IEEE - Aerospace and Electronic Systems - April 2023 - 15

Image licensed by Ingram Publishing
derived from the signal itselfwithout adhering to underlying
approximation mathematical properties, signal features with
new perspectives and the hidden coherence in signals can be
extracted. In addition, the extraction ofthe adaptive features
involves with the process of discarding the residues of the
signals, which mitigates the noise and further improves the
recognition performance. Further, in order to incorporate
more signal characteristics and formulate a comprehensive
representation of the LPI signals for better recognition performance,
the adaptive feature construction (AFC) framework
is proposed to combine the adaptive features with the
predefined analytical features from the traditional techniques
with rigorous mathematical expressions. With the constructed
adaptive feature, the CNN is selected as the
classification method, which has shown excellent performance
of which has been proved in various classification
applications, such as face recognition, image classification,
and the time-frequency image-based signal classification
[24], [25]. The main contributions of the feature adaptive
LPI network-based exploitation (FALPINE) are
summarized as follows.
Utilizing adaptive signal analysis methods to
explore the inherent features of LPI signals for the
waveform recognition, which provides an innovative
and potential effective methodology for extracting
the features ofLPI waveforms.
Proposing an AFC framework to combine the adaptive
features with the predefined analytical features
with adjustable parameters to give the comprehensive
representation of the signals and thus achieves
robust classification performance for the recognition
of the LPI waveforms.
Developing an LPI waveform recognition system is
develop based on the AFC framework and the optimized
CNN.
Demonstrating that the FALPINE method achieves
excellent performance for the simulated LPI
APRIL 2023
waveforms recognition and outperforms the other
benchmark methods.
The rest of this article is organized as follows. The
" System Architecture for LPI Signal Recognition " section
describes the system architecture of the proposed LPI signal
recognition system. The " AFC Framework " section
introduces the details of the proposed AFC framework.
The " CNN-Based LPI Waveform Recognition System "
section presents the experimental results and the comparison
with existing studies. A summary of the study is given
in the " Experimental Results " section. Finally, the
" Conclusion " is provided.
SYSTEM ARCHITECTURE FOR LPI SIGNAL
RECOGNITION
In this section, the LPI signals considered in this study are
introduced and the system architecture for recognizing the
LPI signals is presented.
LPI SIGNALS
In order to verify the recognition performance of the FALPINE
approach, several popularly used LPI radar signals
are selected and the modulation techniques performed on
the waveforms include the noise-contaminated linear frequency
modulation (LFM) or the so-called noise radar and
five polyphase codes including Frank code, P1 code, P2
code, P3 code, and P4 code [1]. As the polyphase-coded
LPI signals are with high similarity and thus are difficult
to be classified from each other, they are selected as a
focus for LPI signal recognition system. These signals are
generated with simulated channel configurations including
random path fading effects and noise contaminations,
which create the training and testing dataset for the recognition
system. The signal generation methods include [1]
the following.
IEEE A&E SYSTEMS MAGAZINE
15
IEEE - Aerospace and Electronic Systems - April 2023

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - April 2023
