Predictive scheduling in multi-carrier wireless networks with link adaptation

Abstract

Abstract-Channel-aware scheduling and link adaptation methods are widely considered to be crucial for realizing high data rates in wireless networks. However, predicting the future channel states, and adjusting the transmission schedules and parameters accordingly, may consume valuable system resources, such as bandwidth, time, and power. This paper considers the trade-offs between the prediction quality and the throughput in a wireless network that uses link adaptation and channel-aware scheduling. In particular, we study the effects of the look-ahead window, i.e., the range of future time slots on which we have channel state estimates, and the reliability of the channel state estimates on the throughput. We develop an online scheduling algorithm for a multi-channel multi-user network that employs predictive link adaptation, and generalize it to incorporate imperfect channel state estimates. We use this heuristic together with performance bounds to the offline version of the problem to evaluate the performance with varying prediction quality. Our results suggest that it may be possible to reap most of the potential channel-aware scheduling benefits with a small look-ahead, and imperfect channel state estimates. Thus, a modest consumption of resources for channel prediction and link adaptation may result in a significant throughput improvement, with only marginal gains through further enhancement of the prediction quality. Our results can provide meaningful guidelines in deciding what level of system resource consumption is justified for channel quality estimation and link adaptation. I. MOTIVATIONThis paper considers the downlink scheduling problem in a multi-carrier wireless network that uses link adaptation. Link Adaptation (LA), which loosely refers to changing transmission parameters over a link, such as modulation, coding rate, power, etc., in response to changing channel conditions is considered to be a powerful means of achieving higher efficiency or throughput in wireless networks. The adaptation of the transmission parameters is performed according to the predicted future quality of the channel, also called as the channel state (CS).While it is desirable to adapt the transmission parameters according to the channel state information (CSI) to capture even small-scale variations, there are practical limitations to channel state prediction and link adaptation. Frequent adaptation increases the number of mode-change messages transmitted over the channel, consuming bandwidth, and time resources [1]. While many aspects of scheduling transmissions over time-varying wireless channels have been studied (see, for example, [2], [3], [4] and the references therein), the penalty induced by LA has not been considered. Moreover, predicting the future channel quality may also consume significant amount of system resources (e.g., time, bandwidth and power), since it may involve transmission of training-sequences, pilot tones, or feedback messages carrying the CSI. Naturally, the additional cost and complication of pred...

Wireless Schedulers with Future Sight via Real-Time 3D Environment Mapping

“…With the dependence on future, and thus uncertain information, a clear avenue for further work is to include the impact of incorrect prediction on the future-based schedulers, a problem considered also in [7], [9], [11]. Within the context of the ViewNet project, since this is producing a functioning hardware system, it would be instructive to use real measurements of throughput tied to position to provide data to the simulators.…”

“…Other predictive scheduling work includes [9] which is effectively a greedy search scheduler over some future timeslots, extended in [10] to consider the impact of re-configuring the system when the scheduler requires a change. An approach for a different family of scheduler is suggested in [11], which alters a simple scheduling metric to include future information in a way reminiscent of this paper, but for the very different metric they use, and less flexibly than here.…”

Predictive Scheduling in Rate-Adaptive Multi-user Relay Channels with Reconfiguration Delays

“…Lemma 1: When ∆ = 1, the off-line multi-rate relay scheduling problem is equivalent to the predictive multi-user multi-channel scheduling problem considered in [3].…”

“…A common assumption made in opportunistic scheduling is that rate adaptation and channel reassignment can be implemented with no delay, based on the instantaneous channel state information. In [3], a predictive version of the problem is considered, where it is assumed that estimates of the channel states over the next K time-slots are available to the system, where K is called the look-ahead. We note that the channel state in this case does not necessarily refer to the instantaneous channel quality for the duration of a symbol, but rather to a high-level estimate that quantifies the expected throughput during a longer period.…”

“…Such estimates may be available, for example, based on the channel characteristics and the mobility pattern of a user. Another assumption made in [3], for the specific architecture considered in [3], is the reconfiguration delay associated with rate and channel reassignment, which was modeled as a one-slot loss of service. Heuristic algorithms were presented for this problem, assuming a look-ahead of 1.…”