A Novel Methodology For Identification of Unclassified Digital Voice

International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: editor@ijaiem.org, editorijaiem@gmail.com Volume 2, Issue 9, September 2013 ISSN 2319 - 4847
A Novel Methodology for Identification of Unclassified Digital Voice

Parthraj Tripathi1, Dr.K.Padma Raju2, N. Satish Kumar3
Scientist, Defence Electronics Research Laboratory, DRDO, Hyderabad, India 2 Principal, University College of Engineering, JNTUK, Kakinada, India 3 Department of Electronics and Communication Engineering, University College of Engineering, JNTUK, Kakinada, India
1
ABSTRACT
Voice signals are encoded with various techniques to exploit communication resources such as source, channel, computation and cost of operation. These encoded signals are generally multiplexed with same/different signal type (data, say). At the receiver end, after the data and voice signals are discriminated, the voice bit stream has to be passed through the respective decoder in order to get meaningful audio. Identification of correct decoder requires a digital voice classifier. The importance of such classification before decoding get especially noticed in the cases of design of universal voice decoder where the unknown/unclassified bit stream is expected as of one of the codec out of many types of codec. We call the set of all expected decoders as an ensemble. In this paper, a concept of operation of digital voice classifier before decoder is proposed .This proposed method is superior to method of subjecting a given unclassified bit stream to each of the decoder in a given ensemble. Also, a novel technique called SWOB (sliding-window on bit stream) is introduced which gives satisfactory results of classification when used along with methods of auto correlation(AC) and central second order moment(SOM). Another parameter for crisp classification is also introduced which is named as BRO (binary ratio). In all, the (SWOBBRO-SOMAC) algorithm is capable of identifying the digital voice signal classes like of PCM-/A law, ADPCM 16-, 24-, 32-, 40 kbps. Using same algorithm, option of coarse classification is open on signals of other type too.
Keywords: Universal decoder design, classification, analysis.
1. INTRODUCTION
To get efficient data transmission in communication systems, needed is to use limited channel resources efficiently. So, the voice and other signals (data) are multiplexed and transmitted through a single channel. At receiver side, this signal is passed through the de-multiplexer [9], [14]. Channel wise, it provides payload to the voice/data discriminator where it discriminates between voice and data and send to their respective decoders after identification [2], [6]. One can decode the received encoded voice signal only if the information is known about encoding technique used at the transmitter side such as PCM-/A law, ADPCM 16-, 24-, 32-, 40 kbps [12], [13]. Subsequently it can be passed through its corresponding decoder and audio up to the intelligence level can be had. In general this information is present in the bit stream itself in the form of headers as indicated by the protocols. But in the cases of loss of headers or non comprehension of the protocol used as in unclassified bit stream, or both; the information about codec used has to be extracted out of the bit stream itself. We exploit the properties of the encoded bit stream to accomplish this. The Figure 1 shows the signal analysis model used at the receiver side.
Figure 1 Signal analysis model at receiver end If the received bit stream is expected to be decoded out of many decoders in the ensemble, sure shot decoding may be observed if the bit stream is subjected to all of the decoders simultaneously. This method has a disadvantage of consuming the computational and electrical power resources unnecessarily running on all of the decoders, all of the time of operation. So, in order to get efficient operation, there is a need to classify the received voice bit stream before actually trying to decode the signal. A digital voice classifier hence is coined as it is useful, in terms of efficiency. Codecs like PCM-/A law, ADPCM 16-, 24-, 32-, 40 kbps constitutes an ensemble in our case [7], [11]. This paper proposes classifier algorithm which is capable of identifying the six (as mentioned above) classes of the voice signal based up on a novel technique called 'SWOB'(sliding window on bit stream) technique, amalgamated with three parameters namely AC (auto correlation), SOM (central second order moment) and BRO (binary ratio). The short form 'SWOBBRO-SOMAC' is indicative of all of the techniques mentioned above [8].
Volume 2, Issue 9, September 2013
Page 297

The main contribution of this paper can be summarized as follows: Introduction of the concept of importance of classification before decoding. Introduction of the novel technique with detailed explanation of the classifier algorithm. Explanation about the parameters used in the classifier. Experimental results are presented.
2. CLASSIFIER ALGORITHM
A digital voice bit stream signal is received from the Voice/Data Discriminator and these bits are processed through independent processing operations [5]. One processing operation applies sliding window technique (SWOB) starting from any bit in the received bit stream, which will convert group of eight bits into pseudo bytes, starting from that bit. Important to note that is these bit streams may contain known and expected protocol information or unclassified protocol information or no protocol information at all. But it is irrelevant as the classifier is exploiting the properties of the payload in the bit stream. We have assumed that header size is very small as compared to payload size. In other words, payload itself contains the information about properties of the expected voice codecs. Also, header will take its meaning only if the position of initial bit is known. This classifier will function even if the bit stream is not byte aligned. These bytes are divided into finite sequence of subsequent bytes, which are called as segments (bytes) and calculate central second order moment and auto correlation sequence on each segment (bytes). In Second processing operation, the received bit stream is divided into finite sequence of subsequent bits, which are called as segments (bits) and processed to calculate the binary ratio on each segment of the received bit stream signal. These independent processing operations results are given to a decision box and finally it will identify the class of received bit stream signal. We experimented with segment length of 20000 bits and observed high accuracy with sufficient response time for classifying the digital voice bit stream.
Figure 2 Classifier algorithm Figure 2 shows the block diagram of the classifier algorithm which is capable of identifying and classifying the digital voice bit stream. By incorporating this classification algorithm in the universal decoder design, its performance can be improved when it analyses the large number of the unclassified signal.
3. CLASSIFICATION TECHNIQUE AND PARAMETERS

3.1 'SWOB' Technique: This technique is novel technique and it is applied on the received bit stream and it will generate psuedo bytes from the group of bits. The mathematical representation of sliding window technique is represented as:
Where, b(i) is received bit stream, S(i) is the decimal values obtained from the window technique, w is the sliding window size and n is the number of received bits.
Figure 3 Process in sliding window Technique
Page 298

The sliding window technique process is shown in figure 3 and as follows: First, choose the size of sliding window and calculate the decimal value from the bits present in the window. Second, slide the window by one bit and calculate its decimal (byte) equivalent and this process continue for remaining all bits in the received bit stream. In our case of analysis, the size of sliding window was taken as eight. After calculating the decimal equivalent values from the sliding window technique, then auto correlation sequence and central second order moment are calculated on these decimal values in order to classify the digital voice bit stream. 3.2 Autocorrelation Sequence: The autocorrelation represents degree of similarity between a given time series and a lagged version of itself over successive time intervals [15]. The mathematical expression of the auto correlation sequence of signal x(n) is represented as follows [1], [4]:
Where,* represents the complex conjugate, k is the lag and R(k) is the auto correlation sequence of the signal. Here, we are not performing the complex conjugate computation. So the ACS equation is rewritten as
Where s(i) is the real values generated from the sliding window technique, N is the segment length, and k is the lag to compute. In order to reduce the multiplications, we use real ACS rather than complex valued one. 3.3 SOM Central second order moment: Central second order moment measures the width of a set of points in one dimension and measures the shape of a cloud of points for higher dimensions [3]. The mathematical representation of central second-order moment of s(i) is given as [1], [4]:
Where,
Here, s(i) is the real values generated from the sliding window technique and N is the segment length. 3.4 'BRO 'Binary Ratio: For the various kinds of codecs, it is observed that the number of lower levels and higher levels in coded bit streams are different [10]. So, the binary ratio (BRO) can be used as a parameter in identifying and classifying the digital voice bit stream. The BRO is calculated on each segment of digital voice bit stream. It is given as If b(i)=0; then ZR=ZR+1. If b(i)>0; then OR=OR+1. Then the binary ratio (BRO) is given as BRO=ZR/OR. Where, ZR is the number of lower levels in digital bit stream. OR is the number of higher levels in digital bit stream.
4. EXPERIMENTAL RESULTS
Prior to testing the algorithm, a large body of test signals was generated. The voice signals are recorded using microphone and sampled at 8000 samples per second. These samples are stored in int16 format and encoded using standard (ITU-T) encoding algorithms. Here six classes of signals are considered: PCM-/A law, ADPCM 16-, 24-, 32-, 40 kbps. All the testing signals are processed using classifier algorithm and observed that six classes can be discriminated.
Figure 4 PCM A-law
Figure 5 PCM -law
Page 299

Figure 6 ADPCM 16 kbps
Figure 4 illustrates the parameter distribution of PCM -law in 3-Dimensional view. It is the plot of AC versus SOM versus BRO. These ranges of values are obtained after testing the large set of testing signals and only few values are shown in plotting for clarity. Similarly, Figure 5-9 illustrates the parameter distribution of: PCM-/A law, ADPCM 16-, 24-, 32-, 40 kbps in 3-Dimensional view. It is observed that the parameters are different for different class of signals. All six classes of signals are differing in at least one of the three parameters. It can be clearly observed in the Figure 10.
Figure 10 AC versus SOM versus BRO of all six classes on single plot. Figure 10 shows the combined plot of all six classes of PCM and ADPCM variants. From this plot we can observe clear the discrimination of all six classes of signals. Figure 11 shows the top view of figure 10. These plots are obtained for segment length of 20000 bits and the auto correlation was calculated at lag zero. We had experimented by varying the parameters such as segment length, sliding window size, number of bits to be slide in sliding window technique and with auto correlation at different lags. But we observed that, the results with combination of segment length of 20000 bits, sliding window size as eight, slide of 1-bit in sliding technique and with auto correlation lag at zero are giving satisfactory for classifying the various all six codecs.
Figure 11 Top view of all six classes on single plot.
Page 300

Table 1: Threshold values of all six classes for three different parameters AC SOM BRO 19500-26100 0.14-0.25 0.47-0.82 24500-35500 0.21-0.35 0.76-1.12 22800-24500 0.48-0.56 1.00-1.09 32900-38000 0.14-0.19 0.45-0.57 26900-31500 0.23-0.29 0.60-0.76 26400-32900 0.27-0.33 0.58-0.80
PCM -law PCM A-law ADPCM 16kbps ADPCM 24kbps ADPCM 32kbps ADPCM 40kbps
Based up on the results obtained after analyzing large set of test body the thresholds are obtained for each class are as shown in table 1. These thresholds values of each class are stored in the decision box and based up on these thresholds it classifies the received bit stream.
5. CONCLUSION AND FUTURE WORK

In this paper, we described a digital voice signal classifier that accurately identifies the digital voice bit stream. This classifier is capable of classifying six signals: PCM-/A law, ADPCM 16-, 24-, 32-, 40 kbps. It will classify the digital voice bit stream by just processing 20000 bits of the given bit stream and that too when the bit stream are not synchronized i.e. start bit is not identified. Procedurally, for a signal captured, the classifier algorithm calculates all three parameters of received signal followed by comparison of threshold values stored in the decision box, it classifies any given bit stream. This classifier algorithm is more reliable and less complex than the traditional and widely used techniques for unknown bit stream identification. The future work can be done on classifying other signal types namely image, video etc.
References
[1] N. Benvenuto, A Speech/Voiceband Data Discriminator. IEEE Trans.on Commun.. Vol. 41, No. 4. April 1993, pp. 539-543. [2] N. Benvenuto, WR. Daumer, Classification of Voiceband Data Signals, Proc. ICC. Atlanta, GA, April 1990. pp. 1010-1013. [3] J.E. Hipp. Modulation Classification Based on Statistical Moments, IEEE MILCOM 86 C d . , Octo& 1986, pp. 20.2.1-20.2.6. [4] J. S. Sewall, Signal Classification in Digital Telephone Networks, M.Sc. thesis, Jan. 3, 1996, Dept. of Electrical and Computer Engineering, University of Alberta, Edmonton,Canada. [5] Xiangqian Xu Bruce F.Cockburn, and Deepak P.Sarda. A DSP-Less Real Time voiceband Signal Classifier Based on a Commodity Personal ComputerPlatform, IEEE Canadian conference on Eletrical and Computer Engineering, Edmonton, Alberta, Canada May 9-12 1999. Pg 90-94. [6] Y.Yatsuzuka, High Sensitive speech detector and high-speed voiceband data discriminator in DSI-ADPCM systems,IEEE Trans. Commun,vol.COM-30 pp739-750,Apr.1982. [7] A text book of Communication System Simon Haykin, fourth Edition. [8] Vaishnavi Tripathi, Kuldeep Goyal, Parthraj Tripathi Relevant Feature Selection for Intrusion Detection using EDIB Methods. 6th IEEE International Conference on Advanced Computing & Communication Technologies (ICACCT 2012). [9] Laws and theory of unified communication, by VashatkarYadav, Controller: Communication Control Chamber. [10] Speech/Data discrimination in communication system, ParthrajTripathi, Dr.K.PadmaRaju, Ashok Kumar Ginni, IOSR, Vol 2, 2012. [11] Thesis "A GUI Based Iteration Approach forEasy and Reliable Speech Codec Detection and Decoding", ParthrajTripathi, Dr.K.PadmaRaju, Manikanta swamy Joka. [12] CCITT. Recommendation G.726 5-, 4-, 3- And 2-BitsSample Embedded AdaptiveDifferentialPulse CodeModulation (ADPCM). [13] ITU-TRecommendation G. 711 Pulse Code Modulation (PCM) of Voice Frequencies. [14] A text book of Communication Systems Engineering,J Proakis and M. Salehi,.Prentice-Hall, 2nd edition, 2002. [15] A text book of Signals and Systems,Alan V.Oppenheim,Alan S.Willsky, S.Hamid Nawab, 2nd edition, 2006.
AUTHOR Parthraj Tripathi obtained BE in Electronics and Communication Engineering from RGPV Bhopal, MP, India in the year 2003. He joined Hindustan Aeronautics Ltd and had 4 years extensive experience in not only in Avionics but also in various Aircraft-systems. In 2008, he joined DLRL, DRDO and working in EW related
Page 301

research. His area of interests are EW, strategic communication, intuitive Analysis, Signal processing. He is a budding defence analyst and published number of articles related to defence systems around world.
Dr K Padma Raju obtained B.Tech in Electronics and Communication Engineering, Nagarjuna University in the year 1989, M.Tech in Electronic Instrumentation, National Institute of Technology Warangal in the year 1992. He received the Doctoral degree from Andhra University in the area of Microwave Antennas in the year 2003. Currently he is the principal and senior professor in Electronics and Communication Engineering in J.N.T.University, Kakinada, A.P, India.
N.Satish Kumar received B.Tech degree in Electronics and Communication Engineering from Aditya Engineering College, JNTUK University, A.P, India in the year 2009. Currently pursuing his masters degree program in computers and communication engineering in J.N.T.University Kakinada, A.P, India.
Page 302

A Novel Methodology For Identification of Unclassified Digital Voice

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

A Novel Methodology For Identification of Unclassified Digital Voice

Diunggah oleh

Hak Cipta:

Format Tersedia

International Journal of Application or Innovation in Engineering & Management (IJAIEM)