Lightweight LSTM-Based Adaptive CQI Feedback Scheme for IoT Devices

. 2023 May 20;23(10):4929. doi: 10.3390/s23104929

Algorithm 1 Lightweight LSTM-based adaptive CQI feedback.

1:
Input:
Actual channel gains $[h_{t - L + 1}, h_{t - L + 2}, . . ., h_{t}, h_{t + 1}, h_{t + 2}, . . ., h_{t + M}]$
Threshold $ϵ$ .
/* Training a lightweight LSTM network */
2:
Build a lightweight LSTM network with N input nodes and M output nodes.
3:
Initialize lightweight LSTM network parameters $W_{i}, W_{h}, B$ , and set training epochs.
4:
while epoch < MAX_EPOCH do
5:
Process $f ([h_{t - L + 1}, h_{t - L + 2}, . . ., h_{t}], W_{i}, W_{h}, B) = [{\hat{h}}_{t + 1}, {\hat{h}}_{t + 2}, . . ., {\hat{h}}_{t + M}]$
6:
Update parameters with loss function $\frac{1}{M} \sum_{i = 1}^{M} {| h_{t + i} - {\hat{h}}_{t + i} |}^{2}$
7:
Update epoch ← epoch $+ 1$
8:
end while
/* Predicting the CQI index */
9:
Calculate the predicted channel gain with the trained neural network
$f ([h_{t - L + 1}, h_{t - L + 2}, . . ., h_{t}], W_{i}, W_{h}, B) = [{\hat{h}}_{t + 1}, {\hat{h}}_{t + 2}, . . ., {\hat{h}}_{t + M}]$
10:
Calculate the predicted SNR ${\hat{γ}}_{t + k_{f d}}$
11:
Calculate the predicted CQI index ${\hat{μ}}_{t + k_{f d}}$
12:
whilet < MAX_STEP do
13:
if $| {\hat{μ}}_{t + k_{f d}} - {\tilde{μ}}_{t + k_{f d} - 1} | > ϵ$ then
14:
Update ${\tilde{μ}}_{t + k_{f d}} \leftarrow {\hat{μ}}_{t + k_{f d}}$
15:
Reports the CQI index to the BS
16:
else
17:
Update ${\tilde{μ}}_{t + k_{f d}} \leftarrow {\tilde{μ}}_{t + k_{f d} - 1}$
18:
end if
19:
Update $t \leftarrow t + 1$
20:
end while