Digital Signal Processing

Russian
Scientific & Technical
Journal


Digital Signal Processing No. 4-2014

In the issue:

- multichannel processing

- empirical mode decomposition
- extremal filtration
- low-density parity check codes
- hybrid voice activity detector
- adaptive clutter suppression
-radiovision and radio monitoring
- technology of CUDA



Weighted Overlap-add Algorithm for Processing Vector Signals in Radio Monitoring Tasks
D.M. Klionskiy, e-mail: klio2003@list.ru
D.I. Kaplun, e-mail: mitya_kapl@front.ru
A.S. Voznesenskiy, e-mail: a-voznesensky@yandex.ru
V.V. Gulvanskiy, e-mail: slava-a-a@mail.ru
Saint Petersburg Electrotechnical University "LETI" (SPbETU "LETI"), Russia, Saint Petersburg

Keywords: weighted overlap-add algorithm, wide frequency band, multichannel signal processing, software-hardware implementation, CUDA technology.

Abstract
The paper is devoted to the development of the weighted overlap-add algorithm (WOLA-algorithm) for processing vector signals in radio monitoring tasks. The processing is performed in the wide frequency band. The algorithm is aimed at operating in radio engineering equipment during the real-time signal processing. The suggested algorithm is compared with the polyphase implementation of a multichannel filter bank. Software-hardware implementation of multichannel signal processing is discussed. We also show the advantages of applying the compute unified device architecture (CUDA) based on computations using graphical processors.

The term "monitoring" means a systematic or continuous data acquisition from some object or system. Monitoring is often performed in a wide frequency band (wideband monitoring) and includes tracking the main system parameters and deviation search in these parameters. Any deviation is usually caused by abnormal functioning of an object and requires immediate action in order to prevent failures.

Wideband monitoring is frequently used in the following areas:

  • hydroacoustics monitoring of water areas (coastal and marine waters), underwater and surface objects, study of navigation canals and near-shore waters, analysis of oceanic seismicity, etc.;
  • radio monitoring time-frequency processing of radiations, noise suppression, signal classification, signal parameter estimation, signal demodulation, direction-finding, etc.;
  • vibration analysis vibration measurements, vibration parameter estimation in the time and frequency domains, resonance frequency estimation; analysis of vibrations is important for spacecraft, aircraft, engines, turbines, machinery, etc.;
  • geophysics monitoring of seismic and geomagnetic activity in various regions of our planet.
The paper is devoted to considering radio monitoring tasks.

We illustrate the performance of WOLA for processing vector signals in real time using the vector discrete Fourier transform. The proposed algorithm can be applied to direction-finding, signal detection in radio channels, sikgnal classification and subsequent signal parameter estimation in the time and frequency domains. As a result of the algorithm comparison with the multichannel polyphase filter bank it has been demonstrated that WOLA has a wider application range due to an arbitrary relation between the number of filter bank channels and decimation factor. We have also made it clear that software-hardware implementation based on CUDA allows us to reduce the amount of time for multichannel signal processing in comparison with CPU.

References

1. A.M. Rembovskiy, A.V. Ashihmin, V.A. Kozmin Radio monitoring: tasks, methods, and tools // 2-nd edition, Moscow: "Telecom", 2010.

2. W. Hirt Ultra-Wideband Radio Technology: Overview and Future Research, Comp. Commun., vol. 26, no. 1, Jan. 2003, pp. 4652.

3. D.I. Kaplun, D.M. Klionskiy, A.L. Oleynik, A.S. Voznesenskiy, N.A. Zhukova, V.V. Gulvanskiy, A.A. Petrovsky Application of polyphase filter banks to wideband monitoring tasks // Izvestia vuzov Rossii. Radio electronics. 2013. 3. pp. 38-43.

4. R. E. Crochiere, L. R. Rabiner Multirate digital signal processing // Prentice Hall, 1983.

5. P.P. Vaidyanathan Multirate Systems and Filter Banks // Prentice Hall. Englewood Cliffs.- NJ, 1993.

6. V.V. Vityazev, S.V. Vityazev, A.A. Zaytsev Multirate signal processing: retrospective review and modern condition (part 1) // Digital signal processing, 1, 2008, pp. 12-21.

7. A.A. Zaytsev Techniques of filter bank construction: subject survey // Digital signal processing. 2003. 1. C. 2-10.

8. V.V. Vityazev Digital frequency selectiv signals. Moscow, Radio I svyaz, 1993. 240 pp.

9. Bellanger M.G., Daguet J.L., Hepagnol G.P. Interpolation, extrapolation and reduction of computation speed in digital filter // IEEE Trans. Acoust., Speech and Signal Processing. V. ASSP-22. Aug., 1974. P. 231-235.

10. Mitra S.K. Digital Signal Processing: a computer-based approach. McGraw-Hill. Comp. Inc., 1998.

11. Rabiner L.R., Crochiere R.E. A novel implementation for narrowband FIR digital filters // IEEE Trans. Acoust., Speech and Signal Processing. V. ASSP-23. Oct., 1975. P. 457-464.

12. Bellanger M.G., Bonnerot G., Coudreuse M. Digital filtering polyphase network: Application to sample rate alteration and filter banks // IEEE Trans. Acoust., Speech and Signal Processing. V. ASSP-24. Apr., 1976. P. 109-114.

13. V.I. Gadzikovskiy Computation of the qyantization error of vector digital filters // Digital signal processing, 2005, 4. pp.24-28.

14. V.I. Gadzikovskiy Complex and vector digital filters // UGTU-UPI "New techniques of information transmission and processing". Ekaterinburg: UGTU-UPI, 2003, pp. 141-155.

15. Al. A. Petrovskiy, A.V. Stankevich, A.A. Petrovskiy Fast design of multimedia systems based on the prototype // Minsk, Bestprint, 2011, 412 pp.

16. J. Sanders, E. Kandrot CUDA by Example: An Introduction to General-Purpose GPU Programming // Addison-Wesley Professional, 2010.


Extraction Method of Spectral Components in Signals by Interpolation via Systems of Integer Shifts
L.A. Minin, e-mail: mininla@mail.ru
Nihad Mahnoud Nasir, e-mail: nihadalnasir1@yahoo.com

E.A. Kiselev, e-mail: evg-kisel2006@yandex.ru
S.D. Kurgalin, e-mail: kurgalin@bk.ru
Voronezh State University, Russia, Voronezh,
University of Diyala, Iraq, Baqubah

Keywords: signal, spectrum, interpolation, algorithm, integer shift, Lorentz line shape.

Abstract

The paper is devoted to the development of the new method for extraction of spectral components in signals that can be modeled by Lorentz line shape. The method is based on the interpolation via systems of integer shifts. The basic concept of such type of interpolation is the node function. The node function in our case is constructed as a linear combination of the Lorentz line shape integer shifts. In our earlier works we have obtained an analytical formula for coefficients of this linear combination. In this paper we discuss computational specificities of the interpolation and its application for digital signal processing.

Lorentz line shape is often used in modeling of different processes and phenomena in physics. As an example we use atomic and molecular spectrums. Often for estimation of the algorithm resolution capability empirical Rayleigh criterion is used. When there is additional information about spectral lines, the resolution capability can be improved as we show in this paper.

The main parameter of the Lorentz line shape is its width. In the paper it is shown that increasing of the Lorentz function width leads to the unstability of interpolation. The theoretical limit of the stability is about 14 discrete points per Lorentz function at half-height. Our experimental calculations confirm this result. Developed in the paper algorithm in this case gives computational precision 0,1% for the signal reconstruction (measured in the characteristic amplitude of peaks in the signal) and allows to extract peaks distant from each other by one step of the discrete grid.

In the paper we also perform experiments with additive random noise. It gives practical estimation of the method precision and its limits of application. Computational results exhibit strong dependence of algorithm resolution capability on the noise level. For example when the noise amplitude reaches 5% (measured in the characteristic amplitude of peaks in the signal) the limit of the algorithm stability falls to 4 discrete points per Lorentz function at half-height. When the noise grows higher the direct application of the algorithm becomes impossible, because for stable extraction of the spectral lines at least 4-5 discrete points per peak are required. Therefore for application of suggested interpolation method for extraction of spectral lines in presence of noise a preprocessing should be carried out. A perspective method in our opinion is preliminary removal of noise by Daubechies wavelets to provide the stability of the interpolation algorithm.

References
1. Born M., Wolf E. Principles of Optics / M. Born, E. Wolf // Moscow, Nauka, 1973, 720 p.

2. Griem H. Broadening of the Spectral lines in Plasma / H. Griem // Moscow, Mir, 1978, 492 p.

3. Drobyshev A.I. Basics of the Atomic Spectral Analysis / A.I. Drobyshev // Saint-Petersburg, Saint-Petersburg University Publisher, 1997, 200 p.

4. Mazya V. Approximate approximations / V. Mazya, G. Schmidt // AMS Mathematical Surveys and Monographs, V. 141, 2007, 350 p.

5. Novikov I.Ya., Protasov V.Yu., Scopina M.A. Wavelet theory / I.Ya. Novikov, V.Yu. Protasov, M.A. Scopina // Moscow, Fizmatlit, 2005, 616 p.

6. Kiselev E. A., Minin L. A., Novikov I. Ya., Sitnik S. M. On the Riesz Constants for Systems of Integer Translates / E.A. Kiselev, L.A. Minin, I. Ya. Novikov, S.M. Sitnik // Mathematical notes, V. 96, 2, 2014, pp. 228-238.

7. Comparative analysis of efficiency of the ECG signal compression by the Daubechies wavelets and the Discrete cosine transform / S.D. Kurgalin, L. A. Minin, E. A. Kiselev, Nihad Nasir // Control Systems and Information Technologies, 3.1(45), 2011, pp. 177-180.


Empirical Mode Decomposition on the Basis of Extreme Filtration
MyasnikovaN.V., e-mail: genok123@mail.ru
Beresten M.P., e-mail: beresten@sura.ru

Keywords: extreme filtering, alternating components, empirical mode decomposition

Abstract
Extremal filtration methods are considered. The method proposed is based on successive selection of the high-frequency components which are locally determine by signal's extrema. It is shown that this method is similar to Empirical Mode Decomposition and has some improvements compared with it.

Theoretical basis of an extremal filtration method are given. Coefficients of two-pole filter are determined. It is proposed to use values of power of two as filter parameters for faster computing.

It is shown that selected components has properties of the empirical modes: sign alternating; zero moving average.

Forms of representation of an extreme filtration results are offered. Decomposition parameters, frequency and amplitude (power) of each component of a signal can be interpreted as the spectrum, and change of this parameters in the sliding window can be interpreted as time-frequency distribution.

It is shown that the used basis of decomposition is finite, convergent, orthogonal and unique.

Extremal filtration method is primarily intended for the express analysis of signals in diagnostic and recognition systems.


References

1. Klionskiy D.M. Empirical Mode Decomposition in the modern digital signal processing.

2. Klionskiy D.M., Oreshko N.I., Geppener V.V. Empirical Mode Decomposition and its use in the analysis of fractional Brownian motion.

3. Klionskiy D.M., Oreshko N.I., Geppener V.V. Empirical Mode Decomposition with parabolic interpolation of envelopes in tasks of cleaning of signals from noise.

4. Myasnikova N.V., Beresten M.P. Extremal digital filtration and its applications.

5. Myasnikova N.V., Beresten M.P., Dolgikh L.A. Methods of signal decomposition based on extreme filtering.

6. Dolgikh L. A., Myasnikova N. V., Myasnikova M. G. Use of empirical modes decomposition in digital signal processing problems.

7. Myasnikova N.V., Beresten M.P. Time-frequency distributions on the basis of extreme filtering.

8. S.L. Marple, Jr. Digital Spectral Analysis with Applications.

9. Myasnikova N.V., Beresten M.P., Stroganov M.P. Multiextremal function approximation and its applications in technical systems.

10. Myasnikova N.V., Beresten M.P. Express-analysis in the technical and information system.

11. Zenov A.U., Beresten M.P. The concept of organization of processing of the information in the systems of diagnostics and detection.


Frequency Domain Data-Aided Channel Estimation for OFDM Signals
Sarana D.V., e-mail: Dmitry.Sarana@gmail.com
Irtyuga V.A., e-mail: virtuga@mail.ru
SAD-COM Ltd. Russia, Moscow


Keywords: OFDM, channel estimation, data-aided estimation, MAP, EM, Wiener filter.

Abstract
The paper is devoted to low-complexity channel estimation algorithm for OFDM signals in frequency domain. Algorithm uses initial channel estimation based on pilot subcarriers, 1-D adaptive Wiener filtering in the time direction and Maximum a-posteriori probability (MAP) estimation that uses both pilot and weighted data subcarriers. The final step is the simplified version of an iterative Expectation-Maximization (EM) algorithm that uses only single iteration to reduce total computational complexity.

Proposed channel estimator is designed for OFDM signal and evaluates the complex frequency response (CFR) of the channel with selective fading. Doppler spread estimation is out of the scope of this paper. At the receiver side pilot pattern and modulation type are supposed to be known.

The channel is assumed multipath with frequency selective fading and an arbitrary dynamics in the time direction. The channel CFR varies smoothly within the signal bandwidth (limited by the number of reflections). The channel CFR is not changed during the symbol length. CFR dynamics (from symbol to symbol) is determined by the receiver speed and supposed unknown.

Channel model is linear. The noise of the receiver input amplifier is assumed as a complex Additive White Gaussian Noise (AWGN) with equal power for all subcarriers and for all symbols within the interval being analyzed.

The paper deals with the data processing at the receiver side after synchronization, removing of a guard interval and transition to the spectral domain for each OFDM symbol.

For the experiment the multipath channel model with 8 paths and 60 km/h receiver speed (channel #8 [6]) was used. An additive noise was generated separately. So an "ideal" channel CFR was known. It allowed the calculation of the estimation accuracy for various signal-to-noise ratios.

The proposed algorithm has a significant (up to 3.6 dB) gain in channel estimation accuracy in comparison with the evaluation based on pilot points only (for 10..18 dB SNR range) at the acceptable computational costs.

References
1. P. Hoeler, S. Kaiser, and P. Robertson, Two-dimensional pilot-symbol-aided channel estimation by Wiener filtering, in Proc. IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Munich, Germany, Apr. 1997, pp. 18451848.

2. M. Necker, F. Sanzi, J. Speidel, An Adaptive Wiener-Filter for Improved Channel Estimation in mobile OFDM-Systems, International Symposium of Signal Processing and Information Technology, IEEE 28 30 December 2001, pp. 213216.

3. Stephan Sand, Armin Dammann, Gunter Auer, Adaptive Pilot Symbol Aided Channel Estimation for OFDM Systems, ISBN: 1-4020-1837-1 In book: Multi-Carrier Spread-Spectrum, For Future Generations Wireless Systems, Publisher: Kluwer Academic Publishers, Editors: null Fazel, K. Kaiser, St Source: DLR

4. Auer, G. ; DoCoMo Euro-Labs, Munich, Germany; Karipidis, E., "Pilot Aided Channel Estimation for OFDM: a Separated Approach for Smoothing and Interpolation", Communications, 2005. ICC 2005. 2005 IEEE International Conference, Volume:4

5. Huang M, Chen X., Xiao L., Zhou S., and authors, "Kalman-filter-based channel estimation for orthogonal frequency-division multiplexing systems in time-varying channels", Communications, IET (Volume:1, Issue:4)

6. ETSI ES 201 980 V4.1.1 (2014-01). Digital Radio Mondiale (DRM); System Specification. (Annex B.2)

7. Report ITU-R BT.2295-0 (12/2013). Digital terrestrial broadcasting systems.

8. GOST R 543092011. Realtime audiovisual information system (RAVIS). Framing structure, channel coding and modulation for digital terrestrial narrowband broadcasting system for VHF band. Technical specification.


Balance Incomplete Block Designs Based LDPC Codes
A.A. Ovinnikov, Ryazan State Radioengineering Univesity, e-mail: ovinnikovalexey@gmail.com
V.V. Vityazev, Ryazan State Radio Engineering University, e-mail: vityazev.v.v@rsreu.ru


Keywords:
channel coding, iterative decoding, low-density parity-check codes, balance incomplete block designs, Steiner systems.

Abstract
The paper is devoted to the development of the low redundancy LDPC codes [1] for high speed data transmission. Code construction methods are based on combinatorics and related subsections. The effectiveness of the obtained code is evaluated using Monte Carlo simulations, as well as by calculating the number of fundamental parameters inherent LDPC codes.

We consider codes, synthesized on the basis of the so-called balanced incomplete block designs (BIBD) [2,3] (ν, k, λ) with different length and rate. The structures of BIBD with λ=1 ensures that no cycles of length 4 are considered in the Tanner graph of LDPC code. Thus, it is the first substantial limitation to the BIBD, which leads to a narrowing set of flowcharts to a subset of the so-called Steiner systems (ν, k, 1) ) [3]. Among all Steiner systems of particular interest is a subset of the cyclic block designs with different number of elements ν = (3,4,5, ...).

The proposed synthesis algorithm of the LDPC code consists of 3 stages. At the first of them Rosa or Skolem combinatorial sequences with modifications, if needed, are created. Sequence length and configuration are selected from input code parameters such as length and rate. At the second stage Steiner system is generated from current combinatorial sequence. At last LDPC code is created from incidence matrix of Steiner sequence from previous stage.

In the paper we create low-density parity check codes from balanced incomplete block desings, which in turn are based on combinatorial sequences Skolem and Rosa. Our codes shown excellent results in comparison with subclass of pseudorandom codes [7], the resulting energy decoding gain lies in the range from 0.7 to 2.56 dB. Generalized synthesis algorithm proposed in this paper, allows to obtain an ensemble of quasi-cyclic LDPC codes with different code lengths and rates, potentially expands the boundaries of using of such structures. One of the areas for further research is to find the most effective codes within the synthesized ensemble.

References
1. R. G. Gallager. Low-Density Parity-Check Codes. Cambridge MA:MIT Press, 1963.

2. W. E. Ryan and S. Lin. Channel Codes. Classical and Modern, Cambridge University Press, 2009.

3. M. Hall, Combinatorial theory, Publishing House MIR, M. 1970.

4. S. J. Johnson and S. R. Weller. Regular low-density parity-check codes from combinatorial designs. In Proc. IEEE Information Theory Workshop (ITW2002), pages 9092, Cairns, Australia, September 2001.

5. T. J. Richardson, M. A. Shokrollahi, and R. L. Urbanke. Design of capacity approaching irregular low-density parity-check codes. IEEE Trans. Inform. Theory, 47(2):619637, February 2001.

6. C. J. Colbourn and A. Rosa. Triple Systems. Oxford University Press, 1999.

7. D. J. C. MacKay. Good error-correcting codes based on very sparse matrices. IEEE Trans. Inform. Theory, 45(2):399431, March 1999.

8. D. M. Arnold, E. Eleftheriou, and X. Y. Hu, Progressive edgegrowth Tanner graphs, in Proc. IEEE Global Telecommun. Conf., San Antonio, TX, Nov. 2001, vol. 2, pp. 9951001.

9. C. E. Shannon. A mathematical theory of communication. Bell Sys. Tech. J.,27:379423, 623656, JulyOct. 1948.

Perspectives of HF-Band Radio CommunicatioN MIMO-Enabled Complexes, Using Spatial and Polarisation Separation
Bukashkin S.A., e-mail: fgup@niia.ru
Ogloblin A.V., e-mail: homealeks@mail.ru

Keywords: HF-band radio communication, multiple input multiple output (MIMO), spatial and polarization separation.

Abstract
This article describes the upsides and downsides of HF-band radio communications. Ways to fix few of the downsides using MIMO technology are listed. Layouts of MIMO radio links and MIMO subchannels are discussed. It is shown that in the end channel speed is limited not only by channel matrix and signal/noise ratio but by mutual impedance of receive and transmit antennas as well. When spatial separation is used impedance could be lowered by increasing distance between antennas, while polar separation necessities taking into account mast and underlying surface influence.

It is suggested to calculate mutual impedance by electrodynamics analysis, and use it to estimate probability of channel speed falling below required minimum.

Antenna selection for MIMO application should be done based on these results optimizing complexity/effectiveness.

References
1. Golovin O.V., Prostov S.P. Sistemy i ustrojstva korotkovolnovoj radiosvjazi / Pod red. professora O.V. Golovina. M.: Gorjachaja linija Telekom, 2006.

2. Minkin M.A. Problemy i perspektivy modernizacii i razvitija sistem DKMV radiosvjazi // Vestnik SONIIR. 2006. 4(14). pp. 4-10.

3. Buzov A.L., Suharev A.S. Voprosy sozdanija universal'nyh bystrorazvorachivaemyh kompleksov tehnicheskih sredstv DKMV radiosvjazi // Vestnik SONIIR. 2006. 2(12). p.12.

4. Suharev A.S. The estimation of efficiency to apply the space-time encoding OFDM for duplex transmission the data in HF radio channel // Radiotehnika (zhurnal v zhurnale). 2006. 10.

5. Erceg, V.; Soma, P.; Baum, D.S.; Catreux, S., "Multiple-input multiple-output fixed wireless radio channel measurements and modeling using dual-polarized antennas at 2.5 GHz," // Wireless Communications, IEEE Transactions on , vol.3, no.6, pp.2288,2298, Nov. 2004.

6. Erceg, V.; Soma, P.; Baum, D.S.; Paulraj, AJ., "Capacity obtained from multiple-input multiple-output channel measurements in fixed wireless environments at 2.5 GHz," // Communications, 2002. ICC 2002. IEEE International Conference on , vol.1, no., pp.396,400, 2002

7. Doluhanov M.P. Rasprostranenie radiovoln; izd. 4-e. M: Svjaz', 1972. 336 p.

8. Salous, S.; Feeney, S. M.; Warrington, E.M.; Gunashekar, S.D.; Abbasi, N. M., "Experimental investigations of MIMO in the HF band," // Ionospheric Radio Systems and Techniques (IRST 2012), 12th IET International Conference on , vol., no., pp.1,4, 15-17 May 2012

9. Kolchugin I.Y. Radiating ring and multi-ring antenna array based on biorthogonal emitters // Radiotehnika. 2014. 4. pp.60-63.

10. Kapishev A.N., Krasilnikov A.D., Nevskij A.V. Research of Complex of Active Receiving HF-Band Antennas with Operated Spatial and Polarizing Characteristics // Antenny. 2012. 6. pp.57-63.

11. Habib, A, "Multiple polarized MIMO with antenna selection," // Communications and Vehicular Technology in the Benelux (SCVT), 2011 18th IEEE Symposium on , vol., no., pp.1,8, 22-23 Nov. 2011

12. Mansoor Shafi and Min Zhang and Aris L. Moustakas and Peter J. Smith and Andreas F. Molisch and Fredrik Tufvesson and Steven H. Simon, Polarized MIMO Channels in 3D: Models, Measurements and Mutual Information // IEEE J. Select. Areas Commun on , vol.24, pp. 514,527, 2006

13. Parshin Yu. N., Komissarov A. V. MIMO Telecommunication System Capacity in Variable Spatial Radio Channel Structure with Simulated Multipath Propagation // DIGITAL SIGNAL PROCESSING. - 2012. - 1. - pp. 50-55.

14. Buzova M.A., Yudin V.V. Proektirovanie provolochnyh antenn na osnove integral'nyh uravnenij: Uchebnoe posobie dlja VUZov. M.: Radio i svjaz', 2005. 172 p.

15. Programmnyj kompleks SAMANT / OAO Koncern Avtomatika. Svidetel'stvo o gosudarstvennoj registracii 2013614026 ot 23.04.2013

Adaptive Clutter Suppression
D.I. Popov., Dr.Sci. (Eng.), Professor, Ryazan state radio engineering university, e-mail: adop@mail.ru

Keywords: auto-compensation, adaptation, adaptive rejector filters, estimation algorithms, training sample, clutter, estimation accuracy.

Abstract
In the article the likelihood function describes, which establishes the dependence of output samples automatic compensator phase Doppler clutter from the correlation matrix clutter. The estimation algorithm of clutter interperiod correlation coefficients on output auto-compensator of clutter Doppler phase has been synthesized using maximum likelihood method. The designing of quasi-optimum (simplified) estimation algorithm is executed and the block diagram appropriate to him device are presented on the basis of the given algorithm.

The comparative analysis of optimum and quasi-optimum estimation algorithms accuracy is conducted on the basis of maximum likelihood estimators asymptotical properties. To this end, expressions are obtained for the variance of the optimal and quasi-optimal estimates of clutter Doppler phase. Ultimately, the analysis showed, that the simplification of optimum algorithm results to insignificant and decreasing with growth of training sample volume losses in estimation accuracy. The legitimacy of the given conclusion is confirmed by results of estimation algorithm computer simulation.

The analysis method of the adaptive rejector filters (ARF) with the valid weight of clutter correlation coefficients is offered. The given analysis method of the allows to take into account clutter correlation properties, type used (optimum or quasi-optimum) device of clutter correlation coefficients and error of Doppler phase clutter auto-compensation and adaptation of weight coefficients ARF, caused in final volume of training sample in estimating the unknown parameters of clutter.

The analysis of losses in efficiency ARF, caused by mistakes auto-compensation and adaptation, is carried out. The results of the analysis have shown technical and economic expediency of use in ARF quasi-optimum device of clutter correlation coefficients.

References
1. POPOV, D.I. SU Patent No. 934816. Byull. Izobret., n. 33, 1998. 20 p.

2. POPOV, D.I. Adaptation of non-recursive rejecter filters. Radioelectron. Commun. Syst., v. 52, n. 4, p. . 46-55, 2009.

3. POPOV, D.I. Auto-compensation of clutter Doppler phase. Digital processing of signals., n. 2, p. 30-33, 2009.

4. POPOV, D.I. SU Patent No. 875960. Byull. Izobret., n. 33, 1998. 11 p.

5. POPOV, D.I. SU Patent No. 1015757. Byull. Izobret., n. 33, 1998. 12 p.

6. POPOV, D.I. SU Patent No. 1098399. Byull. Izobret., n. 35, 1998. 16 p.

Evaluation of Area of Ambiguity of Radio Reflections in Combining Images with Digital Terrain Maps
A.I. Novikov, email:novikovanatoly@yandex.ru
S.A. Yukin, email: yukin-s@yandex.ru
Ryazan State Radio Engineering University
, Russia, Ryazan

Keywords:
combining, radar images of the underlying surface, area of ambiguity.

Abstract
One of the challenges for the effective use of multi-spectral vision system (MSVS) is a combination of real images from different sensors with a synthesized image based on the digital terrain maps (DTM) with the required accuracy. This gives rise to difficulties arising from the difference in the coordinate system relative to which the images are formed from a variety of sensors MSVS, as well as the inability to recognize objects in these images direct signs. In connection with this need indirect methods based on binding of images from different sensors to the DTM. To do this, you must synchronize the resulting images by spatial position, i.e. hold spatial coincidence of all the processed image, the alignment should be performed automatically and in real time. To solve this problem it is necessary to have estimates of the possible spatial position of the sector review of relevant sensors MSVS on the underlying surface.

For the decision of this problem it is necessary to have possible spatial position estimations of the coverage areas of corresponding sensor controls MSVS on underlying terrain. Theoretically such positions - the unlimited set, but actually a range of these positions is defined by accuracy of navigating system and angular position of sensor controls.

In paper are resulted a finding method of a possible (limiting) zone of the radar review on a underlying terrain and its geometrical distortions caused by errors of spatial definition and angular position.

Researches of the review zone in the Earth plane, received spatial positions and distortions of radar images are executed.

The offered models are sufficient for estimation of uncertainty zone radar working in a forward hemisphere. Models can be used in the problems connected with automatic combination of images with the cartographical information. The uncertainty zone estimation of review sector onboard sensor controls reduces a search zone in a problem images registration with DTM at formation of a uniform information field in cockpit, that allows to solve it in real time.

References
1. .Y. Aksenov Combining images // Digital signal processing, 3, 2005, pp. 51-55.

2. R.G. White Change detection in SAR imagery // International Journal of Remote Sensing. vol.12, Issue 2, Feb. 1991.

3. DO-315 Minimum Aviation System Performance Standards (MASPS) for Enhanced Vision Systems, Synthetic Vision Systems, Combined Vision Systems and Enhanced Flight Vision Systems, 2011.

4. E.V. German, E.R. Muratov, A.I. Novikov, Mathematical Model of Uncertainty Zone in Combined Image Problem // Bulletin of the Ryazan State Radio Engineering University. Ryazan, Russia, 2013. Volume 46, Issue 4, Part 2. Pp. 11-16. (In Russian)

Images Improvement with Integer Orthogonal Transforming Matrices
Sheremet I.A., Professor, email: sheremet_ia@apr.gov.ru
Rukin A.P., a senior lecturer at the Institute of Foreign Languages. M. Torez, email: nto@mniti.ru
Lebedev V.D., a chief engineer of JSC MNITI, email: nto@mniti.ru


Keywords: image processing, orthogonal transform matrix, integer, a matrix of Walsh-Hadamard, Haar transforming.

Abstract
This article describes the approach to the problem of digital imaging with advanced transformation and then restore the image by using a new class of orthogonal integer transform matrices.

The JSC "MNITI" have been developed methods and algorithms of spatial correlation eliminate redundancy of digital images using integer transforms that may be applied to the image of an arbitrary format. These transformations have a uniform shape generating transform matrices, which implies their multivariate decorrelated while maintaining the properties, and as a result, the effective use of parallel processing in the hardware implementation.

Said matrix methods have been tested, in particular, under the criterion of degree of compression without further quantization of the digital monochrome test image differ in size and structural complexity. A number of new matrix methods, the degree of compression after the conversion without further quantization results showed no worse, and in some parameters, and better than the classical methods.

This article examines the possibility of using new methods of treatment (improvement) of digital images by transferring them to convert, and after treatment - the restoration using the inverse transform operation.

Generation of a large (in size) of integer orthogonal matrix of decorrelated small can be done in two ways, which differ unifying matrix operations.

If such a unifying operation is the tensor product, then we have decorrelated integer orthogonal matrix, which generalize the Walsh-Hadamard matrix. Generalization Regarding dimension decorrelated matrix, which will be arbitrary, not natural power of two, for Hadamard matrices.

If combining a multiple-scale operation, then as a result we obtain a generalization Haar transformation matrices for the case when the dimension of the matrix is decorrelated arbitrary and is not exclusively natural powers of two, which is typical for the Haar matrices.

Popular methods of Hadamard and Haar longer used due to their low efficiency. Because of their structure created nine matrices used only one. The conversion efficiency is usually compared with the Karhunen-Loeve transformation. This transformation accurately approximate the eigenvectors Toeplitz. This leads to the use of even cosine transform in the processing of photographic images.

The main positive effects of methods Hadamard and Haar are their integer representation and the ability to generate large size of the smaller matrices.

Evaluation of any method can only multiparameter. Close to the ideal method for eliminating spatial correlation Karhunen-Loeve hardly realize, thanks to its ultra-high computational complexity, which negates all his dignity. It requires compromise.

One such trade-offs are the methods based on DCT. These methods are mainly real operating in floating point arithmetic. Methods of decorrelation based on integral decorrelated orthogonal matrices are another embodiment of the compromise.

Developed in the JSC MNITI" generation system integer number of families decorrelated matrix of arbitrary size as the non-orthogonal and orthogonal, showed good results in solving the problem of compression of digital images.

It is shown that the conversion of integer matrices really allow you to change the brightness, contrast and detail of digital monochrome image.

Selection of basic processing parameters - matrix factors, contrast ratio and brightness component is made empirically.

Selecting transforming matrix - a semi-empirical. A graphical comparison of discrete finite functions expansion in orthogonal basis, which is a string integer decorrelated orthogonal matrix with other known methods, the basis on which to make a choice in favor of the corresponding matrix.

A large range of harmonics produced after conversion opens up opportunities for further studies, which are likely to help to establish the exact relationship between the structure and the parameters of its image processing matrix.

References

1. Evstigneev V.G., Bondarenko A.V., Kosharnovsky A.N., Lebedev V.D., Dichotomous and nodichotomous decorrelation methods of digital image, Systems and Communication, Television and Radio Broadcasting, Moscow, 2010, Issue 1, 2, 6.

2. Vilkova N.N., Evstigneev V.G., Lebedev V.D., A device for de-correlation of digital images using integer orthogonal matrix decorrelated video compression systems, RF patent number 2430419, value of 10 June 2010, Bulletin . 27., 12 p.

3. Bondarenko A.A., Evstigneeva O.V., Kosharnovsky N.A., Lebedev V.D., Method of forming integral decorrelated orthogonal matrices and device for its implementation, RF patent number 2509364, May 15, 2012 Priority, Bull. Number 17, -9 p.

4. Evstigneev V.G., Lebedev V.D., Kulikova E.I., Research "KEDR", Research and development of new principles and algorithms of decorrelation and recover digital images TV custom formats based on a new class of integer orthogonal decorrelated matrits arbitrary size, new methods and algorithms for their generation, M, JSC MNITI , 2010, -64 p.

5. Dagman E.E., Kuharev G.A., The rapid transformation of discrete orthogonal transformation, NAUKA, Novosibirsk, 1983, -230 p.

6. Dvorkovich V.P., Dvorkovich A.V., Digital video-information systems (theory and practice), Texnosfera, M, 2012, -108 p.

7. D. Salomon Data, image and noise compression, TEXNOSFERA, M, 2004, -365 p.

8. Pospelov D.A., Arithmetic foundations digital computer, VYSSHAIA Shkola, M .: -1970.

9. Papernov A.A., The logical foundations of digital computers and programming, NAUKA, M., 1968, -591 p.

10. Vilkova N.N., Evstigneev V.G., Sukhachev A.B., Shapiro B.L., Lebedev V.D., Some issues of improving the efficiency of the transmission of information in drone control complexes, M., VEGA, 2012, -13 p.

11. Lebedev V.D., On the orthogonal transformation of the digital image, Household appliances Communication, Technology Television Series, Moscow, 2013, - 98-108 p.

12. Hartmut H.F., The transmission of information by orthogonal functions", Moscow, publishing house "Communication", 1975. - 272 p.

13. Korn G., Korn T., Reference mathematics. - M .: Nauka, 1974., - 720.

14. Golovin L.I., Linear algebra," Nauka ", 1979. - 392 p.

Hybrid Speech Detector
M.A. Kotov,
email: kotov@stel.ru
D.A. Lednov, email:
lednov@stel.ru
T.V. Voznesenskaya, email: tvoznesenskaya@hse.ru

Keywords: speech processing.

Abstract
The speech detector is intended to single out a sequence of segments each containing a word or phrase from an input audio stream. The problem of detecting speech is similar to the problem being solved by the traditional theory of stochastic signals detection and described in many classical publications (e.g. [1,2]). The basic principle of this theory is that the risk of making the wrong decision is minimal, if the decision is taken according to
(1)

where the solution δ=1that the signal contains speech, and the solution δ=0 means that the signal only contains noise. Also in formula (1) the following designations are used:

Ratio λ is usually called the likelihood ratio, y is the observed signal, p(y | θ) the conditional probability density (CPD) of the observed signal depending on a random value, θ={1,0}, p1 is the presence of useful signal in the monitored process prior probability, K(θ,δ) - positive definite loss function chosen empirically, usually in the form of

It is necessary to have an idea about the type of CPD in order to implement this speech detection scheme. Since we do not possess the knowledge about the real form of speech and noise CPD, it is usually assumed that this CPD can be approximated with Gaussian mixture [3]

normal CPD with parameters φi ,where covariance matrix and mathematical expectations vector are used as parameters, αi normal CPD prior probability, n is the number of elements of the mixture.

The next step is calculation of the parameters i | θ} of the Gaussian mixture. To do so it is necessary to create two sets of records. One set should contain speech samples of various speakers, the other should contain samples of different types of noise. Then it is necessary to find the unknown parameters on the basis of collected data using the known EM-algorithm [3].

The algorithm described above is difficult to implement, because it is practically impossible to collect samples of all possible types of non-verbal signals in order to build their statistical models (probability density function). That, in turn, leads to an increase in the number of detector functioning errors. Therefore a question arises: is it possible to construct a detector which would work on the basis of a consistent feature of speech that distinguishes it from other noises?

In this paper we use the following sustainable features: speech holds vocalized intervals, i.e., while pronouncing each syllable we pronounce a vowel sound, and the spectrum of vowel sounds has indications of being linear [4]. Presumably, this approach will enable us to single out speech on the background of a variety of noises with continuous spectrum, while these noises can even be of non-stationary nature.

Undoubtedly, whispered speech does not hold indications of line spectra and music, on the contrary, does. As for whispered speech, it is not of practical importance, because it does not appear in messages which need to be processed. Speaking of music, it will be the source of false call errors for such a detector. However, the experience of practical use of a detector, which is based on statistical characteristics, shows that it copes with the task of detecting music. Thus there arose the idea of creating a hybrid speech detector, including the use of statistical and determinate characteristics of speech.

The article describes methods of determinate characteristics extraction and principles of constructing a speech detector based on these characteristics, as well as principles of building a hybrid detector model. Conclusion contains experimental data for hybrid detector work.

References
1. I.A.Bolshakov. Statistic problems of signals stream distinguishing// Soviet radio, 1969.

2. A.A. Harkevich. Noise control, second edition// Science, . 1965. Jeff A. Bilmes A Gentle Tutorial of the EM algorithm and its Application to Parameter Estimation for Gaussian Mixture and Hidden Markov Models// ICSI Technical Report-021, April 1998.

3. A.V.Agranovsky, D.A.Lednov, S.A.Repalov. .. Method of the text independent speaker identification based on the pronunciation of vowel sounds// Acoustics and applied linguistics. RAO annals, number 3. ., 2002, . 103-115

4. M.A.Kotov, D.A.Lednov and other. The parameters determining of the linear spectra vocalized sounds method and implementation// 2007148606/09(053252) from 27.12.2007

5. Babkin A.A. LPC Speech Coder AT 1000-1200 BPS // In Proceedings of DSPA-2000

6. V.V.Mottle, I.B.Muchnik. .., .. Hidden Markov models in signals structural analysis// .: FIZMATLIT, 1999

7. Steven Greenberg, Hannah Carvey, Leah Hitchcock, Shawn Chang Temporal properties of spontaneous speecha syllable-centric perspective// Journal of Phonetics 2002, 31, pp.465-485.

8. P. Mermelstein (1976), Distance measures for speech recognition, psychological and instrumental in Pattern Recognition and Artificial Intelligence, C. H. Chen, Ed., pp. 374388. Academic, New York.

9. S.B. Davis, and P. Mermelstein (1980), Comparison of Parametric Representations for Monosyllabic Word Recognition in Continuously Spoken Sentences, in IEEE Transactions on Acoustics, Speech, and Signal Processing, 28(4), pp. 357366.

Improving Performance of Genetic Algorithm That Constructs Digital Filters by Using of CUDA Technology
Belobrodskiy V.A., Voronezh State University, Russia, Voronezh, email: belobrodsky@yandex.ru

Keywords: parallel calculations, CUDA technology, mathematic modelling, data processing algorithms, biomedical signals.

Abstract
The fact is that paralleling a genetic algorithm is an effective means of decreasing its performance time. To achieve this goal it first seems obvious to do the calculations using a parallel cluster, but this approach requires a significant amount of preparation and organization work as well as serious technical support and financial investments. Starting with 2006, there appeared an opportunity to increase the performance time of non-visual computing using computational capacity of video cards following CUDA technology suggested by Nvidia company (Compute Unidied Device Architecture)[1,3,4,7-9]. This paper presents the analysis of programme productivity of two versions that implement a, one of those using CUDA technology. We understand productivity of genetic algorithm software W as the number of populations generated within one hour of work. The number of generations obtained within one hour of work (productivity W) is a function of a few variables, such as signal length (N), filter length (M), number of signals (K) and number of filters (L).

Initially, digital filter value construction technology based on genetic algorithm methodology was implemented on Visual # language without using any paralleling methods. Th first computation experiments carried out via the developed programme lead to satisfactory results, but they required rather a long waiting time (about 24 hours). To increase the values of productivity function W (N, M, K, L), a second software version was developed, which implemented the given genetic algorithm and used CUDA technology for convolution computing launched 100*100*50=500000 times in every generation, as there were two groups of input signals with 100 items in each and the number of filters in every generation was set as -50. To do a comparative analysis of the developed programmes productivity there were four series of computation experiment performed. In each of the experiments one of the four values (N, M, K, L) was a variable, and the other three values were taken as a constant.

The productivity analysis of the two realizations of the genetic algorithm aimed at generating digital filter values vividly demonstrated the practical advantage of CUDA technology implementation. The results also proved that the most resource-intensive point of the programme had been identified correctly, as a fivefold increase in CUDA version productivity is achieved.

On the other hand, genetic algorithm can easily be subjected to further paralleling, for instance due to independent optimization parameters of filters it is possible to calculate these values independently from one another. Thus, in the further perspective there is some potential left for further productivity increase through transferring other estimated modules of the genetic algorithm from CPU to GPU computing, such as mating, mutation generation, etc.

The author expresses his special gratitude to S.D. Kurgalin, Ya. A. Turovsky and A.A. Vakhtin.

References
1. A.A. Vakhtin, A.Ya. Turovsky. Implementation of numerical wavelet transformation on graphic adapters of NVIDIA CUDA architecture // Proceedings of Voronezh State Unversity. Series: System Analysis and Information Technologies. - 2012. - 1. - P. 69-72.

2. A.Ya. Turovsky. Developing EEG-state analysis filters based on genetic algorithms // Software Engineering. - 2014. - 6. - P. 23-28.

3. A. V. Boreskov, A.A. Kharlamov. Basics of Using CUDA Technology. M.: DMK Press, 2010. 232 pages.

4. Electronic resource: http://www.nvidia.ru/object/cuda-parallel-computing-ru.html

5. Holland J.H. Adaptation in Natural and Artificial System. MIT Press, 1992. 205 pages.

6. M.Snell, L. Powers. Microsoft Visual Studios #/

7. A.M. Kazyonnov. Basics of Using CUDA Technology.

8. A. Berillo. NVIDIA CUDA Non-visual Computing on Graphic Processors (http://www.ixbt.com/video3/cuda-1.shtml)

9. J. Sanders, E. Kandrot. CUDA By Example: An Introduction to General-Purpose GPU Programming. - Addison-Wesley Professional, 1st edition. 312 pages.

Numerical Modeling of Interaction of a Signal and Noise in the Nonlinear Filter of First Order
Aboelazm M.A., Melchakov V.N., Reshetnyak S.A., Tretyakov G.N.

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

Abstract
The process of passing a weak signal and noise through nonlinear systems are recently increased interest which caused by the discovery of the phenomena of stochastic resonance [1-3] and stochastic filtering [4,5] of signals. In General case these processes can be analyzed on the basis of numerical solution of stochastic differential equations. From a practical point of view, the most interesting effect is the stochastic signal filtering (SF). It allows to reach in radio engineering systems the output signal-to-noise ratio exceeding of the input values [6,7]. Effect of SF due to a signal suppression of the noise and have the threshold nature with fixed noise dispersion. S/N ratio is defined as the ratio of the signal power to the noise spectral density at the frequency of the signal. In this paper the method of numerical analysis of the signal and noise interaction in nonlinear active first-order filters was developed. The order of filter and differential equation associated with the number of reactive elements contained therein. Euler method lays in the basis analysis which was used for the numerical solution of nonlinear stochastic differential equations. It was shown that under the action of harmonic signal and white noise on the input filter output S/N ratio exceeds the input ratio under certain parameters of filters as low and high frequency. In the work a parameters were found for which SF effect is more bright in high-frequency filter compared to the low-frequency filter. It is shown that for monitoring SF there are the optimal values of the signal amplitude and frequency. The analysis results are satisfactorily consistent with experimental data [7].

References
1.Gammaitoni L., Hanggi P., Jung P., Marchesoni F. Rev. of Mod. Phys., 1998, v.70, 1, p.223-287.
2. .., .., ., - . , 1999, .169, 1, .7-38.
3. .. , 1999, .169, 1, .39-47.
4.Hanggi P., Inchiosa M.E.,Fogliatti D., Bulsara A.R. Phys. Rev. E., 2000. v.62, 5, p.6155-6163.
5. .. , .., .. . . . , 2000,5, .12-17.
6. .., .. , 2007, 9, .19-25.
7. .., .. , 2009, .54, 11, .1369-1371.






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