TL;DR
This paper proposes a novel hybrid NOMA-ISAC system enhanced by Reconfigurable Intelligent Surfaces (RIS) to boost sensing performance while meeting communication demands. They introduce an efficient alternating optimization algorithm to jointly optimize power allocation, active beamforming, and RIS phase shifts, achieving significant improvements in minimum beampattern gain.
What Problem Does This Paper Address?
Integrated Sensing and Communication (ISAC) systems aim to combine radar sensing and wireless communication functionalities onto a single platform, promising enhanced spectral efficiency and reduced hardware costs. However, supporting a growing number of devices in ISAC presents challenges, particularly with traditional Non-Orthogonal Multiple Access (NOMA) which can suffer from decoding delays and propagation errors. To address this, the paper explores a hybrid NOMA-OMA approach, where users are clustered and assigned orthogonal sub-resource blocks, with NOMA used within each cluster.
A key challenge in ISAC is optimizing the system's performance for both sensing and communication simultaneously. This paper focuses on maximizing the sensing performance, specifically the minimum beampattern gain (MBPG) at target angles, while ensuring that communication users meet their required data rates. The use of Reconfigurable Intelligent Surfaces (RIS) is investigated as a way to improve signal propagation and enhance both sensing and communication capabilities, especially in scenarios with blocked direct links. The core problem is to jointly optimize the system parameters (power allocation, beamforming, and RIS phase shifts) to achieve this dual objective, which is inherently complex and non-convex.
Methods
The paper proposes a low-complexity Alternating Optimization (AO) algorithm to tackle the non-convex optimization problem. This algorithm decomposes the problem into three sub-problems, iteratively optimizing each variable while keeping others fixed:
- Intra-cluster Power Allocation (Intra-CPA): A closed-form solution is derived for optimizing the power allocated to users within a cluster. This leverages a lemma showing that to maximize the sum rate, minimum power should be allocated to weaker users and maximum to the strongest.
- Active Beamforming Optimization: This sub-problem is addressed using the Successive Convex Approximation (SCA) method. The non-smooth objective function and non-convex constraints are transformed into a series of convex problems that can be solved efficiently. An auxiliary variable is introduced to handle the max-min objective.
- RIS Phase Shift Optimization: Similar to active beamforming, SCA is employed to handle the non-convex constraints. Additionally, a penalty method is used to manage the unit modulus constraint on the RIS phase shifts. This involves reformulating the problem with a penalty term and then iteratively solving it.
The overall algorithm involves nested loops: an outer loop for the penalty factor in RIS phase shift optimization and inner loops for each sub-problem.
Datasets
Not discussed in the paper. The evaluation is based on simulated system parameters.
Key Insights and Ideas
- Hybrid NOMA-OMA for Scalability: Combining NOMA with OMA (in a hybrid fashion) is crucial for supporting a large number of users in ISAC systems, mitigating the delay and error issues of pure NOMA.
- RIS as a Game Changer: RIS can significantly enhance ISAC performance by intelligently reconfiguring the wireless environment, improving signal strength and coverage for both sensing and communication.
- Joint Optimization is Key: Achieving optimal performance requires a holistic approach that jointly optimizes power allocation, active beamforming at the base station, and passive beamforming (phase shifts) at the RIS.
- SCA and Penalty Methods for Tractability: The paper demonstrates the effectiveness of SCA for handling non-convex constraints in beamforming and the penalty method for unit modulus constraints in RIS phase shifts, making the problem solvable.
- Trade-off Between Sensing and Communication: The results highlight an inherent trade-off: improving sensing performance can sometimes come at the cost of communication quality, and vice-versa, especially when communication requirements become more stringent.
Key Results
- The proposed PAO (Proposed AO) algorithm significantly outperforms baseline schemes (NoSIC, MRT, RPS) in terms of Minimum Beampattern Gain (MBPG) at interested sensing angles.
- Increasing the number of RIS elements ($N_s$) leads to a substantial improvement in MBPG, as more elements provide better reflecting paths and higher passive beamforming gain.
- Higher transmit power ($P_t$) also boosts MBPG, as it allows for better signal strength to meet communication rate constraints, leaving more power for sensing.
- The paper shows that more transmit antennas ($M$) at the base station generally improve sensing performance due to higher beamforming gain.
- The simulation results visually demonstrate the trade-off between sensing and communication performance, with a "radar only" system achieving higher peak beampattern gain than the ISAC system, especially under stricter communication rate requirements.
Limitations and Future Work
- Channel State Information (CSI): The paper assumes perfect CSI is available. In real-world scenarios, CSI acquisition is a significant challenge, and future work could explore robust designs under imperfect CSI.
- Complexity: While the proposed AO algorithm is low-complexity compared to brute-force methods, the computational complexity can still be substantial, especially with a large number of antennas and RIS elements. Further research into more efficient algorithms or approximations might be beneficial.
- Two-Layer Iteration for RIS: The RIS phase shift optimization uses a two-layer iteration (outer loop for penalty factor, inner loop for phase shifts). While effective, exploring alternative or more streamlined optimization techniques for RIS could be a future direction.
- Deployment Scenarios: The paper assumes specific deployment geometries. Investigating performance across a wider range of deployment scenarios and user distributions would be valuable.
Summary updated