A MFR Work Modes Recognition Method Based on Dual-Scale Feature Extraction

Zhiyuan Li, Xuan Fu, Chengjian Mo, Tang Jian-long, Ronghua Guo, Wenbo LiRemote 2025 · DOI

TL;DR

This paper introduces a novel hierarchical MFR signal model and a dual-scale feature extraction method for recognizing multi-function radar (MFR) work modes. By segmenting pulse sequences and employing a CNN-LSTM-Attention architecture, the method effectively handles variable-length inputs and complex mode-waveform mappings, achieving superior recognition performance in challenging, non-ideal scenarios.

What Problem Does This Paper Address?

Multi-function radars (MFRs) are increasingly sophisticated, employing complex relationships between their operating modes and waveforms, along with flexible beam scanning. This complexity makes it challenging to accurately identify the MFR's current work mode from intercepted pulse sequences. Existing methods often struggle with the variable durations and wide temporal distribution of features within these sequences, especially when dealing with non-ideal conditions like measurement errors and lost pulses. Furthermore, many current approaches rely on simplified radar models that don't fully capture real-world radar behavior, limiting their applicability.

Methods

The paper proposes a two-pronged approach: a new MFR signal model and a novel recognition method.

1. Hierarchical MFR Signal Model: The authors introduce a more comprehensive hierarchical MFR signal model. This model incorporates:

  • Task Subdivision: Breaking down work modes into fundamental tasks.
  • Reusable Mapping: Establishing relationships between tasks and waveforms that allow for reuse.
  • Spatial Beam Arrangement: Accounting for the radar's spatial scanning behavior. This model aims to better approximate real-world radar behavior by considering both temporal waveform scheduling and spatial antenna scanning, moving beyond simpler one-to-one or one-to-many mappings.

2. Dual-Scale Feature Extraction Recognition Method: The core of the recognition method is a deep learning framework designed to handle variable-length pulse sequences:

  • Pulse Sequence Segmentation: The raw, variable-length PDW (Pulse Description Word) sequences are first segmented into fixed-size chunks. This allows the network to process sequences of any length.
  • Internal-Segment Scale Feature Extraction: A Convolutional Neural Network (CNN) is used to extract local, shallow waveform parameter features from each segment. This is followed by a Bidirectional Long Short-Term Memory (Bi-LSTM) network to capture temporal dependencies within these segments. This CNN-Bi-LSTM combination is shared across all segments, reducing computational complexity.
  • External-Segment Scale Feature Extraction: The features extracted from all segments are then fed into another Bi-LSTM network to capture longer-term temporal patterns and waveform scheduling characteristics across segments.
  • Attention Mechanism: A multi-head attention mechanism is integrated to adaptively assign higher weights to more important waveform segments, allowing the model to focus on crucial information.
  • Classification: Finally, a fully connected layer and a softmax activation function are used to classify the MFR work mode.

Datasets

The study utilizes simulated data for training and evaluation.

  • Signal Models: Two MFR signal models were used:
  • Signal Model 1: A simpler one-to-multiple mapping between work modes and waveforms, based on a two-layer structure.
  • Signal Model 2: A more comprehensive, multiple-to-multiple mapping structure incorporating tasks and spatial beam arrangement, as described in the paper's proposed model.
  • Scenarios: Data was generated under three non-ideal conditions: measurement error, pulse loss, and a combination of both.
  • Dataset Structure: Training datasets were generated for each signal model and scenario. Six test datasets were created for Signal Model 1 and Signal Model 2, with a seventh test dataset (D7) specifically designed to analyze the impact of sample length, segment length, and shift length on recognition performance under complex scenarios. Each category in the test datasets contained 300 samples.

Key Insights and Ideas

  • Hierarchical Modeling is Crucial: The paper strongly suggests that more sophisticated, hierarchical MFR signal models that account for spatial scanning and task scheduling are essential for accurate recognition of modern radars. Simple mapping structures are insufficient.
  • Dual-Scale Feature Extraction is Effective: The proposed dual-scale approach (internal-segment and external-segment) is key to handling the diverse temporal scales present in MFR signals, from individual pulse characteristics to overall waveform scheduling.
  • Segmentation Enables Variable-Length Processing: Segmenting long pulse sequences is a practical way to make them manageable for deep learning models while retaining the ability to process variable-length inputs.
  • Attention Improves Focus: The attention mechanism helps the model prioritize relevant parts of the signal, leading to better discrimination, especially when dealing with complex or overlapping features.
  • Non-Ideal Conditions are a Major Hurdle: The experiments highlight that measurement errors and pulse loss significantly degrade recognition performance for all methods, underscoring the need for robust models.

Key Results

The proposed method demonstrated superior performance compared to several baseline methods (CNN, bi-GRU, bi-LSTM, CNN-LSTM, and DP-ATCN) across various non-ideal scenarios and signal models.

  • Signal Model 1: The proposed method consistently outperformed baselines, maintaining over 90% accuracy even in complex scenarios with measurement error and pulse loss.
  • Signal Model 2: The dual-scale approach showed significantly better recognition performance than comparison methods, achieving over 93% accuracy in all three test scenarios. The paper attributes this to the model's ability to extract dual-scale features and handle the complex waveform mapping.
  • Impact of Sample Length: Recognition accuracy was low for very short samples (100-2000 pulses) as they didn't capture enough temporal information. However, for longer samples (over 2000 pulses), the method achieved over 90% accuracy in complex scenarios.
  • Impact of Segment Length: As segment length and step size increased, accuracy showed a slight decline, particularly for longer samples, suggesting that overly large segments might introduce redundant information.

Limitations and Future Work

  • Real-World Validation: The study primarily relies on simulated data. The authors acknowledge that validating the method with real-world radar data is a crucial next step.
  • Syntactic Representation and Parameter Derivation: Further research is needed on MFR model's syntactic representation and waveform parameter derivation.
  • Computational Cost: The current method uses pulse segments as the minimum unit for feature fusion. Alternative methods like transformers were not explored due to computational costs, which is noted as a limitation.
  • Future Directions: The authors plan to explore methods from Natural Language Processing (NLP) to investigate MFR work mode recognition based on beam position boundary segmentation structures and validate these approaches under more stringent conditions.

Summary updated

Abstract

Multi-function radar (MFR) work modes recognition is an important research component of the electronic reconnaissance field. When facing MFR systems equipped with complex mode-waveform mapping relationships and flexible beam scanning techniques, the intercepted work mode pulse sequences have a wide temporal range of feature distributions and variable durations, which bring significant challenges for accurate recognition. To address this issue, this study constructs a novel hierarchical MFR signal model with waveform multiplexing and waveform scheduling laws with spatial beam arrangement and proposes a work mode recognition method based on dual-scale feature extraction. The recognition method first obtains the variable-length sequence processing capability through pulse sequence segmentation. Then, a structure composed of convolutional neural network (CNN) and long short-term memory (LSTM) is followed to extract the deep time-series features at the internal-segment scale of segments, and the features of each segment are concatenated in the time dimension. Subsequently, an LSTM-Attention network is used to extract the external-segment-scale features while adaptively assigning a higher weight to important waveform segments. Ultimately, the work mode recognition results are obtained. The experimental results show that the proposed method’s performance is advantageous in recognizing work modes under the comprehensive MFR signal model.
BibTeX
@inproceedings{li2025mfr,
  author = {Zhiyuan Li and Xuan Fu and Chengjian Mo and Tang Jian-long and Ronghua Guo and Wenbo Li},
  title = {{A MFR Work Modes Recognition Method Based on Dual-Scale Feature Extraction}},
  booktitle = {Remote},
  year = {2025},
  doi = {10.3390/rs17061054},
}

Loading related content...

Figure