LoRa (Long Range) technique is an improved high-order chirp spread spectrum (CSS) modulation technique, and becomes one of the mainstream modulation techniques for terrestrial internet of things (IoT). Different from the terrestrial IoT, satellite IoT can achieve seamless global coverage, however, the corresponding receiver often works under the extremely-low signal-to-noise ratio (SNR) and also faces large delay and Doppler shift due to long distance and high-speed movement. To this end, this paper applies the LoRa technique to satellite IoT systems and studies the relevant receiver synchronization as well as noncoherent soft demodulation scheme. First, considered system model and data frame structure of the uplink satellite IoT are given. Then, according to the feature of LoRa signals, a joint delay and Doppler shift estimation scheme is proposed. In order to further eliminate the impact of the phase offset as well as other residual terms and consider using soft-decision decoding based coding schemes, a low-complexity noncoherent soft demodulation algorithm is also designed. Simulation results show that the turbo-LoRa system based on the proposed synchronization scheme and soft demodulation algorithm can achieve near ideal error performance under extremely low SNR and small pilot overhead, and can provide 3dB gains over the Hamming-LoRa system at a bit error rate (BER) of 10^-5.