Fig. 2
From: Joint high-dimensional soft bit estimation and quantization using deep learning
Internal architecture for the deep neural networks in Figure 1. The blue blocks are fully connected layers, with the output dimension displayed. The yellow and green blocks represent the \(\mathrm {relu}\) and \(\tanh\) activations, respectively, while the purple block represents a concatenation operator. a The architecture of the quantization encoder \(f_\mathrm {Q}\). b The architecture of the quantization decoder \(g_\mathrm {Q}\). c The architecture of the estimation encoder \(f_\mathrm {E}\), composed of (potentially) multiple residual blocks. The internal architecture of a single residual block is shown here