Two challenges facing adaptive decision feedback equalizers (DFEs) in the underwater acoustic channel are those of the channel changing too rapidly to allow for the stable adaptation of the number of coefficients required to represent the equalizer filters and the high computational complexity of the associated adaptation algorithms. These challenges are particularly acute for multichannel DFEs where a separate filter needs to be adapted for each input signal channel. A multichannel "frequency domain" DFE is proposed in which the feedforward filter coefficients are represented in the frequency domain while the feedback filter coefficients are represented in the time domain. For fractionally sampled input signals, the frequency range over which the feedforward filter coefficients are calculated is limited thus reducing the number of coefficients that need to be calculated. The resulting DFE is shown to have both improved demodulation performance and a reduced complexity when compared to a time domain equalizer.