Le traitement du signal vocal voice signal processing

The speech processing studies have advanced rapidly in recent years spurred on by great progresses in thevlsi technologies and in the digitalization of the networks. This paper offers an overview of the most attractive techniques which have focused the recent researchs and developments in speech coding, recognition and synthesis areas. For speech compression, the emphasis is put on a family of techniques named code-excited linear prediction (celp) which dominates current studies for rates in the range of 4 to 16 kbit/s. In terms of speech recognition, particular emphasis is placed on the following three elements which are essential in order to increase the robustness of the systems : telephone line adaptation, rejection of parasite noise and out-of-vocabulary words, and keyword spotting. In terms of text-to-speech synthesis, thepsola (pitch synchronous overlap and add) technique is outlined herein. This technique gives rise to a new generation of synthesis systems which produce speech with very natural timbre. The analysis of current tendencies for each area allows to suggest attractive directions for future research.     

数字信号处理及DSP芯片的发展前景

简要介绍了DSP的设计流程及其实现方法,着重介绍了DSPs芯片结构特点、运算速度、应用与市场,并展望了DSPs芯片的发展前景。     

Multiple digital signal processor environment for intelligent signal processing

We describe a novel, expandable, multiple digital signal processor (DSP) architecture with a symbolic processing host. A multiprocessor board, called Odyssey, based on this architecture has been developed to combine symbolic and real-time digital signal processing in a single computing environment. Some of the key features of the board are: 20 million multiply/accumulates per second, 512K bytes of data space, and expandability to 16 boards on a NuBus host. The DSPs used are the TMS32020 signal processing chips developed by Texas Instruments, and the host is Texas Instruments' Explorer, a LISP machine workstation. This provides environment to perform many intelligent signal processing tasks by associating meaningful relationships between quantitative (signal processing) and qualitative (symbolic processing) entities to develop inferences using expert system technology. Applications such as grammar-driven connected speech recognition, neural network simulation, EEG analysis, and generation of speech from general English text with natural language processing are some of the tasks that can utilize the computational power of the multiple DSP and/or the associated symbolic processing capabilities. Software development tools to implement applications include the device driver to facilitate communication between the host processor and the Odyssey board, a unique window-based debugger resident on the Explorer that allows for simultaneous state display of all the processors on the board, a FORTH interpreter for high-level language programming, and a cross-assembler/linker for assembly level programming.     

The monogenic signal

This paper introduces a two-dimensional (2-D) generalization of the analytic signal. This novel approach is based on the Riesz transform, which is used instead of the Hilbert transform. The combination of a 2-D signal with the Riesz transformed one yields a sophisticated 2-D analytic signal: the monogenic signal. The approach is derived analytically from irrotational and solenoidal vector fields. Based on local amplitude and local phase, an appropriate local signal representation that preserves the split of identity, i.e., the invariance-equivariance property of signal decomposition, is presented. This is one of the central properties of the one-dimensional (1-D) analytic signal that decomposes a signal into structural and energetic information. We show that further properties of the analytic signal concerning symmetry, energy, allpass transfer function, and orthogonality are also preserved, and we compare this with the behavior of other approaches for a 2-D analytic signal. As a central topic of this paper, a geometric phase interpretation that is based on the relation between the 1-D analytic signal and the 2-D monogenic signal established by the Radon (1986) transform is introduced. Possible applications of this relationship are sketched, and references to other applications of the monogenic signal are given     

基于信号重构和最小二乘的线性调频信号估计

通过多延时离散多项相位变换重构出具有较高信噪比的信号,随后解出瞬时相位曲线,利用最小二乘法估计出线性调频信号的一阶、二阶相位参数.在低信噪比时,估计量均方误差仍达到CRB.另外,该方法信噪比门限低且可调,运算量小,易于实现.仿真验证了其有效性.     

Digital signal processing

Markets have always influenced the central thrust of the semiconductor industry. Beginning in the early eighties, the personal computer (PC) market has been the dominant market influencing the semiconductor industry. Single-chip microprocessors (MPUs) enabled what became the huge PC market, which ultimately overshadowed the earlier minicomputer and mainframe computer markets. The popularity of PCs led to investments in increasingly more powerful MPUs and memory chips of ever-growing capacity. MPUs and DRAMs became the semiconductor industry technology drivers for the data processing needs of the PC. But now, DSP, as opposed to conventional data processing, has become the major technology driver for the semiconductor industry as evidenced by its market growth and the fervour of chip vendors to provide new products based on DSP technology. The increasing need to digitally process analog information signals, like audio and video, is causing a major shift in the semiconductor business. Since DSP is the mathematical manipulation of those digitized information signals, specialized math circuitry is required for efficient signal processing-circuitry that was previously confined to classical DSP chips     

The constrained signal detector

The problem of detecting a material-of-interest in a hyperspectral image is considered. Knowledge of the background materials in the image is assumed. It is also assumed that the stochastic noise in the system has a Gaussian distribution with a known covariance matrix. Using these assumptions, along with the requirement that the material abundances in the pixel must sum to one, a filter called the constrained signal detector (CSD) is derived. The CSD is a variation of the generalized likelihood ratio test (GLRT). Where the GLRT uses maximum-likelihood estimates (MLEs) of the noise in the received signal, the CSD uses constrained least squares (CLS) noise estimates. It will be shown that the CSD is actually a scaling of the CLS target abundance estimate which has been derived elsewhere. However, the CSD computes that estimate much more efficiently then existing methods do. It is proved that the CSD outperforms the orthogonal subspace projection (OSP) detector and that the CSD is the optimal detector when there is only one background material present.     

Dense target signal processing

The problem of determining the density of targets at every range and velocity is addressed. These targets consist of a dense group of reflecting objects moving with different velocities and at different ranges. The problem of how to choose the outgoing signals and process the echoes of those signals from the targets to determine the density function is discussed. The problem is a classical inverse problem. The objective is to reconstruct a function of two variables (range and velocity) from limited information. Two schemes are given. The first of these methods modifies the method of J.R. Klauder et al. (1960) to the case of signals with a large range of frequency components (wideband signals). The second is an improvement on the first that uses a formula of I. Khalil (1974) from affine group theory. Numerical solutions support the conclusions. The method has applications in echocardiography, radar, sonar and fluid flow measurement     

Quantum signal processing

In this article we present a signal processing framework that we refer to as quantum signal processing (QSP) (Eldar 2001) that is aimed at developing new or modifying existing signal processing algorithms by borrowing from the principles of quantum mechanics and some of its interesting axioms and constraints. However, in contrast to such fields as quantum computing and quantum information theory, it does not inherently depend on the physics associated with quantum mechanics. Consequently, in developing the QSP framework we are free to impose quantum mechanical constraints that we find useful and to avoid those that are not. This framework provides a unifying conceptual structure for a variety of traditional processing techniques and a precise mathematical setting for developing generalizations and extensions of algorithms, leading to a potentially useful paradigm for signal processing with applications in areas including frame theory, quantization and sampling methods, detection, parameter estimation, covariance shaping, and multiuser wireless communication systems. We present a general overview of the key elements in quantum physics that provide the basis for the QSP framework and an indication of the key results that have so far been developed within this framework. In the remainder of the article, we elaborate on the various elements in this figure.     

Bilinear signal synthesis

The authors discuss the signal synthesis problem in the general framework of bilinear signal representations (BSRs), thereby obtaining a unified treatment which encompasses, e.g., the Wigner distribution and ambiguity function as special cases. The inclusion of a signal space constraint serves to impart flexibility to the signal synthesis process and to relax mathematical requirements. The characterization of signal spaces either by orthogonal projection operators or by orthonormal bases leads to two different signal synthesis methods. Both methods assume the BSR to be unitary (i.e., satisfy Moyal's formula) on the signal space on which signal synthesis is performed. As an application of the general signal synthesis methods, band-limited signal synthesis in the case of the Wigner distribution is considered     

Seismic signal processing

Seismic prospecting for oil and gas has undergone a digital revolution during the past decade. Most stages of the exploration process have been affected: the acquisition of data, the reduction of this data in preparation for signal processing, the design of digital filters to detect primary echoes (reflections) from buried interfaces, and the development of technology to extract from these detected signals information on the geometry and physical properties of the subsurface. The seismic reflection is genenlly weak, and it must be strengthened by the use of signal summing (stacking) procedures. The determination of depths to a target horizon requires knowledge of the propagational velocities of seismic stress waves, and a wealth of technology has evolved for this purpose. More recently, it has been possible to relate signal amplitude to the physical properties of the medium traversed and, in particular, to make inferences about the oil and gas content of the buried rocks. Much of the exploration effort occurs in offshore areas, where reverberations in the water layer mask reflections from below. The method of predictive deconvolution has been most effective in its ability to attenuate these reverberations, making it possible to detect reflections from structures at depth. Seismic signal processing is neither pure science nor pure art, and offers a continuing challenge to the practitioners of both cultures.     

Low-power MIMO signal processing

In this paper, we present a new adaptive error-cancellation (AEC) technique, denoted as multi-input-multi-output (MIMO)-AEC, for the design of low-power MIMO signal processing systems. The MIMO-AEC technique builds on the previously proposed AEC technique by employing an algorithm transformation denoted as MIMO decorrelating (MIMO-DECOR) transform. MIMO-DECOR reduces complexity by exploiting correlations inherent in MIMO systems, thereby improving the energy efficiency of AEC. The proposed MIMO-AEC enables energy minimization of MIMO systems by correcting transient/soft errors that arise in very large scale integration signal processing implementations due to inherent process nonidealities and/or aggressive low-power design styles, such as voltage overscaling. We employ the MIMO-AEC in the design of a low-power Gigabit Ethernet 1000Base-T device. Simulation results indicate 69.1%-64.2% energy savings over optimally voltage-scaled present-day systems with no loss in algorithmic performance.     

Baseband signal processing

The baseband signal processing of the ALTAIR wireless in-building network (WIN) is described. The discussion covers the 19-GHz oscillator, burst processing, packet detection, symbol clock synchronization and gain and offset correction. The algorithms described are carried out in a single ASIC composed of 60000 active gates. The implemented procedures allow for parallel processing, which significantly reduces the computation time and therefore leads to preserving high bandwidth efficiency. The learning processes that acquire information about packet parameters and the adjustment operations in the receivers are executed in 3 μs     

Soft digital signal processing

In this paper, we propose a framework for low-energy digital signal processing (DSP), where the supply voltage is scaled beyond the critical voltage imposed by the requirement to match the critical path delay to the throughput. This deliberate introduction of input-dependent errors leads to degradation in the algorithmic performance, which is compensated for via algorithmic noise-tolerance (ANT) schemes. The resulting setup that comprises of the DSP architecture operating at subcritical voltage and the error control scheme is referred to as soft DSP. The effectiveness of the proposed scheme is enhanced when arithmetic units with a higher "delay imbalance" are employed. A prediction-based error-control scheme is proposed to enhance the performance of the filtering algorithm in the presence of errors due to soft computations. For a frequency selective filter, it is shown that the proposed scheme provides 60-81% reduction in energy dissipation for filter bandwidths up to 0.5 π (where 2 π corresponds to the sampling frequency f_s) over that achieved via conventional architecture and voltage scaling, with a maximum of 0.5-dB degradation in the output signal-to-noise ratio (SNR_o). It is also shown that the proposed algorithmic noise-tolerance schemes can also be used to improve the performance of DSP algorithms in presence of bit-error rates of up to 10^-3 due to deep submicron (DSM) noise     

Generalized likelihood signal resolution

This paper defines anM-ary generalized likelihood ratio test (MGLRT) that overcomes Root's early objection to the application of generalized likelihood ratio testing to the resolution of correlated signals. The proposed test extends the form of a conventional binary generalized likelihood ratio test (GLRT) in a manner that permits a generalization of the minimax properties of the binary test to theM- hypotheses case. When the estimated signals are orthogonal, the test reduces to a sequence of conventional binary tests duplicating the performance of a narrow-band matched filter envelope-detector receiver.     

Aperture-domain signal processing

An elementary reconsideration of first principles in aperture-domain processing of wave phenomena for reception (and, by reciprocity, for transmission) can yield revealing, and in some cases novel, insights into what can or cannot be achieved. The subjects covered here include direction-selective reception, superdirectivity, direction finding, focused near-field reception, circular arrays and circular modes in such arrays, and the new concepts of arrays composed of `random symmetrical pairs', and of real gain in omnidirectional receiving antennas. The ideas are basic to all wave directional analysis and imagining applications, be they electromagnetic or acoustic, in radar or sonar, communications, navigation, surveillance or medical imaging     

Subsurface radar signal deconvolution

We present two methods of signal deconvolution for systems whose impulse response (wavelet function) can be explicitly determined, and where the goal is to locate short impulses in the presence of strong, reverberation-like interferences.The first method, which we call algebraic deconvolution, differs from other known techniques in two ways: first of all, explicit use of the wavelet function provides more powerful a priori knowledge than the autocorrelation or the power spectrum. Secondly, this method permits to flexibly trade off noise versus resolution.In the second method presented here, we use an analytical model (synthetic wavelet) of the system impulse response to determine an inverse filter.These methods have been developed for video pulse radar signals, and encouraging early results have been obtained.     

ISDN signal distribution network

The author discusses the ISDN Staged Transport System (STS). The STS illustrates how an ISDN network can be used as a flexible distribution medium for directing real-time signals over a WAN to a staged transport server. Although the signal feeds distributed by the STS application are audio voice grade circuits (VGCs), the architecture lends itself to other signal types, such as video. Moreover, ISDN bridges and routers allow the real-time communications infrastructure to distribute bulk recorded signals as well. Finally, the multimedia wide-area connectivity provided by ISDN allows an entire signal processing operation to be remoted, with receivers, storage, and mission control all managed from a single site (or multiple sites) on the network. While the STS application uses the user to user information element (UU-IE) to control signal processing on the edge of the ISDN network, the UU-IE could just as easily be used on the interior of the network by switch adjuncts to augment dialed number routing with intelligent-application-based routing. When combined with the rich set of commercially available applications available on many ISDN switches and adjuncts, sophisticated signal processing, distribution, and management networks could be rapidly configured and deployed with relative ease     

Generation of serial CPMFSK signal

A "serial" or "one channel" approach to the design of an M-ary CPFSK (CPMFSK) generator of arbitrary modulation index has been described. Uniformly spaced quadrant samples of the signaling sinusoids are stored in a ROM, appropriately retrieved by means of a simple digital processor and finally passed through a digital-to-analog converter (DAC) followed by a discretely tracking low-pass filter (LPF) to generate the CPMFSK signal. The design and implementation procedure of the proposed generator has been presented.     

S-K-H algorithm for signal design

An algorithm based on four ideas, two from Simon (hence S) and one each from Kronecker (K) and Hamming (H), has been developed to obtain ternary sequences of large lengths and high merit factors. Operationally, the Hamming scan and the Kronecker product are the two central ideas used