We previously developed unified pixel-based beamforming to generate high-resolution sonograms, based on field pattern analysis. In this framework, we found the transmit wave-shape away from the focus could be characterized by two spherical pulses. These correspond to the maximal and minimal distances from the imaging point to the active aperture. The beamformer uses this model to select the highest-energy signals from backscattered data. A spatiotemporal interpolation formula is used to provide a smooth transition in regions near the focal depth where there is no dominant reflected pulse. In this paper, we show that the unified pixel-based approach is less robust at lower center frequencies. The interpolated data is suboptimal for a longer transmit wave-shape. As a result, the spatial resolution at the focal depth is lower than that in other regions. By further exploring the field pattern, we propose a beamformer that is more robust to variations in beam-width. The new method, named coherent pixel-based beamforming, aligns and compounds the pulse data directly in the transition regions. In simulation and phantom studies, the coherent pixel-based approach is shown to outperform unified pixel-based in spatial resolution. It helps regain optimal resolution at the focal depth while still maintaining good image quality in other regions. We also demonstrate the new method on in vivo data where its improvements over unified pixel-based are demonstrated on scanned objects with a more complicated structure.