It is well known that long-range signal propagation in random-inhomogeneous channels leads to loss of the signal coherence in space, time and frequency domains. As to large-array signal processing, an intrinsic issue arise if the array size is greater then a signal coherence length (for given distance and frequency band). Such a scenario is typical, for example, for long-range sound propagation in underwater channels. These channels are, in fact, multimode refraction-type waveguides, so the problem is to develop the large-array signal processing techniques under the conditions of coherence-reduced multimode signal against the noise background which contains a rather intensive modal component. In this paper, we consider the problem with emphasis on the optimal processing techniques resulted from the output signal-to-noise ratio criterion. Our focus is interrelated effects of the signal coherence loss and the modal noise (which partially or even completely interferes the modal signal) on the SNR array gain. We give also some results of computer simulation to compare the array processors in long-range underwater environments.