Mitra SK, Kaiser JF. Handbook for digital signal processing. John Wiley & Sons; 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....processing. But the question is, what is this signal processing to consist of And how is it to be accomplished One idea is to make use of supervised, adaptive methods. Such methods are commonly used in areas such as artificial intelligence [6] machine learning [7] digital signal processing [8], statistics [9] data mining [10] neural networks [11] and genetic algorithms [12] They aim to produce a mechanism, formula or theory capable of generating appropriate outputs from arbitrary inputs. They do this using a reference set of input output examples (often called a training set) As a ....

Mitra, S. and Kaiser, J. (Eds.) (1993). Handbook for Digital Signal Processing. New York: Wiley. 15

....to its high time complexity and inability to locate the globe optimal solution. In the signal processing and detection field, a variety of techniques were developed to solve the NLS problem of the sinusoidal type without constraint. They are classified into direct and indirect approaches (refer to [20] for details) The direct one involves a multidimensional search of A, b and M for the minimization of E. Although the constraint can be easily incorporated, the direct approach suffers from its extremely high time complexity. The indirect one extracts , and M as accurately as possible without ....

....approach is computationally much more efficient but at the cost of suboptimal. Yet, none of these indirect methods can be applied here because of the constraint 0. Here we propose a heuristic algorithm which is composed of three steps. In the first step, we simply use the standard Prony method [20] to identify an intermediate solution (int, int) without the real nonnegative constraint on int. In the second step, the intermediate solution vector it is much expanded to p with it C p. The extra eigenval ues in Aex 2 are generated through interpolation of the eigen values in Ait. It is based ....

[Article contains additional citation context not shown here]

S. K. Mitra, and J. F. Kaiser, ed., Handbook for Digital Signal Processing,John Wiley & Sons, 1993.

....to date. The generic nonlinear programming techniques cannot be applied here due to its high time complexity and inability to locate the globe optimal solution. Here we propose a heuristic algorithm which is composed of three steps. In the first step, we simply use the standard Prony method [18] to identify an intermediate solution int int without the real nonnegative constraint on int In the second step, the intermediate solution vector int is expanded to a much larger set exp with int exp The extra eigenvalues in exp are generated through interpolation of the eigenvalues in int It is ....

....so far. The last step is to find exp of the given exp with the real nonnegative constraint exp which will minimize the difference between the original correlation sequence and the correlation function of exp exp This is achieved by the well known nonnegative least squares (NNLS) method [18]. Note that most power elements in exp will be zero since the order of the correlation function should not be significantly greater than the order of the correlation function int int Finally, the power vector consists of the nonzero power elements of exp the eigenvalue vector consists of those ....

S. K. Mitra and J. F. Kaiser, Eds., Handbook for Digital Signal Processing. New York: John Wiley, 1993.

....Fig. 1. Motivational example for demonstrating use of transformation for testability: fourth order parallel IIR filter. numerical stability properties (thereby requiring a short wordlength) the parallel structure is one of the most frequently used I1R filters, mainly in audio applications [10] [28]. It is assumed that each operation takes one clock cycle. The available time is six control cycles. The critical path is also six cycles long. To meet the available time of six clock cycles, the minimal resource allocation requires 3 multipliers (three multiplications have to be scheduled in the ....

S. K. Mitra and J. F. Kaiser, Handbook for Digital Signal Processing. New York: Wiley, 1993.

....Aided Design, November 1992 and in Proc. 1993 IEEE ICASSP International Conference on Acoustic, Speech, and Signal Processing, April 1993. page 1 of 38 1.0 Introduction 1. 1 Motivation and Goals Throughput is widely recognized as the single, most important parameter of state of the art designs [Mit93]. Iteration bound, control and data dependences impose fundamental limits on achievable performance Transformations are universally accepted as the most efficient way in overcoming these limitations [Fis88, Kun88, Rab91] However, until now the application of transformations for throughput ....

....efficient implementation for all computations for which maximally fast implementation is achievable, can be produced. The targeted classes include linear computations, polynomial filters and many types of adaptive filters. Note that this class of computation heavily dominates today DSP market [Opp89, Ahm90, Mit93]. The final goal is to build a basis and to give an impetus to the study of the efficient and effective ordering of transformations by studying the underlying synthesis mechanisms behind the proposed procedures. We also discuss how the new approach can be used for increasingly important design ....

[Article contains additional citation context not shown here]

S. K. Mitra, J.F. Kaiser: "Handbook for Digital Signal Processing", John Wiley, New York, NY, 1993.

....can be implemented for a throughput of 10 control steps using 1 multiplier, 1 adder, and 1 transfer unit. Interconnect requirements are shown in Table 1. The more regular version indeed results in less busses and muxes. Consider next the well known wave digital and ladder IIR filter structures [Mit93]. For filters of the same order, both have approximately the same size, except that the ladder structure is sometimes slightly larger. The interconnect requirements are shown in Table 2. The more orderly network of communications that is present in the ladder structure translates into a simpler ....

S. Mitra, J. Kaiser, Handbook of Digital Signal Processing, John Wiley, New York, NY, 1993.

....the literature on selection of features, whitening, fast convolution techniques, extensions, alternate techniques, or applications. The literature on these topics can be approached through introductory texts and handbooks # Current address: Interval Research, Palo Alto CA zilla computer.org [16, 7, 13] and recent papers such as [1, 19] Nevertheless, due to the variety of feature tracking schemes that have been advocated it may be necessary to establish that normalized cross correlation remains a viable choice for some if not all applications. This is done in section 3. In order to make the ....

.... additions, and the direct method is faster than the transform method. The transform method becomes relatively more efficient as N approaches M and with larger M,N . 4. 1 Fast Convolution There are several well known fast convolution algorithms that do not use transform domain computation [13]. These approaches fall into two categories: algorithms that trade multiplications for additional additions, and approaches that find a lower point on the O(N 2 ) characteristic of (one dimensional) convolution by embedding sections of a one dimensional convolution into separate dimensions of a ....

[Article contains additional citation context not shown here]

S. K. Mitra and J. F. Kaiser, Handbook for Digital Signal Processing, New York: Wiley, 1993.

....line spectrum components before performing any spectral noise smoothing because otherwise the lines would lose their sharpness [1] Also in speech and audio coders, an important task is to extract harmonic signals and to calculate masking thresholds for adaptive bit allocation. Many approaches [2] have been proposed for this task, e.g. timedomain Prony s method, subband modeling, least square fitting [3] and frequency domain methods. For the various methods the accuracy and the computational complexity differs significantly. The algorithm presented in [4] is based on the FFT and offers ....

S.K. Mitra and J.F. Kaiser, editors. Handbook for Digital Signal Processing. J. Wiley & Sons, 1993.

....the frequency response the transformation was applied. This allows to eliminate higher harmonics contents through FFT transformation. The further application of the inverse of the relationship above defined reports to the normal frequency range. For the same purpose a median filter was applied [4]. These operations result in a new frequency response with reduced complexity, while the perceptual contents is almost unchanged. Moreover an ideal equalizing system should have a dynamic range of about . This means that the insertion of such a device before the amplifier could overcome the ....

S.K.Mitra and J.F.Kaiser, Handbook of digital signal processing, McGraw-Hill, New--York, 1993.

....routines within the driver are the same for both DSP boards, the differences concern only the data formats and program codes, which are DSP specific. 4. BENCHMARK SUITE The algorithms used to analyze the performance of the system are described in the following paragraphs. Fast Fourier Transform [7] A one dimensional FFT was ran for different input sets from 64 to 2048 points. This input sets represent a sinus function to simplify the verification of the results. Fourier transform is a traditional part of complex media processing algorithms. Viterbi Decoder [8] A 1 3 rate convolutional ....

Mitra, S.K. and Kaiser, J.F. (ed.) Handbook for Digital Signal Processing. John Wiley & Sons, Inc., New York, 1993

....line spectrum components before performing any spectral noise smoothing because otherwise the lines would lose their sharpness [1] Also in speech and audio coders, an important task is to extract harmonic signals and to calculate masking thresholds for adaptive bit allocation. Many approaches [2] have been proposed for this task, e.g. timedomain Prony s method, subband modeling, least square fitting [3] and frequency domain methods. For the various methods the accuracy and the computational complexity differs significantly. The algorithm presented in [4] is based on the FFT and offers ....

S.K. Mitra and J.F. Kaiser, editors. Handbook for Digital Signal Processing. J. Wiley & Sons, 1993.

....[err( P . each specified P . D. Predictability Studies for Real Traffic Traces In this section, we do predictability analysis for some real network traces sampled and aggregated at different levels. A sliding window of size T c is used as the LPF for traffic sampling. Prony method [13], which works well with the absence of noise, is used to get the ARMA transfer function by matching the sampled traffic traces. Furthermore, oe 2 and MPI are derived from (4) and (7) respectively. Our analysis in this section will present 99 percentile traffic predictability since P is fixed ....

S. K. Mitra, and J. F. Kaiser, ed., Handbook for Digital Signal Processing, John Wiley & Sons, 1993.

....its high time complexity and inability to locate the global optimal solution. In the signal processing and detection field, a variety of techniques were developed to solve the NLS problem of the sinusoidal type without constraint. They are classified into direct and indirect approaches (refer to [15] for details) The direct one involves a multidimensional search of , and M for the minimization, which suffers from its extremely high time complexity. The indirect one extracts , and M as accurately as possible without explicit attempt to the minimization, such as Prony, MUSIC and ESPRIT ....

....here because of the constraint 0. l int l exp By NNLS l By Prony Expansion Figure 4: A heuristic algorithm to identify subject to 0. The heuristic algorithm developed in [4] is composed of three steps, as described in Fig. 4. In the first step, we simply use the standard Prony method [15] to identify an intermediate solution ( int ; int ) without the real nonnegative constraint on int . In the second step, the intermediate solution vector int is much expanded to exp with int ae exp . The extra eigenvalues in exp are generated through interpolation of the eigenvalues in ....

[Article contains additional citation context not shown here]

S. K. Mitra, and J. F. Kaiser, ed., Handbook for Digital Signal Processing, John Wiley & Sons, 1993.

....of Walsh code word m m m n bit data word exclusive or (mod 2 adder) reset m two in and m two in and m two in and PSfrag replacements A1cos(2 fct) A2sin(2 fc t) a1 (t) a2 (t) b1 (t) b2 (t) A1 A2 Figure 80: Hadamard matrix row generator, 2 m bits at a time. factor values equal to one [42]. Because all multiplications are by Sigma1, the required hardware software is very simple in comparison to that required for the DFT. Sorrenson and Burris write in [42] Since there are no twiddle factors, the complexity of the Hadamard transform, unlike that of the FFT, cannot be improved by ....

....a2 (t) b1 (t) b2 (t) A1 A2 Figure 80: Hadamard matrix row generator, 2 m bits at a time. factor values equal to one [42] Because all multiplications are by Sigma1, the required hardware software is very simple in comparison to that required for the DFT. Sorrenson and Burris write in [42] Since there are no twiddle factors, the complexity of the Hadamard transform, unlike that of the FFT, cannot be improved by using higher radix or split radix algorithms. It is not clear what Sorenson and Burris meant by complexity. While the computational efficiency of the FHT cannot be ....

[Article contains additional citation context not shown here]

S. Mitra and J. Kaiser, Handbook for Digital Signal Processing, Wiley, New York, 1993.

....First, in section 2, we will introduce the analyzed OFDM structure. The opportunity of FFT structures in OFDM systems is the main aspect of section 3. Here, we present some fundamentals about different FFT algorithms like the Radix 2 [2] the Split Radix and the Winograd FFT algorithm [3]. In section 4, we will define a fixed point model of an OFDM receiver. The derivation of the most important parameters will be shown by simulation results. The impact of quantization to the out of band radiation will be discussed in section 5. Compressed postscript files of our publications are ....

....a power of 2 (N = 2 m ; m2 IN ; m 1) ffl In a similar but more efficient way the Split Radix algorithm (SRFA) computes the DFT. Here, the DFT is divided into a Radix 2 and a Radix 4 part, so the efficiency of a Radix 4 algorithm and the flexibility of the R2FA can be exploited by the SRFA. In [3] the structure of this recursively executable algorithms is shown. For applying the SRFA, all FFT sequence lengths N = 2 m (m 2 IN ; m 2) are possible. ffl A usual unknown FFT representative is the Winograd FFT algorithm (WFTA) Since it is an advancement of the prime factor FFT algorithm ....

[Article contains additional citation context not shown here]

S. K. Mitra and J. F. Kaiser. Handbook for Digital Signal Processing. John Wiley, 1. edition, 1993.

....of the DFT outputs are required which is due to the fact that f n is real and hence the DFT output values are conjugate complex. The computation of realvalued FFT algorithms, especially for N=2 n , has been studied extensively in the literature an excellent survey of which can be found in [5] and [6]. We will concentrate exclusively on the case N=2 n and on the Cooley Tukey or decimation in time approach. This kind of FFT requires its input values to be in bitreversed order which is well suited for an efficient in place computation of the DCT, rendering a Fast Cosine Transform (FCT) The ....

Mitra, S.K. and Kaiser, J.F., Handbook for Digital Signal Processing, John Wiley&Sons, 1993.

....Suite and Experimental Results We considered a range of applications which usually serve as benchmarks for DSP vendors and extracted the algorithmic cores. Note that XC4013 devices consist of 576 cells each, and 113 of them (on average) are consumed by the host interface. Fast Fourier Transform [4]. A one dimensional FFT was ran for an input set of 512 points. We limited all computations to 16 bit wide operands due to the hardware available. In the case of FPGA the butterfly operation was implemented in hardware consuming 526 CLBs. LPC Synthesis Filter [6] Voice synthesis using the LPC10 ....

Mitra, S.K. and Kaiser, J.F. (ed.) Handbook for Digital Signal Processing. John Wiley & Sons, Inc., New York, 1993

....implemented on the FPGAs. The corresponding development environment was ran under Microsoft Windows 95. 3 Benchmark Suite The construction of the program test set was an easy choice. We simply considered some applications which usually serve as benchmarks for DSP vendors: Fast Fourier Transform [1]. A one dimensional FFT was ran for an input set of 512 points. We limited all the computations to 16 bit wide operands due to the hardware available. In the case of FPGAs the butterfly operation is implemented in hardware. Linear Prediction Codec [2] 4] Voice synthesis using the LPC10 ....

Mitra, S.K. and Kaiser, J.F. (ed.) Handbook for Digital Signal Processing. John Wiley & Sons, Inc., New York, 1993

.... = 2 like in W W x X L k L k k i i k k L L k , 1 2 e (32) 12 which have the advantage of being more efficient in terms of implementation. The normalized LMS algorithm (NLMS) is more robust than its unnormalized counterpart and shows an improved convergence behaviour [8]. 5.2 Improved versions of NLMS As has been elaborated above, the NLMS algorithm is based on the assumption R L,k =I, which is not justified for AEC applications. Yet the NLMS algorithm enjoys wide popularity due to its simplicity and robustness, although its convergence speed is unsatisfactory ....

....can be computed in O(L) computations [24] the memory requirement, however, increases with O L L log ( 2 1 6. Additionally L is constrained to be a power of two. A computationally more efficient way to evaluate the sliding DFT as well as DCT can be obtained by using the Goertzel algorithm [8]. Several authors have investigated this possibility [25] 26] 27] and the basic block diagram is sketched in fig. 16 for the DCT. A crucial ingredient in the recursive computation of the sliding transform is the damping factor a which prevents the poles of the recursive filters to lie on the ....

[Article contains additional citation context not shown here]

Mitra, S.K. and Kaiser, J.F., Handbook for Digital Signal Processing, John Wiley, 1993.

....response (IIR) whitening filter is much more complicated than calculating an all zero (FIR) filter (see Section 3) and it is equivalent to moving average spectral estimation. Here, we start from the FIR noise whitening filter given in eq. 9) now denoted by B(z) Using Prony s method (e.g. [15]) in a slight modification, B(z) is approximated by an all pole filter 1=C(z) which is the desired whitening filter H(z) This approach leads to an identical procedure as the approximated maximum likelihood moving average estimation given by Durbin (see [16] Let B(z) 1 P q k=1 b[k]z ....

Mitra, S. K.; Kaiser, J. F.: Handbook for digital signal processing. New York: John Wiley, 1993.

....w, the response of the system to signal x y is v w. Linear systems are the most widely studied type of systems due to their intrinsic conceptual simplicity. More importantly, linear system form by far the most dominant component of today VLSI ASIC and application specific programmable market [16]. For example, portable phone DSP functions are mainly linear computations [16] Linear systems are modeled and optimized using linear computations. Linear computations can be characterized as computations which use only additions subtractions and multiplications with constants. Note that this ....

....most widely studied type of systems due to their intrinsic conceptual simplicity. More importantly, linear system form by far the most dominant component of today VLSI ASIC and application specific programmable market [16] For example, portable phone DSP functions are mainly linear computations [16]. Linear systems are modeled and optimized using linear computations. Linear computations can be characterized as computations which use only additions subtractions and multiplications with constants. Note that this definition is not restricted to traditional arithmetic algebraic structures. What ....

S.K. Mitra, J.F. Kaiser: "Handbook for Digital Signal Processing", John Wiley & Sons, Inc., New York, NY, 1993

....With this formula one can calculate the filter response if the c and d coefficients are given. In order to design a filter, however, one must go the other way: The c s and d s are to be determined for a desired filter response. For further information on filter design, the reader is referred to [HP91,MK93]. Are The Evolved Programs Digital Filters When looking at the algorithm for digital filters, they appear similar to our GP approach for classifying speech data. In an IIR filter there is a feedback of the output. The N formerly produced outputs are being used to produce the current output. The ....

Sanjit K. Mitra and James F. Kaiser, editors. Handbook for digital signal processing. A Wiley-Interscience publication. Wiley, New York, 1993.

....to its high time complexity and inability to locate the globe optimal solution. In the signal processing and detection field, a variety of techniques were developed to solve the NLS problem of the sinusoidal type without constraint. They are classified into direct and indirect approaches (refer to [20] for details) The direct one involves a multidimensional search of , and M for the minimization of E 1 . Although the constraint can be easily incorporated, the direct approach suffers from its extremely high time complexity. The indirect one extracts , and M as accurately as possible ....

....approach is computationally much more efficient but at the cost of suboptimal. Yet, none of these indirect methods can be applied here because of the constraint 0. Here we propose a heuristic algorithm which is composed of three steps. In the first step, we simply use the standard Prony method [20] to identify an intermediate solution ( int ; int ) without the real nonnegative constraint on int . In the second step, the intermediate solution vector int 0 10 20 30 40 50 0 5 10 15 20 25 Frequency (Rad) Original Model 0 20 40 60 80 100 50 0 50 100 150 Frequency (Rad) ....

[Article contains additional citation context not shown here]

S. K. Mitra, and J. F. Kaiser, ed., Handbook for Digital Signal Processing,John Wiley & Sons, 1993.

No context found.

Mitra SK, Kaiser JF. Handbook for digital signal processing. John Wiley & Sons; 1993.

No context found.

S. K. Mitra and J. F. Kaiser. Handbook for Digital Signal Processing. New York: Wiley, 1993.

No context found.

S. K. Mitra, J. F. Kaiser, Handbook for Digital Signal Processing, John Wiley & Sons, 1993

No context found.

S.K. Mitra and J.F. Kaiser, "Handbook of digital signal processing", McGraw-Hill, New York, 1993.

No context found.

S.K. Mitra and J.F. Kaiser, "Handbook of digital signal processing", McGraw-Hill, New York, 1993.

No context found.

S. K. Mitra and J. F. Kaiser, eds., Handbook for digital signal processing. New York: John Wiley & Sons,

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC