Reliability Parameter Estimation Of1/f Noise Of High-Power LD Based On Compressed Sensing

As the rapid development of information technology and network technology, the informationcontent has also had a tremendous increasing. Besids, Signal transmission, data processing andcommunication efficiency requirements are also increasing. On the other hand, the accuracyrequirements of hardware and equipment is also rising. Fourier transform in the application ofinformation science, opened the way for mankind to achieve signal processing. Wavelet transform,using the signal of non-orthogonal transform-based signal processing, image processing and signalanalysis techniques provide a more biased in favor of the application of engineering and technicalapproach.However, when people make the sample of signals in communication theory, theNyquist–Shannon sampling theorem has to be followed. That is, A signal or function isbandlimited if it contains no energy at frequencies higher than some bandlimit or bandwidth B.This objective fact, no doubt brought great challenges for us in the efficient signal transmission.1/f noise is a noise widely exists in nature. It also exists in LDs as a low-frequency electricalnoise. According to the different devices will also be mixed shot noise, the generation-recombination noise and Gaussian white noise.1/f noise frequency characteristics play animportant role in the nondestructive testing of semiconductor lasers. In The low frequency range,1/f noise to a large extent by the Gaussian white noise annihilation. This will inevitably lead todifficulties on the1/f noise parameter estimates. Spectrum analyzer to measure the structures ofthe hardware system will bring the cost of high consumption. Using Spectrum analyzer can exactmeasure the1/f noise, but will cost a large sum of consumer as well. Wavelet analysis theory canaccurately extract the noise, but will consume large computing resources. This article describes thedevelopment of semiconductor laser history and the noise component and the basic principle, twoimportant features.Compressed Sensing tries to use another strategy to sample the signal by measuring it through a RIP matrix, under the condition of signal has a spares decomposition structure or can be sparsedecomposed. Measure y which is a M-dimensional vector whose is able to reconstruct the originalsignal x which is N-dimensional vector perfectly (M <<N). In CS region, one aspect of recentresearches focus on trying to promote the precision of reconstruction by taking known of priorilike variance or some structure information into reconstruct algorithm. Because of one of thesufficient condition of signal reconstruction is x can be sparse decomposed, it is a problem thatchoosing an appropriate base Ψ to transform signal x into a k-sparse signal x’ which means x’ hasonly vary k coefficients which are non-zero. As we known, in real world, most signals are notexactly sparse like image or sound, so as the1/f noise we studied. On the other hand, randommatrices whose entries Gaussian, Bernoulli (±1), i.e., or some others like Toeplitz matrix appearedto be highly probability provided restricted isometry property (RIP).Thus we choose Gaussianmatrix in our simulation.The reconstruct algorithms includes greedy, convex optimizationbased,or some other intergraded algorithm, which are developing faced on engineer application,for instance, the radar sampling or signal denoising.This paper introduced a new method of1/f noise estimation on compressed sensing (CS). Wesuppose wavelet tree structure as its factor of sparse decomposition, use Compressed Sensingbasis pursuit de-noising CS-OMPDN to reconstruct the1/f signal after a rough wavelet de-noising,setting wavelet base as the base of Compressed Sensing. In our algorithm, we use Gaussian matrixas the reconstruction matrix which is follow the RIP. Our work concluded that our algorithm isprecisely to estimate the1/f from the signal of white noise.We use fractal theory to prove the feasibility of our algorithm. We use Matlab to build a series of1/f signal mixed with different power of Gauss White noise. Then, try to recovery the1/f noise.The result proved that our algorithm works well on de-noising of1/f signal mixed with Whitenoise. Meanwhile, we use TMSDM6437DSP of TI company to analyze1/f noise and also had agood performance on recovering the1/f noise.