We study large-alphabet quantum key distribution (QKD) based on the use of weak-coherent states and the time–frequency uncertainty relation. The large alphabet is achieved by dividing time and spectrum into M bins resulting in a frame similar to traditional pulse-position modulation (in time domain). However, the non-uniform occurrence of a photon prepared in a time/frequency bin creates the space for eavesdropping. By analysis, we show that a new intercept-resend attack strategy exists, which is stronger than that has been reported in the literature and hence the secret key rate of time-frequency QKD (TF-QKD) can be more tightly bounded. We then analyse the secret key rates of TF-QKD under various practical issues, such as channel loss, background noise, jitter and atmospheric turbulence in order to better understand the applicability of TF-QKD. Further, we discuss the information reconciliation for TF-QKD. Specifically, we investigate the layered coding scheme for TF-QKD based on quasi-cyclic low-density parity-check codes against jitter and atmospheric turbulence. By simulation, we demonstrate that information reconciliation can be efficiently achieved.