Under sea‐ice noise, in particular that produced by ice floes in the Arctic region, is studied by means of a model simulation and the development of signal processing techniques for its measurement and estimation. This noise is due to the interaction among ice floes that produces non‐Gaussian noise in time and frequency. Based on previous studies, a Gaussian shape acoustic wave pressure generated by a single collision is assumed. The random occurrence of the ice collision process is modeled as a Poisson sequence of impulses. Different sonar array spectrum estimation techniques [Fourier, autoregressive (AR), and minimum variance (MV)] are analyzed for the processing of the resulting acoustic signal. It is found that the minimum variance spectral estimation technique is the overall best for this type of signal for a wide range of the model parameters. Previously reported studies have often used the periodogram technique that as shown here usually gives poor results.