• Journal of KIISE:Computer Systems and Theory
• /
• v.26 no.9
• /
• pp.1064-1072
• /
• 1999

The well-known Dimension Exchange Method(DEM) to quantized loads may result in difference in assigned loads to processors as large as logn~units, in the worst case, after balancing for a hypercube of size n. In order to limit the increase of the accumulation of these differences, this paper proposes a method that limits the accumulation of the difference by redistributing an unit load (odd or even) of the same type, if possible, when there is a load distribution between two processors. We reduced the maximum difference to LCEIL logn over {2} RCEIL . The result shows the simulation experiments which show about 30% improvement in speedup compare to the DEM.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.49 no.5
• /
• pp.7-14
• /
• 2012

The quantized beamforming systems always need the channel state information, which must be quantized into a finite set of vectors (named codebook), and feedback only sends the index representing the desired vector. Thereby it minimized the impact of feedback errors, caused by feedback overhead and delay. In this regard, index assignment (IA) methods, an exhaustive-search and group-based schemes, have been presented for minimizes the performance degradation without additional feedback bits. In this paper, we proposed enhanced group-based IA method, which used the optimal codebook design with chordal distance, having the adaptive properties in application of the existing IA methods. When the number of transmit antennas is 4 and LTE codebook is used, Monte-Carlo simulations verify that the proposed scheme has a power advantage of 0.5~1dB to obtain the same bit error rate than methods without IA, and it has 0.1~0.2 dB better performance compared with the existing IA methods over same environment.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.28 no.11B
• /
• pp.965-971
• /
• 2003

This paper suggests a low-complexity frame timing synchronization system for IEEE802.11a wireless LAN systems. The proposed timing synchronization scheme has been implemented by correlating the received OFDM preamble with quantized coefficients composed of ｛0, ${\pm}$2$^{0}$ , ${\pm}$2$^1$‥‥‥ ${\pm}$2$^{i}$ ), where i is an integer number. The 2$^{i}$ -valued coefficients enable the multipliers in the correlation system to be simplified to i-bit shifters. So we can design the correlation system using shifters instead of multipliers. We estimate the performance of the proposed scheme in comparison with conventional systems under the AWGN and Rayleigh fading channels. In this paper we show that the complexity can be reduced by 90% while still maintaining a performance comparable to that of the conventional system.

• Journal of Advanced Navigation Technology
• /
• v.16 no.5
• /
• pp.810-816
• /
• 2012

In this paper, BER(Bit Error Ratio) performances of the TransferJet system, which is the standard of a close proximity inductive wireless communication system, are presented and analyzed. Comparing to other wireless communication systems, the TransferJet system has some advantages such as short communication range(i.e., high security in the wireless communication environments), fewer effects of multipath distortion, and higher transmission rate. In order to demodulate the received signal, either SC(Soft-decision Combining) or HC(Hard-decision Combining) can apply to the despreader and demodulator of the receiver. When the spreading factor is more than 4, the SC scheme approximately has a minimum signal-to-noise ratio gain of 2 dB over the HC scheme. Moreover, from simulation results, we can conclude that the quantization bits of 3 bits are an optimum value for the SC scheme in the TransferJet system since the 3-bit quantization achieves nearly the performance as that attained by double-precision floating-point.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.11a
• /
• pp.21-23
• /
• 2022

ARM 아키텍처를 사용하는 임베디드 시스템에서 int8, fp16, fp32 데이터를 조합하여 c/c++로 작성된 mixed precision CNN 을 실행시키기 위한 프레임워크 구성으로, 네트워크의 레이어마다 다른 정밀도를 사용하여 네트워크 경량화 및 추론 정확도 향상을 위한 최적의 설정을 탐색하는 실험 및 분석이 가능토록 하는 것을 목적으로 한다. 주요 구성은 network forwarding 중 레이어의 입력이 레이어에 설정된 정밀도와 다를 경우 실행되는 양자화/반양자화를 c/c++로 바인딩된 tensorflow 의 quantization 모듈을 사용하여 진행하고 ARM 시스템에서 c/c++의 fp16 을 사용하기 위해 fp16 를 컴파일이 가능한 ARM compiler 를 사용하는 프레임워크를 제안한다.

• The KIPS Transactions:PartB
• /
• v.9B no.4
• /
• pp.429-436
• /
• 2002

In order to apply the artificial neural network to the practical application, it is needed to implement it with the hardware system. It is most promising to make it with the hybrid VLSI among various possible technologies. When we Implement a trained network into the hybrid neuro-chips, it is to be performed the process of the quantization on its neuron outputs and its weights. Unfortunately this process cause the network's outputs to be distorted from the original trained outputs. In this paper we analysed in detail the statistical characteristics of the distortion. The analysis implies that the network is to be trained using the normalized input patterns and finally into the solution with the small weights to reduce the distortion of the network's outputs. We performed the experiment on an application in the time series prediction area to investigate the effectiveness of the results of the analysis. The experiment showed that the network by our method has more smaller distortion compared with the regular network.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.09b
• /
• pp.162-165
• /
• 2003

본 논문에서는 웨이블렛 계수와 Hidden Markov Model(HMM)이용한 얼굴인식 알고리즘을 제안한다. 입력 영상은 이산웨이블렛을 기반으로 한 다해상도 분석기법을 사용하여 데이터 수를 압축한 후, 각각의 해상도에서 얻어진 웨이블렛 계수를 특징벡터로 사용하여 HMM의 모델을 생성한다. 인식단계 에서는 웨이블렛 변환에 의해 생성된 개별대역의 인식값을 더하여 상호 보완함으로써 인식률을 높일 수 있었다. 제안된 알고리즘의 타당성을 검증하기 위하여 기본적 알고리즘인 벡터 양자화(VQ) 기법을 적용한 경우와 기존 얼굴인식에 제안된 DCT-HMM을 이용한 기법과의 인식률 비교를 한 결과, 제안된 방법이 우수한 성능을 보임을 알 수 있었다.

• 제어로봇시스템학회:학술대회논문집
• /
• 1991.10a
• /
• pp.131-135
• /
• 1991

A quantization error model is suggested for analog to frequency(A/F) converter in strapdown inertial navigation system(SDINS),which is characterized by some white noise exciting the state variables. Also, effects on the performance of SDINS by analog to digital(A/D) converter and A/F converter are analyzed and compared via covariance simulation. As a result, A/F converter turns out to be superior to the A/D converter with respect to the induced navigation error and the difficulty in circuit realization. The quantization error model developed in this paper appears to be useful for optimal filter design.

• Journal of the Korea Computer Graphics Society
• /
• v.10 no.4
• /
• pp.22-26
• /
• 2004

본 논문은 거대한 메쉬를 효율적으로 처리하기 위해 이를 조각화하는 새로운 방법을 제시한다. 메쉬의 특징을 반영하여 데이타를 분할하기 위해서 로이드 양자화 방법을 적용하였으며, 거대한 메쉬 데이터를 다루기 위해 외부 메모리를 이용한 부분적인 차트화 갱신 방법을 제시한다. 차트화 갱신 과정에서 효율적으로 매쉬 데이터를 다루기 위해서 임의접근 가능 메쉬 압축구조를 활용한다.이러한 방법을 통해서 수행된 외부 메모리를 이용한 차트화 결과가 지금까지 개발된 메쉬 차트화 결과와 크게 다르지 않음을 보여준다. 이와 같은 메쉬 차트화 시스템을 통해 거대한 메쉬에 기존의 파라미터화, 몰핑, 매칭, 재매쉬화 등의 매쉬 처리기법을 적용할 수 있다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.8 no.5
• /
• pp.1050-1057
• /
• 2004

The purpose of study is development of a fuzzy controller which independent of a computer and its software for fuzzy control of servo system. This paper describes a method of approximate reasoning for fuzzy control of servo system, based on decomposition of $\alpha$-level fuzzy sets, It is propose that fuzzy logic algorithm is a body from fuzzy inference to defuzzificaion in cases where the output variable u directly is generated PWM. The effectiveness of quantified $\alpha$-levels on input/output characteristics of fuzzy controller and output response of DC servo system is investigated. It is concluded that $\alpha$-cut 4 levels give a sufficient result for fuzzy control performance of DC servo system. The experimental results shows that the proposed hardware method is effective for practical applications of DC servo system.