Digital Signal Processing

Russian
Scientific & Technical
Journal


Digital Signal Processing No. 1-2015

In the issue:

- processing speech and audio signals

- multiresolution wavelet transformation
- multivariate time series analysis
- algorithms of correlational-extremal processing
- method of chains of local extrema
- non-intrusive estimation of noisy speech signal
- effective technology of pattern recognition
- filtering of periodic signals


Listening Enhancement in Noisy Environment Based on Spectral Decomposition and Adaptive Dynamic Range Compression of the Signal
E. Azarov, azarov@bsuir.by
M. Vashkevich, vashkevich@bsuir.by
D. Likhachov, likhachov@bsuir.by
A. Petrovsky
, palex@bsuir.by
Belarusian State University of Informatics and Radioelectronics, Belarus, Minsk, 6, P. Brovky str., 220013


Keywords: listening enhancement, spectral decomposition, dynamic range compression.

Abstract
The paper investigates possibility of automatic listening enhancement in noisy environment. In order to improve subjective quality time-frequency components of the signal are nonlinearly amplified according to spectral components of surrounding noise. The method, proposed in the paper, is applicable to different signals including speech and audio. Practical value of the method is evaluated for speech signals using objective measures.

A general person often experiences problems when listening to audiobooks or music in noisy environment. Audio signal becomes partially or fully imperceptible due to the masking effect. Just increasing playback volume in noisy conditions is excessively tiresome for ears because masked sounds become loud and unmasked sounds become too loud. On the other hand manual adjustment of playback volume is inconvenient especially when noise intensity gradually changes.

The algorithm proposed in the paper is based on frequency decomposition and adaptive dynamic compression of the signal. Dynamic range of each subband is narrowed in order to make it perceivable after mixing with background noise. Response curve of the compressor is adaptively adjusted according to short-time energy of the noise.

We designed a smartphone application that implements the proposed method. The application captures background noise using the microphone and performs playback of the sounds through near end listening enhancement scheme. The subjective listening tests show that algorithm significantly improves listening experience in different noisy conditions including traffic and engine noises. Some objective evaluations have been done as well using sound pressure levels (SPL) and SII measurements.

References

1. Sauert B., Vary P. Near and listening enhancement: Speech intelligibility improvement in noisy environments // Processing of International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Toulouse, France, 1419 May, 2006. Toulouse, 2006. pp. 493496.

2. Sauert B., Vary P. Near and listening enhancement optimized with respect to speech intelligibility index // Processing of 17th European Signal Processing Conference (EUSIPCO 2009), Glasgow, Scotland, 2428 August, 2009. Glasgow, 2009. pp. 18441848.

3. Zorila, T.-C. Speech-in-noise intelligibility improvement based on spectral shaping and dynamic range compression / T.-C. Zorila, V. Kandida, Y. Stylianou // In Proc. Interspeech, 2012. Portland, Oregon, 2012. pp. 635638

4. Vonlanthen, A. and Arndt, H. Hearing Instrument Technology for the Hearing Health Care Professional / Thomson Delmar Learning, Clifton Park, NY, 2006.

5. Blesser B.A. Audio dynamic range compression for minimum perceived distortion / B.A. Blesser // IEEE Trans. Audio and Electroacoustics. 1969. vol. 17. no. 1. pp. 22-32.

6. Vashkevich M.I. Oversampled non-uniform cosine-modulated filter bank design / M.I. Vashkevich, A.A Petrovsky // Informatics. Minsk. 2011. 2 (30). pp. 2139 (in Russian).

7. Vashkevich M.I. Cosine-modulated filter banks with all-pass transform: implementation and application in hearing aids / M.I. Vashkevich, E.S.Azarov, A.A Petrovsky // M: Gorachaya liniya - Telecom, 2014. 210 p (in Russian).

8. Trine T.D., Tasell D.V. Digital hearing aid design: facts vs. fantasy // The Hearing Journal. 2002. Vol. 55, 2. pp. 3642.

9. Koilpillai D. Cosine-modulated FIR filter banks satisfying perfect reconstruction / D. Koilpillai, P.P. Vaidyanathan // IEEE transaction on signal processing. 1992. Vol. 40,  4. pp. 770783.

10. Vashkevich M.I. Application of polynomial algebras and Galois theory for synthesis of fast algorithms / M.I. Vashkevich, A.A Petrovsky // Digital signal processing . 2011. 3. . 210 (in Russian).

11. American National Standard. Methods for the Calculation of the Speech Intelligibility Index. ANSI S3.5-1997, 1997.

12. Kominek J., Black A. The CMU ARCTIC speech databases for speech synthesis research / Language Technologies Institute, Carnegie Mellon University, Pittsburgh, PA, Tech. Rep. CMU-LTI-03-177, 2003.


Modeling of Variations of Cosmic Rays on the Basis of Combination of
Multiresolution Wavelet Expansions and Neural Networks with Variable Structure
O.V.Mandrikova, e-mail: oksanam1@mail.ru
T.L. Zalyaev, e-mail: tim.aka.geralt@mail.ru

Institute of Cosmophysical Research and Radio Wave Propagation of the Far Eastern Branch of Russian Academy of Science (IKIR FEB RAS), Russia, Paratunka, Kamchatka region

Keywords: wavelet decomposition, neuron networks, variable structure, cosmic rays variation, anomalous features.

Abstract

In this paper we propose a new method of modeling variations in the intensity of cosmic rays (CRs), based on a combination of wavelet transform and neural networks. The method allows to perform a detailed analysis of the structure of registered CR variations, to predict their time course and to highlight anomalies in the periods of increased solar activity. During simulation informative components of CR variations, derived from the wavelet transform, are approximated by neural networks with variable structure. To allow construction of adaptive approximations, we propose to use the architecture of feedforward networks with variable structure. For selecting Wavelets we used criteria based on the minimization of error of approximation among the class of orthonormal basis functions. Algorithm for choosing the level of wavelet decomposition and extraction of informative components is based on minimization of errors of obtained model.

Observations of CR intensity variations are used in a number of fundamental and applied research related to monitoring and forecast of the space weather. CR variations observed on the surface of the Earth are the result of various integral solar, heliospheric and atmospheric phenomena and have a complex internal structure. During periods of strong increases in the intensity (GLE-events) extensive flows of high-energy particles create a major problem for space equipment and other technologies, for wireless communication in the polar regions, as well as for astronauts. Therefore, allocation and prediction of such events is very important.

In this paper, based on the proposed method, we researched data of neutron monitor from stations "Cape Schmidt" (Russia, Cape Schmidt) and "Apatite" (Russia, Apatity). Trained neural networks, which approximate CR variations for the analyzed stations, showed that in periods of quiet geomagnetic field, which are characterized by low solar activity, the absolute values of errors does not exceed the value of 0.05, indicating a good quality of approximating properties. In periods of high geomagnetic activity, there was a significant increase in network errors caused by changes in the state of near-Earth space and the decrease in the level of cosmic rays (Forbush decrease). Increase in errors of networks was also observed before and after the analyzed GLE-event. These results confirm the effectiveness of the proposed method and the possibility of its use in the problems of detailed analysis of cosmic ray variations and detection of anomalies that occur during periods of increased solar activity.

References
1. Toptygin I.N. Space Rays in Interplanetary Magnetic Fields. M.: Nauka, 1983. 301 p.

2. Eroshenko E.A., Belov A.V., Kryakunova O.N., Kurt V.G., Yanke V.G. The alert signal of GLE of cosmic rays // Proceedings of the 31st ICRC, 2009.

3. Tyasto M.I., Danilova O.A, Dvornikov V.M., Sdobnov V.E. Bolishie snigenia geomagnitnih porogov cosmicheskih luchei v period vozmushenyi magnitosferi.. Izvestia RAN, seria phizicheskaya, . 73, 3, pp. 385-388. 2009.

4. P. Paschalis, C. Sarlanis, H. Mavromichalaki - Artificial Neural Network Approach of Cosmic Ray Primary Data Processing. Solar Physics, 2013;182(1):303-318.

5. J. Kota, A. Somogyi - Some problems of investigating periodicities of cosmic rays - Acta Physica Academiae Scientiarum Hungaricae, Tomus 27, pp. 523-548 (1969).

6. Mallat S. A Wavelet tour of signal processing [. .]. .: , 2005. 671 .

7. Daubechies I. Ten Lectures on Wavelets. SIAM, 1992.

8. Neuromathematic. School-book for Higher education / A.D. Ageev and other; Galushkin A.I editor. M.: IPRGR, 2002. 448 p.

9. Mandrikova O.V. Mnogokomponentnya model signala so slogenoi strukturoi // Problemi evolucii otkritih system. 2008. Vip. 10. . 2. pp.161172.

10. Polozov U.A. Metod phormirovaniya obuchaushego mnozhestva dlya neironnoy seti na osnove veivlet-philtracii // Izvestia vuzov, Severo-Kavkazkiy region. Rostov-na-Donu. 2010. 3 . pp. 1216.

11. Mandrikova O.V. Optimizatciya protcessa obucheniya neyronnoy seti na osnove primeneniya konstruktcii veyvlet-preobrazovaniya (na primere modelnogo predstavleniya ionosfernogo signala) // Avtomatizatciya i sovremennye tekhnologii. 2009. 3. pp. 14

12. Mandrikova O.V., Polozov Yu.A. Kriterii vybora veyvlet-funktcii v zadachakh approksimatcii prirodnykh vpemennykh ryadov slozhnoy struktury // Informatcionnie tekhnologii. Moskva. 2012. 1. pp. 3136.

13. Mandrikova O.V., Zalyaev T.L., Belov A.V., Yanke V.G. Metod viyavleniya anomaliy v variatciyakh kosmicheskikh luchey na osnove sovmeshcheniya veyvlet-preobrazovaniya s neyronnymi setyami - Sbornik docladov VI mezhdunarodnoy konferentcii Solnechno-zemnie svyazi i fizika predvestnikov zemletriaseniy, 2013 pp.304-310.

14. Mandrikova O.V., Zalyaev T.L Modelirovanie variatciy kosmicheskikh luchey na osnove sovmeshcheniya kratnomasshtabnogo analiza i setey peremennoy struktury .// Sbornik tezisov docladov VI Mezhdunarodnoy nauchno-tekhnicheskoy konferentcii po myagkim vychisleniyam i izmereniyam (SCM`2013). v.2. - SPb, 2013. pp. 111-117.

15. Mandrikova O.V., Glushkova N.V., Polozov Yu.A. Algoritmy vydeleniya i analiza anomaliy v parametrakh kriticheskoy chastoty ionosfery fOF2 na osnove sovmeshcheniya veyvlet-preobrazovaniya i avtoregressionnykh modeley // Tcifrovaya obrabotka signalov. Moskva: RNTORES. 2013. 1. pp. 47-53.

16. Mandrikova O.V., Solovev I.S. Veyvlet-tekhnologiya obrabotki i analiza geomagnitnykh dannykh // Tcifrovaya obrabotka signalov Moskva: RNTORES. 2012 2. pp. 24-29


Short-Term Forecasting Methodology for Dynamic System Characteristics Variability Based on Multivariate Time Series Numerical Analysis

I.F. Zaporozhtsev, (Murmansk State Technical University), e-mail: zaporozhtsev.if@gmail.com
A.-V.I. Sereda, (Saint Petersburg Electrotechnical University "LETI"), e-mail: avis_14@mail.ru

Keywords: short-term forecast, spatial distribution of physical characteristic, multivariate time series, singular spectrum analysis, empirical modes decomposition, TARX model.

Abstract
Short-term temporal variability forecasting methodology for spatially distributed characteristics of natural processes based on multivariate time series numerical analysis is presented in the article. Both its general concept and some stages algorithmic specification are designed. Methodology version introduced currently is formed completely and is implemented as a computer technology. Mathematically proved technology of numerical analysis of natural processes is certain to be interesting for different specialized fields professionals of scientific studies (oceanographers, meteorologists, biologists, ecologists, etc.) and in practical activities, such as fishery, gas-and-oil producing industry, delivery logistics, etc.

Developed methodology can be applied to characteristics forecast with gridded regularly obtained satellite and/or surface measured data as input. Each node of planar regular (uniform) grid defined in rectangular spatial area provides time series of characteristic values. All the time series have the same start time point, time span and length (number of sequential time points). The task is to offer extrapolation to each time series using the information about variability of not only one time series. Forecast horizon is significantly less than available time series fragment length (about several units and several hundreds time points respectively). Our solution is based on multivariate time series analysis, clustering techniques and linear algebra numerical methods. It consists of three stages: grid nodes clustering, initial intracluster forecast and adjusted intracluster forecast.

Clustering uses cross-correlation and spatial neighbouring criterion to divide nodes without metric given directly. Algorithm contains divisive and agglomerative steps. At the first step grid division by four equal rectangles is carried out and continued similarly for each obtained rectangle until any pair of current cluster nodes has corresponding time series cross-correlation coefficient greater than predetermined value and is greater than the same coefficient calculated for this pair time series fragments at nonzero time lag. As the result, set of grid nodes clusters and set of corresponding multivariate cluster series are formed. Finally, cluster characteristics are proposed to estimate cluster structure stability and consistency.

Initial intracluster forecast adopts three modern time series handling techniques such as multivariate empirical mode decomposition (MEMD) and quasi-stationarity condition application to determine length of time series fragment to be used further as two preprocessing tools. Forecast is calculated with so-called K-extension method based on multivariate singular spectrum analysis (MSSA). At this stage cluster series extrapolation is carried out for each cluster separately and it uses only one (current) cluster series dataset. Then series of forecast errors (one-step ahead, horizon is equal to 1) is computed to be used at the next stage.

MSSA forecast results correction (adjusted intracluster forecasting method) is based on univariate time series approximation with threshold autoregressive model with external inputs (TARX). MSSA forecasting method uses special autoregressive model to approximate multivariate time series corresponding to particular cluster. Since the aim here is to employ some spatio-temporal effects and dependancies between different clusters, it appears more relevant to account for some explanatory variables in addition to simple autoregressive model. Therefore, TARX is used for corrected cluster (each cluster in turns) one-step ahead forecast error prediction with regressors being previously calculated errors for both this cluster and some set of the correcting clusters determined with cluster characteristics values analysis.

Developed methodology was applied successfully for sea surface anomalies in the Barents Sea and for sea surface temperature in the Irminger Sea.


References

1. Ashik I.M. Numerical hydrodynamic method of sea level anomalies variability forecasting in the south-eastern part of the Barents sea and in the south-western part of the Kara sea, http://method.hydromet.ru/publ/sb/sb31/sb31.html, 2005.

2. Verbitskaya O.G. Hydrodynamic forecasting method of sea level and current synoptical variability in the Caspian sea, Ph.D. Thesis, 2009, 175 p.

3. Orlov U.N., Osminin K.P. Non-stationary time series. Forecasting methods with financial and materials markets analysis examples, Moscow, 2011, 384 p.

4. Chuchueva I.A. Time series forecasting model over maximum likelihood sample, Ph.D. Thesis, 2012, 153 p.

5. Zagoruyko N.G. Applied aspects of data and knowledge analysis, Novosibirsk, 1999, 270 p.

6. Ngongolo H.K. Tropical precipitation statistical forecasting on basis of ocean surface temperature and quasi-biennial oscillation of zonal flow applied to the eastern Africa datasets, Ph.D. Thesis, 2011, 156 p.

7. Stepanov D.A., Golyandina N.E. SSA-Caterpillar method variants for multivariate time series forecasting. Proceedings of SICPRO'05, Moscow, 2005.

8. Hoppner F., Klawonn F. Compensation of Translational Displacement in Time Series Clustering Using Cross Correlation. Lecture Notes in Computer Science. Advances in Intelligent Data Analysis, vol.5772, pp.71-82, 2009.

9. Liao T.W. Clustering of Time Series Data - a Survey. Pattern Recognition, vol. 38, pp. 1857-1874, 2005.

10. Huang N. E. The Empirical Mode Decomposition and the Hilbert Spectrum for Nonlinear and Non-Stationary Time Series Analysis, http://keck.ucsf.edu/~schenk/Huang_etal98.pdf, 1998.

11. Yang P. at al. The Prediction of Non-Stationary Climate Time Series Based on Empirical Mode Decomposition. Advances in Atmospheric Sciences, vol. 27, pp. 845-854, 2010.

12. Davidov V.A., Davidov A.V. End effects reduction for signals empirical mode decomposition of Hilbert-Huang Transformation, http://www.actualresearch.ru/nn/2011_1/Article/physics.../ davydov2011.pdf?, 2011.

13. Fleureau J. at al. Multivariate Empirical Mode Decomposition and Application to Multichannel Filtering. Signal Processing, vol. 91, pp. 2783-2792, 2011.

14. Rehman N., Mandic D.P. Multivariate Empirical Mode Decomposition. Proceedings of the Royal Society A., vol. 466, no. 2117, pp. 1291-1302, 2010.


Multiprocessor Realization of Correlational-Extremal Signal Processing

A.D Pluzhnikov, e-mail: pluzhnikov@nntu.nnov.ru
N.N. Potapov, e-mail: nick_potapov@nntu.nnov.ru
State Technical University of Nizhny Novgorod, Russia, Nizhny Novgorod


Keywords: correlational-extremal processing, navigation, algorithm, signal processor, multiprocessor systems.

Abstract
Correlational and differential algorithms of correlational-extremal processing are discussed. Required calculation time and coordinates estimation reliability are defined. Recommendations concerning choice of algorithm to solve problem of navigation of aircraft with multiray altimeter radar are given. Possibility of hardware realization of such system based on the integrated module which contains high-performance signal processors is estimated.

At present the most accurate standalone solution of aircraft navigation problem is achieved by map-matching radionavigation systems, especially by systems built on the base of two-dimensional topography maps usage. Comparison of prepared reference map to created during flight measured map performed in such systems by calculating so called proximity index. Global extremum of proximity index defines deviation between real aircraft coordinates and expected coordinates.

Algorithms of correlational-extremal processing differ in the choice of proximity index type. Type of proximity index specifies calculation complexity of particular algorithm. For classic correlational algorithm (CCA) proximity index is represented by mutual correlation function of reference and measured maps. For the family of differential algorithms multiplication is replaced by subtraction, allowing to significantly reduce requirements of calculation unit performance. This paper deals with analysis of the following differential algorithms: algorithm of absolute difference minimum (ADM), algorithm of squared difference minimum with average subtracting (SSM), algorithm of absolute difference minimum with average subtracting (ASM).

Measured maps obtained by processing of information extracted during the area relief scanning is commonly used in map-matching navigation systems. Relief scanning performed by onboard multiray radar. Radar rays are stationary relative to aircraft. Relief scanning is realized due to aircraft motion. Different characteristics of algorithms were compared using mathematical modeling and showed ADM algorithm to provide the worst coordinate estimation reliability. Three other algorithms calculations contain operation of centering which allows to compensate systematic error and to increase the reliability of coordinate estimation. CCA provides worst calculation time, which caused by large amount of multiplication operations. SSM algorithm is preferred with respect to higher coordinate estimation reliability and lower calculation time requirement.

Calculation time required to perform calculation according to SSM algorithm allows to realize this algorithm using modern hardware. In particular, recently developed integrated module with high-performance signal processors can be used for algorithm implementation.

References
1. Bakulev P.A., Sosnovsky A.A. Radiolocation and radionavigation systems: Study guide for high schools. Moscow.: Radio and communication, 1994. 296 p.

2. Andreev G.A., Potapov A.A. Algorithms of spatial-time navigational information processing (part 1) // Foreign radioelectronics. 1989. no. 3. pp. 318.

3. Andreev G.A., Potapov A.A. Algorithms of spatial-time navigational information processing (part 2) // Foreign radioelectronics. 1989. no. 4. pp. 321.

4. Bochkarev A.M. Correlational-extremal navigation systems // Foreign radioelectronics. 1981. no. 9. pp. 2853.

5. Kulikov E.I., Trifonov A.P. Estimation of signal parameters with presence of interference. Moscow.: Sov. radio, 1978. 296 p.

6. Kuzin A.A., Pluzhnikov A.D. and others. Analysis of time relation for signals in design digital modules and availability estimation. // Digital signal processing. 2014. no. 2. pp. 7077.


High-Speed Conversion Algorithm of Orthogonal Signal Components into Amplitude and Phase
V.V. Chekushkin, Murom Institute of Vladimir State University named after A.G. and N.G. Stoletovyuh, e-mail: chekvv@gmail.com
I.V. Panteleev, Aktsionernoye obshchestvo Muromskiy zavod radioizmeritel'nykh priborov (AO MZ RIP). Russia, Murom, e-mail: ilya-panteleev@mail.ru
K.V. Mikheev, Murom Institute of Vladimir State University named after A.G. and N.G. Stoletovyuh, e-mail: kiri-mikheev@yandex.ru

Keywords:
measurement of signals, coordinate transformation, error of the measurement system, polynomial, specialized computer.

Abstract
The high-speed reproduction algorithms of functions , with optimization of the computational process for the accuracy characteristics, speed, software-hardware costs on the examples of measurement of signal amplitudes with an unknown initial phase , a.c. voltages and phase angles have been studied. For the relative error values of the result from 4% to 0.26% the speed gain by 2 to 2.75 times is provided compared with the direct square root computation. Upon this, both the digit grid overflow, and the exponent underflow of numbers A and B do not occur. To find rapidly the value there used the approximate condition check algorithms of ratio of A and B to ensure the consecutive maximum incrementation of the number of significant digits for representation of the result (M) with the minimum increase of computational operations N and memory accesses P corresponding to it.

One condition check algorithm of the ratio ensures the maximum relative error δ=4,08% for 6 operations N+P: comparison, multiplication, addition and constant extraction. It should be noted that only the computation within the interval with the error of 4.6% of the best approximation polynomial requires the implementation of 10 operations: 6 N and 4 P. When using two conditions of ratio comparison the polynomials approximating the function with the error δ of less than 1.4% are obtained. When implementing the algorithm it should be performed 9 N+P operations and stored 5 constants in the memory. Compared with the one condition algorithm when increasing the number of operations by 3 the error δ decreases by 2.9 times. When applying three conditions the error δ decreases to 0.5% that is by 2.8 times less than that for two condition algorithm. Number of N+P operations on the longest branch of the algorithm implementation corresponds to 12, and there should be stored 8 constants in the memory. For four condition algorithm the error does not exceed 0.26% with approximate reproduction of functions for 15 operations. Upon this, 11 constants should be stored in the memory. When increasing the number of operations N+P by 3 the error decreased almost by 2 times. The further increase in the number of conditions complicates the algorithm implementation and for higher accuracy class systems is preferable to use the direct methods of square root computation.

The division operation A/B is substituted by multiplication of the numerator by the reciprocal of the denominator: A/B = A x (1 / B). The function is approximated in the range by three polynomials of best approximation on three subintervals with approximately equal absolute maximum errors. The phase measurement using the predetermined approximation subintervals of orthogonal component computation of the signal amplitude provided the reduction of errors by about 10 times when the other criteria of the computational process are fixed.

References
1. .. Akchyurin Software implementation of mutual transformations of algebraic and exponential representation of a complex signal for digital signal processors / .. Akchyurin // Radio Engineering. 1995. 1-2. pp. 21-23.

2. Ercegovas Milos Dand. Others. Reciprocation square root, inverse square root, and some elementary functions Using small multipliers // IEEE Trans Comput. 2000 / 49, 7 pp. 628-637.

3. Schulte Michall J., Swarzlander Earl E. (Jr). A family of variable precision interval arithmetic processors // IEEE Trans Comput. 2000/ 49, 5 pp. 387-397.

4. V.V. Chekushkin, A.M. Averyanov, A.D. Bogatov A method and apparatus to compute the square root patent RF 2438160 Bul. 36, 2011.

5. A.M. Averyanov, V.V. Chekushkin, I.V. Panteleev Methods of increasing the speed and accuracy characteristics of converters of orthogonal components of a signal into amplitude // Measurement Techniques. November 2012, Volume 55, Issue 8, pp 858-866.

6. USSR. inv. cert. Device for square root extraction from sum of squares./ I. YA. Mironov, YU. V. Malinin, T.G. Lazebnik, L.I. Novikova // 1983. Bul. 8

7. Kahaner D., Mouler K. Nash S. Numerical methods and software: Transl. from English. - M .: Mir, 2001.- 575 p.

8. Avt. svid. USSR. A device for calculating the function / V.V. Chekushkin./1982. Bul. 37.

9. V.V. Chekushkin, Y.Y. Grishin, V.V. Kostrov Development of algorithms division of numbers in the information-measuring systems // Metrology.- 2013.- 11.- pp 3 -14.

10. V.Y. Grishin, I.V. Panteleev, V.V. Chekushkin Improvement of methods, mathematical models of computational processes in the implementation of radar systems // Herald Aerospace Defense. - 2014.- 3.-pp .31-34.

This work was supported by grant RFFI 14-07-00293.


Generalization of the Method of Chains of Local Extrema for Analysis of Signals of Different Origins Abstract
Y. A. Turovsky, e-mail: yaroslav_turovsk@mail.ru
S. D. Kurgalin, e-mail: kurgalin@bk.ru
A.A. Vahtin, e-mail: alvahtin@gmail.com
S. V. Borzunov, e-mail: sborzunov@gmail.com
V. A. Belobrodsky, e-mail: belobrodsky@yandex.ru
Russia, Voronez


Keywords: biomedical signals, spectral analysis, wavelet transformation, the method of chains of local extrema, ribonucleic acid, meteorology, exchange quotation.

Abstract
The method of construction of chains of local extrema for the matrix of squared coefficients of the wavelet transformation, developed and applied previously to analyze electroencephalograms signals ("the method of chains of local extrema"), is summarized in this paper for the research of all one-dimensional signals or data, regardless of their nature. An essential feature of the development of the method is the formulation of a general rule for adding an existing extremum of the local spectrum to the already formed chain by identifying values of the parameters u and v, which determine the characteristics of the formed Chain of Local Maxima or Chain of Local Minima. Selection of parameters must ensure the maximization of the ratio of extrema numbers (only minimums or only maximums) at the time of the existence of the chain. This approach avoids an excessive union of extrema into the chains with large gaps in the time domain. The proposed universal method has been tested on data of different nature: on the recoded RNA genome sequences of hepatitis C virus, on meteorological data - the average monthly temperatures, on information about stock exchange quotations. It may be concluded that the value of k /Δb is sufficiently sensitive criterion for selecting values of u and v necessary for the correct construction of Chain of Local Maxima and Chain of Local Minima signals of different nature.

References
1. Bioelectric brain activity in patients with neurotic and neurosis-like disorders (according to a spectral analysis)  Schultz E.V., Baburin I.N., Karavaeva T.A., Karvasarsky B.D., Slezin V.B. Psychiatry and medical psychology review 2010. 3. pp. 26-31.

2. Relationship between rapid changes in an individual subrange of the electroencephalogram ? wave and heart rate during sleep Vasil'ev E.N., Uryvaev Yu.V.Human Physiology. 2006. . 32. 4. . 389-393.

3. Zenkov L.R. Clinical electroencephalography (with epileptology elements ). .: MEDpress-inform, 2011. 356 p.

4. Analysis the method muscle biopotentials spectrograph Dzehtsiarou Y.G., Osipov A.N., Kovalev V.V., Kulchitskiy V.A. Military medicine. 2013. 3 (28). pp. 94-97.

5. Condition of vegetative nervous system in patients with atopic dermatitis Aksenova O. I., Marchenko V. N., Monakhov K. N. Modern clinical medicine bulletin 2014. Vol. 7. 4. pp. 15-17.

6. The electroencephalogram analysis which based on conversional locals maximums structure of wavelets-coefficients matrix Turovsky Ya. A, Kurgalin S. D., Maksimov A. V., Semenov A. G. Proceeding of Voronezh State University. System analysis and information technology. 2012. 2. pp. 69-73.

7. The electroencephalograms analysis on locals maximums scalegramms chains investigation base . Turovsky Ya. A., Kurgalin S. D., Semenov A. G. Digital signal processing 2013. 2. pp. 20-23.

8. Dynamics of local maxima chains in spectra of human electroencephalogram Turovsky Y.A., Kurgalin S.D., Semenov A.G. Biophysics. 2014. Vol. 59. 1. pp. 148-152.

9. TIME FACTOR AT IMPLEMENTATION OF THE CONTINUOUS WAVELET-TRANSFORMATION FOR THE ANALYSIS OF EEG SIGNALS Turovsky Ya. A., Kurgalin S.D., Vahtin A.A., Maksimov A.V. Information technology in engineering and manufacture. 2012. 2. pp. 61-66.

10. The Research of the Locals Maximums Dynamics in the Wavelet-Spectrums of the Brain Event-Related Potentials Turovski Ya. A., Kurgalin S.D., Semenov A.G Information technology. 2013. 10. pp. 46-50.

11.Vegetative regulation cardio-vascular system fetus and newborn, sustained chronic intrauterine hypoxia Turovski Ya. A.author's abstract dissertation / Voronezh state medical academy. Voronezh, 2005

12. http://www.ncbi.nlm.nih.gov/nuccore/EF032883.1?report=genbank&log

13. http://www.atlas-yakutia.ru/weather/climate_russia-III.html

14. http://www.finam.ru/


Non-intrusive estimation of noisy speech signal intellibility
A.I. Topnikov, e-mail: topnikov@gmail.com
M.S. Nesterov, e-mail: maximator89@mail.ru
S.A. Novoselov, e-mail: sergnovoselov@mail.ru
A.L. Priorov, e-mail: andcat@yandex.ru

P.G. Demidov Yaroslavl State University (YSU)

Keywords: : signal, intelligibility, non-intrusive estimation, denoising.

Abstract
The possibility of non-intrusive estimation with the reference method and subsystems, restoring the reference signal from the distorted was researched. Noise reduction method is this system in case of main distorting factor is the additive noise. A technique is based on the measures of speech intelligibility index SNR loss and noise reduction by Salart method was described.

The calculation of the SNRloss value for the estimated (noisy) signal and the signal, which is obtained by the denoising method proposed by Scalar and Filho, is the main idea for proposed non-reference version of the SNR Loss criterion. The signal is considered reference at the output of squelch method. This is acceptable, as the signal at the output of the denoising method is an estimation of reference signal.

Let`s analyze the dependence of non-intrusive values (which was obtained by the using a denoised signal as a reference and denoted as SNRloss) with the true values SNRloss, which was calculated by the using a pure signal as a reference. Six independent speech fragments were used for modeling. Each fragment was noised by additive white Gaussian noise (AWGN). 230 noisy versions were formed for each fragment: 5 noise implementations for each SNR in the range of -15 to 30 dB. During the simulation SNRloss values were measured for 1380 signals (6 phrases with 230 noise options for each). The set of points in scatter plot was distributed in a way that allows you to make an assumption about the possibility of a linear approximation of the dependence of SNRloss from SNRloss'. On the basis of available data and the method of least squares the relationship can be described by the equation of linear regression:

Coefficients of equation (1) have the following values: b1 = 0,8909; b0 = 0,043 for AWGN. There is a high coefficient of determination (above 0.99) which characterizes the link between true values SNRloss and non-intrusive estimates SNRloss, which was obtained by the substituting of values SNRloss' in (1), in a case of using linear regression.

To verify the proposed non-intrusive method of estimation values SNRloss additional modeling wac conducted. To ensure the reliability the set of speech fragments was selected from different from that set which were used in the first part of the research. Values SNRloss and SNRloss' (for different implementations of noise and SNR) was measured 5520 pairs for 24 speech fragments. The obtained values SNRloss' were substituted into equation (1). Thus, 5520 non-intrusive estimations of values SNRloss was founded. Comparison of non-intrusive estimations with true values of criterion was obtained by using reference method. Proposed method has sufficiently high accuracy of the of non-reference estimation.

When a linear approximation of the average is used, the absolute error value is about 0.008, and its maximum value is equal 0.036. The average value of the relative error is approximately 1.05 %, and its maximum value is 4.72 %. It should be noted that the magnitude of errors can be further reduced through the using a polynomial or piecewise linear approximations.

The obtained results indicate about the prospects of applying the proposed approach to non-intrusive values estimation of SNR Loss criterion. The next step was to test the applicability of the study of this approach to other types of noise, which are more frequently met in solving practical problems. For this purpose, a specialized database of noise "Noisex-92" was used. There are various types of real noise. Since the sampling frequency of speech signal was 8 kHz, the sampling frequency of noise records was also reduced to 8 kHz. The highest standard deviation value and lowest value of the coefficient of determination was observed in the case of finding the linear regression parameters for the noise "Speech babble".

When the modeling with different types of noise confirmed the viability of the proposed non-intrusive technique, the question of its further improvement arose. Analysis of the number of additions and multiplications, which are necessary for estimation of the speech intelligibility, showed that a significant portion of computing operations accounted for the fast Fourier transform (FFT) and inverse FFT (IFFT). However, the used scheme of calculations contains the operations that can was eliminated. However, it is extremely important to bring the parameters of spectral transformation in the block of noise reduction to those that was used in the method of SNR loss because they affect to the reliability of estimations of intelligibility. A proposed modified procedure has a simple structure by eliminating repetitive and mutually exclusive operations and contains one unit for calculating the fast Fourier transform.

The proposed modification was analyzed and compared with the original method by number of computing operations. It was revealed that this modification could reduce the number of operations by approximately 30.5 % according with the original technique.

In addition, the comparison of the procedure speed was made by modeling in Matlab. So a personal computer with the following configuration: Intel (R) Pentium (R) D 930 CPU 3.00GHz, RAM 3,00 GB 400MHz, 64-bit OS Windows 8 was used. The winning in time was 29.9% in the proposed modification.

As simulations have shown, that the proposed modification has a lower accuracy. For example, the mean absolute percentage error (MAPE) increased from approximately 1.3 % to 2.1%. The relative change seems significant in 1.6 times more, but in practice, the above degradation is acceptable. Furthermore, it is possible that further revision of the proposed modification will allow even closer approach to the original accuracy procedure.

Thus, the non-intrusive version for the measure of intelligibility of noisy speech signals SNR loss is proposed. It is based on the application of the original (reference) version of the method SNR loss, squelch method and pair regression. Accuracy of the proposed non-reference measure of intelligibility was explored in the case of impact by different types of noise to the speech signals. The results suggest about the relatively high accuracy of the proposed non-reference estimation methods (the average relative error is 1.05-3.55 %).

Also, a fast version of non-reference technique was presented. From the results of studies, it follows that the modification has a high accuracy, which is inferior to the original, while the speed increased due to a significant reduction in the number of performed computing operations. The proposed modified method can be used for automatic control systems, noise reduction, selection the transmission mode that provides the permissible level of intelligibility in communication systems.

References
1. Didkovskiy V.S., Didkovskaya M.V., Prodeus A.N. Acoustic assessment of speech communication channels. Kiev. Imex-LDT, 2008. 420 p.

2. Novoselov S.A., Topnikov A.I., Savvatin A.I, Priorov A.L. Speech denoising by non-local means // Digital signal processing. 2011. N. 4. pp. 23-28.

3. Collard J. A theoretical study of the articulation and intelligibility of a telephone circuit // Electrical Communication. 1929. V. 7. p. 168.

4. French N.R., Steinberg J.C. Factors governing the intelligibility of speech sounds // The journal of the Acoustical Society of America. 1947. V. 19. Is. 1. pp. 90-119.

5. Kryter K.D. Methods for the calculation and use of the articulation index // The Journal of the Acoustical Society of America. 1962. V. 34. Is. 11. pp. 1689-1697.

6. Kryter K.D. Validation of the articulation index // The Journal of the Acoustical Society of America. 1962. V. 34. Is. 11. pp. 1698-1702.

7. Ma J., Loizou P. SNR loss: a new objective measure for predicting the intelligibility of noise-suppressed speech // Speech Communication. 2011. V. 53. Is. 3. pp. 340-354.

8.Loizou P., Ma J. Extending the articulation index to account for non-linear distortions introduced by noise-suppression algorithms // Journal of the Acoustical Society of America. 2011. V. 130. Is. 2. pp. 986995.

9. Bykov Y.S. The theory of speech intelligibility and efficiency of radio communication Moscow-Leningrad: Gosenergoizdat, 1959. 350 p.

10. Pokrovskiy N.B. Calculation and measurement of speech intelligibility. Moscow: Svyaz'izdat, 1962. 390 p.

11. Sapozhkov M.A. Speech signal in cybernetics and communication. Moscow: Svyaz'izdat, 1963. 452 p.

12. Sapozhkov M.A., Mikhailov V.G. Vocoder communication. Moscow: Radio i svyaz, 1983. 248 p.

13. Houtgast T., Steeneken H.J.M. Evaluation of speech transmission channels by using artificial signals // Acta Acustica united with Acustica. 1971. V. 25. Is. 6. pp. 355-367.

14. Steeneken H.J.M., Houtgast T. A physical method for measuring speech transmission quality // The Journal of the Acoustical Society of America. 1980. V. 67. Is. 1. pp. 318-326.

15. Steeneken H.J.M., Houtgast T. Validation of the revised STIr method // Speech Communication. 2002. V. 38. Is. 3. pp. 413-425.

16. Prodeus A. On possibility of advantages join of formant and modulation methods of speech intelligibility evaluation // Proceedings of VII International Conference on Perspective Technologies and Methods in MEMS Design (MEMSTECH). 2010. pp. 254-259.

17. Prodeus A. Assessment of speech Intelligibility by formant-modulation method // Journal of Basic and Applied Physics. 2013. V. 2 Is. 5. pp. 10-18.

18. Falk T.H., Zheng C., Chan W.Y. A non-intrusive quality and intelligibility measure of reverberant and dereverberated speech // IEEE Transactions on Audio, Speech, and Language Processing. 2010. V. 18. Is. 7. pp. 1766-1774.

19. Santos J.F., Senoussaoui M., Falk T.H. An improved non-intrusive intelligibility metric for noisy and reverberant speech // 14th International Workshop on Acoustic Signal Enhancement (IWAENC). 2014. pp. 55-59.

20. Li F.F. Speech intelligibility of VoIP to PSTN interworking a key index for the QoS. // Proceedings of International Conference "Telecommunications Quality of Service: The Business of Success". 2004. pp. 104-108.

21. Chen F., Hazrati O., Loizou P. C. Predicting the intelligibility of reverberant speech for cochlear implant listeners with a non-intrusive intelligibility measure // Biomedical signal processing and control. 2013. V. 8. Is. 3. pp. 311-314.

22. Veselov I.A., Novoselov S.A. Topnikov A.I. A non-intrusive intelligibility measure of noisy speech // 15th International Conference "Digital Signal Processing and its Applications" (DSPA-2013). 2013. V. 1. pp. 256-259.

23. Scalart P., Filho J. Speech enhancement based on a priori signal to noise estimation // IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP-96). 1996. V. 2. pp. 629-632.

24. Topnikov A.I., Nesterov M.S. Modification of the procedure for non-intrusive speech intelligibility estimation // 16th International Conference "Digital Signal Processing and its Applications" (DSPA-2014). 2014. V. 1. pp. 208-212.


Quality Rating of Television (Thermovision) Coordinators in Observing and Security Systems
J.B. Zubarev, email: osa@mniti.ru
E.P. Arzoumanian, email: arzedouard@gmail.com
D.V. Mikolaichuk, email: amid_mik@mail.ru
Moscow Scientific Research Television Institute (MSRTI) Russia, Moscow


Keywords:
auto tracking system, optoelectronic device, self-regulated circuit.

Abstract
One approach to estimating various characteristics of television (thermovision) auto tracking systems is treated. The solution also provides the area observation within a visual angle of a mobile optoelectronic device.

Stationary surveying and security systems having features of tracking and registration an observed data became widespread. The design of such systems for local event operational surveying is an interesting line of mobile devices development. There are basic requirements such as long distance observation which is met by usage of long-focus lens and wide angled view which is fit by electromechanical drive having high Q-factor.

The article considers double-loop self-regulated circuit which has first order astaticism. It consists of the electronic loop for quick angular displacement response and the external camera drive loop for the wide field of view. Tracking circuits with higher order astaticism are quite sensitive for the fluctuations of camera drive signal, which represent discrete feature of raster image.

The mathematical model has developed in terms of basic system performance criterions. Results with taken into account the influence of image resolution and external actuating factors and different modes of internal tracking circuit are presented. The estimate of tolerant angle velocities and accelerations is given.

References
1. Kaganov V.I. Radio-technical circuits and signals. Computer-aided course: Tutorial // Moscow, Forum: INFRA-M, 2005, 432 p. (Higher education).

2. Alpatov B.A., Babajan, P.V., Balashov O.E., Stepashkin A.I. Methods of automated detection and object tracking. Image processing and control // Radio engineering, 2008, 176 p.

3. Voronin S.G. Electrical drive of aircrafts / Training-methodical set. Version 1.0 // Chelyabinsk, 19952011, 489 files, il., http://epla.susu.ac.ru/vsg_udk.htm.

4. Babenko K.I. Foundation of the numerical analysis. 2 ed., cor. and aug. // Moscow- Izhevsk, RDC Regular and chaotic dynamics 2002, 848 p.

5. Baskakov S.I. Radio-technical circuits and signals // Moscow, Higher school, 1988 448 p.

6. Besekerskiy V.A. Dynamic synthesis of automatic self-regulated systems // Moscow, Science, 1970.


Stochastic Filtering of Periodic Bipolar Signal
V.I. Nefedov, email: nefedov@mirea.ru
S.A. Reshetnyak, email: reshets@bk.ru
G.N. Tretyakov, email: gennady.tretyakov@rambler.ru


Keywords: signal/noise ratio, stochastic filtration, nonlinear radio-engineering filter, meandering signals.

Abstract
Weak signals selection on the intensive noise background is currently a central problem. One of the ways of its solution is to study the processes of signal and noise interaction in nonlinear systems. The importance of these investigations is due to the recently opened effects of stochastic resonance (SR) [1-3] and stochastic filtration (SF) [4]. These effects are described by the equation similar to the one, describing a Brownian particle motion in a particular potential field while being exposed to the deterministic signal and almost white noise. The effect of SR is observed in the systems with a bistable potential function W and the harmonic signal of low power as compared with the noise dispersion . It occurs in the setting of the system where the signal frequency coincides with the average frequency of the crossings over the potential barrier under the noise action. As a result of SR affect, the local signal/noise ratio at the nonlinear system output in the area of its parameters increases as increases because of the noise energy transfer into the signal. According to the studies conducted [1-2] it has been shown that when the SR effect is observed the ratio does not exceed the similar ratio at the system input.

If the signal power is equal or more than noise dispersion in the nonlinear systems, the SF effect may be achieved resulting in . In terms of the selection of weak signals on the background noise this effect is more interesting than the SR one. This was first discovered as a result of the numerical analysis [4] of the stochastic differential equation in the case of the existence of the harmonic signal and the potential of critical species. The system filtering properties were evaluated using the transfer coefficient in relation to the S/N. The theoretical analysis [5] has shown that SF occurs as a result of the noise suppression by the signal. The SF effect was also exhibited by the analog simulation [6-8] for the acoustic signals of the rectangular shape. The results [4] have been experimentally confirmed in [9] for the nonlinear radio-engineering low-frequency filter of the first order. In [10] SF was also found in the nonlinear radio-engineering filter of the second order. In [11] the developed numerical method has shown that SF takes place in the filters of both low and high signal frequencies.

In this paper, based on the [11] approach developed, the SF effect is analyzed when the signal spectrum has not one but a large number of harmonics. As such a signal, the periodic sequence of rectangular pulses, each period of which contained two bipolar pulses of the same amplitude and duration was chosen. The transfer coefficient was calculated on the basis of the integral over spectrum relations: , where and - modules of the signal and noise harmonics complex amplitudes. The results obtained were compared with the data for the linear system. The study carried out has led to the following conclusions. The main regularities of the SF effect in the case of the harmonic signal are also saved in the case of meandering signals. The numerical analysis of the low-frequency nonlinear filter let us to define the filter parameters area, in which its filtering ability is substantially higher than the similar linear filter has. The q transfer coefficient dependences in reference to the signal/noise ratio show the existence of the signal optimal frequencies, its amplitude and noise power for reaching the highest q values. The optimal conditions of the nonlinear filtering can be achieved for the signals of the power being not significantly greater than the noise dispersion.

References

1. Gammaitoni L., Hanggi P., Jung P., Marchesoni F. Stochastic resonance. Rev. of Mod. Phys., 1998, v.70, 1, pp. 223-287.

2. Anishchenko V.S., Neyman A.B., Moss F., Shimanskiy-Guyer L. Stohasticheskiy rezonans kak indutcirovanniy shumom effekt uvelicheniya stepeni poryadka. UFN, 1999, v.169, 1, pp.7-38.

3. Climontovich Yu.L. Chto takoe stohasticheskaya filtratciya i stohasticheskiy rezonans? UFN, 1999, v.169, 1, pp. 39-47.

4. Hanggi P., Inchiosa M.E., Fogliatti D., Bulsara A.R. Nonlinear stochastic resonance: The saga of anomalous output-input gain. Phys. Rev. E., 2000. v.62, 5, pp. 6155-6163.

5. Reshetnyak S.A., Tretyakov G.N. Teoreticheskoe issledovanie effekta stohasticheskoy filtratcii. Radiotekhnika i elektronika, 2013, v. 58, 4, pp. 360-366.

6. Gingl Z., Makra P., Vajtai R. High signal-to-noise ratio gain by stochastic rezonanse in a double well. Fluctuation and Noise Lett., 2001,v.1, No.3, pp. L181-L188.

7. Makra P., Gingl Z., Kish L.B. Signal-to-noise ratio gain in non-dinamical and dinamical bistable stochastic resonator. Fluctuation and Noise Lett., 2002, v.2, No.3, pp. L147-L155.

8. Makra P., Gingl Z., Fulei T. Signal-to-noise ratio gain in stochastic resonators driven by coloured noises. Fhys. Lett., 2003, v.317, No.3-4, pp. L228-L232.

9. Dombrovskiy A.N., Reshetnyak S.A. Stohasticheskiy rezonans i filtratciya signalov v nelineynoy elektricheskoy sisteme vtorogo poryadka. Radiotekhnika, 2007, 9, pp. 19-25.

10. Dombrovskiy A.N., Reshetnyak S.A. O stohasticheskoy filtratcii signalov v nelineynykh elektricheskikh sistemakh. Radiotekhnika i elektronika, 2009, v.54, 11, pp. 1369-1371.

11. Aboelazm M.A., Melchakov V.N., Reshetnyak S.A., Tretyakov G.N. Issledovanie tcifrovoy modeli nelineynogo aktivnogo filtra pervogo poryadka. Tcifrovaya obrabotka signalov, 2014, 4, pp. 62-64.


Local binary patterns median pixel's effective technology of pattern recognition
I.N. Efimov, Samara State University of Transport Communications (SamGUPS), Russia, Samara,
email: Mr.Efimov.IN@gmail.com

Keywords: computer vision, local binary patterns, object recognition, information technology.

Abstract
Face recognition algorithms commonly assume that face images are well aligned and have a similar pose yet in many practical applications it is impossible to meet these conditions. Therefore extending face recognition to unconstrained face images has become an active area of research. To this end, histograms of Local Binary Patterns (LBP) have proven to be highly discriminative descriptors for face recognition. Nonetheless, most LBP-based algorithms use a rigid descriptor matching strategy that is not robust against pose variation and misalignment.

In this paper author, propose the new informative features local binary patterns median pixel's (LBPMP). The author changed the shape and the algorithm for constructing local binary patterns. Author compared the advantages of the new feature in relation to pattern recognition.

The face area is first divided into small regions from which Local binary patterns median pixel's (LBPMP) histograms are extracted and concatenated into a single histogram efficiently representing the face image. The recognition is performed using a nearest neighbor classifier in the computed feature space with function histograms intersection as a dissimilarity measure. Extensive experiments clearly show the superiority of the proposed scheme other local binary pattern on face database tests, which include testing the robustness of the method against different facial expressions, lighting and aging of the subjects. In addition to its efficiency, the simplicity of the proposed method allows for very fast feature extraction.

Features will find their application in various fields of computer vision, such as recognizing traffic signs, identification numbers of vehicles, etc. The presented algorithm for computing informative features can be modified: add new fragments or change the location of the fragments.

References
1. Efimov I.N. Methods to improve the reliability of the identification of students in distance learning / I.N. Efimov // Distance and virtual learning 2013. 3 6273.

2. Efimov I.N. .. / I.N. Efimov, . . // Vestnik transporta povolzhya 2013. 1 6266.

3. Ahonen T. Face Recognition with Local Binary Patterns , 2004. 469481.

4. Ahonen T. Face description with local binary patterns: Application to face recognition / T. Ahonen, A. Hadid, M. Pietikainen // IEEE Trans. Pattern Anal. Mach. Intell. 2006. . 28 12 20372041.

5. Jain V. The Indian face database / V. Jain, A. Mukherjee 2002.

6. OConnor B. Facial Recognition using Modified Local Binary Pattern and Random Forest / B. OConnor, K. Roy // Int. J. Artif. Intell. Appl. 2013. . 4 5.

7. Ojala T. Multiresolution gray-scale and rotation invariant texture classification with local binary patterns / T. Ojala, M. Pietikainen, T. Maenpaa // IEEE Trans. Pattern Anal. Mach. Intell. 2002. . 24 7.

8. Phillips P.J. The facial recognition technology (FERET) database / P. J. Phillips // IEEE Trans. Pattern Anal. Mach. Intell. 2004. . 22.

9. Trefny J. Extended set of local binary patterns for rapid object detection / J. Trefny, J. Matas // Proc. Comput. Vis. Winter 2010.

10.Weber M. Frontal face dataset, california institute of technology // 1999.

Application of Fourier Transform to the Problem of Distance Measurement by Frequency-Modulated Rangefinder in the Presence of Dispersion
V.M. Davydochkin, LLC enterprise KONTAKT-1, Russia, Ryazan, email: skb@kontakt-1.ru

Keywords: frequency-modulated rangefinder, dispersion, spectrum distortion, measurement error, adaptive weighting functions, Nyquist frequency.

Abstract
Many modern industrial control systems require accurate measurements of liquid level in reservoirs. This article considers the task of measuring the level of a liquid in a partially submersed waveguide by means of frequency-modulated rangefinder. Level measurement consists in measuring the length of a part of perforated waveguide that is free of liquid. The chief problem faced by these methods of measurement is the increased error due to waveguide dispersion. Dispersion causes the spurious frequency modulation of the beat frequency during the linear frequency modulation of the transmitter. However, the average value of the beat frequency depends linearly on the distance to the probed liquid surface in the waveguide, which makes it possible to measure the distance based on somehow estimated average frequency.

Frequency of the beat signal that corresponds to the maximum of the amplitude spectral density is frequently taken as the frequency estimate. The main drawbacks of this estimate, stemming from the systematic error of Fourier transform, can be mostly eliminated by smoothing weighting functions (WF), and by increasing the range of frequency modulation. However, the distortions caused by dispersion eliminate the advantages of greater modulation range and reduce the effectiveness of the WF.

This work proposes a variant of integral-discrete Fourier transform that eliminates the negative influence of the dispersion.

Taking into account that many WF, including those that can not be represented in terms of elementary functions, may be represented by the adaptive WF (AWF) or approximated by the AWF to any specified precision, we analyze the properties of the proposed transform using the AWF. One of the spectral properties of AWF is that in the absence of the dispersion and noise it theoretically allows to achieve zero error in the estimate of the frequency of a signal.

Analysis of the spectrum distortions of a signal containing samples had shown that when the proposed transform is used, the main lobe at the frequency and the sidelobe at the frequency are not distorted. At the same time, the sidelobes at the normalized frequencies , are strongly distorted and have diminished amplitudes. The distortions of the consecutive sidelobes increase. Also, the spectral properties of AWF are preserved if a signal's frequency are below the Nyquist frequency by 20 40%.

Another feature of the proposed transform is the possibility of estimating a frequency above the Nyquist limit due to the significant reduction in the amplitudes of spectral sidelobes at the normalized frequencies , and the consecutive ones. Moreover, in the absence of noise and waveguide losses estimating a signal's frequency may be theoretically possible for the average frequency of the beat signal that is much greater than the Nyquist frequency.

References
1. Atayants B.A., Davydochkin V.M., Ezerskiy V.V., Parshin V.S., Smolskiy S.M. Precision FMCW Short-Range Radar for Industrial Applications.- Boston/London, Artech House, 2014. 320 p.

2. Goldshtein L.D., Zernov N.V. Electromagnetic fields and waves. 2nd ed., M.: Sov.Radio 1971. - 664 p.

3. Davydochkin V.M. Methods of reducing the error of estimating the parameters of polyharmonic signals in near-distance frequency modulated radar rangefinding. // Notices of RSREU. Ryazan, RSREU,2006. vol. 18. pp. 63-70.

4. Davydochkin V.M., Davydochkina S.V. Weighting functions for adaptive harmonic analysis of signals with multimodal spectrum. // DSP. 2008. 4. pp. 44-48.

5. Davydochkin V.M., Davydochkina S.V. Weighting functions for digital adaptive harmonic signal analysis // Radiotechnics. 2009. 9. . 11-20.

6. Pat. 2435168 RF, G01R 23/16. A method of harmonic analysis of a periodic multifrequency signal / Davydochkin V.M., Davydochkina S.V. / Applied 09.04.2010, Published 27.11.2011 Bulletin 33.



If you have any question please write: info@dspa.ru