Spectral Efficiency Improvement Using Bi-Deep Learning Model for IRS-Assisted MU-MISO Communication System

. 2023 Sep 11;23(18):7793. doi: 10.3390/s23187793

Algorithm 2 Internal performance of Bi-LSTM model

1: Start
2: The forget gate operation [43]: $f g_{t}$ = $f_{ψ} (Q_{f g} X_{t} + T_{f g} P_{t - 1} + b_{f g})$
3: The input gate operation: $i g_{t}$ = $f_{ψ} (Q_{i g} X_{t} + T_{i g} P_{t - 1} + b_{i g})$
4: The candidate gate operation [43]: $c n_{t}$ = $f_{t a n h} (Q_{c n} X_{t} + T_{c n} P_{t - 1} + b_{c n})$
5: The output gate operation: $o g_{t}$ = $f_{ψ} (Q_{o g} X_{t} + T_{o g} P_{t - 1} + b_{o g})$
6: Updating the prior cell state, $C_{t - 1}$ to the present cell state: $u d_{t}$ = $(C_{t - 1} ⨀ f g_{t}) + (i g_{t} ⨀ c n_{t})$
7: At time step t, the hidden state: $P_{t} = o g_{t} ⨀ f_{t a n h} (u d_{t})$
8: The Bi-LSTM layer’s output hidden state: $H I_{t} = f (Q_{H I \vec{T}} {\vec{T}}_{t} + Q_{H I \overset{\leftarrow}{T}} {\overset{\leftarrow}{T}}_{t} + b_{z})$
9: Stop