Residual Learning and Multi-Path Feature Fusion-Based Channel Estimation for Millimeter-Wave Massive MIMO System

. 2022 Feb 18;24(2):292. doi: 10.3390/e24020292

Algorithm 1 RL-MFF-Net-based channel estimation algorithm

offline training phase:

1:
Initialize $j = 0$ , estimator parameters $Θ$ , stepsize $α$ , learning rate and decay rate.
2:
Input: Training set $({\tilde{Y}}_{t r a i n}, H_{t r a i n})$
3:
while $j \leq G$ do weights update
4:
$g_{j} \leftarrow \nabla_{Θ} L_{j} (θ_{j - 1})$
5:
Compute $M_{j}$ and $V_{j}$ by Equation (27) and (28)
6:
Computer ${\overset{\land}{M}}_{j}$ and ${\overset{\land}{V}}_{j}$ by Equation (25) and (26)
7:
Update estimator parameters $Θ^{j + 1} \leftarrow Θ^{j} - α \frac{{\overset{\land}{M}}_{j}}{\sqrt{{\overset{\land}{V}}_{j}} + ε}$
8:
$j = j + 1$
9:
end while
10:
Output: Well-trained estimator $Ψ_{R L - M F F - N e t} (\cdot)$

online deployment phase:

11:
Input: Test set ${\tilde{Y}}_{t e s t}$
12:
do Channel Estimation with $Ψ_{R L - M F F - N e t} (\cdot)$
13:
Output: the reconstructed mm Wave channel $\overset{\land}{H} = Ψ_{R L - M F F - N e t} ({\tilde{Y}}_{t e s t})$