Sign up

Simple system for testing secure voice communication in wireless networks.
This paper presents the Matlab implementation of a low bit rate communication system for transmitting Mixed-Excitation Linear Predictive vocoder data, based on Matlab's ability to interact from Simulink with external programming languages (such as Java or C Dynamic Link Libraries). Even if the use of external functions in the Simulink Module has some limitations, it allows the rapid implementation of testbeads for secure voice communication between different computers (using the encryption operations provided by the Java Cryptography Architecture). The designed system was implemented and tested in a wireless communication system.

Key words: Matlab, VoIP, MELP, JCA, External functions

Article Type:
VoIP (Network protocol) (Testing)
Wireless networking (Equipment and supplies)
Wireless networking (Technology application)
Ionescu, Valeriu
Sima, Ion
Sofron, Emil
Pub Date:
Name: Annals of DAAAM & Proceedings Publisher: DAAAM International Vienna Audience: Academic Format: Magazine/Journal Subject: Engineering and manufacturing industries Copyright: COPYRIGHT 2009 DAAAM International Vienna ISSN: 1726-9679
Date: Annual, 2009
Event Code: 330 Product information Computer Subject: Voice over IP; Technology application
Geographic Scope: Austria Geographic Code: 4EUAU Austria
Accession Number:
Full Text:

The current tendency in communications and especially in mobile communications is to merge in a single IP-based stream because of communication efficiency over limited available bandwidth. The behavior of these systems can be simulated in a software environment that allows the interaction with real hardware. This paper presents a simple Matlab Simulink implementation of a communication system, based on Matlab's ability to interact with external programming languages, using MELP (Mixed-Excitation Linear Predictive) 2400bps vocoder for good audio quality at a low bit rate, AES (Advanced Encryption Standard) encryption in ECB (Electronic Code Book) Mode for communication security and User Datagram Protocol (UDP) protocol for voice communications. The system uses the Java Cryptography Architecture (JCA) for encryption implementation and C Dynamic Link Libraries (DLL) for MELP implementation and shows that is it easy to design a complex communication system without having to redesign each of its components.


In order to ensure real time simulation, a Real Time Windows Target implementation was considered, but as some incompatibilities with other Matlab components were encountered (as outlined in the Matlab documentation), the final implementation was made in Simulink.

The communication system architecture consists for the transmitting components in: audio data acquisition block, a MELP encoder block, an AES encryption block and a UDP data communication block. The receiving components present the reverse chain as seen in Fig. 1. The disk storage of the voice data output, needed for later analysis, was implemented via a controlled (triggered) subsystem. A processing order had to be established using the "Priority" field of Simulink blocks in order to respect the correct data processing chain.

--Audio Data Acquisition Block Implementation The audio data acquisition block can be implemented in Matlab by either using the "Analog Input" from the Data Acquisition Toolbox, or the "From Audio Device" from the Signal Processing Toolbox.

We chose the "From Audio Device" block, as it allows data acquisition at as low as 1000 samples/sec compared to 5000 samples/sec for "Analog Input", necessary in early stages of AES testing (when the MELP encoding block was not implemented). For the audio output, the "To Audio Device" was used. When Simulink cannot keep up with an audio device that is operating in real time, some audio data will be skipped.

--MELP Vocoder Implementation

The MELP (Mixed-Excitation Linear Predictive) standard was released in 1999 and specified only the 2400 bps mode of operation, but further enhancements in 2001 allowed 1200bps and 600bps, known as MELPe.

While an implementation of the MELP Algorithm has recently become available for Matlab (Basov, 2008), for this system was used initially the output of the executable implementation of MELP from Texas Instruments (TI, 1998). By using the commands: "melp -a -i infile -o bitstream" for the encoding and "melp -s -i bitstream -o outfile" for the synthesis if was possible convert data to MELP and load it for the encryption stage by using temporary hard disk storage for audio bitstream. For real time operation the time spent in the execution of the external program (much time is lost with disk access) has to be lower than the real speech time in order not lead to delays in data processing. By measuring "melp.exe" program execution time was determined for 3 seconds of real time data processing with different storage media. The results showed that the implementation is adequate for faster than real time data acquisition using a fast storage media.

In order to reduce the time lost due to hard-disk access, and still use the C functions, the MELP functions were integrated in a dynamic link library (DLL): Matlabmelp_dll that was called from Matlab. As a guidance for the DLL implementation and function export/import were used the tutorials available at (Webb, 2003) along with the MELP implementation from Texas Instruments (TI, 1998). The library was loaded with: loadlibrary('Matlabmelp_dll', 'melp_dll.h').

The functions exported from the dynamic link library (EXPORT void melp_ana_init();), can be listed in Matlab by using the command: libfunctions Matlabmelp_dll -full:

These speech processing functions are called with calllib and will read and write data into/from arrays resulting in an important advantage over the previous implementation as there is no more dependence on the file system performance. The communication was completed successfully as the system was able to process the audio data in real time.

--Encryption/Decryption Block implementation

The encryption implementation used JCA (Sun, 2006), with the extensive use of extrinsic functions for Simulink implementation. The JCA has support for cryptographic services such as AES, DES, DESede, Blowfish, IDEA, etc. The AES encryption was chosen because it is widely used. NIST has defined 5 modes of operation for AES and other FIPS approved ciphers (Dworkin, 2001). Because of the problem of frequent data loss in mobile communications, in this implementation the Electronic Codebook Mode (ECB) was chosen with a 128 bit key (16 octets) so that the loss of a block will have no impact on the decryption of the following ones. However this mode of operation is not very secure, as encrypting identical data blocks lead to identical encrypted blocks. The encryption used the following code sequence:

The above Matlab functions were then called from the Embedded block with: eml.extrinsic('encryption').We also analyzed the implementation using the "feval" Matlab function but the resulted program complexity was very high.



The implementation was first tested on a single computer using the loopback adapter (with WireShark monitoring loopback packets in the Windows operating system, through the Microsoft Loopback Adapter), then in a real wireless 802.11g network with 4 computers (2.4GHz Core Duo) as seen in Fig. 2: two were participating in the Matlab communication (computers 1 and 2) and two helped simulate high network traffic in order to insure datagram drop (computers 3 and 4). The number of dropped packets was below 1%.

During the simulation and real wireless network communication testing the memory used by the simulator varied rapidly 30MB to 240MB and after about 15 minutes of simulation it has stabilized to approximately 240 MB, for all the rest of the simulation.

The output of the MELP coder compared to the input uncompressed signal is presented in figure 3. The received speech quality was good, with small moments of silence when the datagrams were dropped.



The lag in voice communication (due to both network latency and data processing) increased in the first 5 minutes of the simulation to 4 seconds and then stabilized to about 4 seconds for the rest of the simulation (more than 30 minutes). No such latency increase pattern was detected in the simulation on a single computer, via loopback address.


The results show that a real time implementation of a communication system testbead is possible using only existing external functions/programs for component implementation. This may be necessary when there is not enough time or enough information for a detailed system implementation.

The current simple implementation showed good tolerance to dropped datagrams, even if the resulted audio message becomes hard to understand as the number of dropped datagrams grows. The visual implementation in Simulink, even if it proved to be a challenging task, it is flexible enough to allow future optimizations such as external function integration through the use of MEX files or by using the memory mapping feature of Matlab for disk access. The encryption strength can be improved by using AES - CBC (Cipher Block Chaining) mode but the communication protocol will have to recover the lost data.

In the future the system will be modified for other audio encoders and encryption standards.


Basov O. (2008). MELP encoder and decoder in Matlab, 01/07/2008, Available from: matlabcentral/fileexchange/20530, Accessed: 2009-02-02

Dworkin M. (2001). Recommendation for Block Cipher Modes of Operation, National Institute of Standards and Technology--NIST Special Publication 800-38A, 2001 Edition, Available from: nistpubs/800-38a/sp800-38a.pdf, Accessed: 2009-02-02

Sun Microsystems. (2006). Class MessageDigest, Available from: http://java. sun. com/javase/6/docs/technotes/guides/ security/crypto/CryptoSpec.html, Accessed: 2009-02-02

TI--Texas Instruments Inc. (1998). Specifications for the Analog to Digital Conversion of Voice by 2,400 Bit/Second Mixed Excitation Linear Prediction, Available from: speech/melp/Download/, Accessed: 2009-02-02

Webb P. (2003). Programming Patterns: Calling Generic DLL Functions From MATLAB, The MathWorks Inc., Available from: news_notes/win03/patterns.html, Accessed: 2009-02-02
voidPtr feccode(voidPtr)
[int32, voidPtr] fec_decode(voidPtr, int32)
[singlePtr, voidPtr] melp_ana(singlePtr, voidPtr)
[int32, voidPtr, voidPtr] melp_chn_read(voidPtr, voidPtr)
voidPtr melp_chn_write(voidPtr)
[voidPtr, singlePtr] melp_syn(voidPtr, singlePtr)

k=javax.crypto.spec.SecretKeySpec(key, 'AES');
c=javax. crypto. Cipher.getInstance ('AES');

Tab. 1. External program processing time is lower that the time
being analyzed, therefore real time simulation is possible

 Action        Storage      Program Execution   Simulated
                Media       Time+Disk access    Real Time

Encoding     HDD 7200RPM      0.802 seconds     3 seconds
Synthesis    HDD 7200RPM      1.324 seconds     3 seconds
Encoding    USB 2.0 Stick     2.255 seconds     3 seconds
Synthesis   USB 2.0 Stick     2.785 seconds     3 seconds
Encoding       SD Card        2.562 seconds     3 seconds
Synthesis      SD Card        2.219 seconds     3 seconds

Tab. 2. Evolution of the voice delay with simulation time

Simulation time (min) 1    2    3    4    5    6    7
Delay (s)            1.0  2.1  3.4  4.1  4.2  4.2  4.3
Gale Copyright:
Copyright 2009 Gale, Cengage Learning. All rights reserved.