• IEMEK Journal of Embedded Systems and Applications
• /
• v.17 no.4
• /
• pp.217-228
• /
• 2022

Currently, there are increasing demands for applying deep neural networks (DNNs) in the embedded domain such as classification and object detection. The DNN processing in embedded domain often requires custom hardware such as NPU for acceleration due to the constraints in power, performance, and area. Processing DNN models requires a large amount of data, and its seamless transfer to NPU is crucial for performance. In this paper, we developed a cycle-accurate NPU simulator to evaluate diverse NPU microarchitectures. In addition, we propose a novel technique for reducing the number of memory accesses when processing convolutional layers in convolutional neural networks (CNNs) on the NPU. The main idea is to reuse data with memory interleaving, which recycles the overlapping data between previous and current input windows. Data memory interleaving makes it possible to quickly read consecutive data in unaligned locations. We implemented the proposed technique to the cycle-accurate NPU simulator and measured the performance with LeNet-5, VGGNet-16, and ResNet-50. The experiment shows up to 2.08x speedup in processing one convolutional layer, compared to the baseline.

• Journal of Korea Multimedia Society
• /
• v.5 no.5
• /
• pp.477-487
• /
• 2002

In a VOD system, it is desirable to provide the user with interactive VCR functions such as fast-forward and fast-backward. Typically, these functions can be implemented using the segment interleaving technique in disk away based VOD servers. But the segment interleaving nay not ensure QoS of users because of the concentrated load within some disks. In this paper, we propose an efficient load balancing technique for evenly distributing the retrieval requests across the disks. The proposed technique maintains an access counter in each disk, which has the number of disk accesses. The fast-forward/backward can be achieved by choosing a segment to be accessed from the disk with the minimum number among the access counters. We conduct simulation experiments and then show that our proposed technique outperforms the previous ones.

• Journal of IKEEE
• /
• v.19 no.2
• /
• pp.168-175
• /
• 2015

In this paper, we propose FPGA design technique for rotating LED display device which is capable of displaying videos with the use of the afterimage effect. The proposed technique is made up of image data correction process based on inverse gamma correction and error diffusion, block interleaving process, and data serial output process. The data correction process based on inverse gamma correction and error diffusion is an image data correction step in which image data received are corrected by inverse gamma correction process to convert the data into linear brightness characteristics, and by error diffusion process to reduce the brightness reduction phenomenon in low-gray-level which is caused by inverse gamma correction. In the block interleaving process, the data of the frames entered transversely are first saved in accordance with entrance order, and then only the longitudinal image data are read. The data serial output process is applied to convert the parallel data in a rotating location into serial data and send them to LED Driver IC, in order to send data which will be displayed on high-speedy rotating LED Bar. To evaluate the accuracy of the proposed FPGA design technique, this paper used XC6SLX45-FG484, a Spartan 6 family of Xilinx, as FPGA, and ISE 14.5 as a design tool. According to the evaluation analysis, it was found that goal values were consistent with simulation values in terms of accurate operation of inverse gamma and error diffusion correction, block interleaving operation, and serialized operation of image data.

• Proceedings of the IEEK Conference
• /
• 2002.07a
• /
• pp.706-709
• /
• 2002

This paper describes an effective technique for coding video sequences based on JPEG2000 codec. In the proposed method, multiple frames are combined into one large picture by interleaving each pixel data. A large picture enables images to be coded more efficiently and image quality is improved. A video sequence is efficiently coded by adapting the time correlation of the video sequences to spatial correlation. We demonstrated the effectiveness of this method by encoding video sequences using JPEG2000.

• Journal of Broadcast Engineering
• /
• v.5 no.2
• /
• pp.227-238
• /
• 2000

This paper, at first, provides analysis on loss pattern of Internet based on real experiments of the current Internet. Then, we propose an effective adaptive QoS management technique, in which measured loss pattern as well as PLR (Packet Loss Ratio) are used to select titrate of temporal scalability. level of FEC and retransmission This selection is also incorporated to the MPEG-4 error resilience tools and error concealment techniques. In order to minimize effect of packet loss, multimedia stream is segmented in the unit of group of Pictures (GOP) and interleaving and FEC are applied to the segment. Proposed algorithms are applied to build a VOD system.

• Proceedings of the KOSOMBE Conference
• /
• v.1996 no.11
• /
• pp.310-313
• /
• 1996

In this study, we developed three dimensional FDG brain PET and MRI coregistration technique. The boundaries of the head in PET and MRI were segmented using sinogram of emission PET scan and T1-weighted MRI. We registered both boundaries by minimizing the mean Euclidean distance of those. To display the registered PET and MRI simultaneously, we used weighted normalization method and interleaving method.

• The Transactions of the Korean Institute of Power Electronics
• /
• v.16 no.4
• /
• pp.405-414
• /
• 2011

Recently, the development of several hundreds of kW scale PCS for fuel cell generation is required as commercialization process of distributed generation systems using high temperature fuel cells such as MCFC begins. This paper proposes and optimized topology suitable for MCFC fuel cell generation system and LCL filter design method considering voltage quality of local loads such as MBOP. An interleaving technique is applied to step-up DC-DC converter, optimized number of phases is determined considering efficiency and volume. Also, a LCL filter design method is proposed considering quality of current injected to the grid as well as that of voltage across the local load. The proposed PCS system is validated through reduced 1kW prototype.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.8 no.3
• /
• pp.57-70
• /
• 2008

This paper introduces a new paradigm of physical layer security through channel coding for wireless networks. The well known spread spectrum based physical layer security in wireless network is applicable when code division multiple access (CDMA) is used as wireless air link interface. In our proposal, we incorporate the proposed security protocol within channel coding as channel coding is an essential part of all kind of wireless communications. Channel coding has a built-in security in the sense of encoding and decoding algorithm. Decoding of a particular codeword is possible only when the encoding procedure is exactly known. This point is the key of our proposed security protocol. The common parameter that required for both encoder and decoder is generally a generator matrix. We proposed a random selection of generators according to a security key to ensure the secrecy of the networks against unauthorized access. Therefore, the conventional channel coding technique is used as a security controller of the network along with its error correcting purpose.

• Smart Structures and Systems
• /
• v.20 no.1
• /
• pp.35-42
• /
• 2017

Recently, compressive sensing based data loss recovery techniques have become popular for Structural Health Monitoring (SHM) applications. These techniques involve an encoding process which is onerous to sensor node because of random sensing matrices used in compressive sensing. In this paper, we are presenting a model where the sampled raw acceleration data is directly transmitted to base station/receiver without performing any type of encoding at transmitter. The received incomplete acceleration data after data losses can be reconstructed faithfully using compressive sensing based reconstruction techniques. An in-depth simulated analysis is presented on how random losses and continuous losses affects the reconstruction of acceleration signals (obtained from a real bridge). Along with performance analysis for different simulated data losses (from 10 to 50%), advantages of performing interleaving before transmission are also presented.

• Proceedings of the KIEE Conference
• /
• 2008.10b
• /
• pp.396-397
• /
• 2008

OFDM 송수신 시스템에서 채널성능을 개선하기 위해 CDD(Cyclic Delay Diversity) 기술을 적용하면, Frequency Interlea ving의 영향으로 인해 burst error가 randomize되어, convolutional code/Viterbi decoding FEC의 BER 성능개선 효과를 배가 할 수 있다. 그러나 차세대 통신시스템에 적극 활용되는 Turbo FEC는 자체 interleaving 구조를 가지고 있으므로, CDD기술과의 결합으로 인한 영향을 증명하기 위하여, 본 논문에서는 Turbo FEE를 사용하는 OFDM CDD시스템을 MATLAB을 이용한 computer simulation을 통해서 BER 성능을 비교분석 하였다.