Design And Implementation Of Discrete Cosine Transformer Based On Approximate Calculation

With the rapid development of multimedia technology in portable devices,the demand for low power consumption in signal processing has also increased dramatically.The image and audio compressor can reduce the data volume of the image signal and the audio signal,reduce the communication storage capacity and the transmission bandwidth greatly,and reduce the energy consumption of the signal processing largely.In addition,discrete Cosine Transform(DCT)is the core component of multiple video compression standards.Therefore,low-power and high-precision discrete cosine transformer has become an important research direction to reduce the energy consumption of multimedia devices.This paper makes an in-depth study of low-power discrete cosine transformers in terms of development status,theoretical basis,and design methods.According to the fault-tolerance of discrete cosine transformer,the paper proposed a low-power approximate discrete cosine transformer.Approximate discrete cosine transformer in this paper is achieved by row-column decomposition method,that is to say,the entire discrete cosine transformer is divided into three modules,including row one-dimensional DCT module,column one-dimensional DCT module,and transposed memory module.The innovations of this design mainly focus on row/column one-dimensional DCT modules.Four approximate calculation methods are applied in these two modules,including the classification optimization of DCT coefficients,ignoring the high-frequency output of row one-dimensional DCT,approximate optimization of adjacent pixels,and introduction of approximate adder units.After completing the RTL codes of the approximate discrete cosine transformer,this paper first performs functional simulation of the design and shows the results of image processing.Compared with the original image,the average PSNR of the reconstructed image after processing is 38.93 d B,and the reconstructed picture is almost undistorted.Then,the logic synthesis of the design and the other three reference designs were performed in 0.18?m standard CMOS process,and their performances and precision were measured.By comparing the precision and performances of the four designs,it is concluded that the approximated discrete cosine transformer has the best overall performance and it only requires a small amount of energy to obtain a good calculation accuracy.Compared with the integer DCT structure,the area of the approximate discrete cosine transformer structure is reduced by 25.83%,the maximum frequency is increased by 15.65%,the power consumption is reduced by 46.15%,the energy consumption is reduced by as much as 53.44%,and the performance is greatly improved.And the average PSNR is only reduced by 3.84 d B,but this decline does not cause human perception.Compared with the other two designs,the area,frequency,power consumption,and energy consumption of the proposed design are almost equal,but the image quality is greatly improved.Finally,the paper continues the back-end layout design in the 0.18?m standard CMOS process and elaborates the steps in back-end process.Performing static timing analysis,physical rules check and functional consistency check in layout after the physical implementation.At last,simulating and verifing the back-end layout.