PaperlayerView exact revision on arXiv

Source-linked AI summary

Cell-free Massive MIMO Networks: Optimal Power Control against Active Eavesdropping

Tiep M. Hoang, Hien Quoc Ngo, Trung Q. Duong, Hoang D. Tuan, Alan Marshall

arXiv:1805.04496v1cs.ITmath.IT

TL;DR

For user 1, the paper formulates optimization of achievable data rate or secrecy rate and also considers minimizing AP power under security requirements. Its reported numerical results say secure performance rises with user transmit power and falls with Eve’s average transmit power.

  • Problem

    The paper formulates maximizing user 1’s achievable data rate or secrecy rate in the presence of eavesdropping.

    Source: §III · p. 11

  • Method

    The paper optimizes a power-control matrix for user 1’s data or secrecy rate and formulates power minimization subject to security requirements.

    Source: §III · p. 11, §VII · p. 28

  • Results

    Secure performance increases with Pu and reduces with PE, Eve’s average transmit power.

    Source: §VI · p. 24

  • Takeaways & Limitations

    The equal-coefficient case ηmk=η is reported to have far worse security performance than the compared case.

    Source: §VII · p. 28

Abstract

from arXiv · show

This paper studies the security aspect of a recently introduced network ("cell-free massive MIMO") under a pilot spoofing attack. Firstly, a simple method to recognize the presence of this type of an active eavesdropping attack to a particular user is shown. In order to deal with this attack, we consider the problem of maximizing the achievable data rate of the attacked user or its achievable secrecy rate. The corresponding problems of minimizing the consumption power subject to security constraints are also considered in parallel. Path-following algorithms are developed to solve the posed optimization problems under different power allocation to access points (APs). Under equip-power allocation to APs, these optimization problems admit closed-form solutions. Numerical results show their efficiencies.

I. INTRODUCTION

Cell-free massive MIMO coordinates APs through a CPU, and its security had not been addressed in the cited cell-free studies. Prior work examined active eavesdropping and pilot spoofing in other settings, while this paper focuses on detection and security optimization for cell-free networks.

A. Previous Works

  • In cell-free massive MIMO, APs coordinate through a CPU, making the effective channel an inner product that can converge to a constant as the number of APs grows.The cited passage contrasts this with small-cell systems, which may require both downlink and uplink training for channel estimation.Source: A. Previous Works · p. 2
  • Security for cell-free massive MIMO had not been considered in the cited earlier work, motivating this paper’s focus.Earlier studies addressed user association, interference and power control, and energy efficiency.Source: A. Previous Works · p. 2

2) Pilot spoofing attack:

  • Pilot spoofing occurs when an active eavesdropper sends a spoofed pilot during training; prior studies examined such attacks in distinct wireless settings –.Those studies found active eavesdropping to be a threat across different types of wireless systems.Source: 2) Pilot spoofing attack · pp. 2–3
  • The paper uses one training phase to detect a potential eavesdropper, then focuses on optimization for cell-free systems when a user is suspected of being targeted.The authors distinguish their focus from earlier detection methods and do not treat detection as a major contribution.Source: 2) Pilot spoofing attack · p. 3

B. Contributions

The paper examines active eavesdropping in cell-free massive MIMO and proposes rate-maximization and power-minimization problems. Its model assumes reciprocal channels and an error-free backhaul, with imperfect-backhaul capacity left for future work.

B. Contributions

  • The paper presents its work as the first integration of security with the cell-free massive MIMO architecture.It distinguishes its analysis from prior massive-MIMO security work by using lower and upper SNR bounds to formulate a lower bound on secrecy rate.Source: B. Contributions · p. 3
  • A second group of four comparable optimization problems uses a common power-control coefficient for all APs.The first group instead controls power at each AP and compares secure performance and energy.Source: B. Contributions · p. 4
  • The paper proposes rate-maximization problems for an attacked user and power-minimization problems that preserve security requirements.The formulations include achievable data rate or secrecy rate objectives and security constraints on power consumption.Source: B. Contributions · p. 4

II. CELL-FREE SYSTEM MODEL

  • The system model has M APs, K users, and an active eavesdropper, with uplink training followed by downlink data transmission.The nodes are single-antenna and randomly positioned, and the model assumes channel reciprocity.Source: §II · pp. 5–6
  • The model assumes a sufficiently perfect backhaul for error-free information sharing; capacity limits from imperfect backhaul are left for future work.Channel reciprocity also requires highly accurate hardware calibration.Source: §II · p. 5

A. Uplink training

Uplink training supplies pilots for channel estimation, while an eavesdropper can spoof the targeted user’s pilot. The paper describes detecting abnormal signals and estimating channels with MMSE.

A. Uplink training

  • The first T symbols of the coherence interval are used for pilot training, and the remaining Lint−T symbols are used for data transmission.The paper notes that T≥K permits K orthogonal pilots and gives (T, K) = (150, 100) as a possible setting.Source: A. Uplink training · p. 6
  • Eve targets user 1 by transmitting the same pilot p1, creating a pilot-spoofing attack during training.The paper states that pilot sequences are standardized and public, allowing Eve to know them.Source: A. Uplink training · pp. 6–7
  • APs can identify a potentially harmed pilot from an abnormal sign in the received signal sequence.The paper says APs are unaware of the attack until they realize such an abnormal sign.Source: A. Uplink training · pp. 6–7
  • The AP uses MMSE to estimate the channels, and the paper states that the estimated channels for other users are uncorrelated.The eavesdropper’s attack places ρE in the denominator of the estimate for user 1’s channel.Source: A. Uplink training · p. 7

B. Downlink transmisson

The downlink analysis derives rate bounds for the intended user and Eve from their received signals, then expresses the relevant SNRs through a power-control matrix to define secrecy rate.

B. Downlink transmisson

  • Each AP beamforms using its channel estimate, with power-control coefficients determining the power allocated to each user.The received-signal analysis separates the desired signal, beamforming-gain uncertainty, multiuser interference, and noise.Source: B. Downlink transmisson · p. 8
  • The user’s achievable data rate is lower-bounded by applying a mutual-information bound to the desired signal and treating the other signal terms as noise.The signal terms are pairwise uncorrelated, and the lower-bound SNR is the desired-signal variance divided by the noise variance.Source: B. Downlink transmisson · pp. 8–9
  • Eve’s upper-bound analysis treats the other received-signal terms as noise and assumes Eve perfectly knows the channel gains, representing a worst case for security.The signal terms used in this analysis are pairwise uncorrelated.Source: B. Downlink transmisson · pp. 9–10

3) Achievable secrecy rate:

  • The achievable secrecy rate is expressed using the user and Eve SNRs, which are rewritten as functions of Ψ, a matrix representing AP-to-user power allocation.Each entry of Ψ corresponds to the transmit power an AP uses for a user.Source: 3) Achievable secrecy rate · pp. 10–11

III. SECRECY RATE MAXIMIZATION

The section formulates power-control problems to maximize the attacked user’s data rate or secrecy rate under AP power and security constraints. It uses path-following algorithms built from convex subproblems.

III. SECRECY RATE MAXIMIZATION

  • The optimization targets the attacked user’s achievable data rate or secrecy rate while retaining constraints on Eve’s information and other users’ rates.The AP power constraint also limits each AP’s average transmit power.Source: §III · pp. 11–12
  • The formulation controls each AP’s average transmit power, bounded by its maximum transmit power.The maximum power constraint is expressed as Pmax ≥ E{|xm|²}.Source: §III · p. 11

A. Solving problem (P1)

  • Each proposed path-following algorithm solves one convex quadratic program per iteration and can find at least a locally optimal solution.The method uses convex inner approximations and iterates from a feasible point.Source: §III · p. 12, A. Solving problem (P1) · p. 13

B. Solving problem (Q1)

Problem (Q1) is solved by iterative convex optimization from a feasible point. The power-minimization formulation targets total average AP transmit power subject to security and SNR constraints.

B. Solving problem (Q1)

  • For problem (Q1), the algorithm initializes at a feasible point for convex constraints (23) and (26), then iteratively solves a convex optimization problem to generate the next feasible point.The per-iteration cost is O((MK)^2(ε−1)^2.5 + (ε−1)^3.5).Source: B. Solving problem (Q1) · p. 15

IV. POWER MINIMIZATION

  • The power-minimization formulation minimizes total average transmit power across APs subject to security and SNR-based constraints.The objective is radiated antenna power; consumption by components such as the backhaul and CPU is beyond scope.Source: §IV · p. 16
  • The S1 program uses a designer-selected secrecy-rate threshold rφ, and the resulting secure performances can differ between R1 and S1.The formulation also imposes an SNR threshold for user 1.Source: §IV · p. 16

A. Solving problem (R1)

The section presents path-following solutions for R1 and S1, then specializes the optimization problems to equal power allocation across APs. For P1, satisfying the stated conditions reduces the objective to maximizing the common power value.

A. Solving problem (R1)

V. OPTIMIZATION UNDER EQUAL POWER ALLOCATION AT ACCESS POINTS

  • Under equal power allocation, ηmk is set to η for all APs and users, yielding specialized SNR expressions and reduced forms of P1, Q1, R1, and S1..Source: §V · p. 18

A. Closed-form solutions to (P1

  • For P1 under equal power allocation, once the stated conditions are satisfied, maximizing its objective is equivalent to maximizing η.The passage identifies conditions on the thresholds for this reduction.Source: A. Closed-form solutions to (P1) · p. 19

B. Closed-form solution to (Q1

The appendix derives closed-form solutions for (Q1), (R1), and (S1) by applying feasibility conditions and analyzing separate cases. For (Q1), the derivation uses a Lagrangian and critical-point conditions.

B. Closed-form solution to (Q1

  • For (Q1), condition (56) is necessary before solving, and the problem is then rewritten for the closed-form derivation.The derivation introduces a nonnegative Lagrange multiplier and considers separate cases.Source: B. Closed-form solution to (Q1) · p. 20
  • The optimal solution to (Q1) is obtained from the case analysis of its Lagrangian conditions.The passage states that the resulting optimal solution follows the preceding cases.Source: B. Closed-form solution to (Q1) · pp. 20–21
  • For (Q1), setting the derivative with respect to χ to zero yields two positive-real critical points, χ1 and χ2.For τ > 0, the method solves a system of two equations.Source: B. Closed-form solution to (Q1) · p. 20

C. Closed-form solution to (R1

  • For (R1), condition (56) with k ∈ {1, ..., K} is required, and the solution follows when either condition (58) or (60) holds.The passage identifies these as cases in which (R1) can be solved.Source: C. Closed-form solution to (R1) · p. 21

D. Closed-form solutions to (S1

  • For (S1), condition (56) is required for k ∈ {2, ..., K}, and the third constraint is rewritten as a quadratic inequality in η.The quadratic coefficients are specified in the derivation.Source: D. Closed-form solutions to (S1) · p. 21
  • When the quadratic coefficient is positive, the third constraint holds whenever its discriminant is nonpositive.This is one of the stated cases for satisfying the rewritten constraint.Source: D. Closed-form solutions to (S1) · p. 21
  • When the quadratic coefficient is negative, the constraint holds for positive discriminant and η between the two roots.The endpoints are the roots of the quadratic equation.Source: D. Closed-form solutions to (S1) · p. 21

VI. NUMERICAL RESULTS

The numerical results compare secrecy-rate and AP-power behavior under the proposed optimization problems. They report higher secrecy rates for (Q1) than (P1) under the chosen parameters, while AP-power trends depend on the problem and system size.

Simulation setup

  • The simulations use Hata-COST231 large-scale fading and randomly place nodes over a 1 × 1 km2 square, with 20 MHz bandwidth.The stated setup also uses a 1900 MHz carrier frequency.Source: §VI · p. 22

Secrecy-rate comparisons

  • With the chosen parameters, (Q1) has higher secrecy rate than (P1), and both rates increase with M.The reported secrecy rates also increase with Pu and decrease with Eve's transmit power PE.Source: Figure 4 · p. 24, Figure 5 · p. 24, §VI · pp. 25–26
  • Under the reported comparisons, flat allocation ηmk = η⋆ gives poor security, and secure performance increases with Pu but decreases with Eve's power PE.PE is treated as a measure of active-eavesdropping strength.Source: §VI · pp. 24, 27

Power-minimization results

  • For (R1), total AP power decreases as Ps increases and is lowest at Ps = Pmax, while its secrecy rate remains about 0.0953 nats/s/Hz.The rate is reported as nearly constant across the considered Ps values.Source: Figure 6 · p. 25, §VI · p. 27
  • For (S1), total AP power is about 0.0027 mW across the considered Ps values, while its secrecy rate is approximately zero as Ps changes.The reported (S1) power is lower than that of (R1).Source: Figure 7 · p. 25, §VI · p. 27

Power versus system size

  • For (R1), total AP power falls with M and rises with K; (M, K) = (70, 6) uses less power than (50, 10) while satisfying security constraints.The authors recommend large M and small K when the constraints remain satisfied.Source: Figure 8 · p. 26, §VI · p. 27
  • For (S1), the authors likewise recommend large M and small K, and its power requirement is lower than (R1) under the chosen setup.At sufficiently large M, total power approaches zero while secrecy rate is also around zero.Source: Figure 9 · p. 26, §VI · p. 27

VII. CONCLUSIONS

The paper formulates secrecy-rate maximization under quality-of-service constraints and power minimization under security requirements for a cell-free MIMO network with an active eavesdropper. Numerical results show that equal power-control coefficients perform far worse than coefficients that vary by AP and user; secrecy–energy trade-offs and pilot-intrusion prevention remain future directions.

VII. CONCLUSIONS

  • The paper formulates secrecy-rate maximization subject to quality-of-service constraints and power minimization subject to security requirements.The setting is a cell-free MIMO network with an active eavesdropper.Source: §VII · p. 28
  • Equal power-control coefficients ηmk = η produce far worse performance than coefficients that vary across APs and users.The paper considers both variable coefficients and one common coefficient for all m and k.Source: §VII · p. 28
  • A secrecy-rate and energy-consumption trade-off is proposed as a future problem for cell-free networks.The suggestion is based on numerical results and intuitive observations.Source: §VII · p. 28
  • Preventing Eve from intruding during pilot training is identified as a topic worth considering.Source: §VII · p. 28

APPENDIX

The appendix describes a pilot-training abnormality test that compares a statistic with a known no-attack threshold. It also derives an explicit expression for Eve’s SNR using channel-estimation-error assumptions and substitutions from earlier equations.

A. A Simple Method to Identify Abnormality in Pilot Training

  • APs use the Hermitian inner product of yp,m and pk, then combine information to calculate Y for checking whether Eve attacks user 1.The test compares Y with the known threshold Y|H0; Y > Y|H0 indicates an attack.Source: A. A Simple Method to Identify… · p. 28
  • For ρE > 0, Y|H1 > Y|H0, so comparing Y with Y|H0 supplies the stated attack decision rule.Y|H0 is identified as the only threshold APs need for checking pilot-training abnormality related to p1.Source: A. A Simple Method to Identify… · p. 28

B. Explicit expression for snrE

  • The explicit snrE derivation substitutes the channel estimate plus its error, assumes an independent Gaussian estimation error, and applies earlier equations.The error is distributed as CN(0, βmE − γmE); the derivation also uses γmE = αmγm1.Source: B. Explicit expression for snrE · pp. 29–30
Loading 1805.04496v1…