国际会议演讲稿

国际会议演讲稿

自我介绍

Thank you, Mr./Ms. Chair. /professor

My name is sang qian. I am very honored to be here to do oral presentation.

I am a Master student from Hohai University and I am currently doing some research on physical layer security.

Today, I would like to share with you some of my research on relay selection in cooperative communication. (external /ekˈstərnəl; ɪkˈstərnəl/)

内容安排：

My presentation includes these five parts.

First, some background information about this research;

Second, system model we have done;

Third, NN-based relay selection scheme we have proposed

Forth, Simulation and results analysis

And last, some conclusions we have got

P4

Part one, introduction

Firstly, I would like to give you a bit of background.

Differing from the traditional cryptographic techniques based on secret keys, we can make use of wireless channel characteristics to enhance physical layer security.

Cooperative communication has been widely recognized as an effective way to combat wireless fading and provide diversity gain which is one of the research hot spots.

Machine learning as an emerging technology has been widely applied in image processing, cancer prediction, stock analysis and other fields. So why not try it in wireless communication?

P5:

Next, I want to talk a little bit about present study

Recent studies on deep learning for wireless communication systems have proposed alternative approaches to enhance certain parts of the conventional communication system such as modulation recognition、 channel encoding and decoding、channel estimation and detection and an autoencoder which can replace the total system with a novel architecture

【modulation recognition:An NN architecture for modulation recognition that consists of a 4-layer NN and two two-layer NNs。

channel encoding and decoding：A plain DNN architecture for channel decoding to decode k bits

messages from N bits noisy codewords。

channel estimation and detection ：A dense-Net for symbol-to-symbol detection can adopt long short-term memory (LSTM) to detect an estimated symbol.

Autoencoder：the autoencoder can represent the entire communication system and jointly optimize the transmitter and receiver over an AWGN channel.

】

P6

So why did we conduct this research? Well, we want to exploit the potential benefits of deep learning in enhancing physical layer security in cooperative( /kəʊ'ɒpərətɪv/  wireless communication and reduce the feedback overhead in limited spectrum resouce by our our proposed scheme.

P8

Now let me move onto part two -system model

Here, you can see a figure which is a system model.

This is the source ; these are the relay nodes and this is the destination ,this is the eavesdropper

The whole process of cooperative wireless communication can be divided into two phases.

In the first phase, the source broadcasts the signal to the optimal relay which guarantees perfect security. As shown in Fig 1,represents a fading coefficient of the channel from the source to the relay node( . )

In the second phase, the optimal relay forwards a scaled version of its received signal to the destination in the presence of the eavesdropper, where the optimal relay is considered to adopt amplify-and-forward (AF) relay scheme.

In this figure,

represents a fading coefficient of the channel from the relay to the destination

represents a fading coefficient of the channel from the relay to the eavesdropper.

P9:

Here you can see some following expressions. I am not going to waste our precious time on the lengthy derivation. I would like to invite you to directly take a look at the equation in its final form.

This is the optimal index of the selected relay with the conventional relay selection scheme.Amaong this expression

represents the achievable secrecy rate of system model when the relay is selected.

P11

Now let me move to part three -----NN-based Relay Selection

Here you can see a figure which shows conventional 3-layer neural network . It consists of input layer, hidden layer 1, hidden (/'hɪdn/)layer 2 and output layer. Neural network can learn features from raw data automatically and adjust parameters（/pəˈræmɪtə(r)z/）flexibly( /'fleksəbli/) such as weights and biases.

In complex ( /'kɒmpleks/)conditions (scenarios(/sɪ'nɑːr ɪəʊ/),) Neural network has promising applications in relay selection for several reasons.

First, the deep network has superior(/suːˈpɪərɪə/) learning ability despite(/dɪ'spaɪt/)the complex channel conditions.

Second, Neural network can handle large data sets because of distributed(/dɪ'strɪbjʊtɪd/) and parallel(/'pærəlel/)computing(/kəm'pjuːtɪŋ/s, which ensure computation(/kɒmpjʊ'teɪʃ(ə)n/)speed and processing capacity( /kə'pæsɪtɪ/).

Third, various libraries or frameworks, such as TensorFlow, Theano,and Caffe give it wide applications

In this paper, the problem of the relay selection is modeled as a multi(/'mʌltɪ/,ao)-classification problem. We adopt simple neural network(NN) to select the optimal relay to guarantees perfect secrecy performance of relay cooperative communication system.(enhance physical layer security)

P12

Before training the classification model, we need to make some preparation for deep learning to acquire a training set and a testing set.

First, we need to produce real feature vector for each example according to channel state information;because the channel state information matrices is composed of complex numbers but feature vectors are generally composed of real numbers.So we need to change complex numbers into real numbers with absolute(/'æbsəluːt/) value operation. Moreover, in order to improve the classification performance(precision), it is necessary to normalize the feature vectors.

Second, we need to design key performance indicator (KPI).In order to effectively prevent the eavesdropper from intercepting information, we choose achievable secrecy rate as the KPI of system. This KPI indicates(represents/shows) the difference of the achievable rate from the source to the destination and the achievable rate from the source to the eavesdropper.

Third,we can make labels for examples according to KPI. the index of the relay which obtains the maximum( /'mæksɪməm/) KPI is regarded as the class label of the example.

P13

Classification model

This picture(is about) shows the whole process of building classification model.

The whole process of building classification can be divided into two phases, namely training phase and testing phase. In the first phase, we need to choose suitable hyper ( /'haɪpə/)parameters to train neural network model. In the second phase, we can predict(/prɪ'dɪkt/) labels of optimal relay according to input data and assess classification performance.

P15

Now let me move to part four -----Simulation and Results( /rɪˈzʌlts/) Analysis

Here, you can see a figure which shows the relationship between the average transmit ( /trænzˈmɪt/)power of the source and the achievable secrecy rate with different numbers of relays.

In this figure, the blue line represents the conventional relay selection scheme and the red line represents the NN-based scheme.

In this figure, as the numbers of relay and the average transmit power of the source increases, the achievable secrecy rate increases accordingly,.which means increasing the number of relays can effectively improve the secret performance.

The red line are almost( /'ɔːlməʊst/) close to the blue line, 【whch indicates that our proposed scheme (i.e. the NN-based scheme) achieves almost the same secrecy rates as those of the conventional scheme for all values of , 】which validates effectiveness(/ɪ'fektɪvnɪs/) of our proposed schemes.

P16

This table shows the the normalized(/ˈnɔrməˌlaɪzd/) mean square( /skweə/) error(NMSES)values of dirrerent relay nodes. The value of NMSE means the performance difference between the conventional scheme and our proposed scheme. The values of NMSE are below(/bɪ'ləʊ/) negative('negətɪv/) 20 (), which validates effectiveness of our proposed scheme again.

P17

Now, let me move to the last part -----Conclusion

Okay, now we are going to take a look at the last part -Conclusion.

P18

We have got the following conclusions.

First, In complex (conditions)scenarios, Neural network has promising applications in relay selection for superior learning ability , computation speed and processing capacity.

Second, Compared with the conventional relay selection scheme, our proposed scheme achieves almost the same secrecy performance.

And last, Our proposed scheme has an advantage(/əd'vɑːntɪdʒ/) of relatively small feedback overhead, indicating that proposed scheme can be applied to the conditions (scenarios）where the feedback is limited.

(If the conventional scheme needs feedback of complex numbers, NN-based scheme will only need feedback of real numbers. Therefore, the feedback overhead of our proposed scheme is half(/hɑːf/ ) of that of the conventional scheme,)

Q&A

1、计算复杂度

Computational complexity

The biggest drawback is the highly selection complexities with a small number of relay nodes.

If number of relay node is big, it will have a advantage. This need our further research.

Q: The experiment shows that secrecy rate is almost the same as traditional method and what is the promotion of using NN to relay selection.(what is meaning of introducing NN to relay selection)

A:That our proposed scheme (i.e. the NN-based scheme) achieves almost the same achievable secrecy rate as that of the conventional scheme indicates that our proposed scheme is effective and it can select optimal relay node which obtains maximum achievable secrecy rate.

One reason (the first reason) is that Adopting NN for relay selection is a novel idea.

Another reason is that the spectrum resource is relative limited and our proposed scheme has small feedback overhead.

Q:what’s the meaning of “perfect secrecy performance”? What’s the meaning of “Compared to the conventional relay selection scheme”?

A: “perfect secrecy performance” means the achievable secrecy rate is the biggest one which can enhance physical layer security.

In fact, the conventional relay selection scheme is the exhaustive search. The index of relay selection with this scheme is the best one.

Q:“It is obvious that the feedback overhead of proposed scheme is half of that of the conventional scheme”

A: well, Let's make an assumption. If the conventional scheme needs feedback of complex numbers, NN-based scheme will only need feedback of real numbers. Therefore, the feedback overhead of our proposed scheme is half(/hɑːf/ ) of that of the conventional scheme

Q:1)The training data set and the testing data set are so called "legitimate channel complex matrix and the wiretap channel complex matrix". 2)How these data can be obtained/generated? 3)Why it is important and significant as the input of the NN-based approach and what is the output? The whole process is not clear.

A: 1)the deep learning needs training set and testing set. training set is used to train model and testing set is used to asses the model’s performance. 2) it can be obtained according to channel state information where the legitimate links and wiretap links are modeled as Rayleigh fading channels. 3) the channel state information is important feature(indicator) in wirelesss communication. the communication the NN can learn features from raw date

Q:

A:

Q:

A:

本文来源：https://www.2haoxitong.net/k/doc/ad6c100a65ec102de2bd960590c69ec3d5bbdb3e.html