Effective channel estimation plays a critical role in the overall performance of multi-input multi-output (MIMO) underwater acoustic communications (UAC). This paper compares two closely related channel estimation algorithms developed under different models for sparse and frequency modulated acoustic channels. More specifically, the recently proposed channel estimation algorithm, referred to as the generalization of the sparse learning via iterative minimization (GoSLIM), assumes that each receiver has its own Doppler frequency. This channel model is further simplified in the present paper by assuming that all the receivers experience the same Doppler frequency, and the corresponding channel estimation algorithm is modified accordingly. Both channel estimation algorithms considered address sparsity through a hierarchical Bayesian model. They are user parameter free algorithms and are easy to use in practical applications. By analyzing the in-water experimental measurements recently acquired during the MACE10 experiment, we demonstrate that the employment of the GoSLIM variation not only reduces the overall complexity involved in the channel estimation stage, but also slightly improves the detection performance, as compared with its original GoSLIM counterpart. This paper signifies the importance of establishing proper channel models for enhanced UAC performance.