Online resource allocation for unpredictable large-scale wireless systems – ORACLESS

Greatly raising the bar from previous generation upgrades, fifth-generation (5G) mobile systems are promising fiber-like connectivity and a “faster-than-thought” Internet experience to a diverse ecosystem of users and devices with highly variable attributes and demands. As a result, the successful deployment of 5G systems requires a drastic rethinking of existing resource allocation methodologies with the goal of enabling resource-limited users and devices to adapt “on the fly” to a rapidly – and unpredictably – changing wireless landscape.

So far, the vast majority of works on wireless resource allocation (spectrum, power, etc.) has focused on two limit cases: In the static regime, the attributes of the network are assumed effectively static and the system’s optimality analysis relies on techniques from optimization, game theory and (optimal) control. On the other hand, in the so-called stochastic regime, the network is assumed to evolve randomly following some stationary probability law, and the allocation of wireless resources is optimized using tools from stochastic optimization and control. In practical wireless networks however, both assumptions fail because of factors that introduce an unpredictable variability to the system (such as user mobility, users going arbitrarily on- and off-line, non-random channel fluctuations, etc.). As a result, existing resource allocation schemes do not – in fact, cannot – apply in this setting because “optimum” target states no longer exist, either static or in the mean.

In view of the above, ORACLESS envisions a drastic turn towards a flexible, oracle-less resource allocation paradigm (i.e. without any prior system knowledge) based on the deployment of online optimization protocols at the network’s edge (the system’s wireless devices). Our targeted breakthrough will thus be to develop highly adaptive resource allocation schemes that are provably capable of tracking unpredictable changes in the network: in so doing, the distribution of online optimization protocols at the device end will act as an effective multiplier of wireless resources, ensuring at the same time the system’s robust, self-healing operation in the presence of variabilities and fluctuations.

Specifically, driven by the leading role played by massive multiple-input, multiple-output (MIMO) and cognitive radio (CR) technologies in the ongoing transition to 5G mobile systems, we intend to focus on the following objectives of high practical relevance:

1. Adaptive optimization schemes for massive MIMO systems: in particular, our aim will be to develop adaptive algorithms for tracking optimum transmit attributes in unpredictable MIMO environments, to safeguard the algorithms’ efficient operation under feedback imperfections, and to resolve the semidefinite computational bottlenecks that arise in the case of massively large antenna arrays.

2. Online policies for dynamic and opportunistic spectrum access: namely, methods for following the equilibrium state of a dynamic network with CR capabilities, to ensure the methods’ robustness against asynchronicities, and to mitigate the effects of user heterogeneity on their performance.

Tackling these objectives will require an interdisciplinary blend of techniques from online optimization, learning, game theory, stochastic approximation, and information theory. Thus, given the diverse expertise of the project’s members, the ORACLESS team is uniquely poised to successfully address the challenges identified above and, in so doing, to establish a solid presence in the mature stage of the 5G standardization process where the real identity of 5G will be uncovered.