| In the era of big data,society has an increasing demand for computing power.Due to the problem of transistor heat dissipation and the bottleneck of the bandwidth of electrical computing,the computing speed and computing power of traditional electronic computers are difficult to be further improved,leading to the failure of "Moore's Law".Optical computing is a technique that uses photons to calculate.Compared to electrical computing with insufficient bandwidth and poor parallelism computing ability,light naturally has the characteristics of high bandwidth,parallelism,and high speed.Optical computers surpass ordinary electrical computers that are difficult to match both customized and common computing.Optical computing is a feasible way for heterogeneous computing in the future.The construction of a flexible and simple optical computing system will greatly promote the development of computing technologies.Matrix computing is the foundation of scientific computing.Most scientific computing problems can be converted into matrix computing problems.The vector and matrix multiplication is the basis of matrix and matrix multiplication.Traditional CPU is good at processing logic-controlled tasks but is not suitable for parallel tasks such as matrix computing.Light naturally has the characteristics of parallelism,so it is more suitable for matrix calculation.The problems are how to simply and efficiently realize the optical multiplication of vector and matrix and how to further realize the complete matrix and matrix multiplication.The traditional optical matrix calculation scheme uses spatial light modulation technology,which is sensitive to disturbance and noise,limiting the practicality of the scheme.To efficiently achieve optical multiplication of vector and matrix and complete matrix and matrix multiplication,the main works of this thesis are as follows:(1)This thesis designs an optical cyclic frequency shift calculation system,which can realize the vector and matrix multiplication in the time domain to avoid the problem of spatial light modulation that is sensitive to perturbation.This thesis uses vector and matrix multiplication to calculate the 31-bit m-sequence autocorrelation.The m-sequence autocorrelation sequence waveform should be a periodic pulse.The waveform of the autocorrelation sequence obtained in the experiment is a pulse with a period of 400.6ns,and the round-trip time of the fiber ring is 400.092ns,the error on the period is 0.127%.The theoretical pulse width ?? is 12.9062ns,and the error of the pulse width of a single symbol is measured as 0.2%.This system can realize a special optical vector and matrix multiplier,which provides new ideas for optical heterogeneous computing.(2)To accelerate the calculation of the vector and matrix multiplication,the thesis further improves the proposed system,loading the 31-bit m sequence on 5 different initial carriers.The simulation calculation of the m sequence autocorrelation is accelerated by about 4.4 times.The optical vector and matrix multipliers have been further improved,enhancing the practicality of the system.(3)In order to realize the complete matrix and matrix multiplication in the time domain,this thesis continues to improve the optical computing systems,through multi-wavelength multiplexing and demultiplexing.It can achieve complete matrix and matrix multiplication.A 3*8 matrix X and an 8*10 matrix M are randomly generated,the multiplication result of which is a matrix V.The V matrix obtained by simulation and the theoretical V matrix are flattened and normalized into two sequences,respectively.The standard deviation of two sequence differences is used to measure the error of the matrix.It is 0.050613.This shows that the system can calculate matrix and matrix multiplication with good performance,which enhances the versatility of this optical computing system. |
PDF Full Text Request