One of the most innovative techniques is using convolutional neural networks (CNNs) for predicting and pre-scheduling the different tasks in the receiver chain. Using CNNs as a pre-processor leads to optimizing receiver efficiency to maximize packet transmission reliability while minimizing the complexity of receiver processing. This state of art technique showed high accuracy using a supervised deep learning classification for raw base-band packets. Although the approach is considered a blind technique, in literature it is dependent on the channel power delay profile (PDP). i.e. there are different CNNs for each channel profile with fixed Doppler frequency shifts. Besides that, for some channel profiles, there is an overfitting problem that leads to performance degradation at higher SNR levels. Here we propose a solution for both issues, first by introducing regularization to eliminate the over-fitting problem appeared in literature, then we introduce a channel profile independent approach which significantly reduces the system complexity by replacing multiple CNNs with one and eliminating the need for prior identification of channel's PDP. And finally, we propose also an SNR values-based order training technique that allows tuning the performance vs SNR.