Multi-task Learning Dataset for Automatic Modulation Classification and DOA Estimation

A synthetic signal dataset of 12 different modulations (including PSK, QPSK, 8PSK, QFSK, 8FSK, 16APSK, 16QAM, 64QAM, 4PAM, LFM, DSB-SC, and SSBSC) with different DOAs (discrete angles ranging from -60° to 60° with the step size of 1°) is generated using MATLAB 2021a. Regarding the signal model configuration for the data generation, we specify a uniform linear antenna array of M = 5 elements to acquire incoming signals having N = 1024 envelope complex samples, thus conducting an I/Q data array of size 1024 × 2 × 5. To model real-world phenomena in wireless communication, various impairments are considered: the Gaussian noise in the range [-10, 20] dB (step size of 1 dB), the multi-path propagation with a random number of NLOS signals in the range of [1, 10], the propagation attenuation αp and delay τp randomly distributed in the range [-50, -1] dB and [1, 3000] ns, respectively. The dataset contains 450120 signals covering 121 DOA classes, 12 modulation classes at 31 SNR levels, where each signal has 1024 samples to form into an input data array with the size of 1024x2x5.

Please refer and cite our paper as follows: V. -S. Doan, T. Huynh-The, V. -P. Hoang and D. -T. Nguyen, "MoDANet: Multi-task Deep Network for Joint Automatic Modulation Classification and Direction of Arrival Estimation," in IEEE Communications Letters, doi: 10.1109/LCOMM.2021.3132018.