Foto
21Jul

Network protocol assignment

This is a  solution of network protocol in we discuss about the use and working of the same.

Area of research

The area of interest is voice communications, for which speech quality measurements are done using non-intrusive neural network approach. The paper reviewed proposes a solution to extend the E-model to any new network conditions and for newly emerging codecs in a Voice over Internet Protocol (VoIP) network, eliminating any expensive and time consuming subjective tests.

 Which research method has been applied?

The research method is based on extending the use of the well-known E-model and utilizing an artificial neural network (ANN) approach for measuring speech quality. The research proposal is structured into different sections, where it starts from a review of existing models for the measurement of speech quality. It is then followed by another review of the E-model and the main drawbacks and problems associated with it. In the later sections, various techniques are discusses for extending the E-model to the artificial neural network approach, followed by an outline of the derivations steps for the proposed technique in detail. The last portion of the research focuses on, and discusses the results of the proposed extension, followed by conclusions and future possibilities and the scope of possible further development in the proposed technique.

What is the motivation for this work?

Public switched telephone networks (PSTN) is a traditional application used in the world of communication for transmitting voice over networks. A new technology to this end, Voice over Internet Protocol (VoIP) offers low operating and management expenses, and creates new avenues for innovation by combining voice and data applications into one network. This technology has emerged as an advantageous and attractive application for network security systems. However, for companies using VoIP, it is very important to provide acceptable quality of service to the customers for legal, technical and commercial reasons. As a result, measurement of the quality of voice calls has become extremely detrimental in the networking environment study. There are various existing methods for such measurement, and selection of the most appropriate model depends on the characteristics of the IP network and that of voice calls. The non-intrusive measurement of quality, without any human interference and without any need of speech signal measurement on the sender’s side, is the motivation behind this research. The primary aim is to provide an automated solution to the task, taking into consideration the voice quality measurements in the networks running in a real environment.

What is the proposed solution?

The review of the E-model identifies a primary problem in its applicability, which is the calibration of the existing E-model requires extensive subjective tests, which are costly as well as time consuming. Instead of performing these expensive tests, the proposed solution provided by this paper uses the latest version of the best quality prediction methodology – PESQ, as the intrusive reference criteria for accurately predicting the E-model parameters. The proposed extension of the E-model also uses an ANN approach for approximating the required functions. A 2-state Gilbert model is used for simulating the packet losses and encoding the reference speech signal. These simulations result in a degraded stream which can be decoded to give the degraded speech signal. The reference signal and the retrieved degraded signal can be compared using the PESQ model, and measurement can be done for the speech quality, assuming that these measurements done using PESQ comply with those done with the E-model.

How did the authors evaluate the solution?

The authors have defined MOS value as the measurement of speech quality as perceived by the user. For appropriate mapping of the proposed extension, it is important to compare these values with the E-model values. The success of the ANN prediction, as evaluated by the authors, lies in the extension of the E-model to the derived MOS values using the proposed ANN approach. It is highlighted in the solution that the correlation factor in the E-model’s MOS and the ANN’s MOS values is strongly in agreement and indicates good estimation. Visual comparison done between the two MOS values indicates that there are hardly any significant differences between their characteristics. The scatter diagram of correlation between the two models show a perfect fit and the difference in quality prediction is relatively towards the lower range showing strong correlation. The proposed solution, as evaluated by the authors, is applicable for non-intrusive monitoring of live-traffic and predicting conversational speech quality.

In your opinion how satisfactory was the solution?

The proposed solution can be readily integrated with the networks in a live environment, and a new speech coder can apply the integration to the E-model as soon as the objective tests are performed. Many research studies have done till date, which focus on extending the E-model approach with the intrusive PESQ method of speech quality prediction. However, they did not consider the significance of burstiness in packet loss. Also, the experiments done using silence insertion and repetition packet loss algorithms, incorrectly assumed the equipment impairment factor, therefore, making their results unreliable. Furthermore, the results of the speech quality prediction were not validated against the original E-model, and the choice of fitness curve was not properly justified while extending the E-model. This paper, on the other hand, uses the latest E-model and appropriately takes into account the effect of business of packet loss. Also, the accuracy of the evaluation is justified by comparing the quality of prediction obtained using the proposed model with those obtained from the E-model used in the research.

What are the main contributions of the paper?

The paper highlights a method of extending the E-model for predicting the impairment factor from packet loss, and derives the burst ratios based on the ANN approach. Also, the objective tests conducted with the PESQ method eliminates the need of repetitive subjective tests required for the calibration of E-model, which were expensive and time consuming. Using this method, it will now be possible to extend the E-model towards new network conditions and new speech codecs. The impairment factor and the burst ratio obtained using this method provide the possibility of applying the E-model to a new speech codec, within no time. Most importantly, this method can be widely applied in the rapidly changing communication environment, for estimating speech quality for voice application over a wide range of IP networks.

What is not clear in the paper?

The linear and non-linear regression method of calculating the burst ratio and impairment factor of packet loss is not used while comparing the MOS measurements done with ANN approach to those done with the E-model. Also, the room size, noise level and sound pressure level are not clearly taken into account within the laboratory settings of the experiment. In addition to that, the PESQ method does not take into account the channel distortions and filtering delays of the low-bit rate codecs which may be encountered in a real environment.

References

 Al-Akhras, M. Zedan, H. John, R. and Al-Momani, I. 2007. Non-intrusive speech quality prediction in VoIP networks using a neural network approach. Neurocomputing.72(2009), pp. 2595-2608

Also Read: - Company Business Plan Assignment

Assignment Help Australia provide best quality assignment writing service in affordable prices and we are providing most flexible assignment writing according to Students need, so book your Assignment with us, order now