Dept. of Electronics and Communication Engineering

Indian Institute of Information Technology Sri City, Chittoor, India.

(An Institute of National Importance under an Act of Parliament)

BTP Code : B24RK01

Deep Learning Based Cooperative Relay Selection Policy in Virtual MIMO

Presented by:

Hruday Chowdary G

(S20210020278)

Sai Yaswanth P

(S202100020310)

Under the supervision of

Dr. Rajeev Kumar

Assistant Professor

​

Contents

01

Introduction

Problem Statement

02

Literature Review

03

Previous work

04

Results and Discussion

06

Conclusion

07

08

Timeline

09

Present work flow

05

References

10

Introduction

​

• Previous work in cooperative relay selection for Virtual MIMO systems focused on implementing prediction algorithm models to identify the willingness of inactive users in a VAA (Virtual Antenna Array) cell.

​

• However, these models were limited to scenarios with only 30-50 inactive users per cell. In real-world applications, particularly in rural to sub-urban areas, the number of users in each cell can range from 1000 to 10000.

​

​

Problem Statement

Performance of the cooperative relay selection system in virtual-MIMO using deep learning algorithms.

Literature Review

Previous Work

​

• After the first evaluation, we have decided to generate an artificial dataset with the data given in the base paper and the labels are given accordingly. If the willingness of the user is found then it then we will label them 1 and 0 otherwise.

​

• In the second evaluation, we have discussed briefly about the wireless communication topics mentioned in the project and a DL model to predict the willingness.

​

• In the third evaluation, we achieved a significant reduction in prediction time, decreasing from 636µs to 99µs, which is approximately 84% faster. We also developed a sequential model that not only reduces the prediction time but also offers better accuracy compared to the base prediction model.

Dataset Distribution

Fig(a) - Battery power Distribution v/s no. of inactive users

Fig(b) - Active Days in a week v/s no. of inactive users

Contd.

Fig(c) - Active hours in a Day v/s no. of inactive users

Fig(d) - Incentive Amount Distribution v/s no. of inactive users

Target Distribution

Fig(e) - Willingness of the user to participate v/s no of inactive users

Data Preprocessing

• The dataset only consisted of battery_power, time, day but we also included a new feature to the dataset called the incentive amount with random distribution.

​

• We create binary target labels based on a predefined relationship between the features.

​

​

​

​

Proposed Deep Learning Sequential Model

• Input Layer: Dense(64, activation='relu', input_shape=()): Learns initial patterns from input features.

​

• Dropout Layer(0.2): Reduces overfitting by dropping 20% of neurons.

​

• First Hidden Layer : Dense(32, activation='relu'): Extracts refined patterns from previous layer.

​

​

• - Output Layer : Dense(1, activation='sigmoid'): Outputs probability for binary classification.

Results and Discussion

Contd.

Contd.

(Inference time – time taken by the trained prediction model to predict the new data)

​

We don’t have to worry about the increase in the model size as a eNodeB has a RAM size of 4-16GB.

Establishing secure communication

Nodes generate private and public keys using ECC

Nodes share and receive certificates from the CA

Nodes verify each other’s certificates using ECC math

Nodes compute a shared secret using ECC scalar multiplication

The shared secret is hashed to create an encryption key

Encrypted messages are exchanged using the encryption key

Shared keys are updated periodically for enhanced security

Establishing secure communication

​

​

• Implementation of SAKMS (Secure Authentication and Key Management Scheme) tailored for IETF 6TiSCH industrial wireless networks. The 6TiSCH protocol enables low-power, high-reliability communication in resource-constrained industrial environments. Ref[5]

​

• Certificate (A) = Hash of Device Id (A) + Random Factor(A) * Private Key (C.A)

​

• P_a = Hash of Device Id (A). Public Key(A) + Random Factor(A) * Base Point (G)

​

• Shared Key = Private Key(A). Public Key(B)

​

• Communication is established and the messages are encrypted by the hashed shared key.

​

​

Results

identifying SUCCESSFUL key generation- Checked for 10 communication

Conclusion

​

​

​

• For the prediction model, we have enhanced it’s performance to make it work fast and more precise using the sequential model.

​

• Achieved significant decrease in time to identify faster and with least model size.

​

• A security protocol is used using SAKMS for wireless networks using ECC cryptography. This used random window method where shared key is updated every 10 sec (window length =10 sec in this project)

using a time based factor (example : taken as x^3 in this project).

​

• SAKMS ECC cryptography also focuses on enabling low power and high reliability which suits well for our prediction algorithm where high reliability is achieved through V-MIMO and the security protocol enables no attacks possible in the system.

​

Timeline

Literature Review &

Problem Statement

• Building a model with low interference time
• dataset (Achieved)

Implementation with artificial dataset and use proposed techniques

(Achieved)

Improve prediction algorithm and security protocol (Achieved)

Evaluation

1

Evaluation 2

Evaluation 3

Evaluation 4

References

• S. Dang et al., "Combined Relay Selection Enabled by Supervised Machine Learning," in IEEE Transactions on Vehicular Technology, vol. 70, no. 4, pp. 3938-3943, April 2021, doi: 10.1109/TVT.2021.3065074[1].

​

• Kunal Sankhe, Chandan Pradhan, Sumit Kumar, Garimella Ramamurthy: Machine Learning Based Cooperative Relay Selection in Virtual MIMO. CoRR abs/1506.01910 (2015)[2]

​

• A. A. A. El-Banna et al., "Machine Learning-Based Multi-Layer Multi-Hop Transmission Scheme for Dense Networks," in IEEE Communications Letters, vol. 23, no. 12, pp. 2238-2242, Dec. 2019, doi: 10.1109/LCOMM.2019.2941932.[3]

​

• H. Liu, Y. Xiao et al., "Transmit Antenna Selection for Full-Duplex Spatial Modulation Based on Machine Learning," in IEEE Transactions on Vehicular Technology, vol. 70, no. 10, pp. 10695-10708, Oct. 2021, doi: 10.1109/TVT.2021.3111043.[4]

​

• W. Yang, C. Hou, Y. Wang, Z. Zhang, X. Wang and Y. Cao, "SAKMS: A Secure Authentication and Key Management Scheme for IETF 6TiSCH Industrial Wireless Networks Based on Improved Elliptic-Curve Cryptography," in IEEE Transactions on Network Science and Engineering, vol. 11, no. 3, pp. 3174-3188, May-June 2024, doi: 10.1109/TNSE.2024.3363004[5].

​

�

Thank You