Orthogonal transform processor

Details Orthogonal transform processor Orthogonal transform processor

1999-01-27

2001-08-28

Ngo, Chuong Dinh (Department: 2121)

Electrical computers: arithmetic processing and calculating

Electrical digital calculating computer

Particular function performed

C708S402000

Reexamination Certificate

active

06282555

ABSTRACT:

BACKGROUND OF THE INVENTION
The present invention relates to an orthogonal transform processor suitably used for an image processing, a speech processing, etc.
For example, in a compressing and coding system for image data, a small-size orthogonal transform processor is required for transforming spatial domain signals into frequency domain signals. An encoder adopts a forward orthogonal transform such as a discrete cosine transform (hereinafter referred to as DCT) and a discrete sine transform (hereinafter referred to as DST). A decoder adopts an inverse orthogonal transform such as an inverse discrete cosine transform (hereinafter referred to as IDCT) and an inverse discrete sine transform (hereinafter referred to as IDST).
U.S. Pat. No. 4,791,598 discloses a two-dimensional DCT processor comprising two one-dimensional DCT circuits and a transposition memory interposed therebetween. Each of the one-dimensional DCT circuits adopts a so-called fast algorithm and a distributed arithmetic (DA) method, and comprises a butterfly operation circuit including plural adders and subtracters, and a distributed arithmetic circuit, disposed at the subsequent level, for obtaining vector inner products by using not a multiplier but a ROM (read only memory). The distributed arithmetic circuit includes plural ROM/accumulators (hereinafter referred to as RACs). Each of the RACs includes a ROM for storing, in a form of a look-up table, partial sums of the vector inner products based on a discrete cosine matrix, and an accumulator for obtaining a vector inner product corresponding to an input vector by adding, with the digits aligned, the partial sums successively retrieved from the ROM with bit slice words using as addresses. In a one-dimensional IDCT circuit, a butterfly operation circuit is disposed at a level subsequent to plural RACs included in a distributed arithmetic circuit.
In general, human visual sense is more insensitive to high frequency components than to low frequency components. Therefore, for the purpose of improvement in compression efficiency in a DVC (digital video cassette) and the like, a low frequency component in the one-dimensional DCT result is multiplied by a large weighting and a high frequency component is multiplied by a small weighting before coding. However, multiplication for such frequency-depending weighting of the one-dimensional DCT result requires an additional multiplier. U.S. Pat. No. 5,117,381 describes a one-dimensional eight-point DCT circuit for limited weighting. In this circuit, a multiplier for executing multiplication of input data by elements of a discrete cosine matrix is used also for giving the weighting.
The one-dimensional DCT circuit disclosed in U.S. Pat. No. 5,117,381 is disadvantageously limited to application to the eight-point DCT and cannot be applied to arbitrary weighting. Also, this one-dimensional DCT circuit has a problem that the circuit scale and power consumption are larger, owing to the usage of the multiplier, than those of the one-dimensional DCT circuit described in U.S. Pat. No. 4,791,598.
Furthermore, in dealing with an input data matrix comprising N×N elements in the two-dimensional DCT processor disclosed in U.S. Pat. No. 4,791,598, each of the two one-dimensional DCT circuits executes N-point one-dimensional DCT. In this case, each of the two one-dimensional DCT circuits includes N RACs. Specifically, this two-dimensional DCT processor requires to include 2N RACs in total, which disadvantageously increases the circuit scale and power consumption.
SUMMARY OF THE INVENTION
One object of this invention is providing an orthogonal transform processor which can give arbitrary weighting to the transform result without using any multiplier.
Another object is reducing the number of RACs included in a two-dimensional DCT processor.
According to the invention, for the purpose of achieving the aforementioned objects, with regard to an orthogonal transform, partial sums of vector inner products based on a constant matrix obtained by multiplying respective elements of an orthogonal transform matrix by frequency-depending weighting, and with regard to an inverse orthogonal transform, partial sums of vector inner products based on a constant matrix obtained by dividing respective elements of an inverse orthogonal transform matrix by frequency-depending weighting are stored in look-up tables in a distributed arithmetic circuit. By using the contents of the look-up tables, the transform with weighting given can be realized. In this invention, the orthogonal transform and weighting are simultaneously executed, and the inverse orthogonal transform and removal of weighting are simultaneously executed. Thus, there is no need to use a multiplier for executing multiplication of input data by the elements of the orthogonal matrix or the inverse orthogonal matrix as a result of adopting the DA method. In addition, there is no need to provide a multiplier for giving weighting and a divider for removing weighting. Moreover, the arbitrary weighting can be given to the transform result without restriction in the kind of transform and the transform point.
Furthermore, according to the invention, for example, two-dimensional DCT of 8×8 elements can be realized by using three one-dimensional DCT circuits each for executing four-point one-dimensional DCT. Specifically, a two-dimensional DCT processor for an input data matrix including N×N elements comprises a first circuit for generating a half of N elements resulting from a first N-point transform by successively executing N/2-point one-dimensional DCT on N input vectors each including N elements; a second circuit for generating a half of N elements resulting from a second N-point transform by successively executing N/2-point one-dimensional DCT on N transposition vectors each including N elements; a third circuit for generating another half of the N elements resulting from the first N-point transform by successively executing another N/2-point one-dimensional DCT on the N input vectors and for generating another half of the N elements resulting from the second N-point transform by successively executing another N/2-point one-dimensional DCT on the N transposition vectors; and a transposition memory for storing the N elements resulting from the first N-point transform generated by the first and third circuits and supplying the N transposition vectors to the second and third circuits, and the N elements resulting from the second N-point transform generated by the second and third circuits are output as a two-dimensional DCT result of the input data matrix. In this processor, the first N-point DCT is executed by the first and third circuits, and the second N-point DCT is executed by the second and third circuits. Each of the first, second and third circuits includes N/2 RACs. Accordingly, the number of the RACs included in the two-dimensional DCT processor can be decreased from 2N to 3N/2. According to the invention, a two-dimensional IDCT processor can be similarly configured.
Each of the first and second circuits includes one butterfly operation circuit, and the third circuit includes two butterfly operation circuits. Each distributed arithmetic circuit in the first, second and third circuits includes N/2 look-up tables. Partial sums each of two bits are retrieved from each look-up table of the first and second circuits, and partial sums each of four bits are retrieved from each look-up table of the third circuit. Thus, the processing speed of the third circuit can be twice as high as that of the first and second circuits. In this manner, the third circuit can be operated on a time-sharing basis.


REFERENCES:
patent: 4791598 (1988-12-01), Liou et al.
patent: 5117381 (1992-05-01), Juri et al.
patent: 5299025 (1994-03-01), Shirasawa
patent: 5610849 (1997-03-01), Huang
patent: 5654910 (1997-08-01), Iwata
patent: 5703799 (1997-12-01), Ohta
patent: 5805482 (1998-09-01), Phillips
patent: 366 435 A (1990-05-01), None
patent: 667 583 A (1995-08-01), None
patent: 3-35353 (1991-02-01), None

Affiliated with

Also associated with

Rating

Orthogonal transform processor does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Orthogonal transform processor, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Orthogonal transform processor will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2504222