Study On Feature Recovery From Compressed Measurements Under Low SNR

With the development of wireless communication technology toward to the high frequency point and wide frequency band,the scarcity of available radio spectrum gradually becomes a bottleneck restricting the development of wireless communications technology.How to make more effective use of available spectrum resources is a serious problem,cognitive radio technology using spectrum sharing mechanism to alleviate this problem,specifically,through dynamically sensing spectrum holes on the broadband spectrum to find an unused free band and then using this spectrum band to access the network.Consequently,the cognitive radio system must have an efficient method for dynamic spectrum sensing,however,as fast and dynamically wideband spectrum sensing requires high speed ADC devices and DSP devices,this would lead to high costs and even harder to achieve.But,in wideband communication systems,it often has the following characteristics:?although the spectrum to be sensed has broad bandwidth,the occupied bands used by active users generally sparsely distribute on the whole communication band;?in order to reduce the interference to other active users,the transmitted signal power is often low,thus the SNR of received signal at the receiver is also low.Owning to the first characteristic,we can use the compressed sensing framework to efficiently sample signals within the communication band through random measurement,and store the compressed observations from which can extract the features of signals further.For the second feature,by choosing characteristics which are robust to noise/interference(e.g.cyclic spectrum,higher-order statistics),or by filtering/suppressing noise/interference in compressed measurement samples,we can obtain better performance.With the help of the technology of signal detection/recognition and the theory of compressed sensing,this thesis concentrates on these following points:1)Introduction of compressed sensing and related random measurement matrix,compressed reconstruction algorithms,etc.Analyze and simulate the sparsity of cyclic statistics of typical modulated signal such as MASK,MPSK,MFSK,MQAM,OFDM,etc.2)An efficient method is proposed which can directly reconstruct second statistics of the signal from compressed measurements without first to reconstruct the original signal,which effectively reduces the computational complexity and at the same time improves the compression ratio in the process of random sampling.3)A compressed filtering method is proposed which pre-processes the compressed measurements in compressed domain to suppress noise/interference within it,significant performance improvement can be achieved in low SNR and meanwhile the requirement of compression ratio can also be decreased.