This paper optimizes predictive power allocation to minimize the average transmit power for video streaming subject to the constraint on stalling time, one of the most important factors affecting the experience of users requesting video-on-demand service. Different from the widely used first-predict-then-optimize strategy that regards the predicted channels as the real future channels, we integrate the prediction of large-scale channels into the optimization of power allocation, such that the quality of service constraint can be controlled. Due to the channel prediction errors, stalling is unavoidable and the stalling duration is random. This motivates us to consider an average stalling fraction constraint conditioned on the observed large-scale channel gains, which can be transformed into a conditional probabilistic constraint. The resultant optimization problem is difficult to solve since the probabilistic constraint lacks a closed-form expression. We resort to end-to-end deep learning to optimize the future powers from the past channels. In particular, we propose a method to learn the conditional probabilities in multiple steps with a single neural network. Simulation results show the advantages of the proposed method in reducing average power consumption and in ensuring the probabilistic constraint.

Neural networks (NNs) have been applied to solve various problems in ultra-reliable and low-latency communications (URLLC). Facing the stringent quality of service requirement, the time for training and running NNs is not negligible, and how to ensure the reliability with learning-based solutions is challenging, especially in a dynamic environment. In this paper, we propose an online learning method, which fine-tunes the NNs trained without supervision to ensure the reliability of URLLC for mobile users. A joint power and bandwidth allocation problem, aiming to minimize the bandwidth required for satisfying the quality of service of each user, is considered as an example. A "learning-to-optimize" method with offline training is provided for comparison. Simulation results show that the proposed online learning method can achieve comparable system performance as the offline training method, where the time consumed for online training and inference is about 25% of the 1 ms latency bound for the considered setup. Besides, the online learning method adapts to the abrupt change of average packet arrival rate quickly and can ensure reliability by setting the required overall packet loss probability conservative slightly. By contrast, the offline training method yields much worse reliability when the arrival rate varies.

As one of the key communication scenarios in the fifth-generation and also the sixth-generation (6G) mobile communication networks, ultrareliable and low-latency communications (URLLCs) will be central for the development of various emerging mission-critical applications. State-of-the-art mobile communication systems do not fulfill the end-to-end delay and overall reliability requirements of URLLCs. In particular, a holistic framework that takes into account latency, reliability, availability, scalability, and decision-making under uncertainty is lacking. Driven by recent breakthroughs in deep neural networks, deep learning algorithms have been considered as promising ways of developing enabling technologies for URLLCs in future 6G networks. This tutorial illustrates how domain knowledge (models, analytical tools, and optimization frameworks) of communications and networking can be integrated into different kinds of deep learning algorithms for URLLCs. We first provide some background of URLLCs and review promising network architectures and deep learning frameworks for 6G. To better illustrate how to improve learning algorithms with domain knowledge, we revisit model-based analytical tools and cross-layer optimization frameworks for URLLCs. Following this, we examine the potential of applying supervised/unsupervised deep learning and deep reinforcement learning in URLLCs and summarize related open problems. Finally, we provide simulation and experimental results to validate the effectiveness of different learning algorithms and discuss future directions.

User fairness is an important metric for cellular systems. It has been widely considered for wireless transmission when optimizing radio resource allocation but rarely considered for femto-caching. In this paper, we optimize caching and bandwidth allocation policies to improve long-term user fairness during content placement and content delivery by harnessing heterogeneous user preference. To this end, we maximize the minimal average data rate, where the average is taken over large-and small-scale channel gains as well as individual user requests. This gives rise to a complicated two-timescale optimization problem involving functional optimization. The objective function of the problem does not have closed-form expression due to unknown user preference and channel distributions, and the “variables” to be optimized include a function. To solve such a challenging problem, we first optimize bandwidth allocation policy given arbitrary caching policy, user locations and user requests, whose structure can be found. We next optimize the caching policy given the optimized bandwidth allocation policy. To handle the difficulty of unknown distributions, we resort to stochastic optimization. Simulation results show that optimizing caching policy exploiting user preference can support much higher minimal average rate than optimizing caching policy based on content popularity when user preferences are less similar. Besides, better user fairness can be achieved by optimizing caching policy than by optimizing bandwidth allocation.

Wireless systems are becoming more and more complicated. As a consequence, the expressions of objective function or constraints of many optimization problems are hard or even impossible to derive. In this paper, we propose a model-free framework to learn the mapping from environment parameters to the solutions of generic constrained optimization problems without the labels generated by numerically finding the optimal solution. We use neural networks respectively for parameterizing the policy to be optimized, the Lagrange multiplier function associated with instantaneous constraint, and approximating the unavailable objective function or constraints. We provide learning algorithms to train all the neural networks simultaneously. We reveal the connections of the proposed framework with reinforcement learning, which is a widely recognized tool for model-free problems. Numerical and simulation results demonstrate the efficiency of model-free learning by taking a well-known power control problem as an example.

In this paper, we study how to solve resource allocation problems in ultra-reliable and low- latency communications by unsupervised deep learning, which often yield functional optimization problems with quality-of-service (QoS) constraints. We take a joint power and bandwidth allocation problem as an example, which minimizes the total bandwidth required to guarantee the QoS of each user in terms of the delay bound and overall packet loss probability. The global optimal solution is found in a symmetric scenario. A neural network was introduced to find an approximated optimal solution in general scenarios, where the QoS is ensured by using the property that the optimal solution should satisfy as the ''supervision signal''. Simulation results show that the learning-based solution performs the same as the optimal solution in the symmetric scenario, and can save around 40% bandwidth with respect to the state-of-the-art policy.

Learning the optimized solution as a function of environmental parameters by deep neural networks (DNN) is effective in solving numerical optimization in real time for time-sensitive resource allocation in wireless systems. Existing works of learning to optimize train the DNN with labels, which are generated by solving the optimization problems. The learned solution are often inaccurate and hence cannot be employed to ensure the stringent quality of service. In this paper, we propose a framework to learn the latent function with unsupervised deep learning, where the property that the optimal solution should satisfy is used as the "supervision signal" implicitly. The framework is applicable to both variable and functional optimization problems with constraints, which are respectively formulated to optimize variables and functions of concern. We take a variable optimization problem in ultra-reliable and low-latency communications as an example, which demonstrates that the ultra-high reliability can be supported by the DNN without supervision labels.

In this paper, we aim to find the global optimal resource allocation for ultra-reliable and low-latency communications (URLLC), where the blocklength of channel codes is short. The achievable rate in the short blocklength regime is neither convex nor concave in bandwidth and transmit power. Thus, a non-convex constraint is inevitable in optimizing resource allocation for URLLC. We first consider a general resource allocation problem with constraints on the transmission delay and decoding error probability, and prove that a global optimal solution can be found in a convex subset of the original feasible region. Then, we illustrate how to find the global optimal solution for an example problem, where the energy efficiency (EE) is maximized by optimizing antenna configuration, bandwidth allocation, and power control under the latency and reliability constraints. To improve the battery life of devices and EE of communication systems, both uplink and downlink resources are optimized. The simulation and numerical results validate the analysis and show that the circuit power is dominated by the total power consumption when the average inter-arrival time between packets is much larger than the required delay bound. Therefore, optimizing antenna configuration and bandwidth allocation without power control leads to minor EE loss.

In this work, we study the benefit of retransmission to ultra-reliable and low-latency communications (URLLC). We consider a retransmission policy that employs frequency hopping to improve the retransmission success probability for the packets suffering deep fading. We investigate how to satisfy the quality-of-service (QoS) with retransmission by taking downlink transmission as an example. End-to-end (E2E) delay components, including transmission delay, queueing delay and backhaul latency, and packet loss components, including transmission error probability and E2E delay violation probability are considered. Resource allocation for the retransmission policy is optimized. Numerical results show that the requirement on transmission error probability can be significantly relaxed with retransmission, which does not compromise the E2E delay requirement owing to the short packets in URLLC. Simulation results indicate that more users can be supported with QoS guarantee compared to a counterpart system without retransmission, and the performance gain increases with the times of retransmissions.

Ultra-reliable and low-latency communications (URLLC) is expected to be supported without compromising the resource usage efficiency. In this paper, we study how to maximize energy efficiency (EE) for URLLC under the stringent quality of service (QoS) requirement imposed on the end-to-end (E2E) delay and overall packet loss, where the E2E delay includes queueing delay and transmission delay, and the overall packet loss consists of queueing delay violation, transmission error with finite blocklength channel codes, and proactive packet dropping in deep fading. Transmit power, bandwidth and number of active antennas are jointly optimized to maximize the system EE under the QoS constraints. Since the achievable rate with finite blocklength channel codes is not jointly concave in radio resources, it is challenging to optimize resource allocation. By analyzing the properties of the optimization problem, the global optimal solution is obtained. Simulation and numerical results validate the analysis and show that the proposed policy can improve EE significantly compared with existing policy.