| Project/Area Number |
12044203
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research on Priority Areas
|
| Allocation Type | Single-year Grants |
|---|
| Review Section |
Science and Engineering
|
|---|
| Research Institution | Tohoku University |
|---|
Principal Investigator |
KAWAMATA Masayuki Tohoku University, Graduate School of Engineering, Professor, 大学院・工学研究科, 教授 (70153004)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
ABE Masahide Tohoku University, Graduate School of Engineering, Research Associate, 大学院・工学研究科, 助手 (90312602)
|
|---|
| Project Period (FY) |
2000 – 2002
|
| Project Status |
Completed (Fiscal Year 2003)
|
| Budget Amount *help |
¥46,700,000 (Direct Cost: ¥46,700,000)
Fiscal Year 2002: ¥15,400,000 (Direct Cost: ¥15,400,000)
Fiscal Year 2001: ¥16,300,000 (Direct Cost: ¥16,300,000)
Fiscal Year 2000: ¥15,000,000 (Direct Cost: ¥15,000,000)
|
|---|
| Keywords | Adaptive Signal Processing / Evolutionary Computation / Genetic Algorithm / Evolutionary Digital Filter / Parallel Processing / Multidimensional Signal Processing / 進化論的計算 / 画像処理 / 認識 / ディジタルフィルタ / 並列信号処理 |
|---|
| Research Abstract |
This research implements evolutionary digital filters (EDFs) on the ASIC and the FPGA for its parallel processing, and evaluates its performance. The EDF is an adaptive digital filter (ADF) which is controlled by adaptive algorithm based on the evolutionary strategies of living things. The advantages of the EDF are summarized as follows
The adaptive algorithm of the EDF is a population-based and robust optimization method, especially used to tackle high-dimensional and multi-modal search space problems. It is a non-gradient and multi-point search algorithm. Thus, it is not susceptible to local minimum problems that arise from a multiple-peak surface.
The EDF can adopt the various error functions as the fitness function according to application, for example, the p-power norm error function, the maximum error function and so on.
The adaptive algorithm of the EDF has a self-stabilizing feature whereby unstable poles have a tendency to migrate back into the stable region. In addition, the EDF can search the poles which are near the unit circle.
Since the EDF has multiple inner filters, the parallel realization can reduce the calculation time. In the proposed method, a reproduction algorithm is changed to a simplified algorithm for the hardware implementation. The hardware EDF consists of the Filtering and Fitness Calculation (FFC) module and the Reproduction and Selection (RS) module. The FFC module has the submodule that can operate parallel because the EDF has multiple inner filters. The RS module has the submodule that can perform both asexual reproduction and selection, and sexual reproduction and selection. ASIC implementation results show that the EDF module can be clocked at 20.0MHz, and it contains 63,652 gates.
|
