Physical Layer Security in Downlink Multi-Antenna Cellular Networks

I. INTRODUCTION

• The paper motivates physical-layer security as an alternative to classical network-layer cryptography for protecting wireless communications vulnerable to eavesdropping.It identifies randomly located external eavesdroppers as an additional concern beyond malicious users already considered in prior work.Source: §I · p. 2, A. Motivation and Related Work · p. 2

A. Motivation and Related Work

• Prior cellular study [22] considered single-antenna transmission with orthogonal resource allocation, while this paper models multiple antennas, multiple users, and RCI precoding.It also accounts for exact inter-cell interference and permits all users other than the typical user to cooperate in eavesdropping.Source: A. Motivation and Related Work · p. 3

B. Approach and Contributions

• The model includes eavesdropping by users in the same and other cells, inter-cell interference, and cooperation among malicious users.Base-station and user locations are modeled as independent two-dimensional Poisson point processes, and RCI performance is analyzed using stochastic geometry and random matrix theory.Source: B. Approach and Contributions · p. 4
• RCI precoding can achieve nonzero secrecy rates in some regimes, but unbounded transmit power leads to secrecy outage with probability one.The work also approximates secrecy-outage probability and mean per-user secrecy rate.Source: B. Approach and Contributions · p. 4
• Unlike isolated cells, cellular secrecy rate does not grow monotonically with transmit SNR, and optimal base-station density decreases as SNR increases.The optimal density trades off useful signal power, interference, and information leakage.Source: B. Approach and Contributions · pp. 4–5

A. Network Topology

• Base stations and single-antenna users are modeled by independent Poisson point processes, and each user connects to its closest base station.Each base station has N antennas and serves its connected users.Source: A. Network Topology · p. 5

B. RCI Precoding

• RCI precoding linearly processes confidential messages, and its regularization parameter ξ trades off legitimate-user signal power against crosstalk to co-served users.The method is considered as a low-complexity linear scheme that controls inter-user crosstalk.Source: B. RCI Precoding · pp. 5–6
• The cellular-network optimal ξ is unknown; the analysis instead assumes the value that maximizes large-system secrecy rate in an isolated cell.The results are stated to hold for any ξ, while its cellular-optimal calculation is left for future work.Source: B. RCI Precoding · p. 6

C. Malicious Users

• For a typical legitimate user, all other users in the network may act as eavesdroppers, divided into in-cell and external sets.The base station knows channels of in-cell users but not those of external malicious users.Source: C. Malicious Users · p. 7

A. SINR at a Typical User

• The legitimate receiver treats interference as noise, including crosstalk from other messages sent by its serving base station and signals from other base stations.The SINR expression also includes thermal noise and the useful signal.Source: A. SINR at a Typical User · pp. 7–8

B. SINR at the Malicious Users

• Malicious users within cooperation radius r_c can form a cooperation network; when user density is high enough, percolation yields an infinite cooperating cluster.The analysis adopts the worst case in which malicious users cooperate, cancel other messages, and act as an equivalent multi-antenna eavesdropper using maximal-ratio combining.Source: B. SINR at the Malicious Users · p. 9
• The worst-case malicious-user set can cancel messages intended for other users before eavesdropping on the typical user's message.The equivalent cooperating eavesdropper observes the useful signal in noise after interference cancellation.Source: B. SINR at the Malicious Users · pp. 9–10

C. Achievable Secrecy Rates

• An achievable secrecy rate is the positive part of the difference between the legitimate user's and equivalent malicious user's log-SINRs.The legitimate base station, typical user, and equivalent malicious user form a multi-input, single-output, multi-eavesdropper wiretap channel.Source: C. Achievable Secrecy Rates · p. 11
• The inter-cell interference power gain at the typical legitimate user follows Γ(K_b, 1), while the leakage power gain at an external malicious user follows exp(1).Source: C. Achievable Secrecy Rates · p. 12
• The section defines secrecy-outage probability and mean secrecy rate for the typical user; outage also denotes the fraction of time with no nonzero secrecy-rate transmission.Source: C. Achievable Secrecy Rates · p. 12

IV. LARGE-SYSTEM ANALYSIS

• The analysis approximates secrecy-outage probability and mean secrecy rate using a large-system regime in which user count and antenna count grow together.It replaces random quantities with deterministic equivalents.Source: §IV · p. 12, A. Preliminaries · pp. 13–14

A. Preliminaries

• For tractability, the model equates the average user counts per cell and approximates the tagged Voronoi region by an equal-area ball.It also approximates connected-user distances using the typical user's distance to the tagged base station.Source: A. Preliminaries · pp. 12–13
• The simplified model is presented as retaining key security characteristics and providing insight into how secrecy metrics depend on transmit power and base-station density.The paper identifies tractability as the reason for its assumptions.Source: A. Preliminaries · p. 13
• For N = 20, the simulated ergodic secrecy rate R and its approximation ˆR versus SNR follow the same trend, while cellular secrecy rate does not increase monotonically with SNR.Source: A. Preliminaries · p. 14, Figure 3 · p. 15
• The interference and leakage CDFs are well approximated by lognormal distributions across the tested base-station densities.The comparison uses ρ = 10 dB, N = 20, K = 20, ∥c∥ = r, and η = 4.Source: Figure 4 · p. 17, B. Characterization of ˆI and ˆL · p. 17

C. Probability of Secrecy Outage

• Theorem 1 provides an approximation for secrecy-outage probability using the interference and leakage distributions and the typical user's distance to its nearest base station.The result allows outage evaluation without computationally expensive Monte Carlo simulations.Source: C. Probability of Secrecy Outage · pp. 17–18
• As transmit power grows unbounded, cellular networks tend to secrecy outage with probability one, unlike isolated cells where sufficient antennas can drive outage probability to zero.Inter-cell interference and leakage also cause outage in cellular networks and cannot be controlled by the base station.Source: C. Probability of Secrecy Outage · pp. 18–19

D. Mean Secrecy Rate

• Theorem 2 approximates mean secrecy rate, enabling its evaluation without computationally expensive Monte Carlo simulations.A separate corollary supplies a lower bound that does not require the interference and leakage densities.Source: D. Mean Secrecy Rate · pp. 19–20
• Cellular mean secrecy rate can be nonzero, but interference and leakage limit it, so it cannot grow unbounded with SNR regardless of antenna count.The analysis states that mean secrecy rate approaches zero in the unbounded-SNR limit.Source: D. Mean Secrecy Rate · p. 19
• At fixed average users per base station, an optimal deployment density maximizes mean secrecy rate, and that density decreases as SNR rises.The paper describes the optimum as a tradeoff between useful power and intra-cell and inter-cell interference and leakage.Source: D. Mean Secrecy Rate · pp. 19–20

V. NUMERICAL RESULTS

• For N = 20, Fig. 5 compares simulated secrecy-outage probability ˆPo with the analytical result from Theorem 1, which is accurate for all λb values at relatively low SNR.Due to the lognormal approximations, the result is slightly less accurate.Source: §V · pp. 20–21, Figure 5 · p. 21
• For N = 20, Fig. 6 compares simulated mean secrecy rate with Theorem 2; the curves follow the same trend and the analysis is accurate across BS densities at relatively low SNR.At relatively high SNR, the analytical curve is less accurate because of the lognormal approximations.Source: §V · p. 21, Figure 6 · p. 22
• RCI can achieve a nonzero per-user secrecy rate, and more transmit antennas increase it, but the rate does not grow unbounded with SNR, even with many antennas.This contrasts with the isolated-cell secrecy-rate behavior.Source: §V · pp. 22–23
• RCI achieves a probability of secrecy outage below one, and more transmit antennas reduce outage; however, outage tends to one as transmit power grows unbounded, regardless of antenna count.This differs from an isolated cell, and Fig. 7 examines outage versus SNR for K = 10 users per BS and varying antenna counts.Source: §V · p. 22, Figure 7 · p. 23
• A BS deployment density maximizes secrecy rate, and the maximizing density is smaller at higher SNR.The rate is plotted against BS density for multiple SNR values.Source: §V · p. 23

• RCI precoding can achieve a nonzero secrecy rate with secrecy-outage probability below one despite neighboring-cell users cooperating to eavesdrop.Source: §VI · pp. 23, 25
• The secrecy rate does not grow monotonically with SNR, and the network tends toward secrecy outage as transmit power grows unbounded.Source: §VI · p. 25
• An optimal base-station density maximizes secrecy rate, and this density decreases as SNR increases.Source: §VI · p. 25
• The authors identify heterogeneous networks and exact characterization of interference and leakage power distributions as future research directions.Source: §VI · p. 25

APPENDIX A

• Under RCI, each confidential message is weighted by its user precoder; the derivation gives exponential single-link gains and an aggregate gain distributed as Gamma(Kb, 1).Source: Appendix A · p. 25

APPENDIX B

• The interference Laplace transform is derived using the independence of channel powers and base-station locations, the PPP probability generating functional, and the gain distribution.Source: Appendix B · p. 26

APPENDIX C

• Campbell’s theorem is used to obtain the means and variances of interference and leakage.Source: Appendix C · pp. 26–27

APPENDIX D

• Theorem 2’s proof substitutes approximation (24) into (16) and bounds the inner integration using 0 ≤ L̂ < τ(Î, ||c||).Source: Appendix D · p. 27

APPENDIX E

• The lower bound in (42) follows from Jensen’s inequality and Parseval’s theorem applied to the respective Fourier transforms.Source: Appendix E · p. 28