Enhancing time‐frequency resolution via deep‐learning framework

The fixed window function used in the short-time Fourier transform (STFT) does not guarantee both time and frequency resolution, exerting a negative impact on subsequent study of time-frequency analysis (TFA). To avoid these limitations, post-processing method that enhances resolution using deep-learning (DL) framework is proposed. Initially, deconvolution theoretical formula derived operation performed representation (TFR) STFT via deconvolution, calculation to obtain ideal (ITFR). Then, aiming at adverse influence function, novel fully-convolutional encoder-decoder network trained preserve effective features acquire optimal kernel. In essence, generation kernel can be regarded as process. authors conducted qualitative quantitative analyses numerical simulations, with experimental results demonstrate proposed achieves satisfactory TFR, possesses strong anti-noise capabilities, exhibits high steady-state generalisation capability. Furthermore, comparative experiment several TFA methods indicate yields significantly improved performance terms energy concentration, computational load.