Recognition of PRI modulation using an optimized convolutional neural network with a gray wolf optimization based on internet protocol and optimal extreme learning machine

Abstract

In the modern electronic warfare (EW) landscape, timely and accurate detection of threat radars is a critical and necessary issue in electronic support Measure (ESM) and electronic intelligence (ELINT) because these radars' correct and timely detection plays an essential role in electronic countermeasures strategies. The PRI (pulse reputation interval) modulation type is one of the main parameters in radar signal analysis and identification. However, recognizing PRI modulation is challenging in a natural environment due to destructive factors, including missed pulses, spurious pulses, and large outliers, which lead to noisy sequences of PRI variation patterns. This paper presents a new four-step real-time approach to recognize six common PRI modulation types in noisy and complex environments. In the first step, an optimal convolutional neural network (CNN) structure was formed by a gray wolf optimization (GWO) based on the Internet Protocol (IP-GWO) according to the simulated PRI data set, which acts as a feature extractor. In the second step, the last fully connected layers are replaced by an extreme learning machine (ELM) to improve the time complexity of the proposed model. After that, in the third stage, GWO was introduced to adjust the biases and weights of ELM to reduce the complexity of the suggested method space. In the fourth step, the efficacy of the suggested approach is compared with the results of 5 CNNs based on transfer learning. The suggested strategy performs superior to other benchmarks, with 98.23% final accuracy on the simulated PRI dataset and 99.20% on the real PRI dataset. Also, the training time of the proposed method for 42,000 images takes 69.45 s, which indicates that the proposed method is real-time. The results show the optimal performance of the suggested approach for both simulated and real data sets. Compared to traditional DCNN models, which require significantly more computation and exhibit reduced performance under noisy conditions, the proposed method demonstrates enhanced resilience and efficiency for both simulated and real data sets. Therefore, the proposed method can be a suitable option in systems where accuracy and time are critical for detecting threat radars.

Keywords: Convolutional neural network; Extreme learning machine; Gray wolf optimizer; Pulse repetition interval modulation.