Energy-Efficient Transmission Scheduling with Strict Underflow Constraints

TL;DR

This paper develops optimal transmission scheduling policies for wireless media streaming that minimize power and holding costs while satisfying strict buffer underflow constraints across multiple users.

Contribution

It characterizes optimal policies for single and multi-user scenarios with various power-rate curves, extending to practical multi-user cases.

Findings

Modified base-stock policy is optimal for single user with linear power-rate curves.

Finite generalized base-stock policy is optimal for piecewise-linear convex power-rate curves.

Structural analysis of policies for two-user case and methods for multi-user approximation.

Abstract

We consider a single source transmitting data to one or more receivers/users over a shared wireless channel. Due to random fading, the wireless channel conditions vary with time and from user to user. Each user has a buffer to store received packets before they are drained. At each time step, the source determines how much power to use for transmission to each user. The source's objective is to allocate power in a manner that minimizes an expected cost measure, while satisfying strict buffer underflow constraints and a total power constraint in each slot. The expected cost measure is composed of costs associated with power consumption from transmission and packet holding costs. The primary application motivating this problem is wireless media streaming. For this application, the buffer underflow constraints prevent the user buffers from emptying, so as to maintain playout quality. In the case of a single user with linear power-rate curves, we show that a modified base-stock policy is optimal under the finite horizon, infinite horizon discounted, and infinite horizon average expected cost criteria. For a single user with piecewise-linear convex power-rate curves, we show that a finite generalized base-stock policy is optimal under all three expected cost criteria. We also present the sequences of critical numbers that complete the characterization of the optimal control laws in each of these cases when some additional technical conditions are satisfied. We then analyze the structure of the optimal policy for the case of two users. We conclude with a discussion of methods to identify implementable near-optimal policies for the most general case of M users.