1 Linear Precoder Performance for Massive MIMO Systems in near LOS Environments: Application to mmWave Transmission A. Rozé 1 , M. Hélard 1,2 , M. Crussière 1,2 and C. Langlais 1,3 1 IRT b<>com, Rennes, France 2 Université Européenne de Bretagne, INSA, IETR, CNRS UMR 6164, Rennes, France 3 Electronics Department, Institut Mines-Telecom, Telecom Bretagne, CNRS UMR 6285 Lab-STICC, Brest, France Abstract—As the wireless communication network is driven toward densification with small cell deployments, massive MIMO technology shows great promises to boost capacity and energy efficiency. This paper aims at showing how raising carrier frequencies impacts the performance of the linear precoders used in Massive MIMO systems. By means of a geometrical determin- istic channel model, we simulate a dense outdoor scenario and highlight the influence of the direct and multi-paths components. More importantly we prove that, in a Line-of-Sight configuration, the discriminating skills of the well known Zero Forcing precoder is much more sensitive to the antenna array architecture and the user location than the Conjugate precoder. Index Terms—Massive MIMO, mmWave, Zero Forcing, Con- jugate Beamforming, LOS, antenna array architecture I. I NTRODUCTION Today spectrum is scarce and this scarcity greatly un- dermines the improvement in capacity targeted for future 5G communications. Significant technological advances have thus to be made to meet the 5G challenges. Among the considered technologies, multi-user systems through massive MIMO (Multiple Input Multiple Output) and Millimeter Wave (mmWave) transmissions are both showing great promises to boost the capacity of broadband networks. On the one hand, mmWave transmissions offer true opportunities due to the large bandwidths potentially available at such frequency ranges, but also because the related path-loss attenuations are well suited to small cell deployments. On the other hand, the underlying spatial filtering or beamforming concepts in massive MIMO systems are expected to efficiently deal with multi-user access issues. Besides, the fact that mmWave will keep large-scale antenna systems to a reasonable size makes the combination of the two strategies worthy of interest, especially in dense network scenarios. One can find lots of contributions on massive MIMO aspects in the literature, evaluating the performance of various beamformers in the form of baseband precoding schemes [1], [2]. Although it is understood that massive MIMO would be interesting at high frequencies (e.g. in [3]), studies describing the precoding performance at such frequencies have not yet surfaced. Moreover, theoretical performance almost systemat- ically assume decorrelated fading environments [1], [4], which may not be very realistic for mmWave transmissions. So that precoder performance can be better understood at these high frequencies, let us sum up the main mmWave characteristics: 1) Delay spread is reduced due to the often directive nature of the used antennas and the path losses of the scattered paths [6]. This in turn means a greater coherence bandwidth. 2) High beamforming gains are needed to deliver high rate transmissions and counteract the drastic path losses [7]. 3) Channels often have a Line of Sight (LOS) component and still high capacities can be reached [8]. 4) The higher the frequency, the wider the gap between the power of the direct path and the reflected ones [6], hence driving the channel behaviour toward pure LOS (no multi-paths reflections). As of today it is well understood that the choice of a particular precoding technique depends on the scenario in which the solution is to be implemented. The authors in [9] emphasise this aspect by comparing two widely used precoding algorithms, the Conjugate and Zero Forcing (ZF) beamforming, both of which are implemented in the first massive MIMO prototype, ARGOS [10], working at 2.4 GHz. In this paper we investigate the impact that raising the carrier frequency up to 60 GHz has on the performance of these precoders in a multi-user massive MIMO system. In that perspective we derive a deterministic channel model based on a geometrical environment description, more realistic than theoretical models commonly used in the literature. The authors in [10] highlight how ZF precoding falls behind Conjugate in performance when the number of users increases. We provide a further insight into the reason for that drop by showing analytically the ZF limitation in highly LOS scenarios. Furthermore we show how the antenna array configuration influences the performance. Section II serves as background for our analytical descrip- tion of the Zero Forcing limitations in highly LOS configu- ration, which is derived in section III. Section IV describes our chosen channel model, the deterministic approach taking into account antenna correlation and a few signal reflections. Then follows the performance metric description before our results are unrolled in section V. There, we confirm through