<p>Owing to the widespread application of analogue circuits in electronic systems, fault detection and localization are crucial. However, during the early stages of circuit faults, the fault characteristics are weak, making diagnosis difficult. This paper presents an incipient soft fault diagnosis method for analogue circuits based on a multiscale fault diagnosis network. Building on the continuous wavelet transform, a similarity matrix between fault signals and fault-free signals is constructed on the basis of the cosine similarity. By combining phase and amplitude information, a stacked matrix is formed and used as the input to the model. Moreover, a multiscale fault diagnosis network is designed; the primary component is a channel feature aggregation module, which enables multiscale extraction and fusion of fault features to complete the fault diagnosis task. This method was validated through simulation experiments with the Sallen-Key bandpass filter circuit and four-op-amp biquadratic filter circuit, and the results indicate that the method is accurate and reliable.</p>