Kompiuterių garso sistemų tyrimas ir taikymas ; Investigation and application of computer audio systems

Administrator

Description

VILNIUS GEDIMINAS TECHNICAL UNIVERSITY Gediminas Gražulevi čius INVESTIGATION AND APPLICATION OF COMPUTER AUDIO SYSTEMS Summary of Doctoral Dissertation Technological Sciences, Electrical and Electronics Engineering – 01T Vilnius “Technika” 2004 1 Doctoral dissertation was prepared at Vilnius Gediminas Technical University in 2000-2004.

Sujets

• Analog-to-digital converter

Informations

• Publié par: vilnius_gediminas_technical_university
• Publié le: 01 janvier 2004
• Nombre de lectures: 13
• Langue: English

Extrait

    VILNIUS GEDIMINAS TECHNICAL UNIVERSITY

Gediminas Graulevičius

INVESTIGATION AND APPLICATION OF COMPUTER AUDIO SYSTEMS  

                            Summary of Doctoral Dissertation   Technological Sciences, Electrical and Electronics Engineering  01T                             Vilnius Technika 2004    1

 Doctoral dissertation was prepared at Vilnius Gediminas Technical University in 2000-2004.    Scientific Supervisor:    Prof Dr Habil Julius SKUDUTIS   (Vilnius Gediminas Technical University, Technological Sciences, Electrical and Electronics Engineering  01T)     The dissertation is defended at the Council of Scientific Field of Electrical and Electronics Engineering at Vilnius Gediminas Technical University:    Chairman:   Prof Dr Habil Albinas MARCINKEVIČIUS (Vilnius Gediminas Technical University, Technological Sciences, Electrical and Electronics Engineering  01T)    Members:  Prof Dr Habil Danielius EIDUKAS (Kaunas University of Technology, Technological Sciences, Electrical and Electronics Engineering  01T)  Prof Dr Habil Raimundas KIRVAITIS (Vilnius Gediminas Technical University, Technological Sciences, Electrical and Electronics Engineering  01T)  Prof Dr Habil Stanislovas TARAS (Vilnius Gediminas Technical University, Technological Sciences, Electrical and Electronics Engineering  01T)  Assoc Prof Dr Rimantas PUPEIKIS (Institute of Mathematics and Informatics, Physical Sciences, Informatics  09P)    Opponents:  Prof Dr Habil Romanas MARTAVIČIUS (Vilnius Gediminas Technical University, Technological Sciences, Electrical and Electronics Engineering  01T)  Assoc Prof Dr Antanas Leonas LIPEIKA (Institute of Mathematics and Informatics, Technological Sciences, Informatics Engineering  07T)    The dissertation will be defended at public meeting of the Council of Scientific Field of Electrical and Electronics Engineering in the Senate Hall of Vilnius Gediminas Technical University at 1 p. m. on September 10, 2004.  Address: Saulėtekio av. 11, LT10223 Vilnius - 40, Lithuania  Tel. [phone]. Fax. [phone],  e-mail [e-mail]    The summary of doctoral dissertation was distributed on August 10, 2004.  A copy of the doctoral dissertation is available for review at the Library of Vilnius Gediminas Technical University (Saulėtekio av. 14, Vilnius).     2

                             

    VILNIAUS GEDIMINO TECHNIKOS UNIVERSITETAS

Gediminas Graulevičius

KOMPIUTERIŲGARSO SISTEMŲ   TYRIMAS IR TAIKYMAS

Daktaro disertacijos santrauka Technologijos mokslai, elektros ir elektronikos ininerija  01T                             Vilnius Technika 2004   3

 Disertacija rengta 2000-2004 metais Vilniaus Gedimino technikos universitete.     Mokslinis vadovas:    prof. habil. dr. Julius SKUDUTIS   (Vilniaus Gedimino technikos universitetas, technologijos mokslai, elektros ir elektronikos ininerija  01T).     Disertacija ginama Vilniaus Gedimino technikos universiteto elektros ir elektronikos ininerijos mokslo krypties taryboje:    Pirmininkas:   prof. habil. dr. Albinas MARCINKEVIČIUS (Vilniaus Gedimino technikos universitetas , technologijos mokslai, elektros ir elektronikos ininerija  01T).    Nariai:  prof. habil. dr. Danielius EIDUKAS (Kauno technologijos universitetas, technologijos mokslai, elektros ir elektronikos ininerija  01T),  prof. habil. dr. Raimundas KIRVAITIS (Vilniaus Gedimino technikos universitetas, technologijos mokslai, elektros ir elektronikos ininerija  01T),  prof. habil. dr. Stanislovas TARAS (Vilniaus Gedimino technikos universitetas, technologijos mokslai, elektros ir elektronikos ininerija 01T),   doc. dr. Rimantas PUPEIKIS (Matematikos ir informatikos institutas, fiziniai mokslai, informatika  09P).      Oponentai:  prof. habil. dr. Romanas MARTAVIČIUS (Vilniaus Gedimino technikos universitetas, technologijos mokslai, elektros ir elektronikos ininerija  01T),  doc. dr. Antanas Leonas LIPEIKA (Matematikos ir informatikos institutas, technologijos mokslai, informatikos ininerija  07T).    Disertacija bus ginama   vieame elektros ir elektronikos ininerijos mokslo krypties tarybos posėdyje 2004 m. rugsėjo mėn. 10 d. 13 val. Vilniaus Gedimino technikos universiteto senato posėdiųsalėje.    Adresas: Saulėtekio al. 11, LT10223 Vilnius - 40, Lietuva  Tel.: [phone]; faksas [phone];  el. p. [e-mail]    Disertacijos santrauka isiuntinėta 2004 m. rugpjūčio mėn. 10 d.    Disertaciją galima periūrėti Vilniaus Gedimino technikos universiteto bibliotekoje (Saulėtekio al. 14, Vilnius).       VGTU leidyklos Technika 1019 mokslo literatūros knyga ©Vilniaus Gedimino technikos universitetas, 2004   4

INTRODUCTION 

   Topicality of the thesis.Computer technologies due to their fast development are finding ever-increasing use in various fields of human activity. Educational, measurement, monitoring and voice-controlled systems, voice recognition, text reading, writing during dictation, acoustic signal registration and analysis in medicine, criminology, engineering, are among the fields of the computer audio systems (CAS) application. Quite a number of the audio signal processing software means are in the MSWindowsoperational system as well.  Modern computer information technologies impose far much stricter requirements on the CAS, especially when it is used for measurement purposes as a convenient input/output port of measurement signals in the audio frequency range. Therefore, it is relevant to know the quality of a CAS.  At present, the choice of CAS is rather wide. However, the manufacturers of audio systems provide little information on their quality. Analog-to-digital and digital-to-analog converters (ADC and DAC) have essential influence on the CAS quality. The ADC effective number of bits (ENOB) is one of the main parameters allowing objective evaluation of dynamic properties of the CAS input channel. A comparatively complicated algorithm of the ENOB calculation and a large calculation extent in applying modern computers and their software are no longer topical. However, the calculation methodology, where the ideal ADC quantization error distribution at a sine-shaped test signal is assumed as that of uniform probability, raises some doubts. Moreover, the requirement for the test signal that it use all ADC codes is not achieved easily, which affects the calculation accuracy.  For the ADC investigation, analog signals are needed. By comparing, the form of these signals in the ADC input and output, static and dynamic characteristics of converters can be evaluated. It is understandable that for such evaluation the precise signal form in the ADC input must be known. The sine- shaped signal is recently used in the ADC testing. Such a signal is good in that the linear input circuit does not change its form. Only its amplitude and initial phase change. The choice of generators of sine-shaped oscillations is rather wide but in their generated oscillation there are always higher harmonics. It is especially difficult to obtain a sine-shaped oscillation of good spectral purity in the audio frequency range, where generators most often areRCtype.    Transitional processes in linear electric circuits can also be used as test signals but these problems have not been considered in the literature.  Modern CAS usually consists of two main parts: an audio coding-decoding device (Codec) and a digital controller. These devices are interconnected by a special link (AC-link). Codec is the main part of a new architecture CAS. The main function of Codec, the sound recording and reproduction through specially designed ports, can be applied for measurement purposes in the audio frequency range. A great number of virtual devices are created: oscilloscopes, generators, multimeters and others, that use this possibility. An oscilloscope and functional

 

 5

generator are in the program package MATLAB as well. Programmers of such virtual devices pay much attention to their shaping. These devices are not functional and easy in application, and important functions such as the result copying and documentation by an image or graphs are most often not foreseen.  The objective and tasks of the thesis.The object of the thesis was to develop a simple method of the ADC quality evaluation of CAS, which does not require a special apparatus, and software for its realization. To correct the ADC ENOB measurement methodology recommended in the IEEE Std 1241-2000. By applying the recommended method, to investigate possibilities of formation and application of new test analog signals for ADC investigation. To show the possibilities of untraditional CAS application for measurements. For this purpose, the following tasks should be solved:   1.   To analyze the methods of the quality evaluation of CAS and software intended for this. 2.   To investigate the classical calculation method of the ADC ENOB of CAS. 3.   By applying the program package MATLAB, to develop software intended for calculations of amplitude-frequency characteristics of the ADC ENOB of CAS. 4.   To investigate characteristics of reproducing channels of CAS. 5.   To determine possibilities of formation and application of new test analog signals in the ADC investigation. 6.   develop examples of untraditional CAS application for measurements.To   Methodical principles of the thesis.In the doctoral thesis, analytic, digital and experimental investigation methods are applied. A digital model of an ideal ADC is developed, the calculation method of the ADC ENOB of CAS and the calculation algorithms are corrected. By applying this model as well as the suggested method and algorithm, software for calculation of amplitude-frequency characteristics of the ADC ENOB of CAS has been developed by the MATLAB program package. The obtained results are analyzed and compared with the results obtained by applying a classical calculation methodology and the experimental investigation results.  Scientific novelty and use of results.It is proved that the classical method of the ADC ENOB calculation introduces a methodical error. A new, more precise method of the ADC ENOB calculation, which does not require a special apparatus, has been suggested. Possibilities of forming and applying new test analog signals (exponential, binary aperiodic signal and decay periodic oscillations) have been proved theoretically and experimentally.  The obtained results enabled us to make a suggestion on the IEEE Std 1241- 2000 standard improvement.  MATLAB programs, using ADC of a CAS, of a low-frequency oscillograph, spectrograph, frequency meter, acoustic tachometer, acoustic tachometer-recorder and acoustic noise recorder, have been developed.

 

 6

 The virtual devices developed by the author (an oscillograph, spectrograph, frequency meter, acoustic tachometer and acoustic tachometer-recorder) took part in the exhibition INFOBALT '2002 and aroused interest among visitors. At present, these virtual devices are improved and their metrological characteristics are investigated.  Approbation of the work and publications.The main results of the thesis are reported at 11 scientific conferences: at the republican conference Electronics in 1999, 2000, 2001 (two reports), 2002, 2003 and 2004, at the republican conference of young scientists Lithuania without science  Lithuania without future in 1999 and 2001, at the scientific-practical conference Sound Card 2001 (Kaunas Technological University. Kaunas, 2001), at the international scientific-technical conference Problems of Electronics (Kiev, 2003).  On the subject of the thesis, 10 scientific articles were published, among which 5 articles in Lithuanian technical peer-reviewed journals, 2 articles in foreign peer-reviewed journals and 3 articles in proceedings of conferences.  It is presented for defense:   1.   The programmable model of ideal ADC. 2.   A simple method of the CAS ADC quality evaluation, which does not require a special apparatus, and software for its realization. 3.   formation and devices for CAS ADCNew principles of the test signal investigation.  4.   Examples of untraditional application of CAS for measurements.  The structure and contents of the work.The thesis includes introduction, four chapters, generalization of the results, references, publications of the author and appendixes.  The thesis consists of 122 pages, 6 tables, 46 figures, 120 bibliographic items and 2 appendixes.   CONTENTS OF THE THESIS     tcoi.nIntrodu In the introduction, the topicality of the chosen subject is analyzed, the main objective and tasks as well as the structure of the doctoral thesis are presented.   Chapter 1.   methods for the computer audio systemAnalytical survey of quality evaluation. In chapter 1, the structure and types of CAS, dynamic parameters of their main parts, ADC and DAC, and their evaluation methods as well as software used in the investigation of the CAS quality are surveyed. The application of CAS is described. The objective and tasks of the work are formulated.    Chapter 2.   The procedure of investigating the computer audio systems quality. Inpossibilities of application of the program package chapter 2, MATLAB in the investigation of the CAS quality are described. The developed

 

 7  

procedure of the CAS quality investigation in the MATLAB environment as well as the performed CAS quality investigations of its application are presented. Selection and substantiation of the investigation method of the computer audio systems quality.  CAS will be evaluated according to its main function  the sound recording and reproduction quality. The main parts  ADC and DAC− of the CAS sound coding and decoding device play the decisive role here. Most of modern CAS allows a duplex operation regime, i.e. sound reproduction and recording simultaneously. We will use ADC ENOB as the evaluation criterion.  At first, using a special MATLAB program, a standard fileEtalon.wav, imitating the sine-shaped oscillation discretization and conversion to the idealN- bit ADC by numerical codes, is developed. Then this file is reproduced by the investigated CAS DAC and at the same time is re-recorded into another file Real.wav ADC (in byFull Duplex regime). For the reproduction and recording procedures the MSWindows Sound Recorder is used. The re-recorded program file differs from the standard file in that both converters ADC and DAC had influence on the signal which passed them.  According to the re-recordedReal.wav file data, for the quantitative evaluation of the CAS reproduction and recording path quality the ADC ENOB is calculated by a known formula:   N ef =N−log 2 S S r i , (1) whereNis the physical number of ADC bits,S r   andS j are mean square errors of conversion of a real and ideal ADC.  The calculatedN ef   value will depend on properties not only of ADC but also of DAC, therefore it can be considered as a complex CAS quality criterion. The practice shows that if we want to ignore the DAC influence the number of its bits should be by one bit larger than the number of ADC bits.  In the case of an ideal ADC, the quantized analog signal momentary value gets into any quantization interval place with an almost uniform probability. Therefore, the standard estimate of a conversion error is calculated by evaluating the uniform probability distribution:   S i = Q 12 , (2) whereQis the ADC lowest bit value.  The standard conversion error estimate of a real ADC is calculated in respect of the sine-shaped oscillation by the least-squares method:   S r min ( E ) , (3) 2 where sin 2π E= k 1 −1 i k ∑ =1 ⎧ ⎩ ⎪ ⎨ ⎪ x i − ⎡ ⎢ ⎢ ⎣ A ⎝ ⎛ ⎜ f f s iϕ+ ⎠ ⎞ ⎟ +C ⎥ ⎦ ⎥ ⎪ ⎭ ⎫ ⎬ ⎤ ⎪ , (4)

 

 8

f, Hz x10   φ,rad Fig. 1.General graphical view of the objective function  (6)

x i is thei-th number in the ADC output;A,ϕandCare the sine-shaped oscillation amplitude, initial phase and constant component;f andf s  the signal and are discretization frequencies.  The objective function (4) is of multiextremal character (Fig. 1), therefore by minimizing it, measures directing search for the global minimum determination should be taken. Moreover, by forming the search algorithm it is important to determine the relationship between variables.  The optimal values of frequency fand the initial phaseφ be cannot found independently of each other because they are interdependent. The amplitudeAand constant componentC have no influence on the optimalfandφ so they values, can be found separately applying the classical method of partial derivatives:    E A  (5) ⎩ ⎨ ⎧ ∂ ∂ E∂ ∂ C= = .0 ,0  By solving equation system (5), we obtain:  ,    A opt = k⋅B1B3⋅B4 . k⋅B2− (B3) 2 wherekis the number of samples. ⋅ B1A t. B2 =   C opt. B o 3 p ,    (7) = k ⎛ ⎜ f i +   B1 i= ∑ 1 x i sin ⎝ 2π f s ϕ ⎞ ⎟ ⎠ , (8)   ⎜ ⎞ ⎟ B2= i= ∑ k 1 sin 2 ⎝ ⎛ 2π f f s i ⎠ +ϕ , (9)   B= k ∑ ⎜ ⎛ f + ⎞ ⎟  (10) 3 =1 sin ⎝ 2πi f ⎠ ϕ , i s k   B4= ∑ x i . (11) i=1  The optimal frequency and initial phase values found separately must be incorporated into equations (8)(10).  By using the above presented equations and the minimization procedure of MATLAB, the Ess program for the ADC ENOB calculation was developed.

 

 9

The program calculates ENOB and deduces graphs of the signal which passed through the DAC ADC path and of the signal approximating it.   Since the objective function is multiextremal, the global extreme search means are foreseen in the program.   Selection of the number of samples. is a classical problem of It mathematical statistics, the solving of which ensures the precision of estimate of the chosen random numerical characteristic. The number of samples is found according to the desired precision of the mean determination. If the number of samples is large enough (k>100), the distribution of mean is close to normal and then we can use the known formula:   k ⎡ = ⎢ ⎣ t ε ( P ⎦ ⎤ ⎥ ) 2 σ 2 , (12) wheret=t(P)  the Stjudent coefficient is found from equation 2Φ(t) =P;Φ(t) is the error function;Pis the confidence probability;εis the ENOB standard mean deviation (experimental);σ 2 is the average square value of random error.  In the case considered the random value is ENOB. Its dispersion is determined not only by the chosen sample number (sample quantity) for the ENOB calculation but also by characteristics of a real signal  noise, frequency, dynamic ADC characteristics, etc. Therefore, we will find the sample number ensuring the desired precision by measuring ENOBk times and statistically processing the results.  It will allow calculating the ENOB limit relative error:   RSP= ± 3S ⋅100% , (13) N ef whereS is the estimate of the standard ENOB measurement error; N ef  the is ENOB mean.  After calculation of dependence of the ENOB limit relative error on the number of samples by equation (13), according to the number of samples, we can choose the desired ENOB calculation precision. For the determination of the number of samples, in the considered case equation (12) cannot be applied because the mean square valueσ random error is of unknown. In this case, the needed number of samples can be chosen from the table.  Thus, after processing of results the chosen sample number with a probability P, we can assert that:   N ef0 −N ef <qS , (14) whereN ef0 is the real ENOB value.  In order to perform these calculations and to obtain the ENOB distribution histograms, the MATLAB program Dsp was developed.

 

10  

 ENOB calculations are performed at random places of fileReal.wav, so different ENOB values are obtained. Thus, ENOB is the random (more precisely, pseudorandom) number having its distribution.  The calculated probability distribution asymmetry coefficient and excess coefficient do not contradict the normal distribution hypothesis, therefore for the RSP the tolerance field is assumed to be equal to 6 mean square calculations deviations.   Thus, in performing a single ENOB measurement, a random error is inherent. In order to reduce it, the chosen number of samples must be increased.   Selection of the recording level. measurements of ADC ENOB it is In required that the double amplitude of the input signal involve the whole scale of the converter codes. However, at a large number of converter bits (16 and more) it is rather complicated to achieve this, thus the input voltage is set a bit lower. The author attempted to evaluate the influence of the recording level on the audio range ADC ENOB in his works. But later investigations have shown that the conclusions drawn in respect to some converters do not suit other converters.  For the investigation of the recording level influence, the MATLAB program consisting of blocks of polyamplitude signal modeling and ENOB calculation was created. The polyamplitude signal modeling program Tsta models fixed 1000 Hz frequency and tantivy increasing sine-shaped oscillations (Fig. 2) and records them into the test fileEtalon.wav.  The second program Essa 1 calculates the ADC ENOB amplitude 0.8   characteristics.  0.6    The experiments have shown that 0.4   most of the investigated CAS are not 0.2   qualitative. ENOB of real ADC of 0 these CAS substantially differs from -0.2     the number of bits of the -0.4 corres ding ideal ADC which -0.6   pon -0.8   should be equal to 16 according to the -1   0 200   400 600   800   1000   1200   1400   1600   1800   2000   user manual of CAS. ENOB Number of samples itude charac ampl teristics have a Fig. 2.Fra s s l t i e g e h p t n ly e   s d s  r i o s  p t p h i e n  s g  m c a h l a le r r a   c t t h e e r   .  hi T g h h e e   r d   r t o he p     signal for the investigation of recording level dependence ADC quality, i.e. the larger ENOB value. The ADC ENOB drop of not qualitative CAS at larger values of the input signal can be explained by the fact that, when the increasingly larger scale of the converter codes is covered, the mean square value of noise increases. By extrapolation the ENOB value can be found for each ADC at the maximum value of the input signal. To calculate the general correction coefficient estimating the ENOB decrease at increasing input voltage would be not correct.

 

11