2017 Fiscal Year Annual Research Report
Curved computer generated holography for 3D displays by development of paraxial solutions
| Project/Area Number |
16K20948
|
|---|
| Research Institution | National Institute of Information and Communications Technology |
|---|
Principal Investigator |
ボワズ ジャキン 国立研究開発法人情報通信研究機構, 電磁波研究所電磁波応用総合研究室, 研究員 (00726325)
|
|---|
| Project Period (FY) |
2016-04-01 – 2018-03-31
|
| Keywords | Cylindrical holography / Computer holography / Parallel computing / Diffraction theory |
|---|
| Outline of Annual Research Achievements |
1. Fast and accurate computations of cylindrical computer-generated holograms was realized using Bessel functions expansion. The computation method requires less sampling compared to conventional techniques and hence saves a lot of calculation time and memory usage. This method will be very useful in realizing holographic head up displays in the future. The research findings were published as a peer reviewed article.
2. The developed computational solutions have to implemented on a high performance computing device to achieve reasonable computation time. From analysis it was found that multi GPU clusters is the best candidate to satisfy the hologram computation requirements. A 16-node PC cluster with each node having one GPU was setup to realize the calculations.
3. We found that data dependency in the computational solutions affected the performance due to communication bottle neck in a cluster environment. A new algorithm was developed with the aim of avoiding computation bottleneck. We used the shifting property of Fourier transform to realize computations in totally independent blocks (in each node) with out any need for inter-nodal communication.
4. The important feature of this method is its ability to decompose both object and hologram plane during calculation and we have named this method as the 'decomposition method'. By totally avoiding communication between nodes the potential of the computing devices can be utilized maximum. We were able to achieve 20-times improvement in a 8-node multi GPU cluster compared to the conventional technique.
|
