E2E-Aware Multi-Service Radio Resource Management for 5G New Radio: Radio Access and Resource Management Solutions

The fifth generation (5G) of the cellular technology offers greater support for three main service classes; the ultra-reliable and low-latency communications (URLLC), enhanced mobile broadband (eMBB), and the massive machine type communication (mMTC). URLLC services require the transmission of sporadic and small-payload packets with stringent radio latency and reliability targets. The eMBB applications demand wide-band transmissions with extreme peak data rates. Finally, for mMTC, the network is required to simultaneously serve a large number of connected devices, each is associated with strict energy consumption constraints. However, there is a fundamental tradeoff between the achievable latency, reliability, and network spectral efficiency. Concurrently optimizing the quality of service (QoS) of those service classes is one of the major challenges of the 5G new radio and neither been addressed for the former wireless standards. Furthermore, the 5G new radio is designed to support both the frequency and time division duplexing (FDD, TDD) modes. And due to the abundantly available bandwidth at the 3.5 GHz unpaired spectrum, most of the early 5G deployments are envisioned with the TDD duplexing technology. However, achieving such an efficient multi-service-aware resource management is further challenging with TDD. The broader scope of this PhD. project is to research and develop novel and multi-service-aware radio resource management algorithms for multi-QoS 5G networks, spanning both FDD and TDD modes.

The first part addresses the multi-QoS (URLLC-eMBB) multiplexing problem. A QoS-aware multi-user multiple-input multiple-output (MU-MIMO) downlink scheduler is developed based subspace projections. The key idea is to eliminate the scheduling queuing delay of the newly-arriving URLLC packets in case the sufficient radio re-sources are not immediately available. The incoming URLLC transmissions are instantly paired with the active eMBB users which spatial signatures are closest possible to a pre-defined subspace. To control the inter-user interference at the critical URLLC users, the co-scheduled eMBB transmissions are spatially projected on-the-fly into an arbitrary spatial sub-space, to which the paired URLLC users align their respective transceivers into the orthonormal subspace, exhibiting substantially zero eMBB interference. Moreover, we have developed several variants of the proposed scheduler for eMBB capacity recovering and spectral efficiency optimization. We adopt highly-detailed system level simulations, with a high degree of realism in line with 3GPP NR assumptions, to evaluate the performance of the proposed schemes. Our simulation results demonstrate considerable improvements of the URLLC outage latency and the network capacity, e.g., minimizing the URLLC outage latency by 50 percent while enhancing the network capacity by 79 percent, compared to Rel-15 standard URLLC scheduler.

In the second part of the study, we target achieving the stringent URLLC outage targets in TDD 5G networks. We first demonstrate that the URLLC QoS is further harder to achieve in TDD deployments, mainly due to the TDD frame structure, i.e., no simultaneous downlink and uplink transmissions are possible, and the severe cross-link interference (CLI) when neighboring base-stations or users are adopting opposite transmission directions. A diversity of novel inter-cell coordination schemes are developed for mitigation of the critical CLI. Those schemes incorporate a new set of TDD system design
improvements such as semi-static frame configuration, sliding frame-book design, joint hybrid frame design and slot-aware user scheduling, and coordinated transceiver design. Accordingly, developed coordination techniques offer a wide variety of the required inter-cell signaling over-head, TDD frame adaptation flexibility, and the achievable URLLC outage performance. Our results show a no-table URLLC outage improvement compared to standard dynamic TDD setups, e.g., 80 percent URLLC outage latency reduction.

Backed by our former conclusions, the last part of the PhD project demonstrates the potential of adopting a machine learning (ML) algorithms for real-time selection of the TDD radio frame structure. A simple, but efficient, Q-reinforcement-learning (QRL) approach for distributed online TDD frame optimization is proposed. First, a QRL network is utilized to estimate the near-optimal numbers of downlink and uplink transmission opportunities for a balanced traffic handling. A secondary QRL instance is selects the corresponding downlink and uplink symbol structure that minimizes the directional URLLC tail latency. The QRL-based solution is evaluated for both macro networks and newly emerging indoor industrial wireless deployments with dense small cell layouts. The proposed solution offers a significant URLLC outage gain in terms of autonomization of the TDD frame design on a real-time basis, URLLC outage latency reduction, and CLI-avoidance.