Kuninobu Tanno

Parallel FFT algorithms using radix 4 butterfly computation on an eight-neighbor processor array

Abstract Fast Fourier transform (FFT), which has wide and variety application areas, requires very high speed computation. Since parallel processing of FFT is very attractive for high speed FFT computation, many processor arrays and multiprocessor systems have been proposed with efficient FFT algorithms. As a result of the recent development of VLSI technology, several massively parallel computers have been implemented on commercial basis. The MasPar, which is one of the SIMD type massively parallel computers, consists of an eight-neighbor processor array. This paper discusses parallel 1-D FFT algorithms on an eight-neighbor processor array. We propose three algorithms according to various data allocation methods. Then we estimate and evaluate their processing time. With the number of processors N = N r × N r , processing time is estimated to be 2( N r − 2) t c + ( log 2 N r ) t b , where t c is the communication time between neighbor processors, and t b is the execution time for the radix 4 butterfly computation. We also compare these algorithms with the conventional radix 2 FFT algorithm implemented on a mesh processor array. It is shown that the radix 4 FFT algorithms are faster than the radix 2 algorithms. These algorithms get high speed FFT computation by combining the radix 4 FFT algorithm with the characteristics of the eight-neighbor processor array.

Register insertion/self-token protocol for high speed ring LANs

A new ring access control scheme adopting a multiple-token protocol is investigated. In this protocol, each station distributed on a ring has a private token, called self-token, and a fixed-length register to carry out the register insertion protocol. The protocol shows good properties for shorter packets, becoming more superior to the token-ring and the FDDI-I protocols as the packet length decreases. Simulation results show that the protocol is more suitable for high-speed LANs than the Fiber Distributed Data Interface (FDDI) protocol.<<ETX>>

A fuzzy policing mechanism for still picture in ATM networks

The asynchronous transfer mode (ATM) technique has been accepted as a basis for the future B-ISDN networks. In ATM networks, all information is packetized and transferred in small packets of fixed length, called cells. The packetized information transfer, without flow control between the user and the network and the use of statistical multiplexing, results in the need of a policing mechanism to control the traffic parameters of each virtual connection in order to guarantee the required quality of service (QoS). The proposed parameters for monitoring the source traffic characteristics are the mean cell rate, the peak cell rate or the peak burst duration. Policing of the peak cell rate is generally not complex and can be achieved by using a cell spacer or other policing mechanisms (PMs). Monitoring the mean cell rate is more difficult. It is intended to improve the link utilization when it has to handle bursty traffic sources. We propose a policing mechanism called the fuzzy policing mechanism (FPM), by applying fuzzy set theory to police the mean cell rate of the still picture source in ATM networks. The performance evaluation via simulation shows that the FPM efficiently control the mean cell rate of the still picture source. The proposed FPM shows a good response behavior against parameter variations. The selectivity characteristics of the FPM approach the ideal characteristics required for a PM.

Parallel computing algorithm of neural networks on an eight-neighbor processor array

The authors describe a parallel computing algorithm to simulate the backpropagation (BP) model and Kohonens self-organizing feature map (SOFM) upon an eight-neighbor processor array. Taking account of the parallelism intrinsically found in neural networks, algorithms are presented which minimize the transmission overhead among processors, so that high-speed simulation of neural networks becomes feasible. The processing time required for one learning of BP or Kohonens SOFM for one input vector is estimated.<<ETX>>

Parallel 2-D FFT algorithms on an eight-neighbor processor array

The two-dimensional fast Fourier transform (2-D FFT), which has a wide application in areas such as digital signal processing, speech recognition and image processing, require quite high-speed computing to handle a huge amount of data. Many parallel hardware FFT algorithms have been proposed. The paper discusses 2-D FFT hardware algorithms which are implemented on an eight-neighbor processor array. They allow 2-D FFT computation to be processed by iterating a 2-D butterfly operation and a data transfer alternately. These algorithms combine effectively the direct 2-D FFT algorithm and the communication capability of the 8-neighbor processor array.<<ETX>>

Radix R parallel FFT algorithms with a global interconnection networks and its evaluation

In order to efficiently compute Fast Fourier transform (FFT) various parallel algorithms and their implementation to multiprocessors and multicomputers have been developed. In general, the local interconnection network is more high speed than a global one, but its capability depends on network architecture. On the other hand, the global interconnection network is not so high speed, but it does not depends on network architecture. It provides a flexible communication interface to the programmer. In this paper, we discuss parallel radix R FFT algorithms on a multiprocessor or multicomputer system with a global interconnection network. We propose two algorithms a stage-by-stage method and a multi-stage method. We also estimate the communication time. Then we show that the communication time is very sensitive to and affected by data exchange strategy. Finally, we implement these algorithms on two commercial massively parallel computers (nCUBE/2 and CM5) and measure the communication time.

Self-token protocol for high-speed multimedia LANs

The authors deal with a new channel access scheme for ring LANs, referred to as the self-token protocol. They describe the basic operation of the protocol, priority transmission and the packet error detection/recovery method. The performance of the protocol was evaluated by using computer simulation. It was shown to have superior characteristics by combining the merits of the multiple-token and the register insertion channel access methods. Asynchronous and synchronous communications and a bidirectional ring for realizing high-speed and multi-media-oriented LANs are discussed.<<ETX>>

A load adaptive bandwidth allocation method for high-speed LANs

Recently, Local Area Networks (LANs) have been a basic technology for information exchange between high-speed peripheral devices and multimedia communication facilities. Furthermore, they are considered a fundamental element (domain) of internetworking to construct high-tech information processing systems. Therefore, it is highly desirable to develop a new LAN protocol supporting super high-speed multimedia communications and preparing an easy interconnection methodology for LANs. In this paper, we propose an adaptive bandwidth allocation (ABA) method which promises a high bandwidth utilization rate and adequate fairness for dual ring LAN topology. The communication protocol using this method not only realizes a gigabit LAN but also can support the multimedia communications. Performance evaluation via simulations shows that the ABA protocol has better characteristics than the MetaRing method.

Performance Evaluation of the Fuzzy Policing Mechanism for Still Picture in ATM Networks

The Asynchronous Transfer Mode (ATM) technique has been accepted as a basis for future B-ISDN networks. In ATM networks, all information is packetized and transfered in small packets of fixed length, called cells. The packetized information transfer, without flow control between the user and the network and the use of statistical multiplexing, results in a need of a Policing Mechanism (PM) to control the traffic parameters of each virtual connection in order to guarantee the required Quality of Service (QoS). Policing of the peak cell rate is generally not complex. The control of the mean cell rate is more difficult, but is intended to improve the link utilization when it has to handle bursty traffic sources. Conventional PMs, such as the Leaky Bucket Mechanism (LBM) and the Window Mechanisms (WMs), are not well suited to the bursty nature of sources supported by ATM networks, therefore intelligent PMs are needed. In this paper, we propose a Fuzzy Policing Mechanism (FPM) to police the mean cell rate of the still picture source. We consider the case when the peak cell rate of the still picture source is not controlled separately by other PMs. The performance evaluation via simulations shows that the FPM efficiently controls the mean cell rate of the still picture source. The FPM has better responsiveness and selectivity characteristics than the LBM.

Buffer insertion/self-token (BIST) protocol for multimedia LANs

This paper describes a new multimedia LAN protocol, referred to as the buffer insertion/self-token (BIST) protocol, combining buffer insertion and the multiple-self-token ring protocols. The basic buffer insertion ring has the maximum spatial reuse on a ring by allowing stations to transmit packets concurrently. However, it is well known that the buffer insertion method suffers from the possibility of starvation. BIST prevents starvation and maintain fairness by assigning each station individual tokens, called self-tokens. Then a station is allowed to send packets only if it holds its own tokens, that is, each packet transmitted by the station is attached a self-token belonging to the station. That is why packet flow on a ring is regulated, and fairness is adjustable. Asynchronous and synchronous transmissions are supported by using two kinds of self-token, one is for asynchronous transmission and the other is for synchronous transmission. The effectiveness of the proposed protocol is demonstrated via simulation.