A novel compressed sensing-aided generalised spacefrequency index modulation (CS-GSFIM) scheme is conceived for the large-scale multiuser multiple-input multiple-output uplink (LS-MU-MIMO-UL). Explicitly, the information bits are mapped both to the spatial-and frequency-domain indices, where we treat the activation patterns of the transmit antennas (TAs) and of the subcarriers separately. Specifically, our indexing strategy strikes a flexible trade-off between the throughput (Tp), performance and complexity. In order to further increase the system's achievable rate, CS-aided pre-processing is applied to the subcarriers. An upper bound of the average bit error probability (ABEP) of the proposed system using the optimal maximum likelihood (ML) detector is derived, which is shown to be tight by our simulation results at moderate to high signal-to-noise ratios (SNRs). Then we design a CS-aided reduced-complexity detector, namely the reduced search-space based iterative matching pursuit (RSS-IMP), which significantly reduces the detection complexity compared to the ML detection and makes the proposed design a feasible one for LS-MU scenarios. Furthermore, the simulation results presented in this paper demonstrate that the proposed RSS-IMP detector significantly reduces the detection complexity, while attaining better performances than both the conventional MU-MIMO-OFDM system using the ML detector and the proposed system using the minimum mean square error (MMSE) detector. We also characterise the performances of the proposed system in the presence of channel estimation errors. Our simulation results show that the proposed CS-GSFIM system is more robust to imperfect channel than the conventional MU-MIMO-OFDM system. In order to achieve a near-capacity performance, softinput soft-output (SISO) decoders are designed for the proposed CS-GSFIM system using both the ML and the RSS-IMP multiuser detectors (MUDs) for detecting all users.