Trust-aware buffer-aided relay selection for secure communications in cooperative wireless systems

Abstract

Cooperative wireless networks have been facing security challenges due to the inherent openness of wireless channels. Previous studies on the physical layer security (PLS) of cooperative networks mainly focus on scenarios with only trusted relays. In practice, however, untrusted relays may exist in the network and such relays may refuse to cooperate or even work as a helper of an eavesdropper for packets eavesdropping/interception. To address this issue, we consider in this paper a cooperative wireless network with untrusted relays and a passive eavesdropper attempting to wiretap packets transmitted in the network. To select a secure and trusted relay, we first propose two buffer-aided relay selection schemes, one for perfect eavesdropper channel state information (CSI) case and another for no eavesdropper CSI case. We then conduct theoretical analysis to derive the models for the performance metrics of secrecy outage probability (SOP) and expected queuing delay. Finally, simulation and numerical results are provided for the validation of our theoretical models. The results indicate that our proposed schemes can outperform the typical max-ratio scheme under perfect eavesdropper CSI case and can outperform the max-link and the max-max relay selection schemes under no eavesdropper CSI case in terms of the SOP and expected queuing delay.

Introduction

Cooperative wireless networks, where network nodes cooperate with each other for information transmission, have attracted considerable attention due to their advantages of improving the network performances such as the network throughput and security performances [1]. However, the openness of wireless channels enables the launching of information eavesdropping/interception by unauthorized users [2]. Thus, ensuring security of communications has been pivotal to studies on cooperative wireless networks.

To secure the information transmitted in wireless networks, the traditional approach is to employ cryptographic algorithms [3]. However, these algorithms are complex and assume that unauthorized users (e.g., eavesdroppers) are with limited computation capability, which is challenged with the emergence of quantum computing [4]. To ensure an enhanced security in wireless networks, physical layer security (PLS) approach has been explored as a promising method. By utilizing the natural randomness of physical layer channels, PLS approach can ensure an information-theoretic security (i.e., perfect security) with a low complexity and regardless of the eavesdropper’s computational capability [5], [6]. Thus, PLS approach holds great potential to guarantee secure communications in wireless networks [7], [8].

To improve the PLS performance of wireless networks, lots of PLS schemes have been proposed. In these schemes, the mainly adopted PLS techniques include cooperative jamming, artificial noise injection, beamforming, traditional relay selection [9], [10] and buffer-aided relay selection [11], [12]. Among these techniques, the buffer-aided relay selection has showed its great potential to improve the PLS performances of wireless network due to its flexibility of being able to adjust the transmission schedule. In the traditional relay selection, packets are transmitted with a fixed source–relay–destination transmission schedule. This schedule generally selects the same message relay for packet transmissions in the two hops and packets received by the message relay have to be delivered to the destination right away even if the quality of the current link is too weak to ensure a secure delivery. With the aid of relay buffers, the buffer-aided relay selection, however, enables packets to be kept in the relay buffers currently for a secure delivery. Due to this, different message relays can be employed for the packet transmission in the two hops. This enables several possible transmission schedules in the buffer-aided relay selection , i.e., schedules of the source–relay–destination, the source–relay transmission, the relay–destination transmission and the source–destination transmission. Therefore, buffer-aided selection is more flexible and can be applied as a promising PLS technique to obtain significant security improvement in wireless networks.

Existing works to guarantee the PLS of cooperative wireless networks with buffer-aided relay selection mainly concentrate on networks with trusted relays [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29]. These studies can be summarized into network scenarios with a single relay and network scenarios with multiple relays respectively. When considering a single trusted relay, the authors in [13] considered cooperative networks with a trustful decode-and-forward (DF) half-duplex (HD) relay. Under assumptions of a fixed transmission rate and an adaptive transmission rate respectively, they proposed two link selection schemes to ensure secure and reliable communications in the network, and explored the secrecy outage probability (SOP) and secrecy throughput performances. In [14], the authors extended the work in [13] by considering cooperative networks with a trustful DF full-duplex (FD) relay. They aim to maximize the secrecy throughput performance of the concerned network by choosing the link with a better quality. In [15], the authors proposed a mechanism to create a better link by conducting artificial noise injection to the eavesdropper link. In [16], the authors explored the end-to-end (E2E) security performance of the scheme proposed in [13]. Considering the same network scenario as in [16], the authors in [17] explored the secrecy throughput maximization issue by combining the link selection and the power control method. In [18], the authors proposed a link selection scheme based on the link quality and the relay’s transmit power. In [19], the authors proposed a secure communication mechanism by applying the simultaneous wireless information and power transfer (SWIPT) to increase the secrecy throughput of the network. In [20], the authors proposed an energy-harvesting based link selection scheme to explore the achievable secrecy rate region of relaying wireless network.

Considering cooperative networks with multiple trusted relays, the authors in [21] proposed a relay selection scheme to choose the message relay with the maximum signal to noise ratio (SNR). Focusing on cognitive radio networks (CRN) with multiple trusted relays and a passive eavesdropper, the authors in [22] and [23] explored the secrecy throughput maximization issue by adopting the max-ratio scheme in [21] and the reinforcement learning and double Q-network methods. In [24], the authors considered MIMO systems with trusted relays and a passive eavesdropper. They selected the message relay by considering CSI of the legitimate channels and analyzed the SOP of the network. In [25], the authors focused on MIMO systems with trusted relays and multiple passive eavesdroppers. They proposed an algorithm by exploiting both the buffer-aided relay selection and jammer selection and examined the security performance regarding the secrecy rate. Considering the same network model as in [21], the authors in [26] proposed a relay selection scheme depending on the relay buffer states to achieve a reduced packet queuing delay performance. In our previous works [27] and [28], we explored the E2E security and delay performances of the max-ratio scheme by using the Markov chain theory and the queuing theory. For networks with trusted relays and a diversity-combining eavesdropper, we proposed in [29] a novel buffer-aided relay selection scheme to prevent the packets from the eavesdropper’s combining decoding. We also evaluated the security performance of our new scheme regarding the SOP.

We can summarize from the related work that buffer-aided relay selection can achieve an enhanced PLS performances for cooperative wireless networks. However, available works only consider ideal scenarios with trusted relays, the more practical scenarios with untrusted relays have been largely unexplored yet. In practical cooperative wireless networks, however, untrusted relays may also exist as relays can be mobile devices of human users. If a user has no social trust with the transmitter, he/she may refuse to help the information transmission. In an even worse scenario, he/she may exist as a helper of the eavesdropper with the information eavesdropping/interception. Although there are many existing studies that considered untrusted relays, the buffer-aided relays [30], [31] and the security issue (e.g., eavesdroppers) [32] have not been addressed therein. Therefore, we combine in this paper the social trust and PLS as the social trust between nodes greatly affects the buffer-aided selection and thus the PLS performances of the concerned system. Especially, we aim to explore new buffer-aided relay selection schemes to achieve PLS in cooperative wireless systems with multiple untrusted relays. We assume two cases of the eavesdropper CSI cases, i.e., perfect eavesdropper CSI ($\mathrm{PE}$) case (i.e., the instantaneous CSI of all eavesdropper channels are available) and no eavesdropper CSI ($\mathrm{NO}$) case (i.e., the eavesdropper’s CSI is totally unavailable). Table 1 illustrates differences between this paper and the existing works. These differences can be concluded from the following three aspects. (1) Different network scenario: this paper considers network scenarios of untrusted relays. Existing works, however, consider network scenarios of trusted relays; (2) Different relay selection scheme: due to the different assumptions of the relays, this paper selects the message relay based on CSI of the legitimate and eavesdropper links and the trust level between the source and relays, while existing works select the message relay without considering the trust level between the source and relays; (3) Different theoretical analysis: as we consider a special network scenario where the relays are untrusted relays, the theoretical analysis is different from that in existing works. The main contributions of the paper are summarized as follows.

• We first propose two buffer-aided relay selection schemes to select a secure and trusted message relay under the $\mathrm{PE}$ case and the $\mathrm{NO}$ case respectively. For the $\mathrm{PE}$ ($\mathrm{NO}$) case, our scheme chooses the relay with the maximum instantaneous secrecy capacity (with the maximum instantaneous channel gain of legitimate channels) from all candidate relays as the message relay.

• We then conduct theoretical analysis to derive expressions of the SOP and expected queuing delay, for exploring the security and delay performances of our proposed relay selection schemes under both $\mathrm{PE}$ and $\mathrm{NO}$ cases. To this end, we first model the packet transmission process from the transmitter to the receiver by using a state transition matrix. With the state transition matrix, we then derive the stationary probability of each state, the SOP and the average queuing delay at a state. Finally, the expressions of the SOP and expected queuing delay are obtained based on the stationary probabilities of all states, the SOP at a state and the average queuing delay at a state.

• Finally, we conducted extensive simulation and provided numerical results to confirm the efficiency of our theoretical model for the SOP and expected queuing delay. We studied how some key parameters affect the SOP and expected queuing delay performances. We also compared our new schemes with the typical max-ratio, max-link and the max-max relay selection schemes based on results of the SOP and the expected queuing delay under various network settings. The results show that our new schemes can outperform the typical max-ratio scheme (max-link and the max-max schemes) under the $\mathrm{PE}$ ($\mathrm{NO}$) case regarding the SOP and expected queuing delay.

The rest of this paper is structured as follows. Section 2 introduces the system model and the trust model. Section 3 provides the buffer-aided relay selection schemes, the transmission process and the performance metrics. Section 4 derives expressions of the SOP and expected queuing delay performances. Section 5 illustrates the numerical results and discussions. Finally, Section 6 concludes our work.