Search

Analyzing radar signals is an important task in operating electronic support measure systems. The received signals in the real electromagnetic environment often originate from multiple emitters and must be separated for further processing. Pulses from important target emitters with known parameters should be picked out first. To solve the problem, time-of-arrival (TOA) deinterleaving may be performed to extract signals from a certain emitter by learning the pulse repetition interval (PRI) modulation that makes up the signal. However, conventional deinterleaving methods only work with simple PRI modulations; their performance degrades in noisy environments. A novel approach based on denoising autoencoders for TOA deinterleaving was developed in this article. The inner patterns of pulse-of-interest sequences were learned by the proposed denoising autoencoders to generate output sequences from well-trained autoencoders. Simulation results show that the proposed method outperforms conventional methods, especially in environments with high lost and spurious pulse ratios. [ABSTRACT FROM AUTHOR]

In modern electronic warfare environments, there are multiple-radar transmitting signals. For an electronic support system, it is essential to recognize the modulation of pulse repetition intervals (PRIs), since it is directly related to the indication of radar emitters. However, PRI modulations are more difficult to recognize in modern electronical environments due to the high ratio of lost and spurious pulses. Therefore, a fully automatic approach for recognizing seven PRI modulation types using a convolutional neural network (CNN) is proposed in this letter. Simulation results show that our CNN-based recognition method not only promotes performance but is also robust to the environment with lost and spurious pulses. The recognition accuracy is 96.1% with 50% lost pulses and 20% spurious pulses in simulation scenario. [ABSTRACT FROM AUTHOR]

Unintentional modulation on pulse (UMOP) is the basis for specific emitter identification and has been studied for decades. However, the isolated distribution of UMOP in frequency domain has not been investigated. This Letter first gives the signal model of UMOP, and then evaluates its frequency‐domain distribution based on sinc‐function fitting. Experiments on real‐world data validate the effectiveness of the proposed method, and further indicate that UMOP mainly distributes in low frequencies with dramatically limited bandwidth. Furthermore, the proposed distribution density can also serve as a practical UMOP feature which exhibits better recognition performance than conventional spectrum features. [ABSTRACT FROM AUTHOR]

Automatic modulation classification (AMC) is an important and challenging task that aims to discriminate modulation formats of received signals, such as military communications, cognitive radio and spectrum management. With the development of deep learning techniques, research in AMC has gained promising results because of its powerful representation and classification abilities. In this Letter, the authors present a new network architecture that combines a frequency selection module and a convolutional neural network (CNN). This scheme not only processes raw signal data with carriers to increase the in‐band signal‐to‐noise ratio but also converge faster than traditional CNN. Experiments demonstrate the effectiveness and efficiency of the proposed model. [ABSTRACT FROM AUTHOR]

The modulated wideband converter (MWC) is an attractive analogue compressed sensing technique that has been proposed in recent years. Regrettably, the MWC has high hardware complexity owing to its parallel channel structure. To reduce the complexity, a novel compressive wideband spectrum sensing scheme based on a single channel for band‐limited wide‐sense stationary (WSS) signals is proposed. In the proposed scheme, a special periodic waveform in the time domain is designed and a power spectrum sensing model is presented to detect the occupied frequency spectrum of such signals from sub‐Nyquist samples of which the sample rate is only MWC's single‐channel rate. Simulation results show that the proposed scheme has a perfect performance on wideband WSS signal spectrum sensing. [ABSTRACT FROM AUTHOR]