In this letter, a machine learning approach is proposed to cancel the interference from multiple sources in the concurrent spectrum access (CSA) model of the cognitive Internet of Things (C-IoT). We assume that the C-IoT system is non-cooperative with the primary user (PU) system, and has little knowledge on the interference caused by PU transmitters. In order to recover the C-IoT signal under power strong multi-interference, we employ an iterative receiver consisting of a linear estimation module, and a demodulation-and-decoding module, as well as a clustering module by following the idea of iterative detection and interference cancellation. Conventional clustering algorithms, such as the K-means algorithm and the Gaussian mixture model (GMM) based expectation maximization (EM) algorithm, can potentially be used to realize the clustering module. However, their performance is poor under the existence of multiple interferers since the performance of these algorithms heavily relies on the initialization of cluster centroids. To address this problem, we use the affinity propagation (AP) algorithm, which works well without the initialization of cluster centroids, to realize the clustering module. Numerical results demonstrate that the AP algorithm has much better performance than the K-means and GMM-EM algorithms in handling multi-interference.

中文翻译：

---

基于机器学习的认知物联网迭代检测和多干扰消除

在这封信中，提出了一种机器学习方法来消除认知物联网 (C-IoT) 的并发频谱访问 (CSA) 模型中多源干扰。我们假设 C-IoT 系统与主用户 (PU) 系统不合作，并且对 PU 发射机造成的干扰知之甚少。为了在强强多重干扰下恢复C-IoT信号，我们采用迭代接收器，该接收器由线性估计模块和解调解码模块以及聚类模块组成，遵循迭代检测的思想和干扰消除。传统的聚类算法，例如 K-means 算法和基于高斯混合模型 (GMM) 的期望最大化 (EM) 算法，可以潜在地用于实现聚类模块。然而，由于这些算法的性能在很大程度上依赖于簇质心的初始化，因此在存在多个干扰源的情况下它们的性能很差。为了解决这个问题，我们使用亲和传播（AP）算法来实现聚类模块，该算法无需初始化聚类质心即可运行。数值结果表明，AP算法在处理多干扰方面比K-means和GMM-EM算法有更好的性能。