Automatic Modulation Classification for MASK, MPSK, and MQAM Signals Based on Hierarchical Self-Organizing Map

Zerun Li, Qinglin Wang, Yufei Zhu, Zuocheng XingSensors 2022 · DOI

TL;DR

This paper introduces a novel hierarchical Self-Organizing Map (SOM) for Automatic Modulation Classification (AMC) that effectively distinguishes between MASK, MPSK, and MQAM signals. By leveraging high-order cumulants and amplitude moments, and employing a unique discrete transformation for MQAM signals, the proposed method achieves higher accuracy with fewer computational resources compared to existing clustering models.

What Problem Does This Paper Address?

Automatic Modulation Classification (AMC) is crucial for modern communication systems, enabling signal decoding and resource management, especially in complex, non-cooperative environments. Existing methods often struggle with a large number of modulation types, leading to lower classification accuracies and requiring significant computational power. Specifically, differentiating between similar modulation schemes like different orders of MPSK and MQAM, particularly under noisy conditions, presents a significant challenge. This paper aims to address this by developing a more efficient and accurate AMC system capable of handling a wider range of modulation types.

Methods

The core of this paper is a hierarchical Self-Organizing Map (SOM). Unlike traditional single-layer SOMs, this approach uses a two-layer structure:

  • Root Layer: This layer performs a coarse clustering of modulation signals based on their intrinsic differences. It utilizes features derived from high-order cumulants (HOC), specifically $|C_{42}|$ and $|C_{40}|$, which are known to be robust to noise and effectively differentiate between MASK, MPSK, and MQAM signals.
  • Leaf Layers: For finer discrimination, especially between similar modulation orders within MPSK and MQAM, leaf layers are generated. These layers further cluster the samples. For MQAM signals, a novel discrete transformation method based on modified activation functions (mod-arctangent, mod-logistic, and mod-hard-logistic) is applied to the amplitude moment feature ($\mu_{42}$). This transformation aims to create more distinct cluster boundaries, improving classification accuracy.

The feature space is constructed using normalized HOC and amplitude moment features. The hierarchical structure allows the model to break down a complex classification problem into smaller, more manageable sub-problems, leading to improved accuracy and efficiency.

Datasets

The paper does not explicitly mention the use of pre-existing public datasets. Instead, it describes the generation of synthetic signals for simulation and evaluation. The generated signals simulate real-world conditions by including:

  • Modulation Types: 4ASK, 8ASK, 2PSK, 4PSK, 8PSK, 16QAM, and 64QAM.
  • Environmental Factors: Sampling rate deviation, carrier frequency offset, and additive white Gaussian noise (AWGN).
  • Signal-to-Noise Ratio (SNR): Ranging from -5 dB to 10 dB.

A total of 14,000 samples were used for clustering, with 2000 samples per modulation type.

Key Insights and Ideas

  • Hierarchical Structure for Granular Classification: The hierarchical SOM effectively breaks down the complex AMC problem. The root layer handles broad distinctions, while leaf layers specialize in differentiating subtle variations, leading to better overall accuracy.
  • Feature Robustness: High-order cumulants (HOC) and amplitude moments prove to be effective features for distinguishing between MASK, MPSK, and MQAM signals, even in the presence of noise. Their theoretical values are independent of initial phases, enhancing robustness.
  • Discrete Transformation for MQAM Clarity: Applying discrete transformations to MQAM features is crucial. This technique sharpens cluster boundaries, making it easier for the SOM to separate closely related MQAM signals (like 16QAM and 64QAM), which are otherwise difficult to distinguish. The mod-hard-logistic function showed particularly promising results in this regard.
  • Unsupervised Learning Advantage: The use of an unsupervised learning approach (SOM) means the model can learn from data without requiring pre-labeled examples, which is beneficial in non-cooperative communication scenarios.

Key Results

  • High Classification Accuracy: The proposed hierarchical SOM achieved a classification accuracy of 97.6% in the root layer at 14,000 iterations. In the leaf layers, accuracies for MPSK reached 99.5%, and for MQAM, they reached up to 97.9% without discrete transformation. With discrete transformation, MQAM leaf layer accuracies exceeded 98%, with the mod-hard-logistic function achieving the highest accuracy.
  • Improved MQAM Discrimination: The discrete transformation significantly improved the clustering of MQAM signals, creating clearer boundaries between 16QAM and 64QAM.
  • Computational Efficiency: The paper claims that the hierarchical SOM model requires fewer computational resources compared to other methods like Maximum Likelihood (ML) and Stacked Auto-Encoder Deep Neural Network (SAE-DNN) classifiers, primarily due to fewer multiplications and exponential operations.
  • Broader Modulation Support: The proposed method can handle more modulation types (7 in this study: MASK, MPSK, MQAM) compared to previous works that typically focused on fewer categories.

Limitations and Future Work

  • Iteration Dependency: The performance of the hierarchical SOM is highly dependent on the number of iterations. Achieving high accuracy requires a significant number of iterations (over 10,000), and determining the optimal number can be challenging.
  • Dataset Sensitivity: The clustering results can be influenced by the distribution of the original dataset. Similar samples from different categories might lead to overlapping clusters, potentially decreasing accuracy. While discrete transformation helps mitigate this, it doesn't entirely eliminate the issue.
  • Future Work: The authors suggest exploring the application of this hierarchical SOM model to a wider range of modulated signals or other types of similar data. They also note that model training parameters require professional experience.

Summary updated

Abstract

Automatic modulation classification (AMC) plays a fundamental role in common communication systems. Existing clustering models typically handle fewer modulation types with lower classification accuracies and more computational resources. This paper proposes a hierarchical self-organizing map (SOM) based on a feature space composed of high-order cumulants (HOC) and amplitude moment features. This SOM with two stacked layers can identify intrinsic differences among samples in the feature space without the need to set thresholds. This model can roughly cluster the multiple amplitude-shift keying (MASK), multiple phase-shift keying (MPSK), and multiple quadrature amplitude keying (MQAM) samples in the root layer and then finely distinguish the samples with different orders in the leaf layers. We creatively implement a discrete transformation method based on modified activation functions. This method causes MQAM samples to cluster in the leaf layer with more distinct boundaries between clusters and higher classification accuracies. The simulation results demonstrate the superior performance of the proposed hierarchical SOM on AMC problems when compared with other clustering models. Our proposed method can manage more categories of modulation signals and obtain higher classification accuracies while using fewer computational resources.
BibTeX
@inproceedings{li2022automatic,
  author = {Zerun Li and Qinglin Wang and Yufei Zhu and Zuocheng Xing},
  title = {{Automatic Modulation Classification for MASK, MPSK, and MQAM Signals Based on Hierarchical Self-Organizing Map}},
  booktitle = {Sensors},
  year = {2022},
  doi = {10.3390/s22176449},
}
Figure