Circuit structure using number theoretic transform for calculating cyclic convolution

Abstract

The invention belongs to the technical field of integrated circuit design, and particularly relates to a circuit structure using number theoretic transform for calculating cyclic convolution. The circuit structure comprises an input data buffer memory RAM (random-access memory), an output data buffer memory RAM, an operation RAM, a counter, a main controller, an address generating module, a butterfly-shaped operation module, a mold take-up operation unit and the like. According to the circuit structure, only addition (subtraction) and displacement operation is needed in the transform, and multiplication is not needed, so the operation speed is accelerated, the round-off error caused by FFT (fast Fourier transform) is also eliminated through FNT (Fermat number theoretic transform), and the high-precision convolution can be obtained; and in addition, the base function access is not needed, so the storage space is saved, the circuit structure belongs to a circuit structure capable of realizing fast cyclic convolution at smaller chip area and lower power consumption, and the circuit structure can be preferably applied to the fields of image filtering and image processing.

Description

technical field [0001] The invention belongs to the technical field of integrated circuit design, and in particular relates to a new circuit structure using number theory transformation to calculate circular convolution. Background technique [0002] Convolution is a linear operation whose essence is the idea of ​​moving average. It is widely used in image filtering. The common mask operation in image processing is convolution. In addition, convolution has many other applications in engineering and mathematics. In statistics, the weighted sliding average is a kind of convolution. In probability theory, the probability density function of the sum of two statistically independent variables X and Y is the convolution of the probability density functions of x and Y. In acoustics, an echo can be represented by the convolution of the source sound with a function that reflects various reflection effects. In electronic engineering and signal processing, the output of any linear sy...

AI Technical Summary

Problems solved by technology

However, FNT also has shortcomings, mainly because it has no obvious physical meaning; the transformation {Xk} of the sequence {xn} is no longer a spectrum, so the intermediate process cannot be used for frequency measurement like FFT; plus the word length is limited, it is not flexible enough

Images