Due to the poor energy aggregation of multi-sideband amplitude modulation and frequency-modulated amplitude modulation signals，it is often challenging to distinguish between signal singular values and noise singular values. Constructing attractor trajectory matrices，selecting effective rank orders for singular values，and setting iteration termination conditions are key challenges in multiscale singular value decomposition denoising applications. In order to maximize the improvement of signal-to-noise ratio for wideband signals without causing over-denoising，a broadband multi-resolution singular value decomposition denoising method based on energy criterion was proposed. Firstly，different Hankel matrix rows were set for noisy signals for phase space reconstruction，and each attractor trajectory matrix was subjected to singular value decomposition. Then，the first half of the larger singular values were selected as the principal singular values，and the embedding dimension was obtained by searching the number of row in the matrix where the principal singular values energy proportion increment began to decrease，and the effective rank order of the singular value decomposition was obtained by rounding 75% times the embedding dimension. Finally，the iterative denoising was performed on the noisy signal，and the iteration was terminated when the approximate signal singular value energy proportion increment was less than the setting threshold. The verification of multi-sideband amplitude modulation simulation signal and frequency-amplitude modulation signal show that this method does not experience excessive noise reduction，the modulation characteristics are highlighted，and the signal-to-noise ratio gain can reach 11.286 6 dB and 6.904 4 dB，respectively. The verification of frequency-amplitude modulation test signal shows that this method can eliminate noise effectively. Multiple components of the signal are preserved. Some components having obvious kurtosis characteristics are convenient for the interpretation of the failure. This method has certain engineering application value.