A novel method of acoustic holography is introduced by expanding the equivalent source method (ESM) to include multipole sources under the higher-order Ambisonic (HOA) paradigm and obtaining an optimal solution in a two-stage manner: First, sparse Bayesian learning (SBL) is used as a multiple-dictionary feature selector to estimate a sparse set of broadband spherical harmonic sources. Next, a low dimension inversion is performed on the sparse source set over all frequencies to yield a final estimate, which can be stored, manipulated, and rendered using multipole techniques. The method is used to estimate a complex directivity radiating source in free-field and stochastic reverberant environments, and reconstruction accuracies using several microphone array layouts and reverberant conditions are compared.