Source-linked AI summary
Performance Analysis and Sensing-Aware Resource Allocation for Full-Duplex Massive MIMO ISCC Networks
Vo Phi Son, Van-Dinh Nguyen, Nhan Thanh Nguyen, Tuan-Vu Truong, Symeon Shatzinotas
TL;DR
In FD massive-MIMO ISCC, radar probing interferes with uplink reception while user transmissions degrade target estimation, coupling sensing accuracy with offloading latency. The paper derives rate and sensing bounds and jointly allocates resources; its algorithm achieves lower latency than the considered fixed-allocation and half-duplex benchmarks.
Problem
In FD ISCC, radar probing interferes with data reception while user transmissions degrade target estimation, coupling sensing accuracy with task-offloading latency.
Source: §I · p. 1Method
The paper derives an imperfect-CSI ZF uplink-rate bound and angular CRLBs, then uses an SCA-based algorithm to jointly allocate offloading, transmit powers, and edge CPU resources.
Source: B. Motivation and Main Contributions · p. 2, §VI · p. 11Results
The SCA-based joint allocation achieves lower latency than the considered fixed-allocation and half-duplex benchmarks.
Source: §VI · p. 11Takeaways & Limitations
When user powers decrease inversely with receive-array size, the analysis identifies a finite uplink-rate limit and motivates coordinating radar power with user-power reduction.
Source: §VI · p. 11
Abstract
from arXiv · showhide
Emerging edge applications must sense their environment and process data within tight latency budgets. Full-duplex (FD) integrated sensing, communication, and computing (ISCC) supports these functions concurrently, but radar probing interferes with uplink reception while user transmissions degrade target estimation. These coupled effects call for joint sensing and offloading decisions. We investigate a FD massive multiple-input multiple-output (mMIMO) ISCC network in which users partially offload tasks to an edge server while an access point probes a target. For zero-forcing uplink reception with imperfect channel estimates, we derive a tractable rate bound and target-angle Cramer-Rao lower bounds (CRLBs) under the stated interference and beam-alignment assumptions. We then jointly optimize offloading fractions, user and radar transmit powers, and edge computing resources to minimize average end-to-end task latency subject to sensing-accuracy, communication-rate, energy, deadline, and resource constraints. A successive convex approximation algorithm is developed to efficiently address the resulting nonconvex problem. The analysis shows that larger arrays improve sensing and uplink performance, but residual radar interference creates a finite rate ceiling when user powers scale down with array size. Simulation results show joint allocation reduces radar transmit power by approximately 86% relative to equal communication-user power at a -40 dB CRLB threshold and lowers latency compared with fixed-allocation and half-duplex schemes.
I. INTRODUCTION
FD-mMIMO ISCC lets an access point sense while receiving uplink data, but radar and user transmissions interfere with each other and couple sensing accuracy to offloading latency. The paper analyzes these links and jointly allocates transmission, offloading, and computing resources in a network with MEC and conventional users.
A. Related Work
- Prior FD-ISAC work addresses sensing–communication interference, while MEC studies coordinate offloading; here, shared-array users and uplink interference add coupling among power, sensing, and task completion,,.Massive arrays are motivated by their spatial gain and interference suppression,.Source: A. Related Work · pp. 1–2
B. Motivation and Main Contributions
- The analysis derives a ZF uplink rate bound with imperfect CSI and angular CRLBs that account for channel-estimation errors, residual self-interference, target echoes, and stated beam-alignment assumptions.The expressions characterize how radar and user powers affect both offloading rates and target-estimation accuracy.Source: B. Motivation and Main Contributions · p. 2
- With fixed user powers, uplink rate increases with receive-array size, but approaches a finite limit when user powers scale as 1/Nr with radar power and transmit-array size fixed.Transmit-array gain and receive-array resolution play different roles in improving angular accuracy.Source: B. Motivation and Main Contributions · p. 2
- The proposed SCA-based allocation jointly chooses offloading fractions, user and radar powers, and edge CPU resources to minimize average end-to-end latency under sensing, rate, energy, deadline, and resource constraints.The formulation uses large-scale channel information and approximates coupled rate, delay, and energy terms through convex subproblems.Source: B. Motivation and Main Contributions · p. 2
C. Paper Organization and Notations
- The paper compares its analytical results with Monte Carlo simulations and evaluates convergence and latency against fixed-allocation and half-duplex schemes across sensing thresholds, antenna configurations, task sizes, and user loads.The introduction describes the paper’s progression from system model and performance analysis to optimization, numerical results, and conclusions.Source: B. Motivation and Main Contributions · p. 2, C. Paper Organization and Notations · p. 2
II. SYSTEM MODEL
- The system uses partial offloading: MEC users send task fractions to the edge server while executing the remainder locally, with local processing parallel to the network phases.The access point has transmit and receive UPAs and serves both MEC and conventional communication users.Source: §II · pp. 2–3
1) Uplink pilot training and imperfect CSI:
- Each user sends a pilot sequence of length at least K for channel estimation, with MMSE estimates and errors modeled as independent.The estimation and error variances depend on pilot power, pilot length, large-scale fading, and noise.Source: 1) Uplink pilot training and imperfect… · p. 3
2) FD-mMIMO ISCC communication signal model:
- During phase 2, MEC and conventional users transmit uplink signals while the access point concurrently sends a beamformed radar probe toward the target.The modeled received signal includes user signals, the target echo, and residual self-interference.Source: 2) FD-mMIMO ISCC communication signal… · p. 3
B. Radar Sensing Model
The sensing model treats the access point as tracking a target using a prior location estimate to steer its radar beam. It forms an equivalent signal from multiple snapshots with interference and noise, alongside bounds for angle estimation accuracy.
B. Radar Sensing Model
- In steady-state tracking, the access point uses a prior location estimate to steer a directional radar beam toward the target.The sensing model considers target angles θ and ϕ and derives CRLBs for their estimation accuracy.Source: B. Radar Sensing Model · p. 3, §III · p. 4
III. COMMUNICATION AND SENSING PERFORMANCE
- The model vectorizes L received snapshots into an equivalent radar signal whose interference-plus-noise term includes residual self-interference and receiver noise.The resulting rate lower bounds and angular CRLBs evaluate communication and sensing performance under the stated assumptions.Source: B. Radar Sensing Model · p. 4, §III · p. 4
A. Communication Performance
The section derives closed-form achievable uplink rates for MEC and communication users under imperfect channel estimates. It characterizes how radar and user interference affect those rates and how receive-array growth behaves under different user-power regimes.
A. Communication Performance
- Theorem 1 gives a closed-form achievable offloading rate for priority MEC users using the use-and-then-forget bound and an inverse-Wishart property.The section also gives the corresponding rate for communication users by symmetry.Source: A. Communication Performance · p. 4
- Increasing radar power raises residual self-interference and target-echo interference, reducing uplink rates.The rate expressions account for inter-user interference from channel-estimation errors and structured residual self-interference.Source: A. Communication Performance · p. 4
- With fixed user powers, the achievable rate grows as B log2 Nr + O(1), while radar interference remains in the limiting rate under the stated user-power-scaling regime.The rate expression also includes a co-site radar cross-interference factor.Source: A. Communication Performance · p. 4
B. Sensing Performance
The sensing analysis models aggregate interference and derives closed-form CRLBs for the target angles. It shows that beam-pointing error and the competing effects of radar and user powers matter for sensing accuracy.
1) Multi-source interference:
- For large L, the analysis upper-bounds colored residual self-interference as equivalent spatially white Gaussian interference, yielding a scaled-identity aggregate covariance.The resulting interference scales with the number of UEs K and the residual self-interference level χSI.Source: 1) Multi-source interference · pp. 4–5
2) The CRLB analysis:
- Theorem 2 provides closed-form CRLB expressions for estimating target angles θ and ϕ in the FD-mMIMO ISCC system.The derivation uses the Fisher information matrix and a decoupled 2×2 angular block.Source: 2) The CRLB analysis · p. 5
- The angular CRLBs increase with beam-pointing error ϵ over 0°–10°, and the analysis adopts perfect alignment for tractable resource allocation.Under perfect alignment, the normalized sensing beamformer and geometric trace constants simplify.Source: 2) The CRLB analysis · p. 5, Figure 2 · p. 5
- Under perfect alignment, the CRLBs are proportional to uplink interference and inversely proportional to radar power, making joint power allocation important for preserving sensing performance.The transmit-array size scales gain, whereas the receive-array size governs the angular terms.Source: 2) The CRLB analysis · pp. 5–6
- At finite array sizes, radar power improves sensing but reduces uplink rates, while user power increases sensing interference.These competing effects motivate joint resource allocation.Source: 2) The CRLB analysis · p. 6
IV. ISCC RESOURCE ALLOCATION
The resource-allocation formulation models local and edge-assisted task execution and jointly minimizes average end-to-end latency. It accounts for task deadlines, energy budgets, communication rates, sensing accuracy, and system resource limits.
1) Computational model:
- With asynchronous execution, end-to-end latency is the maximum of local execution time and combined offloading-plus-edge execution time.Downlink latency is treated as negligible because results are small and the access point transmits at high power.Source: 1) Computational model · p. 6
- Communication-user energy use comes only from data transmission and pilot training, while MEC users also incur local-computation, offloading, and pilot-training energy.Communication users do not execute computational tasks.Source: 1) Computational model · p. 6
2) Resource allocation problem:
- The optimization jointly selects offloading fractions, user and radar powers, and edge CPU allocations to minimize average end-to-end task latency.The constraints include deadlines, energy budgets, minimum communication rates, sensing accuracy, CPU and power limits, and offloading limits.Source: 2) Resource allocation problem · p. 6
B. Proposed Solution
The proposed solution applies successive convex approximation to the coupled rate, latency, and energy terms. It iteratively replaces nonconvex relationships with convex approximations and expresses the sensing constraint in affine form.
B. Proposed Solution
- SCA is applied to the coupled rate, latency, and energy terms, while the sensing constraint is rewritten in affine form.The resulting reformulation provides the iterative convex approximation used to solve the resource-allocation problem.Source: B. Proposed Solution · p. 6
- A concave lower bound around feasible iteration points is used to convexify the rate constraint.The method rewrites the rate expression in ln(1 + x/y) form before approximation.Source: B. Proposed Solution · pp. 6–7
- The latency constraint is convexified by approximating the coupled offloading and edge-execution terms at each iteration.The approximation addresses products involving offloading fractions, rate-related slack variables, and CPU allocations.Source: B. Proposed Solution · p. 7
- The energy constraint is iteratively replaced with a convex approximation to handle its coupling among user power, offloading fraction, and rate-related slack.The communication-rate constraint is also convexified using the rate approximation.Source: B. Proposed Solution · p. 7
5) Sensing constraint:
For ρ > 0, substituting (39)–(40) into (46e) gives affine constraints. The proposed SCA algorithm repeatedly solves a convex program and uses its solution as the next feasible point until convergence.
5) Sensing constraint:
- For ρ > 0, substituting (39)–(40) into (46e) gives affine sensing constraints.Source: 5) Sensing constraint · p. 7
C. Algorithm and Complexity Analysis
- At each iteration, the convex-program solution becomes the next feasible point, and the SCA process repeats until convergence.The algorithm initializes with randomly generated feasible points and outputs the final solution.Source: C. Algorithm and Complexity Analysis · pp. 7–8
- The cubic terms in constraint (61) are represented with a constant number of auxiliary variables and SOC constraints.Source: C. Algorithm and Complexity Analysis · pp. 7–8
- Problem (67) has 4KMEC+Kcom+1 scalar variables and 4KMEC+2Kcom+4 linear, quadratic, and SOC constraints.Source: C. Algorithm and Complexity Analysis · p. 8
V. NUMERICAL RESULTS
The numerical evaluation specifies a cell-based FD-mMIMO ISCC simulation setting and compares the proposed framework against fixed and equal-power allocations, brute-force search, and half-duplex operation.
A. Simulation Setting
- Users are uniformly distributed in a 1000-meter-radius cell around a central AP, with none closer than 100 meters; the target is 400 meters away.Large-scale fading uses log-normal shadowing with 7 dB standard deviation and path-loss exponent 3.2.Source: A. Simulation Setting · p. 8
B. Benchmark Schemes
- Brute-force search explores candidate resource allocations at high computational cost, with accuracy dependent on search resolution.Source: B. Benchmark Schemes · p. 8
- HD-ISCC divides the resource equally between uplink and radar phases, eliminating SI while halving users’ effective achievable rates.Sensing is free from multiuser uplink interference in this benchmark.Source: B. Benchmark Schemes · p. 8
- EqMEC fixes MEC-user transmit power at Pmax_k/2, while EqCM tests communication-user powers of Pmax_k/2 and Pmax_k.These equal-power schemes assess the effects of optimizing MEC and communication-user power, respectively.Source: B. Benchmark Schemes · p. 8
- Fixed-offloading benchmarks hold MEC users’ task-offloading fractions at 20%, 50%, or 70%.Source: B. Benchmark Schemes · p. 8
C. Numerical Results and Discussion
The numerical results validate the derived rate and sensing analyses and show that joint allocation achieves lower latency than the compared schemes. They also characterize how array size, sensing thresholds, task size, and user loading affect performance.
1) Validation of asymptotic performance and power scaling:
- Analytical rates closely match Monte Carlo results; power-scaled uplink rates saturate near Nr = 1000, while angular CRLBs decay faster with Nr than with Nt.Source: 1) Validation of asymptotic… · pp. 8–9, Figure 3 · p. 9, Figure 4 · p. 9
2) Convergence analysis of Algorithm 1:
- Algorithm 1 converges in approximately seven iterations across Nr = 25, 36, and 64, with an objective gap of around 4 × 10^-3 versus BFS.Source: 2) Convergence analysis of Algorithm 1 · p. 9, Figure 5 · p. 9
3) Offloading and sensing performance:
- Analytical and Monte Carlo CRLBs tightly lower-bound the MLE’s MSE for both target angles under the evaluated sensing model.Source: 3) Offloading and sensing performance · p. 9, Figure 6 · p. 9
- Relaxing the CRLB threshold from −40 to −10 dB reduces required radar power and interference, enabling more offloading and lower latency.Source: 3) Offloading and sensing performance · p. 9, Figure 7 · p. 10
- Larger receive arrays reduce latency, radar power, and UE energy; the latency gap between joint and fixed-power schemes narrows as the array grows.Source: 3) Offloading and sensing performance · p. 10, Figure 8 · p. 10
- As task size rises from 5 to 15 kbits, latency increases; Algorithm 1 attains the lowest latency at 66.68% average offloading, below the fixed 70% benchmark.Source: 3) Offloading and sensing performance · p. 10, Figure 9 · p. 10
- Latency increases with user loading, while joint allocation outperforms the benchmarks and has a larger advantage over EqMEC at higher loads.Source: 3) Offloading and sensing performance · pp. 10–11, Figure 10 · p. 11
VI. CONCLUSION
- Joint allocation lowers latency and reduces radar power by approximately 86% relative to EqCM at a −40 dB CRLB threshold; the sensing analysis assumes the adopted interference approximation and beam-alignment assumptions.Source: §VI · p. 11
APPENDIX A: PROOF OF THEOREM 1
The appendices detail the rate-bound and sensing-covariance derivations, then establish how uplink rate behaves under array-size power scaling. They also use an interference approximation to make the sensing analysis tractable.
APPENDIX A: PROOF OF THEOREM 1
- The UatF rate-bound proof evaluates separate interference expectations, including inter-user and radar cross-interference, and uses the ZF inverse-Wishart property to obtain (11).For large N_r, it treats the ZF direction as statistically isotropic when evaluating cross-interference.Source: Appendix A · p. 11
APPENDIX B: PROOF OF REMARK 2
- When user powers scale as 1/N_r, the desired-signal term vanishes while radar interference and receiver noise remain fixed, so the SINR converges and the rate saturates.This limit is specific to that power-scaling regime; with fixed user powers, the rate grows logarithmically with N_r.Source: Appendix B · p. 12
APPENDIX C: PROOF OF LEMMA 1
- Because the exact FIM for colored LoS interference is intractable, the proof bounds it as spatially white Gaussian noise to derive the sensing-noise covariance.The resulting approximation gives expressions for the residual self-interference covariance and the received-signal covariance.Source: Appendix C · p. 12
APPENDIX D: PROOF OF REMARK 3
Under perfect beam alignment, array orthogonality decouples the angular metrics and supports closed-form angle CRLBs. The bounds improve with transmit-array size and decay quadratically with receive-array size for a square UPA.
APPENDIX D: PROOF OF REMARK 3
- With perfect transmit-beam alignment, transmit-array orthogonality decouples the spatial metrics and the spatial angles from amplitude, enabling closed-form angle CRLBs.Receive-array orthogonality further removes the coupling between the two spatial angles and amplitude.Source: Appendix D · p. 12
APPENDIX E: PROOF OF REMARK 4
- For a square UPA, both angle CRLBs scale as 1/[N_r(N_r−1)], giving quadratic decay with receive-array size; CRLB_θ also scales as O(1/N_t).The transmit-array scaling is stated for N_t → ∞.Source: Appendix E · p. 12