Fig. 5
From: Joint high-dimensional soft bit estimation and quantization using deep learning
Impact of latent feature dimension on different instances of EQ-Net without numerical quantization in an i.i.d. Rayleigh fading, \(2 \times 2\), 64-QAM scenario using an (324, 648) LDPC code. A feature dimension of six corresponds to the upper bound in Theorem 2, which is proven explicitly for diagonalized channels, but is shown here to be practically achievable for arbitrary channels