Dependable wireless connectivity: insights and methods for 5G and beyond

In MIMO transmission CSIT is an essential ingredient. CSIT for downlink transmission in cellular networks is commonly obtained either from uplink measurements, relying on channel reciprocity [15], or from quantized channel estimates provided by the users over dedicate limited capacity feedback links [16]. Either way, the CSI estimate at the transmitter is prone to imperfections, causing residual multi-user interference when applying spatial multi-user multiplexing schemes. This is especially problematic in high-mobility scenarios, where CSIT imperfections are aggravated by fast temporal channel variations. High-mobility situations are therefore not suited for so-called coherent spatial multi-user multiplexing schemes, which rely on destructive multipath interference to mitigate multi-user interference (e.g., zero-forcing beamforming). Better suited are incoherent spatial multiplexing schemes, such as, space division multiple access (SDMA), which attempt to spatially separate multiple users in the angular (azimuth and elevation) domain, by forming transmit beams that are non (or only weakly) overlapping in the angular domain. Such SDMA schemes are, however, limited to situations in which the number of significant multipath components of each user is small and where users are spatially relatively far apart, to enable angular separation of users at the transmitter.

In [3], we develop robust limited CSI feedback methods and non-coherent beamforming techniques for FD-MIMO multi-user transmission that are suitable for high-mobility situations. These methods enable reliable and spectrally efficient multi-user transmission with modest amount of CSI feedback overhead. The method is based on quantizing the angular directions of the most significant multipath components of each user, in order to form transmit beams that steer the signal energy towards the respective intended user while limiting the interference leakage caused to other users. An example of transmit beams obtained in that way is shown in Fig. 1; in this figure, the fully white circle illustrates the angular position of the intended user, while the black circles/rectangles show the positions of other users. The leakage bounded beamforming method developed in [3] allows to incorporate angular uncertainty in the transmit beam design. The left and right plots of Fig. 1 show the resulting transmit beams without angular uncertainty (left) and with angular uncertainty (right); the amount of angular uncertainty is represented by the black rectangles in the right plot. Clearly, accounting for angular uncertainty allows for more robust and reliable multi-user transmission, since imperfections in the angular position estimation are compensated by the more forgiving transmit beam.

Channel measurement campaigns in the mmWave band have revealed that mmWave transmissions exhibit very limited multipath propagation [17, 18]. The main reason for this behavior is that mmWave transmissions require high-gain directive antennas to compensate for the pathloss, which act as spatial filters and thereby effectively reduce the number of significant multipath components. In many cases, the mmWave channel can be characterized by very few (one or two) dominant multipath components. Due to the small wavelength of millimeter waves, the relative phases of these multipath components vary significantly already for tiny variations of the propagation environment. We have observed these effects also empirically through measurements in indoor scenarios. Figure 2 shows an exemplary result of the correlation of the phase of the measured channel as a function of the displacement of the receiver (position \((\Delta x, \Delta y) = (0,0)\) is the reference position). The noticeable phase correlation pattern in Fig. 2 can largely be explained by two waves interfering at our receiver. We observe that the phase varies significantly even when moving only a few millimeters. This implies that CSI uncertainty of the relative phases of multipath components are practically unavoidable even in virtually static scenarios. Therefore, the so-called two-wave with diffuse power (TWDP) channel fading model [19] is well-suited to describe the fading behavior of mmWave transmissions. This is also evidenced by our measurement results reported in [20], where we observe TWDP fading in an indoor mmWave transmission scenario. The TWDP channel distribution can lead to fading behavior that is worse than Rayleigh fading, which significantly impairs the reliability of mmWave links. As we show in [21], smart beamforming at the transmitter, even without knowledge of the relative phases of the multipath components, can reduce the severeness of multipath fading and therefore enhance the reliability. Yet, to achieve high-degree dependable wireless mmWave transmissions, macroscopic diversity in the form of multi-point and/or multi-band connectivity should be provided by the mobile communications system.

The most important property related to favorable channel conditions in massive MIMO systems is that channels of different users become asymptotically orthogonal with growing number of transmit antennas. This is, e.g., fulfilled in rich scattering environments (independent Rayleigh fading), but also under line of sight (LOS) conditions provided users can be separated by the transmitter in the angular domain [22]. Favorable channel conditions theoretically allow to serve an unbounded number of users in parallel when the number of transmit antennas grows to infinity. Practically, however, a possibly large but limited number of antenna elements needs to be arranged within a given form factor. Scaling up the number of antenna elements with a fixed form factor does not lead to favorable propagation conditions, since the channels seen by different antenna elements become more and more correlated [23]. This then leads to the important issue of the optimal choice of antenna geometry (i.e., how many antenna elements should be used and how should they be arranged) to achieve the best possible performance. The optimal choice is on the one hand dictated by the achievable spatial resolution of the array, which determines spatial separability of users in the azimuth and elevation domain [24] and therefore affects whether or not such users can reliably be served in parallel, but on the other hand also by impairments of the radio frequency hardware, such as, power losses in phase shifters, phase quantization, etc. [25]. In our research lab, we attempt to address these issues by means of virtual antenna array measurements. In a virtual array approach, in the simplest case, a single antenna element is mounted on a mechanical guide system and is connected via a single cable to a single channel transmitter. By sequentially re-positioning this single antenna element in space and transmitting sounding sequences at each position, an arbitrary antenna array geometry can be virtually formed in space. Yet, such virtual arrays can naturally not incorporate mutual antenna coupling effects, which can lead to overestimation of the performance of full arrays (realizing all antenna elements simultaneously) [26]. To compensate for this effect, we have developed a mutual coupling model [27], which, when applied to virtual array measurement data, allows to closely approximate the performance of full arrays without having to physically build them. This enables us to investigate and compare the performance of different array geometries in our future work, in order to identify geometries that enable reliable multi-user spatial multiplexing in practical FD-MIMO systems.

Network densification runs like a common thread through the success story of cellular networks [34]. This trend of ever increasing spatial reuse of bandwidth will persist within 5G networks, especially since mmWave transmissions are mostly limited to relatively short distance communication. Despite deploying autonomous small cells to densify network architectures, distributed antenna systems (DASs) are another important technology that can enhance network performance by reducing the access distance between transmission/reception points of the network and users. DASs extend the BSs’ antenna ports by connecting remote radio heads (RRHs) over dedicated high-bandwidth low-latency connections to the BSs, utilizing e.g. radio-over-fiber technology or mmWave fronthaul [35], thereby enabling spectrally efficient single-user and multi-user MIMO operation by coherent transmission from spatially distributed antenna arrays [36]. Especially the combination of mmWave fronthauling and CRAN is promising, since it enables on-demand formation of DASs by dynamically connecting cloud base band units (BBUs) to RRHs to satisfy user-requirements.

In [37], we investigate a variant of such dDAS, in which existing macro BSs are dynamically connected to RRHs via on-demand fronthaul connections. This approach has the advantage that basic coverage can be provided by the layer of existing macro BSs, whereas the actual high-performance data transmission to the user equipments (UEs) is supported by additional RRHs. The challenge in such systems is to optimize the dynamic allocation of RRHs amongst multiple BSs. In [37] we propose joint RRH allocation and distributed beamforming methods for dDASs with to goal of minimizing the outage probability of UEs. We show an exemplary result of [37] in Fig. 4, where we compare the outage performance of autonomous small cells to DASs with fixed allocation of RRHs amongst BSs (allocation to closest BS) and to the proposed dDAS architecture. Clearly, dDAS can provide a significant improvement in terms of user outage probability, because the dynamic user-centric coordination of RRHs provides a gain in macroscopic diversity. Notice that in combination with mmWave fronthauling such a dDAS architecture can be realized relatively cheaply, since no wired fronthaul is required.

Since release 14 of LTE, vehicular communications is considered as an integral part of mobile cellular networks [39]. Vehicular communications is an important enabler for enhancing the safety on roads by supporting mutual awareness of vehicles, as well as, for improving the efficiency of transportation through intelligent transport systems. Yet, vehicles featuring radio access equipment can also be useful for enhancing the performance of the mobile network itself; specifically, utilizing vehicles as relaying nodes between BSs and conventional UEs (e.g. smart-phones) has significant potential for coverage extension of mobile networks, especially in urban areas where vehicles are basically omnipresent [38]. Thus, equipping vehicles with mobile network access equipment could be a relatively cheap enabler for network densification without requiring dedicated on-premise infrastructure. Vehicular relaying can also reduce the energy consumption of conventional UEs by decreasing the transmission distance, thereby potentially even enabling outdoor mmWave transmissions.

In [38], we investigate the coverage improvement of vehicle-relay-enhanced micro-cellular networks in urban environments. We show an exemplary result of [38] in Fig. 5, which exhibits the reduction of signal outages as a function of the average density of vehicles on the road. We observe that the coverage of the network improves with increasing density of vehicle-relays, due to increased macroscopic diversity. Notice that in the relay-assisted communication scenario relaying is only activated if it is beneficial in terms of coverage; that is, if the BS-to-relay or the relay-to-UE link has worse performance than the direct BS-to-UE link, relaying is not used. In our investigation we only consider relaying by a single vehicle-relay; the performance can potentially be further improved by enabling multi-relay connectivity, thereby enhancing the reliability of the wireless connection by making it resilient even w.r.t. microscopic fading.