Biblio

The fifth-generation and beyond mobile networks should support extremely high and diversified requirements from a wide variety of emerging applications. It is envisioned that more advanced radio transmission, resource allocation, and networking techniques are required to be developed. Fulfilling these tasks is challenging since network infrastructure becomes increasingly complicated and heterogeneous. One promising solution is to leverage the great potential of Artificial Intelligence (AI) technology, which has been explored to provide solutions ranging from channel prediction to autonomous network management, as well as network security. As of today, however, the state of the art of integrating AI into wireless networks is mainly limited to use a dedicated AI algorithm to tackle a specific problem. A unified framework that can make full use of AI capability to solve a wide variety of network problems is still an open issue. Hence, this paper will present the concept of intelligence slicing where an AI module is instantiated and deployed on demand. Intelligence slices are applied to conduct different intelligent tasks with the flexibility of accommodating arbitrary AI algorithms. Two example slices, i.e., neural network based channel prediction and anomaly detection based industrial network security, are illustrated to demonstrate this framework.

Joint transmission coordinated multi-point (CoMP) is a combination of constructive and destructive superposition of several to potentially many signal components, with the goal to maximize the desired receive-signal and at the same time to minimize mutual interference. Especially the destructive superposition requires accurate alignment of phases and amplitudes. Therefore, a 5G clean slate approach needs to incorporate the following enablers to overcome the challenging limitation for JT CoMP: accurate channel estimation of all relevant channel components, channel prediction for time-aligned precoder design, proper setup of cooperation areas corresponding to user grouping and to limit feedback overhead especially in FDD as well as treatment of out-of-cluster interference (interference floor shaping).

Massive MIMO and tight cooperation between transmission nodes are expected to become an integral part of a future 5G radio system. As part of an overall interference mitigation scheme substantial gains in coverage, spectral as well as energy efficiency have been reported. One of the main limitations for massive MIMO and coordinated multi-point (CoMP) systems is the aging of the channel state information at the transmitter (CSIT), which can be overcome partly by state of the art channel prediction techniques. For a clean slate 5G radio system, we propose to integrate channel prediction from the scratch in a flexible manner to benefit from future improvements in this area. As any prediction is unreliable by nature, further improvements over the state of the art are needed for a convincing solution. In this paper, we explain how the basic ingredients of 5G like base stations with massive MIMO antenna arrays, and multiple UE antennas can help to stretch today's limits with an approximately 10 dB lower normalized mean square error (NMSE) of the predicted channel. In combination with the novel introduced concept of artificially mutually coupled antennas, adding super-directivity gains to virtual beamforming, robust and accurate prediction over 10 ms with an NMSE of -20 dB up to 15 km/h at 2.6 GHz RF frequency could be achieved. This result has been achieved for measured channels without massive MIMO, but a comparison with ray-traced channels for the same scenario is provided as well.