Effects of architecture implementation on DFT algorithm performance

Mehalic, M.A.; Rustan, P. L.; Route, G.

doi:10.1109/tassp.1985.1164593

Cited by 22 publications

(6 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The reasons why certain algorithms perform better on certain computers were briefly explained in terms of four different instruction categories used. Further research is being conducted on the implementation issues of more FCT algorithms on many modem computers and DSPs by incorporating the correlation coefficients (Mechalic et al 1985) between the different instruction categories and the execution times. n 2 = N < < I ; for (k=1; k<m; k t e = PI/n2; n2=n2> > 1; n4=n2> >2:…”

Section: Discussionmentioning

confidence: 99%

Split-radix fast cosine transform algorithm

SltODRAS

Christopoulos

1993

International Journal of Electronics

View full text Add to dashboard Cite

An extension of an existing fast algorithm for the computation of the discrete cosine transform is presented. The proposed algorithm has a structure similar to the split-radix algorithm used for the computation of the FFT, the same computational complexity as the direct fast cosine algorithms and a high regularity which facilitates its implementation in VLSI technology. A comparison on different computer architectures shows that the proposed algorithm is superior to existing algorithms in terms of execution time.

show abstract

Section: Discussionmentioning

confidence: 99%

Split-radix fast cosine transform algorithm

SltODRAS

Christopoulos

1993

International Journal of Electronics

View full text Add to dashboard Cite

show abstract

“…Overheads like data movement, address generation and constants generation often significantly affect the speed of computation, especially in high-level language implementations. In fact, it has been shown that (Mehalic et al 1985) the eficiency of a fast algorithm is strongly affected by the architecture of the target machine. On the other hand, popular algorithms like the WFTA and the FFT required large sets of multiplication constants which are not practical to pre-compute and store as program parameters.…”

Section: Software Realizationmentioning

confidence: 99%

Realization of discrete Fourier transforms using a nesting algorithm

Siu¹,

Wong²

1991

International Journal of Electronics

View full text Add to dashboard Cite

A nesting discrete Fourier transform technique is proposed for computing discrete Fourier transforms. This technique relies on only two primitive modules and other modules are generated by a standard nesting procedure. In software realization, the speed of computation of this approach is comparable to the speed of computation of Winograd Fourier transform algorithms, whereas the program size of the present approach is smaller than that of WFTAs. This approach is most suitable for cases where there are restrictions on the memory size. For a hardware realization, two simple systolic cells are suggested for the realization of long DFTs using a pipeline systolic structure. This new architecture is most suitable for realization using VLSI techniques and requires significantly fewer devices compared to methods reported before. IntroductionRecently, numerous efforts have been devoted to the high performance realization of digital signal processing algorithms using parallel computer architectures. However, the performance of parallel computing systems is strongly affected by the complexity of inter-processor communication and the exploitation of concurrency. The key issue to obtain high efficiency is to keep a good balance among system control, data communication and computation complexity. Systolic array (Kung 1982) aim at making a good compromise among these factors while maintaining simple and highly regular architectures. This allows the huge computing capacities offered by contemporary VLSI technologies to be more readily utilized. Unfortunately, it is often difficult to map conventional fast algorithms to efficient parallel computing hardware structures. Problems usually arise from complicated dependency structures. A commonly used approach is to parallelize a high level operation directly from its basic definition. The speed-up is then achieved by employing multiple functional units. This results in increased hardware cost that is often not justifiable when compared to the drastic reduction in computation complexity of fast algorithms, especially in application specific systems.The discrete Fourier transform (DFT) is one of the most important tools in modern digital signal processing applications. Efficient algorithms for the realization of the discrete Fourier transform are desirable and critical to the success of many systems. The fast Fourier transform (FFT), which allows a tremendous saving in the calculation of discrete Fourier transforms was introduced by Cooley and Tukey (1965). The number of multiplications is still relatively large owing to the numerous twiddle factors required. There is a further penalty that either a large storage memory or a time consuming run-time evaluation must be used to supply these twiddle factors in actual implementation. On the other hand, the global data communication nature of its butterfly structure imposes many

show abstract

“…The effects of the type of architecture on the performance of the FFT are studied in [3]. Computer architectures are divided into three classes: floating-point processors, data transfer processors and vector processors.…”

Section: Theoretical Performancementioning

confidence: 99%

“…Results obtained by [3] showed that the WFTA performs the best on a vector processor because of its matrix structure. The floating point processor is the least suited for the WFTA because of the large number of data transfers necessary to perform the WFTA The radix-2 FFT algorithm performs the best on the floating-point processor because it have very few data transfers to make.…”

Section: Theoretical Performancementioning

confidence: 99%

“…A theoretical performance evaluation is given in section 4 on the major FFI' algorithms according to their arithmetic complexity. When using the CFM any one of the conditions, but not both, in (3) can be used to compute (2), while the PFM allows one to choose both conditions in (3) to simplilj (2). Two types of algorithms are formed using these maps, the prime factor algorithms (PFA) and the common factor algorithms (CFA).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation