In this paper, we propose a generic receiver based on optimum maximum likelihood sequence estimation (MLSE) employing a generalized Ungerboeck metric with significantly reduced complexity for the general class of M-ary quasi orthogonal signaling (MOS) in severe multipath fading channels. First, by using the factor graph (FG) of the obtained generalized Ungerboeck metric for MOS and sum-product algorithm (SPA) framework, a highly efficient reduced state sequence estimation (RSSE) algorithm is proposed that operates on forward and backward directions and is directly applied to unwhitened channel matched filter (CMF) and code matched filter outputs at symbol rate. The proposed structure, generating the a posteriori probabilities (APP) by bidirectional RSSE recursions, are substantiated with symbol rate bidirectional decision feedback (BDF) based on surviving paths in order to eliminate the post-and pre-cursor inter-symbol interference (ISI) as well as multi code interference (MCI) due to the non-ideal properties of the signaling waveforms and multipath channel. It appears as the unification of Ungerboeck type reduced trellis equalization framework for MOS and achieves near-optimum performance especially for channels with large dispersion. Second, we identify a bias term through an error probability analysis which helps us improve the proposed receiver's performance. Furthermore, a tight approximation for the bit error rate (BER) of the proposed receiver is derived. These analyses lead to significant insight on the selection of system parameters and clearly demonstrate the high performance and efficiency of the proposed scheme by a proper choice of the signaling parameters.