Abstract
We propose a scheme to reduce the overhead associated with channel state information (CSI) feedback required for opportunistic scheduling in wireless access networks. We study the case where CSI is partially overheard by mobiles and thus one can suppress transmitting CSI reports for time varying channels of inferior quality. We model the mechanism of feedback suppression as a Bayesian network, and show that the problem of minimizing the average feedback overhead is NP-hard. To deal with hardness of the problem we identify a class of feedback suppression structures which allow efficient computation of the cost. Leveraging such structures we propose an algorithm which not only captures the essence of seemingly complex overhearing relations among mobiles, but also provides a simple estimate of the cost incurred by a suppression structure. Simulation results are provided to demonstrate the improvements offered by the proposed scheme, e.g., a savings of 63-83% depending on the network size.
Introduction
In modern wireless networks, such as those based on IEEE 802.11n, Access Points (APs) utilize opportunistic scheduling to maximize throughput by prioritizing users with the strongest instantaneous signal quality. This requires precise Channel State Information (CSI) for every user across multiple frequency subcarriers. However, as the number of users and subcarriers increases, the CSI feedback itself becomes a significant source of overhead, consuming bandwidth that could otherwise be used for data transmission.
Traditional scheduling assumes users are isolated. In reality, mobile devices in proximity can often "overhear" each other's transmissions. This paper introduces a scheme where devices exploit this overhearing to suppress redundant feedback: if a device hears a neighbor report a channel quality superior to its own, it remains silent.
Methodology: Exploiting Overhearing and Bayesian Networks
The proposed Greedy Quantile-Based Sequential Feedback (G-QBSF) scheme leverages the "silence is golden" principle. The AP does not need to know every user's channel state; it only needs to identify the user with the best channel. By establishing a reporting order, the network performs a distributed "max" operation.
Bayesian Network Modeling
The core of the G-QBSF scheme involves modeling the "who-overhears-whom" relationships. These relationships are represented as a Directed Acyclic Graph (DAG), where nodes are mobile devices and edges represent overhearing capabilities. The probability of a node transmitting feedback is then modeled as a marginalization problem within a Bayesian Network.
Max-Quantile (MQ) Scheduling and Fairness
To ensure fairness, the system uses "quantiles" rather than raw Signal-to-Noise Ratios (SNR). A quantile represents a signal's strength relative to its own historical distribution. This prevents users physically closer to the AP (who naturally have higher SNR) from permanently silencing distant users, maintaining opportunistic gain while ensuring all users have a chance to transmit.
The G-QBSF Algorithm
Finding the optimal reporting order is computationally NP-hard. To address this, the authors identified that for "singly connected" DAGs (polytrees), the overhead cost can be computed in polynomial time. The G-QBSF algorithm uses a greedy heuristic to build a reporting sequence step-by-step, selecting the next node that maximizes the estimated reduction in total network overhead.
Evaluation and Results
The scheme was validated using synthetic numerical simulations. Overhearing graphs were generated randomly, and channel conditions were modeled as i.i.d. uniform random variables representing SNR quantiles.
Key findings include:
- Significant Overhead Reduction: The scheme reduced feedback overhead by 63% to 81% in uniform scenarios.
- Heterogeneous Gains: In networks with a mix of device types (non-uniform weights), savings reached up to 83%.
- Algorithmic Efficiency: G-QBSF outperformed random reporting orders by 24% to 36% and performed near-optimally in small-scale tests.
Conclusion and Future Work
By exploiting overhearing and modeling feedback as a Bayesian Network, this research demonstrates that significant bandwidth can be reclaimed from CSI overhead. While the current model assumes a relatively static topology, future work could explore highly mobile environments and the application of these techniques to data fusion in sensor networks.
Summary updated