In the absence of complex astrophysical processes that characterize the reionization era, the 21-cm emission from neutral hydrogen (H i) in the post-reionization epoch is believed to be an excellent tracer of the underlying dark matter distribution. Assuming a background cosmology, it is modelled through (i) a bias function b(k, z), which relates H i to the dark matter distribution and (ii) a mean neutral fraction () which sets its amplitude. In this paper, we investigate the nature of large-scale H i bias. The post-reionization H i is modelled using gravity only N-body simulations and a suitable prescription for assigning gas to the dark matter haloes. Using the simulated bias as the fiducial model for H i distribution at z≤ 4, we have generated a hypothetical data set for the 21-cm angular power spectrum (Cℓ) using a noise model based on parameters of an extended version of the Giant Metrewave Radio Telescope (GMRT). The binned Cℓ is assumed to be measured with S/N ≳ 4 in the range 400 ≤ℓ≤ 8000 at a fiducial redshift z= 2.5. We explore the possibility of constraining b(k) using the principal component analysis on these simulated data. Our analysis shows that in the range 0.2 < k < 2 Mpc−1, the simulated data set cannot distinguish between models exhibiting different k-dependences, provided 1 ≲b(k) ≲ 2 which sets the 2σ limits. This justifies the use of linear bias model on large scales. The largely uncertain is treated as a free parameter resulting in degradation of the bias reconstruction. The given simulated data are found to constrain the fiducial with an accuracy of ∼4 per cent (2σ error). The method outlined here could be successfully implemented on future observational data sets to constrain b(k, z) and and thereby enhance our understanding of the low-redshift Universe.