This paper investigates the problems of interference prediction and sensing for efficient spectrum access and link adaptation. The considered approach for interference prediction relies on a parametric model. However, we assume that the number of observations available to learn theses parameters is limited. This implies that they should be treated as random variables rather than fixed values. We show how this can impact the spectrum access and link adaptation strategies. We also introduce the notion of "interferer-coherence time" to establish the number of independent interferer state realizations experienced by a codeword. We explain how it can be computed taking into account the model uncertainty and how this impacts the link adaptation.

翻译：本文研究面向高效频谱接入与链路自适应的干扰预测与感知问题。所采用的干扰预测方法依赖于参数化模型，但我们假设用于学习这些参数的观测样本数量有限。这意味着应将参数视为随机变量而非固定值。我们阐述了这一特性对频谱接入和链路自适应策略的影响，并引入"干扰源相干时间"概念来确定码字所经历的独立干扰源状态实现次数。本文进一步解释了如何考虑模型不确定性来计算该相干时间，以及它对链路自适应的具体影响。