Fig. 1
From: Joint high-dimensional soft bit estimation and quantization using deep learning
High-level block diagram of the EQ-Net architecture. In the first training stage, and at test time for the soft bit quantization task, the dashed blocks are learned and used for inference, respectively. In the second training stage, the decoder \(g_\mathrm {Q}\) is kept frozen while training the estimation encoder \(f_\mathrm {E}\). During deployment, the top or bottom signal paths are used to output the reconstructed (for quantization) or estimated (for estimation) soft bits \(\tilde{\mathbf {\Lambda }}\), respectively