Narrowband Internet of Things (IoT) is broadly utilized in scenarios involving large-scale low-power terminal access. The accuracy of channel state acquisition in complex propagation environments directly affects system reliability and spectral efficiency. To address the problems of high pilot overhead and poor stability in traditional pilot-based channel estimation methods under low signal-to-noise ratio and small sample conditions, a semi-blind channel estimation model integrating density clustering and narrowband IoT architecture is constructed. The reference signal is only used to obtain a low-overhead preliminary channel response, while DBSCAN-based density clustering performs sample denoising, effective path separation and weighted parameter recovery. This method establishes a frequency domain model based on the orthogonal frequency division multiplexing signal structure and multipath channels, utilizes density-based spatial clustering to group and denoise received samples, and achieves weighted estimation of channel parameters within effective clusters. Simulation results show that in comprehensive testing, the normalized mean square error is 0.028, the bit error rate is reduced to 2.1 × 10 −4 , the silhouette coefficient reaches 0.98, the effective sample retention rate is 96.8%, and the average runtime is 214 ms. In generalization testing on the COST259 dataset, the normalized mean square error is further reduced to 0.026, and the bit error rate is as low as 1.9 × 10 −4 , significantly outperforming the comparative methods. The studies show that density clustering-based blind estimation methods can effectively extract path features in complex multipath and noisy environments, significantly improving channel estimation accuracy and system reliability.
Fu et al. (Wed,) studied this question.