Hybrid NOMA Assisted Heterogeneous Semantic and Bit Users Communication

Technical Summary: Hybrid NOMA Assisted Heterogeneous Semantic and Bit Users Communication

Problem Statement
The paper addresses the challenge of integrating traditional bit-based communication with emerging semantic communication within next-generation wireless networks. While semantic communication offers advantages in low Signal-to-Noise Ratio (SNR) environments by focusing on information meaning, traditional bit-based communication remains superior for near-lossless recovery at higher SNRs. The coexistence of these two distinct user types (heterogeneous users) creates specific design challenges, particularly regarding Successive Interference Cancellation (SIC). In standard Non-Orthogonal Multiple Access (NOMA), SIC relies on channel conditions (stronger users decode weaker users' signals). However, in a joint system, bit users lack the pre-trained neural networks required to decode semantic signals, rendering traditional channel-based SIC ineffective. Furthermore, existing hybrid NOMA schemes often fail to dynamically adapt user pairing and decoding orders to accommodate these heterogeneous signal processing requirements.

Methodology
The authors propose a downlink hybrid NOMA framework that dynamically combines Orthogonal Multiple Access (OMA) and NOMA to serve one bit user (B) and two semantic users (S1, S2) over Rayleigh fading channels.

• System Model: The system utilizes a DeepSC (Deep Semantic Communication) model where semantic rates are determined by a semantic similarity function, approximated via a generalized logistic function. The bit user's rate follows Shannon's capacity formula. To enable comparison, the bit user's rate is transformed into an equivalent semantic rate.
• Adaptive User Pairing and Decoding: The core of the methodology is a dynamic slot allocation strategy (Algorithm 1):
  • Heterogeneous Slots (Bit + Semantic): If a bit user and a semantic user share a slot, the system employs a bit-to-semantic decoding order. The bit user decodes its signal first, treating the semantic signal as interference. The semantic user then decodes the bit user's signal, applies SIC to remove it, and decodes its own signal interference-free.
  • Semantic-Only Slots: If two semantic users share a slot, the user with the stronger channel gain is decoded first (treating the weaker as noise), followed by SIC for the weaker user.
  • OMA Transition: If a semantic user has a stronger channel gain than the bit user in a potential pairing, the slot is allocated exclusively to the bit user (OMA mode) to avoid inefficient resource usage.
• Optimization: A power allocation problem is formulated to maximize the expected equivalent ergodic semantic Spectral Efficiency (SE) subject to a total power constraint ($P_{max}$). Due to the non-convex nature of the semantic similarity function, the authors leverage the "time-sharing" condition to establish strong duality. They solve the problem using the Karush-Kuhn-Tucker (KKT) conditions, deriving closed-form approximations and transcendental equations for optimal power distribution.

Key Contributions

1. Hybrid NOMA Framework: Proposal of a hybrid NOMA architecture specifically designed for heterogeneous users, dynamically assigning users to time slots based on real-time channel conditions.
2. Adaptive Decoding Strategy: Introduction of a tailored decoding order that prioritizes a bit-to-semantic sequence in mixed slots (contrary to traditional channel-based SIC) and a stronger-to-weaker sequence in semantic-only slots.
3. Power Allocation Optimization: Formulation and solution of a non-convex power allocation problem that maximizes equivalent ergodic semantic SE, accounting for the distinct rate characteristics of bit and semantic users.
4. Performance Validation: Demonstration through simulations that the proposed hybrid approach outperforms OMA-only schemes in terms of ergodic semantic SE.

Simulation Results
The authors validate their approach using Monte Carlo simulations over $10^5$ channel realizations with a 100-meter cell radius.

• SE Performance: The proposed hybrid NOMA setup consistently achieves higher equivalent ergodic semantic SE compared to a heterogeneous OMA setup across various maximum power ($P_{max}$) levels and semantic similarity thresholds ($\epsilon_{th}$).
• Power Allocation Behavior: As $P_{max}$ increases, the system allocates significantly more power to the bit user than to the semantic users. The authors note this results in an "unfair" optimality, driven by the difference between the logarithmic growth of bit rates and the saturating nature of the semantic similarity function.
• Threshold Sensitivity: While increasing the semantic similarity threshold ($\epsilon_{th}$) generally reduces SE due to reconstruction limitations, the hybrid NOMA setup maintains a performance advantage over OMA, attributed to the bit user's ability to aid in accurate signal reconstruction through efficient resource utilization.

Significance and Claims
The paper claims that its proposed framework effectively bridges the gap between semantic and bit-based communications by resolving the SIC incompatibility inherent in heterogeneous networks. By dynamically transitioning between OMA and NOMA and enforcing a bit-to-semantic decoding order, the system enhances overall Spectral Efficiency. The authors assert that this approach provides a viable solution for future wireless networks requiring the coexistence of diverse communication paradigms, ensuring efficient resource utilization while maximizing the equivalent semantic performance of the network. The work is presented as a step toward optimizing heterogeneous user networks without inventing new physical layer technologies, but rather by adapting existing NOMA/OMA and DeepSC concepts into a cohesive, optimized framework.