causes correlated noise at the receiver can signiﬁ- cantly affect predicted performance in practical transmis- sion scenarios, when this is not taken into account in the modeling; which is the case for current channel models. Such transmitter noise can be typically introduced in OFDM by the non-linearity of ampliﬁer and phase noise [8] and/or quantization and clipping by digital to analog converters (DACs) [9]. The performance of ZF detector is seriously underestimated (more than 4 dB was observed) if the transmitter noise is neglected. The performance of LSD is less affected by this correlated noise. • We develop realistic prediction of measured bit error probability (BEP) performance by simulation based on measured MIMO-OFDM channels and measured SNR. Previously, implementation of MIMO-OFDM for high spec- tral efﬁciency employing more than two transmitters has been reported, but not achieving both high coverage and high spectral efﬁciency, as we do here. In [10], a spectral efﬁciency of 8 bit/s/Hz with 162 Mbit/s PHY data rate was demon- strated by 3 × 3. Per-antenna-coding (PAC) with soft-output maximum likelihood detection (MLD) as well as PAC with successive interference cancellation (SIC) were implemented by off-line signal processing. 4 × 4 was implemented in [11] achieving 208 Mbit/s and 10 bit/s/Hz using real-time signal processing. Another 4 × 4 with 10 bit/s/Hz using real-time signal processing but achieving 1 Gb/s was reported in [12] where MLD-based signal detection and turbo decoding were utilized. QR decomposition with M-algorithm (QRM-MLD) was used to reduce the high computational requirement of MLD in [12]. A higher spectral efﬁciency of 14 bit/s/Hz was demonstrated in [13] with 281 Mbit/s using 3 × 3 by off- line processing. The authors of [13] reported that the bit error probability (BEP) was close to 10 -5 at a signal-to-noise ratio 0733-8716/08/$25.00 c  2008 IEEE