eduzhai > Applied Sciences > Engineering >

Unsupervised Deep Learning for Optimizing Wireless Systems with Instantaneous and Statistic Constraints

  • Save

... pages left unread,continue reading

Document pages: 30 pages

Abstract: Deep neural networks (DNNs) have been introduced for designing wirelesspolicies by approximating the mappings from environmental parameters tosolutions of optimization problems. Considering that labeled training samplesare hard to obtain, unsupervised deep learning has been proposed to solvefunctional optimization problems with statistical constraints recently.However, most existing problems in wireless communications are variableoptimizations, and many problems are with instantaneous constraints. In thispaper, we establish a unified framework of using unsupervised deep learning tosolve both kinds of problems with both instantaneous and statistic constraints.For a constrained variable optimization, we first convert it into an equivalentfunctional optimization problem with instantaneous constraints. Then, to ensurethe instantaneous constraints in the functional optimization problems, we useDNN to approximate the Lagrange multiplier functions, which is trained togetherwith a DNN to approximate the policy. We take two resource allocation problemsin ultra-reliable and low-latency communications as examples to illustrate howto guarantee the complex and stringent quality-of-service (QoS) constraintswith the framework. Simulation results show that unsupervised learningoutperforms supervised learning in terms of QoS violation probability andapproximation accuracy of the optimal policy, and can converge rapidly withpre-training.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×