An Efficient Automatic Modulation Classification Approach Using Improved Migration Algorithm-Based Ensemble Machine Learning Network

In the realm of intelligent radio communications, Automatic Modulation Classification (AMC) plays an important role in
various applications such as identifying transmitters, allocating spectrum resources and facilitating industrial automation.
Both in contemporary military engagements and in civilian electromagnetic regulation, AMC is pivotal for ensuring effective
communication signal management on the Internet. It provides the scientific foundation and confidence needed for smart
signal reception and processing. Typically, AMC involves extracting and utilizing certain characteristics of the received signal
for classification purposes. The efficiency of AMC can be greatly improved by selecting the appropriate features. However, in
non-cooperative communication scenarios, the presence of noise in the received signal can pose challenges for existing AMC
algorithms, making it difficult to achieve a balance among classification accuracy and model complexity. To solve the problem, a
novel Optimized Ensemble Network (OENet) is proposed in this paper. Initially, the required data is gathered from the standard
data resource. After that, the features from the input data are optimally selected utilizing the Fitness-based Update in Migration
Algorithm (FUMA). Subsequently, the resultant optimal feature is subjected to the classification phase. Here, the OENet is used to
classify the class with the help of Naive Bayes (NB), Multilayer Perception (MLP), Capsule Network (CapsuleNet) and Adaboost.
These classification outcomes are further averaged by adopting the Fuzzy ranking method. The parameters from the OENet model
are optimally tuned using improved FUMA optimization to enhance the performance. Through the simulation experiment, it
is determined that the performance of the designed OENet model surpasses that of conventional models. The results obtained
indicate superior performance of the OENet model contrasted to existing approaches