Design Of CMOS Image Sensor For Spectrum Reconstruction And System Verification

Different from traditional spectrometer for the spectrum acquisition of spatial point,the imaging spectrometer focuses on the spectrum acquisition of spatial plane,which has important applications in satellite remote sensing,computer vision and medical imaging.Inspired by MSFA(Multi-Spectral Filter Array),this paper establishes the spectral reconstruction method by finding a finite number of orthogonal basis vectors in spectrum space as the filtering curves of filters,and the incident spectra can be reconstructed by a linear combination of these orthogonal basis vectors.Orthogonal basis coefficients are the output electrical signals of photosensitive pixels under the filters with correspongding basis curve.This method gets rid of complex least-squares algorithm which can not be avoided by the traditional wideband-filtering spectral reconstruction method,is realtime and efficient in computing resources,and is more conducive to implement spectral reconstruction system with high integration level,low power consumption and low cost.Finally,this paper innovatively proposes an imaging-level spectral reconstruction scheme based on the CMOS image sensor MSFA method,and completes the work from bottom-level circuit design,algorithm verification,to top-level system integration.This paper firstly completes the circuit design,post-simulation and test verification of CMOS image sensor with basic imaging function based on TSMC 0.18 um CIS process.Then,based on the Munsell data set,the first four principal components are extracted using the PCA method as the original filtering basis vectors,and the validity of these basis vectors is verified on the test data set.Considering the real filtering characteristics that can be made,the filtering curve of actual filter to be made is built as the original basis split and fitted into Gaussian functions and multiplied by a certain amplification factor.Combining the specific requirements of spectral reconstruction,the second-version CMOS image sensor circuit is designed to use a block with 4x4 pixels as structural unit,which has a column-level signal processing capability to calculate the basis coefficients during pixel signal output process.The designed CMOS image sensor has a certain degree of versatility.It can simultaneously address pixels in the intersection points of arbitrary rows and arbitrary columns in 2x2 blocks,and perform addition and subtraction operations on the output signals with coefficients.The coefficient ranges from-3.1 to 3.1 with 0.1 step length.Post-simulation shows that one computation operation can be done within 3us with a power consumption of 309 uW in average.Finalize the system test scheme,including test light source and test object selection,new basis calculation based on the test light source,filter waveform design by splitting the original basis vectors into gaussian functions and filter arrangement,as well as the determination of actual test light path structure and output data acquisition methods.For three different filter bases: the original basis with light source information,the basis formed by fitted gaussian functions of original basis and RGB basis of a commercial image sensor,the system simulation results show that for 24 color patches,for the sum of squared error(SSE)between the incident spectrum and the spectrum rebuilt by the three different bases,the mean of SSE of the first two bases are only 1.6% and 2.04% of that of the RGB basis.