©2026 ANSYS, Inc.
4
New Signal Models
LSTM
• Parameters are repeated and stacked with the
index "t" times
• The LSTM is used to predict one element at a time
until the entire signal is predicted
• Current output is fed back into the cell at the next
step -> introduce a memory behavior
LSTM
P1 , P2, .. Pn, X1
.
.
.
P1, P2, .. Pn, Xt
n: number of parameters
t: signal length
S1
.
.
.
St
NN
P1 , P2, .. Pn, X1
IndexNN
• The problem is scalarized: every signal element is
modeled as a single scalar problem
• The index is appended as a parameter at the input
• The NN predicts an element at a time
• The approximation function is responsible for
reconstructing the initial signal
P1 , P2, .. Pn, X2
P1 , P2, .. Pn, Xt
S1
S2
St
