Since the LFM signal occupies a wide band in frequency domain

it’s impossible to use singular value decomposition to separate the noisy LFM signal into signal subspace and noise subspace. To counter this drawback

a new subspacedecomposition algorithm based on the rotation of time-frequency plane is presented in this paper along with its correspondingperformance. A new concept

namely

the "pseudo signal subspace"

and a new transform for detecting the tilting angle calledRadon-STFT transform are proposed. Theoretical predictions and simulation results indicate that the strategies advasted arefeasible for denoising a class of LFM signals.