The invention discloses a soft-decision spectrum sensing method based on compression non-reconstruction. The method mainly comprises the steps that firstly, a sensing model is established, secondary users establish a
local spectrum sensing model, a binary presumptive model is adopted, and the energy detecting method is adopted under the non-cooperative mode; secondly,
compressed sampling is carried out, and the secondary users utilize a two-dimensional
Gaussian measurement matrix to carry out
compressed sampling on a receiving
signal; thirdly, under the situation that the
compressed sampling data are not reconstructed, the secondary users directly utilize the compressed sampling data to establish the test statistics amount based on energy, whether a primary user exists or not is not judged, and the test statistics amount is transmitted to a
fusion center; fourthly, the soft decision of the
fusion center is carried out, and after the
fusion center collects the test statistics amount sent by all the secondary users, the unified test statistics amount is established and compared with the detecting threshold value, and a final decision result is given.