Full-duplex Processing

• Full-duplex processing is defined as the simultaneous transmission and reception on the same frequency band, offering nearly double the spectral efficiency compared to half-duplex systems.
• It introduces challenges like self-interference (SI) and cross-link interference (CLI), which are mitigated using a combination of passive, analog, digital, and beamforming cancellation techniques.
• Its applications span cellular networks, massive MIMO systems, and speech-domain processing, demanding multi-domain optimization for practical deployment.

Full-duplex processing refers to simultaneous transmission and reception over the same time-frequency resources in wireless and speech communication systems. This modality contrasts with half-duplex (HD) protocols such as time-division duplexing (TDD) and frequency-division duplexing (FDD), which separate transmit and receive phases either in time or frequency. Full-duplex offers theoretical doubling of spectral efficiency but introduces severe interference management challenges, notably self-interference (SI) and cross-link interference (CLI), necessitating advanced signal processing techniques, architectural solutions, and novel optimization frameworks (Han et al., 2014, Korpi et al., 2014, Niu et al., 2023, Li et al., 2016).

1. Principles of Full-Duplex Operation

A full-duplex node concurrently transmits and receives on the same frequency band and time slot. The SI problem originates from the node's own transmit signal leaking into its receive chain, often exceeding the desired signal by 50–100 dB. CLI encompasses mutual interference in multi-user and multi-cell scenarios, including uplink UE emissions impacting downlink UE reception or base station (BS)-to-BS interference in cellular networks (Han et al., 2014, Jr. et al., 2020).

Full-duplex systems can nearly double link-level spectral efficiency compared to HD systems, as both uplink and downlink utilize the full bandwidth concurrently. The theoretical sum-rate for a symmetric FD link is

$$C_{FD} = \log_2(1 + \mathrm{SINR}_{DL}) + \log_2(1 + \mathrm{SINR}_{UL})$$

as opposed to the HD sum-rate

$$C_{HD} = \frac{1}{2}\left[\log_2(1 + \mathrm{SINR}_{DL}) + \log_2(1 + \mathrm{SINR}_{UL})\right]$$

where SINR expressions reflect residual SI, CLI, and noise (Li et al., 2016, Ngo et al., 2014).

2. Self-Interference and Cross-Link Interference

Self-Interference (SI)

In full-duplex wireless, SI consists of the node's own transmitted signal entering the receiver via electromagnetic coupling, antenna proximity, or RF circuit leakage. SI must be suppressed below the receiver noise floor for correct demodulation. The baseband SI model is

$$r(t) = h_{SI}\cdot x_{tx}(t) + h_{SR}\cdot s_{rx}(t) + n_{rx}(t)$$

with $h_{SI}$ as the SI channel, $x_{tx}$ as the transmit waveform, $h_{SR}$ as the desired channel, and $n_{rx}$ as noise.

Cross-Link Interference (CLI)

CLI occurs in networked scenarios, where simultaneous uplink and downlink operations create additional interference paths: UL UE to DL UE, BS-to-BS, and inter-cell or intra-cell UE-to-UE. The aggregate CLI terms limit performance in large-scale deployments (Jr. et al., 2020, Li et al., 2016, Goyal et al., 2014).

3. Interference Cancellation Architectures

A multi-domain cancellation approach is essential:

• Passive Suppression: Antenna isolation, polarization, shielding, and ground-plane loops, achieving 40–55 dB SI isolation in compact RF designs (Korpi et al., 2014).
• Analog Cancellation: Injection of amplitude- and phase-adjusted transmit replicas into the RF or baseband receiver path before the LNA, utilizing vector modulators or tapped delay lines for alignment. Analog techniques provide 20–70 dB further cancellation (Kaufman et al., 2013, Kaufman et al., 2013).
• Digital Cancellation: Subtraction of the estimated SI waveform post-ADC, employing least-squares channel estimation, adaptive filters, or neural network-based nonlinear cancellation; up to 40 dB is achievable (Korpi et al., 2014, Niu et al., 2023).
• Beamforming/Spatial SIC: In MIMO or massive-MIMO, transmit and receive beamformers are jointly designed to suppress SI spatially, steering nulls and exploiting array degrees of freedom (Korpi et al., 2014, Ngo et al., 2014, Sharma et al., 2017).

A typical aggregate SI suppression exceeding 100 dB is necessary for practical operation with reasonable transmit powers and modulation schemes.

4. Full-Duplex Networking and Resource Management

Full-duplex introduces new interference patterns in networked deployments, transforming the resource management problem:

• Cellular Networks: In pico-layer or small-cell systems, advanced FD UE pairing and binary power control maximize sum-rate utility while controlling mutual interference. Hybrid scheduling allocates FD timeslots only when throughput gains outweigh CLI penalties (Goyal et al., 2014, Li et al., 2016). Suboptimal greedy UE selection and geometric programming for power allocation are employed for scalable resource optimization.
• Relay and Massive MIMO: In multi-pair full-duplex relaying, massive antenna arrays, MMSE channel estimation, and zero-forcing preprocessing substantially mitigate loop-back interference and interpair CLI. Analytical lower bounds for spectral efficiency demonstrate nearly 2$\times$ gain over HD once the antenna regime is sufficiently large (Sharma et al., 2017, Ngo et al., 2014).

5. Semantic and Speech-domain Full-Duplex Processing

Recent advances extend full-duplex into the speech and semantic communication domains:

• Full-Duplex Speech Agents: Architectures such as FlexDuo and LLM-Enhanced Dialogue Management decouple turn-taking control from language generation, introducing explicit Idle states, semantic buffering mechanisms, and fine-grained dialogue actions. Modular control layers enable robust interruption handling, significant reductions in false interruption rates (up to 24.9%), and productivity improvements in human–machine conversational systems (Liao et al., 19 Feb 2025, Zhang et al., 19 Feb 2025, Lin et al., 30 Jul 2025).
• Semantics-Division Duplexing (SDD): Integrates deep joint source-channel coding and semantic k-domain suppression to enhance robustness against severe residual SI, enabling reliable source reconstruction at SINR down to –50 dB and extending the feasible region beyond conventional IBFD (Niu et al., 2023).

6. Optimization Formulations and Trade-Offs

Optimization of full-duplex operation involves multidimensional trade-offs:

• Beamforming and Sensing-Communication (S&C) Trade-off: Bidirectional ISAC systems frame FD versus HD as an optimization over rate and sensing accuracy (CRB), solved via one-layer successive convex approximation (SCA) algorithms, yielding KKT-optimal solutions. In certain regimes (e.g., sensing-prior or LOS channels), HD can outperform FD even with perfect SI cancellation (Wang et al., 2022).
• Power Scaling: In massive-array relays, optimal power allocation via geometric programming yields substantial energy-efficiency improvements. Power can be scaled inversely to the antenna count (1.5–3 dB savings per doubling), subject to pilot power constraints and loop-interference thresholds (Ngo et al., 2014).
• Physical-layer Security: Deliberate manipulation of relay processing delay exploits inherent SI for secure artificial-noise injection in FD amplify-and-forward channels, balancing secrecy rate with cyclic-prefix overhead and null-space dimensionality (Marzban et al., 2021).

7. Practical Challenges and Future Directions

Despite theoretical gains, full-duplex faces several engineering hurdles:

• Aggregate SI cancellation exceeding 110 dB across domains is required for meaningful gains, especially in outdoor or macro-cell scenarios where path-loss is extreme.
• Hardware limitations include phase noise, ADC dynamic range, component nonlinearity, and analog filter imperfections.
• Cross-layer approaches are needed to harmonize FD operation in networks with legacy HD devices and evolving standards.
• Research thrusts include the unification of TDD/FDD paradigms into flexible FD frameworks, joint design of full-duplex and massive-MIMO scheduling/precoding, integration with mmWave, AI-optimized RF/antenna structures, and multidomain semantic processing for speech applications (Han et al., 2014, Niu et al., 2023).
• FD does not universally outperform HD; network topology, traffic patterns, and sensing priorities determine the preferable duplexing scheme (Wang et al., 2022).

In conclusion, full-duplex processing encompasses a diverse set of architectural, algorithmic, and optimization methodologies that collectively enable simultaneous bidirectional communication. While theoretical capacity doubling is attainable, practical realization is contingent on multi-domain SI suppression, advanced resource management, and explicit trade-off analysis across networking, relay, sensing, and semantic domains.