In this paper, to improve bandwidth efficiency and error performance, partial response signaling (PRS) and trellis coded modulation (TCM) are combined together and named as QPR-TCM. The emphasis in this paper is on the jitter and error performance of QPR-TCM and neural network equivalent models in the fading environment with ideal channel state information (CSI) and no side information exist on the phase noise process. Analytical upper bounds are derived using Chernoff bounding technique, combined with the modified generating fuctional approach. One interesting result is that the bit error probability upper bounds of the proposed scheme is better than 4QAM-TCM for SNR values greater than a threshold, which have the best error performance till now. Neural network models of QPR-TCM schemes are introduced the first time in literature for fading channels and the bit error event curves are derived using lack propogation algorithm. It is shown that neuro-computing results confirm analytical approach.