TMCnet News
A Novel Digital Signal Modulation Mode Recognition Algorithm [Sensors & Transducers (Canada)](Sensors & Transducers (Canada) Via Acquire Media NewsEdge) Abstract: Digital modulation signal instantaneous characteristics are susceptible to noise interference, and instantaneous frequency extraction is easy to be folded and distorted. In this paper, a novel digital signal modulation mode identification algorithm under low SNR is proposed. In the algorithm, empirical mode decomposition (EMD) has been employed to extract signal instantaneous characteristics firstly; and then instantaneous characteristics has been preprocessed by using wavelet de-noising; finally five characteristic parameters have been calculated to recognize seven kinds of digital modulation signal. The simulation results show that the instantaneous frequency can get actual physical meaning using proposed algorithms with higher recognition efficiency under low SNR. Copyright © 2014 IFSA Publishing, S.L. Keywords: Experience mode decomposition, Instantaneous characteristics, Wavelet de-noising, Digital modulation signal, Low SNR. (ProQuest: ... denotes formulae omitted.) 1. Introduction With the rapid development of modem communication technology, digital signal automatic identification has become more and more important. At present, there are many kinds of digital signal modulation mode automatic identification methods, such as the method based on decision theory and the method based on statistical theory. Nandi and Azzouz used statistical theory to extract the instantaneous amplitude, phase and frequency of the signal information, because the instantaneous characteristic can be used to structure characteristic parameters to identify the signal modulation mode; Assaleh modeled the signal as a time-varying autoregressive process, through the regression model parameters to estimate the signal frequency and bandwidth; Hong used wavelet transform to extract PSK and FSK signal's instantaneous phase and instantaneous frequency information [1-4]. These methods have their advantages and disadvantages, and the main problem of modulation recognition is the identification of signal under low SNR. In this paper, first, the empirical mode decomposition (EMD) is used to overcome the influence of noise on the signal features and the emergence of frequency distortion that directly used Hilbert transform to extract the instantaneous parameters. And then wavelet de-noising is used to preprocessed instantaneous characteristics. Finally, five characteristic parameters are got to recognize 2ASK, 2PSK, 2FSK, 4ASK, QPSK, 4FSK and 16QAM signals. The algorithm that achieved the way of modulation identification under low SNR has the certain reference value in practical application. 2. The Instantaneous Characteristic Parameters Extraction The EMD first decomposed the signal into a limited number of Intrinsic Mode function (IMF), The characteristic of IMF is that it has effective definition of instantaneous frequency, can represent the intrinsic oscillation modes in data and can give the instantaneous frequency meaningful physical interpretation. The Hilbert spectrum analysis is made the Hilbert transform to each IMF component, get the instantaneous frequency signal. The physical meaning of the instantaneous frequency shows that it has many ineffective instantaneous frequencies. When producing a method of Hilbert transform that can make the data analysis, make the signal expression type (1), or complex signal as type (2), it is the time to calculate the instantaneous frequency. ... (1) ... (2) 2.1. Instantaneous Characteristics of the Modulation Signal Signal instantaneous characteristics include instantaneous amplitude, instantaneous frequency and instantaneous phase. The most commonly transient feature extraction method that be used is the Hilbert transform method. This kind of method that be used to extract the signal's instantaneous amplitude and instantaneous frequency is more effective under high signal noise ratio (SNR), but for instantaneous frequency extraction, that is easy to produce fold and distortion [6]. Therefore, in this article, the method of empirical mode decomposition is adopt to overcome the problem that influenced by noise and the distortion. 2.2. The Steps of EMD Decomposition The characteristic of IMF is that it has effective definition of instantaneous frequency can represent the intrinsic oscillation modes in data and can give the meaningful physical interpretation of instantaneous frequency. Any signal can be decomposed into x(f) intrinsic mode components and remaining components such as formula (7) [7-8]. Decomposition processes are as follows: 1) Find out all local maximum values in the signal and using cubic spline function to connect into the envelope; in the same way, using the cubic spline interpolation function to connect all the local minimum value constitute the envelope; 2) Beg out the average of the upper and lower envelope mx , and beg out the difference between original signal and envelope h1; ... (3) 3) If h\ meet the IMF's conditions, so that the first IMF component h\ is obtained; otherwise it would be h\ as an original signal (1) ~ (2) steps, until after the k iteration difference /?/,*(£) become a the IMF, remember to: ... (4) The above termination criterion k iteration is to: ... (5) Lies between [0.2, 0.3] [5]; 4) Subtract ci(t) from the original signal, the first order signal is r\{t). ... (6) 5) The residual signal r,(i) as the original signal to finish step (1) - (4), when residual signal r"(t) is so much so that it can't extract the IMF, the circulation is end. To sum up, the original signal x(i) is decomposed into: ... (7) where ci is the i lh intrinsic mode components, rn is the remaining components. Signal after the EMD is decomposed into several IMF basis function, to the Hilbert transform of each IMF component to get its instantaneous characteristic value. For every IMF after the Hilbert transform, x(i) is as following: ... (8) Then to make the Hilbert transform ofx(i), to get instantaneous characteristic information ofx(i). 1) Instantaneous amplitude: ... (9) 2) Instantaneous frequency: ... (10) 3) Instantaneous phase: ... (11) 3. Modulation Recognition Algorithm 3.1. Characteristic Parameters The core of statistical pattem recognition method based on feature extraction is characteristic parameters, Characteristic parameters used in this article are as follows [9-10]: 1) The maximum value of an estimate of the power spectral density of the amplitude sequence ymax ... (12) Ns is the sampling number, acn(i) = an(i)-l , ... is the instantaneous amplitude of the signal, ma is the mean value of instantaneous amplitude. 2) The standard deviation of the nonlinear component of instantaneous phase of the zero-center non-weak signal odp. ... (13) c determines the number of signals required in all sampling values. 3) The Standard deviation of the absolute value of the nonlinear component of instantaneous phase of the zero-center non-weak signal cap. ... (14) 4) The standard deviation of the absolute value of instantaneous amplitude of the zero-center non-weak signal iaa. ... (15) 5) The standard deviation of the absolute value of instantaneous frequency of the zero-center non-weak signal car ... (16) 3.2. Process of Modulation Recognition After the analysis of the characteristic parameters, they can be used to identify the signal, the process is shown in Fig. 1. 1) First, estimate the digital modulation signal parameter, and then calculate the signal's /max , through the comparison with the threshold method ¿(/max)' if /max > ¿(/max)' ^ P^Cate the Signal ÍS ASK or 16QAM, otherwise predicate the signal FSK or PSK; to set the decision threshold ) again, if /max > ¿'(/max ) ' the ^g113! is ASK' otherwise is 16QAM. 2) If the signal is one of FSK and PSK, calculate Gap, compared with t(cap), ^Gap<t(<Jap), the signal is 2PSK, otherwise is 4PSK, 2FSK or 4FSK. 3) If the signal is belong to 4PSK, 2FSK or 4FSK, calculate <jdp compared with t(aJ, if Gdp<t{<Jdp), the signal is 4PSK, otherwise is 2FSK or 4FSK. 4) If the signal is belong to 2ASK or 4ASK, calculateG^ , compared with t(caa), '^Gaa <t(cm), the signal is 2ASK, otherwise is 4ASK, this is the end of recognition 5) If the signal is belong to 2FSK or 4FSK, calculate oaf, compared with t(Gaf), if Gaf < ?(<7a/), the signal is 2FSK, otherwise is 4FSK, this is the end of recognition. The process is shown in Fig. 1. 3.3. Simulated Analysis In this paper, the simulation environment is under the environment of Matlab 7.0. In this paper, digital signal modulation mode identification under low SNR is is mainly studied. First, extracted signal instantaneous characteristics using empirical mode decomposition. Then preprocessed instantaneous characteristics using wavelet de-noising. Finally got five characteristic parameters to identifícate 2ASK, 2PSK, 2FSK, 4ASK, QPSK, 4FSK and 16QAM. Simulation parameter is set to the carrier frequency of 1 kHz, the sampling frequency is 8 kHz, symbol speed rate of 500 Hz. First of all, simulate the characteristic parameters, the five characteristic parameters corresponding to the signal trends along with the change of SNR is shown in Fig. 2, abscissa for SNR value of dB, and ordinate with the value of the corresponding characteristic parameters (no unit). In Fig. 2, each of graph (a) to (e) represents a kind of digital modulation signals, can be seen from the figure in the five kinds of characteristic parameters of distinguish ability is very strong, especially when the signal-tonoise ratio is higher than 5 dB, the modulation signal corresponding to the characteristics of the parameter values have obvious differences. After determining the characteristic parameters threshold, according to the above parameters, to simulate each recognition signal independently, the noise is additive white Gaussian noise, the simulation is in MATLAB, seven kinds of digital modulation signal recognition rate along with the change of signal to noise ratio as shown in Fig. 3, the modulation signal when the signal to noise ratio greater than or equal to 5 dB can achieve above 95 %, shows that the recognition algorithm in low SNR has high recognition efficiency. In this paper, the signal processing method and parameter extraction method these are proposed are compared with recognition accuracy traditional method which was proposed by A.K.N and E.E.A zzouz et al [11-12], as shown in Table 1. It can be seen that the traditional method of the identification accuracy of recognition rate in low signal noise ratio (SNR) is low, but using the proposed based on wavelet de-noising combined with EMD transform, the method of extracting instantaneous information when SNR is 5 dB recognition accuracy up by more than 20 %, the biggest can prove that the presented method better recognition ability, especially in low SNR when stronger recognition ability. 4. Conclusion In this paper, EMD and wavelet transform denoising method are used to preprocess the signals. Using this method not only can make the instantaneous frequency has the actual physical meaning, but also can extract more real instantaneous characteristic parameters under low SNR. Digital modulation signal mode recognition realizing under the condition of low SNR has been laid certain foundation for the signal pattem recognition applications in military and civilian. Acknowledgments The work was supported by the Scientific Research Foundation of the Higher Education Institutions of Gansu Province, China (Grant No. 2013A-048). References [1] . Swami A., Sadler B. H., Hierarchical digital modulation classification using cumulants, IEEE Transactions on Communications, Vol. 48, Issue 3. 2000, pp. 416-429. [2] . Ebrahimzadeh, A., Automatic modulation recognition using RBFNN and efficient features in fading channels, in Proceedings of the IEEE Conference on Networked Digital Technologies, Ostrava, Czech, July, 28-31, 2009, pp. 485-488. [3] . E. E. Azzouz, A. K. Nandi, Automatic identification of digital modulation types, Signal Processing, Vol. 47, Issue 1, pp. 56-58. [4] . Tan X. H., Liu J., and Hu Y. Q., A new algorithm for digital modulation recognition under the low SNR, Journal of Systems Engineering and Electronics, Vol. 31, Issue 6,2009, pp. 60-64. [5] . Norden E. Huang, et al., The empirical mode composition and the Hilbert spectrum for nonlinear and non-stationary time series analysis, Proc R, Soc. Lond.A, 1998. [6] . Li Hua-Ying, Signal modulation recognition using HHT, China Science and Technology Information, Vol. 11, Issue 3,2009, pp. 113-114. [7] . Zou Bao-Juan, Li Chi-Sheng, The application of digital modulation signal instantaneous characteristic parameters extraction and modulation recognition, Signal Processing, Vol. 24, Issue 2, 2008, pp. 201-203. [8] . Zhang Li-Na, Guo Ming, Wheat clash recognition research based on HHT, Computer Engineering and Applications, Vol. 30, Issue 1,2013, pp. 35-40. [9] . Sun Jian-Cheng, Zhang Tai-Yi, Liu Feng, Automatic Multi-Class Digital Modulation Recognition Algorithms Based on Support Vector Machines, Journal of Xi'An Jiaotong University, Vol. 6, Issue 38,2004, pp. 619-622. [10] . Tang Zhi-Ling, Yang Xiao-Niu, Li Jian-dong, Automatic Digital Modulation Recognition Using MDA and Fractal Dimension, Journal of Si Chuan University, Vol. 3, Issue 44,2012, pp. 117-120. [11] . Taira S., Murakami E., Automatic classification of analogue modulation signals by statistical parameters, in Proceedings of the Military Communications Conference, 1999, pp. 202-207. [12] . Li Ji, He Chen, Chen Jie, An Automatic Digital Modulation Recognition Based on Combined Statistic Parameters, Journal of Shang Hai Jiaotong University, Vol. 5, Issue 41,2007, pp. 693-697. 1 Chunlei Zhang,2 Hui Wu,2 Huanyu Ning 1 School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou, 730070, China 2 School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou, 730070, China 1 Tel.: 13359442222 1 E-mail: [email protected] Received: 13 June 2014 /Accepted: 29 August 2014 /Published: 30 September 2014 (c) 2014 IFSA Publishing, S.L. |