Abstract
Estimating the precise frequencies in multi-sine signals, particularly when they have closely spaced frequency components, is quite challenging. In response to the aforementioned challenge, we introduce EFreqNet, a super-resolution frequency estimation algorithm grounded in the complex-value neural network (CVNN) framework. The proposed EFreqNet differentiates itself from previous algorithms such as DeepFreq and cResFreq by incorporating complex modules. The proposed modules include an adaptive matched filtering module (AMFM), a frequency estimation module (FEM), and a frequency super-resolution module (FSM). Furthermore, EFreqNet successfully decouples the processes of denoising and frequency estimation by concurrently training the model with denoising losses and frequency estimation losses. EFreqNet's superior performance across diverse signal-to-noise ratio (SNR) environments is evidenced by comparative experiments. As a result, EFreqNet surpasses five other advanced algorithms and demonstrates enhanced target detail in both synthetic and real radar system signals, underscoring its significant potential.
Funding Information
  • Special Foundation for Hunan Innovation Province Construction (2020RC3004)

This publication has 31 references indexed in Scilit: