Hardware implementation of the pseudo-spectral time-domain method

• US 7,295,939 B2
• Filed: 06/24/2003
• Issued: 11/13/2007
• Est. Priority Date: 06/24/2002
• Status: Expired due to Fees

• Current Assignee: EM Photonics, Inc.
• Original Assignee: EM Photonics, Inc.
• Inventors: Prather, Dennis W., Durbano, James P., Humphrey, John R.
• Primary Examiner(s): Ramos-Feliciano; Eliseo
• Assistant Examiner(s): Gutierrez; Anthony
• Application Number:
  US10/517,224
• Publication Number:
  US 20050154546A1
• Time in Patent Office: 1,603 Days
• Field of Search: 702/182, 702/57, 702/196, 702/77
• US Class Current: 702/77
• CPC Class Codes: G06F 17/13 Differential equations using digital differential analysers G06F7/64

Abstract

A computer hardware configuration for performing the pseudo-spectral time-domain (PSTD) method on data. The hardware configuration includes a forward fast Fourier transform (FFT) unit that calculates a forward fast Fourier transform (FFT) from the data, and a complex multiplication unit that receives the FFT-processed data and calculates a spatial derivative in the frequency domain from the FFT-processed data. The hardware configuration further includes an inverse fast Fourier transform (IFFT) unit that converts the spatial derivative in the frequency domain from the complex multiplication unit into the time domain, and a computation engine that solves a PSTD equation based upon the spatial derivative in the time domain received from the IFFT unit.

Claims

1. A system for performing the pseudo-spectral time-domain (PSTD) method on data, comprising:

a forward fast Fourier transform (FFT) unit calculating a forward fast Fourier transform (FFT) from the data;

a complex multiplication unit receiving the FFT-processed data and calculating a spatial derivative in the frequency domain from the FFT-processed data;

an inverse fast Fourier transform (IFFT) unit converting the spatial derivative in the frequency domain from the complex multiplication unit into the time domain;

a computation engine solving a PSTD equation based upon the spatial derivative in the time domain received from the IFFT unit; and

a memory subsystem that provides input values to the FFT unit and receives output results from the computation engine,

wherein input values for the FFT unit are stored in increasing-x, increasing-y and increasing-z patterns in the memory subsystem that allows a burst read of the memory subsystem to maximize throughput.

2. A system as recited in claim 1, wherein the PSTD equation takes the form:

$E_{\mathrm{ab}}=A<br><br><br><br>E_{\mathrm{ab}}+B<br><br>\frac{\partial <br><br>H_c}{\partial <br><br>b}+C<br><br><br><br>E_{\mathrm{ab}}^{\mathrm{inc}},$

where a, b, and c are directions (x, y, and z), A, B, and C are coefficients based on material properties of a medium, and E_ab^inc is the incident field associated with the node.

3. A system as recited in claim 1, wherein as the FFT is being calculated, primary fields, incident fields, and coefficients are being fetched by the system.

4. A system as recited in claim 1, wherein the FFT and IFFT units are provided inside a field-programmable gate array (FPGA).

5. A system as recited in claim 4, wherein the FFT and IFFT calculations are performed by a digital signal processing (DSP) chip.

6. A system for performing the pseudo-spectral time-domain (PSTD) method on data, comprising:

a plurality of forward fast Fourier transform (FFT) units, each FFT unit calculating a forward fast Fourier transform (FFT) from the data;

a plurality of complex multiplication units, each complex multiplication unit receiving the FFT-processed data from a corresponding FFT unit and calculating a spatial derivative in the frequency domain from the FFT-processed data;

a plurality of inverse fast Fourier transform (IFFT) units, each IFFT unit converting the spatial derivative in the frequency domain from a corresponding complex multiplication unit into the time domain;

a plurality of computation engines, each computation engine solving a PSTD equation based upon the spatial derivative in the time domain received from a corresponding IFFT unit; and

a memory subsystem that provides input values to the FFT unit and receives output results from the computation engine,

wherein the input values for the FFT unit are stored in increasing-x, increasing-y and increasing-z patterns in the memory subsystem that allows a burst read of the memory subsystem to maximize throughput.

7. A system as recited in claim 6, wherein the PSTD equation takes the form:

$E_{\mathrm{ab}}=A<br><br><br><br>E_{\mathrm{ab}}+B<br><br>\frac{\partial <br><br>H_c}{\partial <br><br>b}+C<br><br><br><br>E_{\mathrm{ab}}^{\mathrm{inc}},$

where a, b, and c are directions (x, y, and z), A, B, and C are coefficients based on material properties of a medium, and E_ab^inc is the incident field associated with the node.

8. A system as recited in claim 6, wherein as the FFT is being calculated, primary fields, incident fields, and coefficients are being fetched by the system.

9. A system as recited in claim 6, wherein the plurality of FFT and IFFT units are provided inside a field-programmable gate array (FPGA).

10. A system as recited in claim 9, wherein the FFT and IFFT calculations are performed by a digital signal processing (DSP) chip.

11. A computer hardware configuration for performing the pseudo-spectral time-domain (PSTD) method on data, comprising:

a forward fast Fourier transform (FFT) unit calculating a forward fast Fourier transform (FFT) from the data;

a complex multiplication unit receiving the FFT-processed data and calculating a spatial derivative in the frequency domain from the FFT-processed data;

an inverse fast Fourier transform (IFFT) unit converting the spatial derivative in the frequency domain from the complex multiplication unit into the time domain;

a computation engine solving a PSTD equation based upon the spatial derivative in the time domain received from the IFFT unit; and

a memory subsystem that provides input values to the FFT unit and receives output results from the computation engine,

wherein the input values for the FFT unit are stored in increasing-x, increasing-y and increasing-z patterns in the memory subsystem that allows a burst read of the memory subsystem to maximize throughput.

12. A computer hardware configuration as recited in claim 11, wherein the PSTD equation takes the form:

$E_{\mathrm{ab}}=A<br><br><br><br>E_{\mathrm{ab}}+B<br><br>\frac{\partial <br><br>H_c}{\partial <br><br>b}+C<br><br><br><br>E_{\mathrm{ab}}^{\mathrm{inc}},$

where a, b, and c are directions (x, y, and z), A, B, and C are coefficients based on material properties of a medium, and E_ab^inc is the incident field associated with the node.

13. A computer hardware configuration as recited in claim 11, wherein as the FFT is being calculated, primary fields, incident fields, and coefficients are being fetched by the system.

14. A computer hardware configuration as recited in claim 11, wherein the FFT and IFFT units are provided inside a field-programmable gate array (FPGA).

15. A computer hardware configuration as recited in claim 14, wherein the FFT and IFFT calculations are performed by a digital signal processing (DSP) chip.

16. A computer hardware configuration for performing the pseudo-spectral time-domain (PSTD) method on data, comprising:

a plurality of forward fast Fourier transform (FFT) units, each FFT unit calculating a forward fast Fourier transform (FFT) from the data;

a plurality of complex multiplication units, each complex multiplication unit receiving the FFT-processed data from a corresponding FFT unit and calculating a spatial derivative in the frequency domain from the FFT-processed data;

a plurality of inverse fast Fourier transform (IFFT) units, each IFFT unit converting the spatial derivative in the frequency domain from a corresponding complex multiplication unit into the time domain;

a plurality of computation engines, each computation engine solving a PSTD equation based upon the spatial derivative in the time domain received from a corresponding IFFT unit; and

a memory subsystem that provides input values to the FFT unit and receives output results from the computation engine,

wherein values for the FFT unit are stored in increasing-x, increasing-y and increasing-z patterns in the memory subsystem that allows a burst read of the memory subsystem to maximize throughput.

17. A computer hardware configuration as recited in claim 16, wherein the PSTD equation takes the form:

$E_{\mathrm{ab}}=A<br><br><br><br>E_{\mathrm{ab}}+B<br><br>\frac{\partial <br><br>H_c}{\partial <br><br>b}+C<br><br><br><br>E_{\mathrm{ab}}^{\mathrm{inc}},$

where a, b, and c are directions (x, y, and z), A, B, and C are coefficients based on material properties of a medium, and E_ab^inc is the incident field associated with the node.

18. A computer hardware configuration as recited in claim 16, wherein as the FFT is being calculated, primary fields, incident fields, and coefficients are being fetched by the system.

19. A computer hardware configuration as recited in claim 16, wherein the plurality of FFT and IFFT units are provided inside a field-programmable gate array (FPGA).

20. A computer hardware configuration as recited in claim 19, wherein the FFT and IFFT calculations are performed by a digital signal processing (DSP) chip.