Several important problems in signalprocessing, such as linear prediction, linear regression, or spectrum factorization, need close-to-Toeplitz matrices to be factored. To solve these problems, several fast algorithm...
详细信息
ISBN:
(纸本)0819406945
Several important problems in signalprocessing, such as linear prediction, linear regression, or spectrum factorization, need close-to-Toeplitz matrices to be factored. To solve these problems, several fast algorithms have been derived. They differ by the kind of adaptivity (block processing or exponential weighting of the data) and by the kind of recursion (order or time recursion), but they all have a common vector ladder recursion, involving an hyperbolic rotation. These algorithms are therefore well suited for implementation on an array processor. But there exists a number of applications where an efficient parallel implementation of these algorithms on a DSP network would be very attractive. The redundancy suppression, which is performed in the equations to get a fast algorithm, destroys the processing regularity of the corresponding standard algorithms, which prevents efficient high level parallel implementation. Auxiliary quantities, such as generalized reflections coefficients, are introduced that don't have the same dimensions as primary quantities. As a result there is a loss of efficiency in such tasks processing, and this leads to use array partitioning in as many sub-arrays as the number of DSPs available. If the number m of DSPs is low compared to primary quantities of dimension p (m << p/m) (i.e. a low level parallelism) and if the dimension a of reflection coefficients is also low ((alpha) << p/m), the global efficiency of a parallel implementation on a DSP network may still be interesting, with in addition, the advantages associated to such a network : simple design, simple control. To illustrate this, an application of the method to a Fast Recursive Least Squares algorithm is presented.
Media signalprocessing requires high computing power and the algorithms exhibit a great deal of parallelism on low precision data. The basic components of multi-media objects are usually simple integers with 8, 12, o...
详细信息
ISBN:
(纸本)0819437611
Media signalprocessing requires high computing power and the algorithms exhibit a great deal of parallelism on low precision data. The basic components of multi-media objects are usually simple integers with 8, 12, or 16 bits of precision. In order to support efficient processing of media signals, Instructions Set Architecture (ISA) of the traditional processors requires modifications. In this paper, we present the quantitative analysis and the computational complexity required to perform media processing. Main classes of instructions that are needed for the required level of performance of the Media Processor are identified. Their efficient implementation and effect on the processor data-path is discussed. The main operations required in media processing are Addition (with or without saturation), Multiplication (with or without rounding), Sum of Products, and Average of two numbers.
Information processing theory aims to quantify, how well signals encode information and how well systems process information. Time-frequency distributions have been used to represent the energy distribution of time-va...
详细信息
ISBN:
(纸本)0819454974
Information processing theory aims to quantify, how well signals encode information and how well systems process information. Time-frequency distributions have been used to represent the energy distribution of time-varying signals for the past twenty years. There has been a lot of research on various properties of these representations. However, there is a general lack of quantitative analysis in describing the amount of information encoded into a time-frequency distribution. This paper aims to quantify how well time-frequency distributions represent information by using information-theoretic distance measures. Different distance measures, such as Kullback-Leibler distance, Renyi distance, will be adapted to the time-frequency plane. Their performance in quantifying the information in a given signal will be compared. A sensitivity analysis for different distance measures will be carried out to assess their robustness under perturbation. Different example signals will be considered for illustrating the information processing in time-frequency distributions.
This is a review of bootstrap methods, concentrating on basic ideas and applications. It begins with an exposition of the bootstrap principle and gives several examples of its use. Bootstrap methods for testing statis...
详细信息
ISBN:
(纸本)0819412767
This is a review of bootstrap methods, concentrating on basic ideas and applications. It begins with an exposition of the bootstrap principle and gives several examples of its use. Bootstrap methods for testing statistical hypotheses are then reviewed and an analysis of accuracy of bootstrap tests is provided. We discuss how the bootstrap can be used to estimate variance stabilizing transformations that are crucial for the level of accuracy of bootstrap tests. Finally, we describe an application of multiple bootstrap tests to the problem of finding optimum locations of vibration sensors for knock detection in spark ignition engines.
In this paper we introduce a. multi-scale deconvolution technique performed in the scale-domain. In sensor array applications such as in radar, sonar and seismic processing, the sensor outputs are modeled as the convo...
详细信息
ISBN:
(纸本)0819432938
In this paper we introduce a. multi-scale deconvolution technique performed in the scale-domain. In sensor array applications such as in radar, sonar and seismic processing, the sensor outputs are modeled as the convolution of the unknown source signal with various unknown system impulse responses that are scaled versions of each other with unknown scale parameters. In many applications these signals or the scaling parameters are needed to be estimated only from the sensor outputs. In our earlier work, we estimated the unknown scale parameters by using properties of the scale transform and then employed existing deconvolution algorithms. Here, we derive the multiscale blind deconvolution algorithm in the scale transform domain. The performance of the method is illustrated using simulation examples.
This paper considers ''blind beamforming'' operations on a wireless network of randomly distributed MEM sensors. Maximum power collection criterion is proposed and results in array weights obtained fro...
详细信息
ISBN:
(纸本)0819425842
This paper considers ''blind beamforming'' operations on a wireless network of randomly distributed MEM sensors. Maximum power collection criterion is proposed and results in array weights obtained from the eigenvector corresponding to largest eigenvalue of a matrix eigenvalue problem. Theoretical justification of this approach to an extension of Szego's asymptotic distribution of eigenvalues is provided. Numerical results on propagation time delay estimation and loss of coherency due to propagation disturbances are presented.
In this paper, under the assumption that noise correlation is spatially limited, using two separated arrays, we propose a new parametric approach for consistent directions of arrival (DOA) estimations in unknown noise...
详细信息
ISBN:
(纸本)081940943X
In this paper, under the assumption that noise correlation is spatially limited, using two separated arrays, we propose a new parametric approach for consistent directions of arrival (DOA) estimations in unknown noise environments. The theoretical performance analysis of the proposed DOA estimations is also presented. With the use of the theoretical performance, the best weighting matrices of the parametric criteria have been derived. More significantly, it has been shown that within the best weighted criteria, using canonical decomposition, we can achieve optimal performance of the DOA estimation among a large set of eigendecompositions.
Higher-order Wigner distributions are not unique: definitions differ in the lag separations between the terms used in the time-domain product, as well as in how many of the terms are conjugated. We study a class of th...
详细信息
ISBN:
(纸本)081940943X
Higher-order Wigner distributions are not unique: definitions differ in the lag separations between the terms used in the time-domain product, as well as in how many of the terms are conjugated. We study a class of third-order WDs (TWD), parameterized by a single parameter (alpha) , and show that there is a duality between the choices of (alpha) equals -1/3 (Gerr's definition) and (alpha) equals -1. Interesting signal attributes, such as the instantaneous frequency, the derivative of the log-magnitude, and the group delay can be recovered from the TWD. Important issues such as aliasing problems and sampling requirements, and whether or not the analytic form of a real signal should be used, are addressed. It is shown theoretically that the TWD with (alpha) equals -1 is particularly useful for the detection of transients in the presence of colored Gaussian noise.
In a very recent past, new techniques, referred to as time-scale methods and making use of the so-called wavelet transform, have been proposed for the analysis of nonstationary or time-varying signals. They are basica...
详细信息
ISBN:
(纸本)0819404098
In a very recent past, new techniques, referred to as time-scale methods and making use of the so-called wavelet transform, have been proposed for the analysis of nonstationary or time-varying signals. They are basically devoted to the description of signal time evolutions at different observation scales; this is achieved by using shifted and dilated versions of some elementary analyzing waveform along the time axis. The purpose of this paper is twofold: it is intended (1) to provide a brief overview of linear wavelet techniques (continuous and discrete transforms) and bilinear time-scale methods (time-scale energy distributions), and (2) to put them in some common perspective with existing signalprocessing tools (Gabor decompositions, constant-Q analysis, quadrature mirror filters, wideband ambiguity functions, time-frequency energy distributions). Existing or potentially relevant applications are also pointed out.
A new high-performance systolic architecture for calculating the discrete Fourier transform (DFT) is described which is based on two levels of transform factorization. One level uses an index remapping that converts t...
详细信息
A new high-performance systolic architecture for calculating the discrete Fourier transform (DFT) is described which is based on two levels of transform factorization. One level uses an index remapping that converts the direct transform into structured sets of arithmetically simple four-point transforms. Another level adds a row/column decomposition of the DFT. The architecture supports transform lengths that are not powers of two or based on products of coprime numbers. Compared to previous systolic implementations, the architecture is computationally more efficient and uses less hardware. It provides low latency as well as high throughput, and can do both one- and two-dimensional DFTs. An automated computer-aided design tool was used to find latency and throughput optimal designs that matched the target field programmable gate array structure and functionality.
暂无评论