Successive Interference Cancellation for ISAC in a Large Full-Duplex Cellular Network

TL;DR

This work addresses how to order successive interference cancellation (SuIC) at the base station in a large full-duplex ISAC cellular network with monostatic radar detection. Using a stochastic-geometry model with Poisson-distributed base stations and users, the authors derive SINR-based outage probabilities for both decode-first and detect-first SuIC orders, incorporating intercell interference and residual self-interference. The analysis yields threshold behaviors: a target-distance threshold and a UE-power threshold that determine which SuIC order is advantageous, and shows that intercell interference can reverse intuitive advantages inferred from single-cell thinking. The results underscore the importance of carefully selecting SuIC order in dense ISAC deployments and highlight vulnerability sensitivities to RSI and interference, offering design guidance for practical ISAC-enabled networks.

Abstract

To reuse the scarce spectrum efficiently, a large full-duplex cellular network with integrated sensing and communication (ISAC) is studied. Monostatic detection at the base station (BS) is considered. At the BS, we receive two signals: the communication-mode uplink signal to be decoded and the radar-mode signal to be detected. After self-interference cancellation (SIC), inspired by NOMA, successive interference cancellation (SuIC) is a natural strategy at the BS to retrieve both signals. However, the ordering of SuIC, usually based on some measure of channel strength, is not clear as the radar-mode target is unknown. The detection signal suffers a double path-loss making it vulnerable, but the uplink signal to be decoded originates at a user which has much lower power than the BS making it weak as well. Further, the intercell interference from a large network reduces the channel disparity between the two signals. We investigate the impact of both SuIC orders at the BS, i.e., decoding $1^{st}$ or detecting $1^{st}$ and highlight the importance of careful order selection. We find the existence of a threshold target distance before which detecting $1^{st}$ is superior and decoding $2^{nd}$ does not suffer much. After this distance, both decoding $1^{st}$ and detecting $2^{nd}$ is superior. Similarly, a threshold UE power exists after which the optimum SuIC order changes. We consider imperfections in SIC; this helps highlight the vulnerability of the decoding and detection in the setup.

Figures (6)

• Figure 1: A snapshot realization of the network. All the nodes of interest in the typical cell have been shown. Solid (dashed) arrows show signals of interest (interfering signals). The color of the link distance matches it arrow.
• Figure 2: The probability of detection or decoding vs. $\theta=\theta_b=\theta_u$; $R_1=5v$ and $\zeta=10^{-12}$. Lines (markers) represent the analysis (simulations).
• Figure 3: The probability of detection or decoding vs. $R_1$. We use $\theta_u=-30$ dB, $\theta_b=-60$ dB and $\zeta=10^{-9}$.
• Figure 4: The probability of detection or decoding vs. $P_u$. We use $R_1=7v$, $\theta_u=-30$ dB, $\theta_b=-60$ dB and $\zeta=10^{-9}$.
• Figure 5: The probability of detection or decoding vs. $R_1$ and $P_u$, using $\theta_u=-40$ dB, $\theta_b=-60$ dB and $\zeta=10^{-9}$.