• Journal of Digital Convergence
• /
• v.18 no.2
• /
• pp.261-268
• /
• 2020

Machine learning is achieved by making a cost function from a given digital signal data and optimizing the cost function. The cost function here has local minimums in the cost function depending on the amount of digital signal data and the structure of the neural network. These local minimums make a problem that prevents learning. Among the many ways of solving these methods, our proposed method is to change the learning step-size. Unlike existed methods using the learning rate (step-size) as a fixed constant, the use of multivariate function as the cost function prevent unnecessary machine learning and find the best way to the minimum value. Numerical experiments show that the results of the proposed method improve about 3%(88.8%→91.5%) performance using the proposed method rather than the existed methods.

• Atmosphere
• /
• v.33 no.4
• /
• pp.331-341
• /
• 2023

A numerical forecasting models usually predict future states by performing time integration considering fixed static time-steps. A time-step that is too long can cause model instability and failure of forecast simulation, and a time-step that is too short can cause unnecessary time integration calculations. Thus, in numerical models, the time-step size can be determined by the CFL (Courant-Friedrichs-Lewy)-condition, and this condition acts as a necessary condition for finding a numerical solution. A static time-step is defined as using the same fixed time-step for time integration. On the other hand, applying a different time-step for each integration while guaranteeing the stability of the solution in time advancement is called an adaptive time-step. The adaptive time-step algorithm is a method of presenting the maximum usable time-step suitable for each integration based on the CFL-condition for the adaptive time-step. In this paper, the adaptive time-step algorithm is applied for the Korean Integrated Model (KIM) to determine suitable parameters used for the adaptive time-step algorithm through the monthly verifications of 10-day simulations (during January and July 2017) at about 12 km resolution. By comparing the numerical results obtained by applying the 25 second static time-step to KIM in Supercomputer 5 (Nurion), it shows similar results in terms of forecast quality, presents the maximum available time-step for each integration, and improves the calculation efficiency by reducing the number of total time integrations by 19%.

• Journal of applied mathematics & informatics
• /
• v.4 no.2
• /
• pp.341-378
• /
• 1997

We consider three classes of numerical methods for solv-ing the semi-explicit differential-algebraic equations of index 1 and higher. These methods use implicit multistep fixed stepsize methods and several iterative processes including simple iteration, full a2nd modified Newton iteration. For these methods we prove convergence theorems and derive error estimates. We consider different ways of choosing initial approximations for these iterative methods and in-vestigate their efficiency in theory and practice.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.15 no.1
• /
• pp.131-137
• /
• 2015

This paper relates with the VSS-MMA (Variable Step Size-Multiple Modulus Algorithm) adaptive equalization algorithm which is possible to improving the equalization performance by use the nonlinear fuction of error signal in the MMA adaptive equalization algorithm that are used for the minimization of the intersymbol interference due to the distortion which occurs in the time dispersive channel for the transmission of QAM signal in the system.. In the conventional MMA, we obtains the tap coefficient of adaptive equalizer using the fixed step size, but in the VSS-MMA, we obtains the tap coefficient of adaptive equalizer using the variable step size based on a nonlinear function of error signal. By adapting the variable step size, it was confirmed that the improved equalization performance were obtained by computer simulation. For this, the equalizer output signal constellation, residual isi, maximum distortion, MSE and SER were used in the performace index.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.16 no.2
• /
• pp.107-112
• /
• 2016

This paper proposes the improving the equalization performance using the variable step size in the S-MMA (Sliced-Multi Modulus Algorithm) equalization algorithm in order to minimize the effect of intersymbol interference which occurs at the nonlinear transfer function of communication channel. The S-MMA were showned for the improving the steady state equalization performance and misadjustment compared to the MMA present algorithm, this two algorithm has a limitation of performance improvement due to the adapting the fixed step size according to the error signal amplitude. In order to solving the abovemensioned problem, the proposed algorithm was adopting the variable step size proportional to the error signal amplitude and the computer simulation was performed for showing the performance improving. As a result of simulation, the proposed VSS S-MMA algorithm has more superior equalization performance compared to the present S-MMA.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.19 no.3
• /
• pp.67-72
• /
• 2019

This paper evaluates the performance of CCA (Compact Constellation Algorithm) adaptive equalization algorithm by varying the step size for minimization of the distortion effect in the communication channel. The CCA combines the conventional DDA and RCA algorithm, it uses the constant modulus of the transmission signal and the considering the output of decision device by the power of compact slice weighting value in order to improving the initial convergence characteristics and the equalization noise by misadjustment in the steady state. In this process, the compact slice weight values were fixed, and the performance of CCA adaptive equalization algorithm was evaluated by the varing the three values of step size for adaptation. As a result of computer simulation, it shows that the smaller step size gives slow convergence speed, but gives excellent performance after at steady state. Especially in SER performance, the small step size gives more robustness that large values.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.19 no.5
• /
• pp.101-106
• /
• 2019

This paper propose the VSS-FC-MMA algorithm that is possible to improve the equalization performance based on varying step size to the FC-MMA adaptive equalization algorithm in order to reducing the intersymbol interference effect occurred in the nonconstant modulus signal transmission, and improved performance were confirmed. The FC-MMA is possible to improve the convergence speed, and degrades the steady state performance based on the fixed step size and modified dispersion constant considering the level number of signal symbol for obtain the error signal in adaptive equalization compared to MMA. The proposed VSS-FC-MMA uses varying step size and current FC-MMA possible to improve the steady state equalization performance, it was confirmed by computer simulation. For this, the signal recovery capabilities and residual isi, MSE, SER were applied for performance comparison index in the same channel and signal to noise ratio. As a result of computer simulation, the proposed VSS-FC-MMA improve the risidual value in steady state and SER performance than the FC-MMA, but has 1.7 times slow convergence time by using varying step size.

• Transactions of the Korean Society for Noise and Vibration Engineering
• /
• v.18 no.3
• /
• pp.284-292
• /
• 2008

The active control technique mostly uses the least-mean-square(LMS) algorithm, because the LMS algorithm can easily obtain the complex transfer function in real-time, particularly when the Filtered-X LMS(FXLMS) algorithm is applied to an active noise control(ANC) system. However, FXLMS algorithm has the demerit that stability of the control is decreased when the step size become larger but the convergence speed is faster because the step size of FXLMS algorithm is fixed. As a result, the system has higher probability which the divergence occurs. Thus the Co-FXLMS algorithm was developed to solve this problem. The Co-FXLMS algorithm is realized by using an estimate of the cross correlation between the adaptation error and the filtered input signal to control the step size. In this paper, the performance of the Co-FXLMS algorithm is presented in comparison with the FXLMS algorithm. Simulation and experimental results show that active noise control using Co-FXLMS is effective in reducing the noise in duct system.

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2007.05a
• /
• pp.598-603
• /
• 2007

The active control technique mostly uses the Least-Mean-Square (LMS) algorithm, because the LMS algorithm can easily obtain the complex transfer function in real-time, particularly when the Filtered-X LMS (FXLMS) algorithm is applied to an active noise control (ANC) system. However, FXLMS algorithm has the demerit that stability of the control is decreased when the step size become larger but the convergence speed is faster because the step size of FXLMS algorithm is fixed. As a result, the system has higher probability which the divergence occurs. Thus the Co-FXLMS algorithm was developed to solve this problem. The Co-FXLMS algorithm is realized by using an estimate of the cross correlation between the adaptation error and the filtered input signal to control the step size. In this paper, the performance of the Co-FXLMS algorithm is presented in comparison with the FXLMS algorithm. Simulation results show that active noise control using Co-FXLMS is effective in reducing the noise in duct system.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.19 no.5
• /
• pp.115-121
• /
• 2019

In this paper, we propose MSAG-SCS-MMA-I adaptive blind equalization with double step size with very small residual ISI and MSE at steady-state while significantly improving the convergence speed of the traditional SCS-MMA-I algorithm in 256-QAM system. And we evaluate the equalization performance for this algorithm. Different step sizes according to the absolute value of decision-directed error instead of a fixed step-size are applied to the tap update equation of MSAG-SCS-MMA-I, which is controlled by binary flags of '1' or '0' obtained from SCS-MMA-I and decision-directed algorithms. This makes for excellent equalization performance. As a result of computer simulation, we confirmed that the proposed algorithm has more better performance than the MMA, SCS-MMA-I, and MSAG-SCS-MMA-I algorithms in terms of the performance index such as residual ISI, MSE, and MD.