KANN-DBSCAN algorithm for blind channel estimation in NB-IoT

Blind channel estimation in Narrowband Internet of Things links is challenged by weak coverage, multipath superposition, and locally non-uniform complex-domain observations. Manually tuned density clustering and global-bandwidth kernel density estimation may therefore produce unstable cluster structures and oversmooth weak propagation modes. To address these problems, a K-Nearest Adaptive Neighborhood and Density-Based Spatial Clustering of Applications with Noise blind channel estimation method is proposed. K-nearest adaptive neighborhood statistics and expected neighborhood counts are used to generate paired Eps and MinPts candidates, after which a cluster-count stability and noise-ratio rule determines the clustering parameters. Robust cluster-specific bandwidths are then assigned for local kernel density estimation, and the equivalent channel responses are recovered through posterior weighting and block-energy rescaling. At an SNR of 0 dB, the proposed method achieves an NMSE of −13.14 dB and a BER of 9.71%, outperforming the strongest external baseline by 1.10 dB and 0.91 percentage points, respectively. At an SNR of 12 dB, its NMSE and BER reach −23.11 dB and 0.31%. In the SDR playback test under ETU-300 Hz, the proposed method retains an NMSE of −13.14 dB and a BER of 7.03%, compared with −12.02 dB and 8.76% for GM-LAMP. The ablation, sensitivity, and density-model comparisons further confirm the complementary contributions of adaptive parameter selection and cluster-specific bandwidth estimation. The method is suitable for base-station or edge-assisted NB-IoT channel recovery, although its quadratic distance computation introduces additional runtime and memory costs.