Table 2 Hyperparameters of interest for the proposed DU-PDA

Training set size

\(10^5\) samples

Layers

\(L = 4N_t\)

Input dimension

\({\mathbb {R}}^{2N_r}\text {,} \ {\mathbb {R}}^{2N_r \times 2N_t}\text {,} \ {\mathbb {R}}^{2N_t \times \sqrt{M}}\text {,} \ {\mathbb {R}}^{2N_t}\)

Output dimension

\({\mathbb {R}}^{2N_t \times \sqrt{M}}\)

Number of

learnable

\(\#\{w_\ell \}_{\forall \ell } = 4N_t\)

parameters

Activation function

\(\text {softmax}\left( \cdot \right) , \ \forall \ell\)

Learning rate

\(10^{-3}\)

Solver

Adam