• Proceedings of the IEEK Conference
• /
• 1998.10a
• /
• pp.1215-1218
• /
• 1998

본 논문은 Digital Audio Compression(AC-3) Standard 인 A-52를 기반으로 하였으며 Borland C++3.1 Compiler를 사용하여 AC-3 Decoding Algorithm 구현하였다. Input Stream은 DVD VOB File에서 AC-3 Stream만을 분리하여 사용하며 최종 출력은 16 Bit PCM File이다. AC-3의 Frame구조는 Synchronization Information, Bit Stream Information, Audio Block, Auxiliary Data, Error Check로 구성된다. Aduio Block 은 모두 6개의 Block으로 나뉘어져 있다. BSI와 Side Information을 참조하여 Exponent를 추출하여 Exponent Strategy에 따라 Exponent를 복원한다. 복원된 Exponent 정보를 이용하여 Bit Allocation을 수행하여 각각의 Mantissa에 할당된 Bit수를 계산하고 Stream으로부터 Mantissa를 추출한다. Coupling Parameter를 참조하ㅕ Coupling Channel을 Original Channel로 복원시킨다. Stereo Mode에 대해서는 Rematrixing을 수행한다. Dynamic Range는 Mantissa와 Exponent의 Magnitude를 바꾸는 것으로 선택적으로 사용할 수 있다. Mantissa와 Exponent를 결합하여 Floating Point coefficient로 만든 후 Inverse Transform을 수행하면 PCM Data를 얻을 수 있다. PC에서 듣기 위해서는 Multi Channel을 Stereo나 Mono로 Downmix를 수행한다. 이렇게 만들어진 PCM data는 PCM Data를 재생하는 프로그램으로 재생할 수 있다.

• Korean Journal of Computational Design and Engineering
• /
• v.7 no.2
• /
• pp.81-88
• /
• 2002

Transmission of 3D shape model through Internet has become one of the hottest issues in these days. Presented in this paper is a new approach for the rapid transmission of the geometry data of the shape model. By analyzing the important three factors, the shape fidelity, the file size, and the decompression time, for the compression, we point out the potential problems of previous approaches of using the deltas between consecutive vertices and propose an alternative of directly using the position values of vertices of the model. It turns out that the proposed approach has smaller file size, has lesser distortion in the model, and the decompression is faster.

• Journal of information and communication convergence engineering
• /
• v.19 no.1
• /
• pp.22-28
• /
• 2021

Convolutional neural networks (CNNs) are one of the most frequently used artificial intelligence techniques. Among CNN-based applications, small and timing-sensitive applications have emerged, which must be reliable to prevent severe accidents. However, as the small and timing-sensitive systems do not have sufficient system resources, they do not possess proper error protection schemes. In this paper, we propose MATE, which is a low-cost CNN weight error correction technique. Based on the observation that all mantissa bits are not closely related to the accuracy, MATE replaces some mantissa bits in the weight with error correction codes. Therefore, MATE can provide high data protection without requiring additional memory space or modifying the memory architecture. The experimental results demonstrate that MATE retains nearly the same accuracy as the ideal error-free case on erroneous DRAM and has approximately 60% accuracy, even with extremely high bit error rates.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.1 s.307
• /
• pp.91-100
• /
• 2006

A scheme is proposed for implementing the block LMS algorithm in a block floating point framework that permits processing of data over a wide dynamic range at a processor complexity and coat as low as that of a fixed point processor. The proposed scheme adopts appropriate formats for representing the filter coefficients and the data. Using these and a new upper bound on the step size, update relations for the filter weight mantissas and exponent are developed, taking care so that neither overflow occurs, nor are quantifies which are already very small multiplied directly. It is further shown how the mantissas of the filter coefficients and also the filter output can be evaluated faster by suitably modifying the approach of the fast block LMS algorithm

• Journal of Korea Society of Industrial Information Systems
• /
• v.18 no.6
• /
• pp.31-37
• /
• 2013

Modern graphic processors, multimedia processors and audio processors mostly use floating-point number. High-level language such as C and Java use both single precision and double precision floating-point number. In this paper, an algorithm which computes the reciprocal of double precision floating-point number using a 32 bit multiplier is proposed. It divides the mantissa of double precision floating-point number to upper part and lower part, and calculates the reciprocal of the upper part with Newton-Raphson algorithm. And it computes the reciprocal of double precision floating-point number with calculated upper part reciprocal as the initial value. Since the number of multiplications performed by the proposed algorithm is dependent on the mantissa of floating-point number, the average number of multiplications per an operation is derived from some reciprocal tables with varying sizes.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.15 no.5
• /
• pp.3093-3099
• /
• 2014

Modern graphic processors, multimedia processors and audio processors mostly use floating-point number. Meanwhile, high-level language such as C and Java uses both single-precision and double precision floating-point number. In this paper, an algorithm which computes the reciprocal of double precision floating-point number using a 32 bit multiplier is proposed. It divides the mantissa of double precision floating-point number to upper part and lower part, and calculates the reciprocal of the upper part with Goldschmidt's algorithm, and computes the reciprocal of double precision floating-point number with calculated upper part reciprocal as the initial value is proposed. Since the number of multiplications performed by the proposed algorithm is dependent on the mantissa of floating-point number, the average number of multiplications per an operation is derived from some reciprocal tables with varying sizes.

• Journal of IKEEE
• /
• v.14 no.2
• /
• pp.64-68
• /
• 2010

This paper proposes efficient log and exponent calculation methods using a dual phase instruction set without additional ALU unit for a mobile enviroment. Using the Dual Phase Instruction set, it extracts exponent and mantissa from expression of floating point and calculates 24bit single precision floating point of log approximation using the Taylor series expansion algorithm. And with dual phase instruction set, it reduces instruction excution cycles. The proposed Dual Phase architecture reduces the performance degradation and maintain smaller size.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.5 no.3
• /
• pp.517-524
• /
• 2001

This paper describes a design of radix-2 SRT divider unit, which supports IEEE-754 floating-point standard, using redundant binary number system (RBNS). With the RBNS, the partial quotient decision logic can operate about 20-% faster, as well as can be implemented with a simple hardware when compared to the conventional methods based on two's complement arithmetic. By using a new redundant binary adder proposed in this paper, the mantissa divider is efficiently implemented, thus resulting in about 20% smaller area than other works. The divider unit supports double precision format, five exceptions and four rounding modes. It was verified with Verilog HDL and Verilog-XL.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2010.05a
• /
• pp.320-323
• /
• 2010

This paper proposes efficient log and exponent calculation methods using a dual phase instruction set without additional ALU unit for a mobile enviroment. Using the Dual Phase Instruction set, it extracts exponent and mantissa from expression of floating point and calculates 24bit single precision floating point of log approximation using the Taylor series expansion algorithm. And with dual phase instruction set, it reduces instruction excution cycles. The proposed Dual Phase architecture reduces the performance degradation and maintain smaller size.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.37 no.3
• /
• pp.45-55
• /
• 2000

In this paper, we proposed the hardware architecture of wavelet transform digital filter for an image processing. Filter bank pyramid algorithm is used for wavelet transform and each fillet is implemented by the FIR filter. For DWT computation, because the memory controller is implemented by hardware, we can efficiently process the multisolution decomposition of the image data only input the parameter. As a result of the image Processing in this paper, 33㏈ PSNR has been obtained on 512$\times$512 B/W image due to 11-bit mantissa processing in FPGA Implementation. And because of using QMF( Quadrature Mirror Filter) properties, it reduces half number of the multiplier needed DWT(Discrete Wavelet Transform) computation so the hardware size is reduced largely. The proposed scheme can increase the efficiency of an image Processing as well as hardware size reduced. The hardware design proposed of DWT fillet bank is synthesized by VHDL coding and then the test board is manufactured, the operating Program and the application Program are implemented using MFC++ and C++ language each other.