Beamforming with an array of microphones on a sphere is an attractive tool for doing noise source localization in cabin environments. In order to achieve acceptable angular resolution, the array must have some minimum diameter, implying that many microphones are needed to obtain low sidelobe level over the frequency range of interest. For electric cars there is an increased need to cover high frequencies. The present paper describes a method to significantly reduce the sidelobe level over a broad frequency range relative to Spherical Harmonics Beamforming (SHB). For each focus point, a set of Finite Impulse Response (FIR) filters are optimized to minimize the highest sidelobe in a Filter And Sum (FAS) beamformer, while maintaining sensitivity at the focus point and limiting the White Noise Gain (WNG). Similar approaches have been published, but a main issue in connection with applications for noise source mapping is the computational load in performing the optimization of FIR filters for each focus point. The paper describes a method to handle that challenge. Basically, filter coefficient are pre-optimized for a pre-defined 3D mesh and stored in a database. When focusing has to be performed at an arbitrary point in space, an interpolation is performed, typically on the filter vectors related to a set of neighboring mesh points. Results are presented for simulated and real measurements. These show that the optimized FAS method provides very significant improvement relative to SHB on sidelobe suppression and thus on the dynamic range available for source identification.