Scheduling with limited information in wireless systems

Abstract

Opportunistic scheduling is a key mechanism for improving the performance of wireless systems. However, this mechanism requires that transmitters are aware of channel conditions (or CSI, Channel State Information) to the various possible receivers. CSI is not automatically available at the transmitters, rather it has to be acquired. Acquiring CSI consumes resources, and only the remaining resources can be used for actual data transmissions. We explore the resulting trade-off between acquiring CSI and exploiting channel diversity to the various receivers. Specifically, we consider a system consisting of a transmitter and a fixed number of receivers/users. An infinite buffer is associated to each receiver, and packets arrive in this buffer according to some stochastic process with fixed intensity. We study the impact of limited channel information on the stability of the system. We characterize its stability region, and show that an adaptive queue length-based policy can achieve stability whenever doing so is possible. We formulate a Markov Decision Process problem to characterize this queue lengthbased policy. In certain specific and yet relevant cases, we explicitly compute the optimal policy. In general case, we provide a scheduling policy that achieves a fixed fraction of the system's stability region. Scheduling with limited information is a problem that naturally arises in cognitive radio systems, and our results can be used in these systems.

Energy and Bursty Packet Loss Tradeoff Over Fading Channels: A System-Level Model

“…Specially, availability of CSI at the transmitter side in a fast mobility scenario is very complex and the cost is enormous. This leads to a tradeoff between exploration and exploitation [17], [18].…”

“…The channel distribution for the scheduled VUs is computed using fading distribution in (18) and the path loss distribution in (2). Thus, the optimization problem is formulated as…”

Network utility maximization over partially observable Markovian channels

“…Namely, let s n (t) denote the state of channel n in slot t. Then ω n (t) Pr [s n (t) = ON | channel observation history] . (3) We will show later ω n (t) takes values in a countably infinite set. Thus computing Λ and solving (1)-(2) seem to be infea- normalized slots t ∈ Z + .…”

“…The problem (1)- (2) we consider here generalizes the network utility maximization framework in [1] to networks with limiting channel probing capability (see [? ], [2] and references therein) and delayed/uncertain channel state information (see [3]- [5] and references therein), in which we shall take advantage of channel memory [6] to improve network performance. In sequential decision making, (1)-(2) also captures an important class of restless bandit problems [7] in which each Markovian channel represents a two-state restless bandit, and packets served over a channel are rewards from playing the bandit.…”

On the benefits of edge caching for MIMO interference alignment

“…In frequency-division duplex (FDD) systems, the imperfect case with CSI quantization process for single-antenna receivers [15], and multipleantenna receivers [16], [17] are studied, showing that the degree-of-freedom (DoF) can be achieved at high signal-to-noise ratio (SNR) regime by using a specific quantization scheme with optimal number of feedback bits. The IA methods that exploit channel reciprocity in timedivision duplex (TDD) systems are studied (see [18]- [21] for instance), assuming that the CSI acquisition cost is independent of the transmission rate and is linear in the number of probed receivers. In fact, most of aforementioned IA methods rely on CSI exchange over the backhaul links and do not consider the implications of data traffic on the limited backhaul links and exchange process.…”