The primordial CMB $B$-mode search is on the spotlight of the scientific community due to the large amount of cosmological information that is encoded in the primeval signal. However, the detection of this signal is challenging from the data analysis point of view, due to the relative low amplitude compared to the foregrounds, the lensing contamination coming from the leakage of $E$-modes, and the instrumental noise. Here, we studied the viability of the detection of the primordial polarization $B$-mode with a ground-based telescope operating in the microwave low-frequency regime (i.e., from 10GHz-120GHz) in a handful of different scenarios: i. the instrument's channels distribution and noise, ii. the tensor-to-scalar ratio ($r$) detectability considering different possible $r$ values and degrees of delensing, iii. the effect of including a possible source of polarized anomalous microwave emission (AME), iv. the strengths and weaknesses of different observational strategies and, v. the atmospheric and systematic noise impact on the recovery. We focused mainly on the removal of galactic foregrounds as well as noise contamination by applying a full-parametric pixel-based maximum likelihood component separation technique. Moreover, we developed a numerical methodology to estimate the residuals power spectrum left after component separation, which allow us to mitigate possible biases introduced in the primordial $B$-mode power spectrum reconstruction. Among many other results, we found that this sort of experiment is capable of detecting Starobinsky's $r$ even when no delensing is performed or, a possible polarized AME contribution is taken into account. Besides, we showed that this experiment is a powerful complement to other on-ground or satellite missions, such as LiteBIRD, since it can help significantly with the low-frequency foregrounds characterization.