Simultaneous acquisition of channel state information (CSI) of many devices with high accuracy is crucial to provide the Internet-of-Things (IoT) connectivity in cellular networks. A traditional channel estimator in cellular networks adopts the orthogonal pilot structure, which cannot provide high scalability for many IoT devices without a significant increase of pilot overhead. To overcome such a limitation, we propose a novel channel estimator based on non-orthogonal pilot signals for frequencydivision duplex (FDD) based uplink cellular IoT networks. The proposed scheme provides an interferencecanceled environment during the channel estimation procedure by reconstructing the non-orthogonal pilot signals and removing their effects. Consequently, the proposed scheme can increase the number of available pilot signals without any increase of pilot overhead, which makes the BS spectral-efficiently accommodate more IoT devices with full exploitation of antenna technique. The numerical results verify the effectiveness of the proposed scheme on supporting more IoT devices at the expense of a slight loss of the channel estimation accuracy, compared to the conventional discrete Fourier transform (DFT)-based channel estimator, from the viewpoint of a mean squared error, a bit error rate, and a network throughput.INDEX TERMS Channel estimation, non-orthogonal pilot signals, Zadoff-Chu sequence, the Internet-of-Things, cellular networks.