This paper introduces HiLoRa , a new LoRa system that multiplies per-node throughput by replacing standard chirps with sliced chirps. A sliced chirp is a segment of a standard chirp. Transmitting more of these partial chirps in a packet raises symbol density and hence data rate. Although slicing reduces the anti-interference capability of each symbol, we observe that interference rarely corrupts a complete symbol; at least one segment remains “clean”. We exploit a neural network model to demodulate the sliced chirps, and carefully design the input features for the model to avoid interference in low SNR scenarios. It is worth mentioning that HiLoRa can be successfully integrated into existing concurrent technologies and supports both orthogonal and non-orthogonal concurrent transmissions. We implement and validate HiLoRa in a 8 \(\times \) 10 km \(^2\) outdoor environment, the empirical evaluation demonstrates that the median throughput of HiLoRa is 6 \(\times \) higher than that of the state-of-the-art approaches.