Please use this identifier to cite or link to this item:
|Energy-Efficient Secure Transmission Techniques for 5G-Enabled HetNet
Tekchandani, Raj Kumar
|Wireless networks;5G;Physical layer security;AI;DRL
|Heterogeneous Networks (HetNets) play an essential role in enhancing the qualityof-service (QoS) for end-users by increasing the spectral efficiency of the network and reducing the power consumption of user equipment (UE). With an exponential increase in the number of Internet of Things devices (IoTD), data traffic flow demands, and the complex network structure of 5G, the HetNets are also growing rapidly to increase the spectral efficiency of the wireless network. The market size of HetNets is expected to reach about 51.1 billion USD by the year 2027 as compared to 18.3 billion USD in the year 2020 with a compound annual growth rate (CAGR) of 15.2%. In contrast to traditional homogeneous networks, HetNets allow small cells to collaborate in macrocell networks, which increases the possibility of spatial resource reuse and improves the quality-of-service (QoS) for user equipment (UE). However, the dynamic and distributed nature of HetNets makes them susceptible to various types of attacks (e.g., eavesdropping, jamming). Also, new technologies of 5G such as Massive MIMO, mmWave, NOMA brings unique security concerns to 5G HetNets, which were not present in pre-5G HetNets. Implementing traditional security techniques such as access control, encryption, and network security seems to be insufficient for 5G HetNets and their inherent vulnerabilities. Also, HetNet’s architecture is more open and varied than traditional single-tier cellular networks, making information sharing more vulnerable to security threats. Thus, designing and implementing effective eavesdropping countermeasures is essential for secure wireless transmissions in 5G HetNets. Although, cryptography-based solutions have been widely used to provide network security at the upper levels. But, these solutions are limited in their ability to meet the security needs of 5G-and-beyond networks due to the following constraints i) It is extremely difficult to use cryptographic approaches using public keys in large, decentralized networks ii) Public-key infrastructure (PKI) has remained unbreakable until now in light of the usage of extremely long key pairs; however, advances in computing power, such as strong quantum systems, can now crack the cryptographic keys. PLS is based on the fundamentals of information theory and focuses on the security of propagation channel. It can be used in the 5G HetNets to efficiently degrade signal transmission efficiency at unauthorized receivers and applications to prevent them from obtaining sensitive data from the received signal. It ensures safe and efficient communications, even though eavesdroppers (illegitimate smart devices) are fitted with powerful computational devices in these networks. Some of the commonly used anti-eavesdropping techniques in HetNets include secure beamforming, cooperative jamming, and physical layer authentication (PLA). Beamforming (BF) is one of the promising PLS techniques to solve the issues mentioned above. At the transmitters and receivers, BF matrices can be used to shape the beam patterns of antennas to maximize a specific security parameter, such as signal-to-interference noise ratio (SINR), secrecy rate. Direct-sequence spread spectrum (DSSS) and frequency hopping (FH) techniques have been widely adopted as antijamming strategies in literature to mitigate the aforementioned issues. Particularly, FH is a sophisticated and commonly used technique which allows user equipment (UEs) to change their operating frequency to another frequency spectrum, which in turn avoids malicious jamming assaults. Besides DSSS and FH, power control is also an effective antijamming technique. Jamming cognition, decision-making, and joint optimization of beamforming and power allocation are the three fundamental phases of the power control aided anti-jamming communication cycle Although the existing proposals include efficient usage of various signal-processing PLS techniques, but these techniques suffer from the following constraints: i) Most of the pre-existing PLS techniques require prior and accurate knowledge of channel state information (CSI) values for effective PLS designs. However, it is difficult to obtain instantaneous global CSI of dynamic HetNet, since global CSI often varies frequently ii) Usage of large number of relays and active antennas in PLS techniques increases the power consumption. Also, cooperative jamming and broadcasting artificial noise require increased transmit power to achieve perfect secrecy iii) Most of the above mentioned studies utilize classical optimization techniques to optimize beamforming and power allocation vectors, which are less efficient in dynamic largescale networks. Also, FH and DSSS based anti-jamming techniques are constrained by their inherent dependency on pre-shared secrets (i.e., spreading codes and hopping sequences) between the communicating parties. Also, by using intelligent radio devices such as software-defined radio (SDR), the jammers can work together to block the wireless channels and disrupt the transmissions of FH-based UEs. Moreover, eavesdropping attacks and spectrum sensing on the control channel of 5G boost the jamming strength of malicious jammers in FH-based HetNets. As evidenced by widespread use of AI in different application areas of PLS, it is undoubtedly one of the most necessary elements for enhancing the PLS of 5G HetNets. It can be used to learn about normal and aberrant behaviors of HetNets based on how users and base stations communicate with one another. AI techniques can successfully anticipate future new instances by learning from existing instances. AI techniques can also be used to forecast new attacks, which are usually mutations of previous attacks. AI has been used in various PLS applications such as security ori- ented beamforming, cooperative jamming, PLA, secure handover schemes, etc. vi Deep learning (DL) methods, in general, rely on training and experience to improve task completion performance. This learning approach, which is a subset of machine learning (ML), analyzes the data for categoriza- tion and decision-making without programming. Also, Reinforcement learning (RL) algorithms can be used to design an optimal policy using the Markov decision process (MDP). Although RL-based techniques are viable solutions in designing PLS schemes, but the use of Q-learning method for the large state and action spaces suffers from stagnant learning speed, which may result in performance degradation of PLS techniques. DL has recently been integrated into RL techniques, allowing them to tackle a wide range of complicated problems. Deep reinforcement learning (DRL) is a set of approaches for estimating value functions or policy functions using deep neural networks. It uses Markov decision models to help choose between several actions based on state transition models. DRL have recently piqued the interest of the research community in designing intelligent PLS techniques for wireless networks. In this research work, the following schemes have been proposed to rectify the aforementioned issues: • Firstly, we propose a secrecy-aware energy-efficient scheme for a two-tier heterogeneous network (HetNet), consisting of a sub-6 GHz macrocell and multiple millimeter wave (mmWave) picocells. Each picocell is assumed to have several users and an eavesdropper (Eve) which intercepts the signal of the picocell users. In the proposed scheme, firstly, to maximize the secrecy energyefficiency (SEE) of picocell users, a joint optimization problem of power control, channel allocation, and beamforming is formulated by considering the minimum secrecy rate and signal-to-interference-plus-noise ratio (SINR) constraints. Due to the non-convex nature of the aforementioned optimization problem in a highly dynamic HetNet environment, we transform it into a reinforcement learning (RL) problem using the Markov decision process (MDP). Then, a multi-agent reinforcement learning (MARL) technique is used to obtain the maximum long-term reward. Moreover, we propose a multi-agent cooperative deep reinforcement learning (DRL) scheme known as SecBoost to solve the MDP with large number of action and state spaces. It uses the dueling and double-Q architecture of dueling double deep Q-network (D3QN) to optimize power control, channel allocation, and beamforming vectors to maximize the SEE of picocells. Also, prioritized experience replay is used to increase the sampling efficiency of SecBoost. The SEE performance of SecBoost is compared with MARL, multi-agent deep Q-network (MA-DQN), state-of-the-art joint beamforming based secrecy energy efficiency maximization (JBF-SEEM) scheme, and one-time pad based encrypted data transmission (O-EDT). Simulation results demonstrated that the proposed SecBoost vii scheme achieves 14.7%, 8.33%, 30%, and 69% better average SEE in comparison to MARL, MA-DQN, JBF-SEEM, and O-EDT schemes, respectively, which reveals its effectiveness in improving SEE of picocells. • Further,we propose a federated deep reinforcement learning (DRL) based antijamming technique for two-tier 5G HetNets. In the proposal, each femtocell of 5G HetNets is assumed to have multiple single antenna femto users (FUs) and a multi-antenna jammer used to jam the downlink signals from femto base station (FBS) to FUs. Aiming to improve the achievable rate at FUs in the presence of jammers, a joint optimization problem of beamforming and power allocation at FBSs is formulated by considering the quality-of-service (QoS) requirements of FUs. Due to the non-convex nature of the aforementioned optimization problem, we have used the Markov decision process (MDP) to transform the optimization problem into a multi-agent reinforcement learning (MARL) problem. Then, to solve this MDP with large number of states and action spaces, a federated deep reinforcement learning (DRL) scheme is proposed to maximize the achievable rate at FUs. The proposed scheme uses federated learning and dueling architecture of dueling double deep Q network (D3QN) to optimize the beamforming vectors and power allocation jointly at FBSs. The achievable rate performance of the proposed federated DRL scheme is compared with double deep Q network (DDQN) and deep Q network (DQN). Simulation results show that the proposed federated DRL scheme achieves 19.39% and 23.85% better achievable rate in comparison to DDQN and DQN schemes.
|Appears in Collections:
Files in This Item:
|901803020_Himanshu Sharma_PhD Thesis.pdf
|View/Open Request a copy
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.