Processing device and method for performing a stage of a Fast Fourier Transform
View Patent ↗A data processing device and a method for performing second or next stage of an N point Fast Fourier Transform is suggested. The processing device comprises an input operand memory unit and an input buffer comprising a plurality of addressable memory cells arranged in lines and columns. Furthermore, the device comprises a number of radix-P operation units for producing output operands that are buffered in an output buffer. Input operands are read from the input operand memory unit and buffering into the input buffer. The input operands are stored and fetched from the input buffer according to a reordering scheme that allows efficient parallel processing of the operands by the butterflies and the buffering of subsequent input operands.
1. A data processing device for performing a stage of an N point Fast Fourier Transform, the stage comprising computing N output operands on the basis of N input operands by applying a set of N/P radix-P butterflies to the N input operands, with N being a positive integer and P being a value equal to 2 or 4, wherein the data processing device comprises:
an input operand memory unit arranged to store a plurality of input operands addressable in blocks of K operands, with K being a positive integer;
an input buffer comprising a plurality of addressable memory cells arranged in lines and columns;
K/P radix-P operation units for calculating the N/P radix-P butterflies, each operation unit being connected to the input buffer;
a logic circuit arranged to control the input operand memory unit and the input buffer according to an addressing scheme so as to perform the following actions:
read P subsequent blocks of K input operands from the input operand memory unit;
buffer the P subsequent blocks into P subsequent lines of the input buffer;
transfer K column oriented input operands from K/P subsequent columns of the input buffer to the radix-P operation units for processing by the radix-P operation units;
repeat transferring of the K column oriented input operands from the K/P subsequent columns of the input buffer to the radix-P operation units for processing until K of the columns of the input buffer are transferred and processed;
read P further subsequent blocks of K input operands from the input operand memory unit;
buffer the P further subsequent blocks into P subsequent columns of the input buffer;
transfer K line oriented input operands from K/P subsequent lines of the input buffer to the radix-P operation units for processing by the radix-P operation units; and
repeat transferring of the K line oriented input operands from the K/P subsequent lines of the input buffer to the radix-P operation units for processing until K of the lines of the input buffer are transferred and processed;
wherein at least two actions are performed in parallel.
2. The device according to claim 1 , wherein the logic circuit is arranged to perform the following two actions in a single clock cycle:
read a P th block of K input operands from the input operand memory unit and buffer the P th block into a P th line of the input buffer and
transfer K input operands from the first K/P columns of the input buffer to the radix-P operation units.
3. The device according to claim 1 wherein the logic circuit is arranged to perform the following two actions in a single clock cycle:
transfer K input operands from K/P columns of the input buffer subsequent to the first K/P columns, to the radix-P operation units, and
read a (P+1) th block of K input operands from the input operand memory unit and buffer the (P+) th block into a first column of the input buffer.
4. The device according to claim 1 , wherein the logic circuit is arranged to perform the following two actions in a single clock cycle:
read a (2×P) th block of K input operands from the input operand memory unit and buffer the (2×P) th block of K input operands into a P th column of the input buffer;
transfer K input operands from the first K/P lines of the input buffer.
5. The device according to claim 1 , wherein the device comprises:
an output operand memory unit arranged to store a plurality of output operands;
an output buffer comprising a plurality of addressable memory cells arranged in lines and columns;
a further logic circuit arranged to control the output operand memory unit and the output buffer so as to:
buffer operands processed by the radix-P operation units, into the output buffer;
write the processed operands from the output buffer into the output operand memory unit.
6. The device according to claim 5 , wherein the further logic circuit is arranged to address the output buffer according to the addressing scheme used by the logic circuit for addressing the input buffer.
7. The device according to claim 1 , wherein the input operand memory unit is a random-access memory unit.
8. The device according to claim 1 , wherein the input buffer comprises a set of K 2 -(K-4) 2 individually addressable buffer cells, each cell being capable of buffering one input operand.
9. The device according to claim 1 , wherein K is a multiple of 4.
10. The device according to claim 1 , wherein K is 8.
11. A method for performing a stage of an N point Fast Fourier Transform, wherein each stage comprises computing N output operands on the basis of N input operands by applying a set of N/P radix-P butterflies to the N input operands, with N being a positive integer and P being a value equal to 2 or 4, wherein a logic circuit is arranged to control an input operand memory unit and an input buffer to perform the method comprising:
reading P subsequent blocks of K input operands from the input operand memory unit, with K being a positive integer;
buffering the P subsequent blocks into P subsequent lines of the input buffer having a plurality of addressable memory cells arranged in lines and columns;
transferring K column oriented input operands from K/P subsequent columns of the input buffer to K/P radix-P operation units for calculating the N/P radix-P butterflies;
processing the K column oriented input operands in radix-P operation units;
repeating the transferring of K column oriented input operands from the K/P subsequent columns of the input buffer and the processing of the K column oriented input operands in the radix-P operation units until K of the columns of the input buffer are transferred and processed;
reading P further subsequent blocks of K input operands from the input operand memory unit;
buffering the P further subsequent blocks into P subsequent columns of the input buffer;
transferring K line oriented input operands from K/P subsequent lines of the input buffer to the radix-P operation units;
processing the K line oriented input operands in the radix-P operation units; and
repeating the transferring of K line oriented input operands from the K/P subsequent lines of the input buffer and processing the K line oriented input operands until K of the lines of the input buffer are addressed and processed;
wherein at least two actions are performed in parallel.
12. The method according to claim 11 , wherein the following two actions are performed in a single clock cycle:
reading a P th block of K input operands from the input operand memory unit and buffering the P th block into a P th line of the input buffer and
transferring K input operands from the first K/P columns of the input buffer to the radix-P operation units.
13. The method according to claim 11 , wherein the following two actions are performed in a single clock cycle:
transferring K input operands from K/P columns of the input buffer subsequent to the first K/P columns, to the radix-P operation units, and
reading a (P+1) th block of K input operands from the input operand memory unit and buffering the (P+1) th block into a first column of the input buffer.
14. The method according to claim 11 , wherein the following two actions are performed in a single clock cycle:
reading a (2×P) th block of K input operands from the input operand memory unit and buffering the (2×P) th block of K input operands into a P th column of the input buffer;
transferring K input operands from the first K/P lines of the input buffer.
15. The method according to claim 11 , wherein the method comprises:
buffering operands processed by the radix-P operation units, into an output buffer having a plurality of addressable memory cells arranged in lines and columns;
writing processed operands from the output buffer into an output operand memory unit.
16. The method according to claim 11 , wherein the input operand memory unit is a random-access memory unit.
17. The method according to claim 11 , wherein the input buffer comprises a set of K 2 -(K-4) 2 individually addressable buffer cells, each cell being capable of buffering one input operand.
18. The method according to an claim 11 , wherein K is a multiple of 4.
19. The method according to claim 11 wherein K is 8.