Scribd
Upload
4 views1 page

4

Automatic Modulation Recognition (AMR) is essential for optimizing communication systems, utilizing both traditional machine learning and advanced deep learning techniques. Traditional methods like Support Vector Machines (SVM) rely on feature engineering but struggle under certain conditions, while deep learning approaches, particularly Convolutional Neural Networks (CNNs), offer improved accuracy and adaptability without manual feature extraction. Hybrid models combining DNNs and CNNs further enhance classification capabilities for complex modulation schemes.

Uploaded by

24126663zsp
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
4 views1 page

4

Automatic Modulation Recognition (AMR) is essential for optimizing communication systems, utilizing both traditional machine learning and advanced deep learning techniques. Traditional methods like Support Vector Machines (SVM) rely on feature engineering but struggle under certain conditions, while deep learning approaches, particularly Convolutional Neural Networks (CNNs), offer improved accuracy and adaptability without manual feature extraction. Hybrid models combining DNNs and CNNs further enhance classification capabilities for complex modulation schemes.

Uploaded by

24126663zsp
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 1

3.

Automatic Modulation Recognition Research Based on Different Methods

Automatic Modulation Recognition (AMR) is a critical function in modern communication systems, enabling

efficient utilization of spectrum and improving the adaptability of communication networks. Research in AMR

spans traditional machine learning methods, advancements in deep learning, and innovations in data

preprocessing and enhancement techniques.

3.1 Based on Traditional Machine Learning Methods

3.1.1 SVMh

Support Vector Machines (SVM) have been a cornerstone in early AMR systems due to their robustness in

handling high-dimensional feature spaces. SVMs rely heavily on feature engineering, where domain-specific

features such as high-order cumulants, cyclostationary features, or wavelet transforms are extracted from the

received signals. These features allow the SVM to separate different modulation types effectively ([1])([2])([3]).

For instance, feature-based methods combined with SVM classifiers have demonstrated high accuracy in

additive white Gaussian noise (AWGN) environments. However, their performance often degrades under fading

channels or low SNR conditions, primarily due to the challenges in extracting noise-resilient features ([1])([4])

([5]). Hybrid approaches that integrate SVM with preprocessing techniques, such as Kolmogorov-Smirnov tests,

have been explored to improve robustness([2])([6]).

3.2 Based on Deep Learning Methods

Deep learning methods have revolutionized AMR by providing end-to-end solutions that do not require manual

feature extraction. These methods have achieved significant breakthroughs in classification accuracy and

adaptability to diverse communication environments.

3.2.1 CNN

Convolutional Neural Networks (CNNs) are extensively used for AMR due to their ability to process high-

dimensional data and learn hierarchical features. CNNs are particularly effective in extracting spatial and

temporal patterns from spectrograms, constellation diagrams, and IQ data. For instance, a CNN-based AMR

system achieved over 93% accuracy on the RadioML dataset by leveraging data preprocessing and advanced

network architectures such as residual blocks([4])([7])([8]).

Bispectrum-based CNNs, which use amplitude spectrums of bispectrum (ASB) as input, have shown outstanding

classification performance in low SNR environments. These networks achieve recognition rates exceeding 97.7%

under moderate noise conditions([7])([8]). Feature fusion approaches, which combine multiple representations

such as IQ sequences, amplitude-phase data, and constellation diagrams, further enhance CNN performance by

leveraging complementary features([9])([10]).

3.2.2 DNN/CNN Models

The combination of fully connected Deep Neural Networks (DNNs) and CNNs has been shown to improve

classification accuracy, especially for distinguishing between modulation types with subtle differences. For

example, a DNN-CNN hybrid system was proposed to handle complex modulation schemes like QPSK and

16QAM. This

You might also like