When a multi-user communication system over a block-fading MIMO channel utilizes Tomlinson-Harashima precoding in the downstream direction, to eliminate the interference between the spatially multiplexed data streams, the conventional detection, involving a modulo operation at the receiving terminals, is known to yield a performance degradation that becomes considerable at low SNR. In this contribution, we propose a novel detection method that exploits sending one bit of extra information per user and per frame to the receiver, which indicates whether or not the considered user can detect all its data within the frame without performing a modulo operation. Moreover, in the case of M-PAM transmission it is possible to optimize the rotation of the constellations at the transmitter, maximizing for each frame the number of users for which no modulo operation is required. Numerical results show that in the case of 2-PAM the novel detection algorithm is able to completely recover the modulo loss experienced by the conventional detection method without an increase in transmit power, and to outperform 4-QAM (with conventional or novel detection) in terms of mutual information at low SNR.