The aim of this manuscript is to propose and evaluate a novel nonlinear precoding scheme for distributed MIMO based systems. We consider a high-speed backhaul network where the remote antenna units are transparently linked by optical fiber to a central unit. In the considered architecture we can assume the knowledge of the data and channel state information of all users, at the central unit. The precoder design is divided in two phases: first a solution based on minimization of bit-error-rate is computed by assuming that all remote antenna units can jointly pool their power, i.e., a total power constraint is imposed instead of a per remote antenna unit power constraint; the second phase consists in scaling the obtained transmit signal by a factor η to satisfy the individual per remote antenna unit power constraint. Numerical results show that the algorithm achieves significant gains over the linear zero-forcing and minimum mean square error based approaches, with a slight increase in complexity, and that its performance is close to the optimum.