Optimal Maximum Likelihood (ML), narrow-band direction finding cannot be easily initialized in coherent and low signal to noise ratio environments. Sparse under-determined solvers are considered as viable solutions to this problem, since they drastically reduce the dimensionality of the search space by exploiting the array model sparseness. However, because of quantized locations, conventional sparse solvers present some ambiguity problems. In this work, we propose a novel boosting scheme for ML-type estimators, referred to as Parallel BOOSTer (P-BOOST), where a set of generalized MUSIC solutions provides pre-estimates of the directions and the number of coherent paths for arbitrary sensor array geometry and noise covariance. P-BOOST delivers improved and reliable coarse parameter estimates to a further ML or sparse optimization stage even in coherent and/or high noise scenarios. Moreover, its dataflow is highly parallel, which is essential in foreseen remote sensing and telecommunication applications and fully justifies its acronym.