---
title: 'Duplex: Bidirectional Communication & Applications'
url: https://www.emergentmind.com/topics/duplex-aa48752a-b8ff-4f01-9d02-42d65546170f
type: topic
---

# Duplex: Bidirectional Communication & Applications

Searching arXiv for recent and foundational papers on “duplex” across communications and related domains to ground the article in published work.
arXiv search query: all:duplex

In the cited technical literatures, **duplex** denotes a family of paired or bidirectional constructions in which two directions, streams, or complementary components are jointly represented and often active concurrently. In wireless systems, the term covers **full duplex**, **half duplex**, **flexible duplex**, and hybrid extensions of TDD and FDD for simultaneous or dynamically partitioned uplink and downlink operation [1410.5326]. In speech modeling it denotes direct modeling of **simultaneous user and agent streams** [2505.15670]. In quantum information it denotes **two-way quantum communication** through a single spin chain [1106.3502]. In optics it denotes **spliced absolute optical instruments** such as duplex Mikaelian and duplex Maxwell’s fish eye lenses [2002.03122]. In graphics it has also been extended to a **dual-hierarchy** construction in Gaussian splatting [2508.03180]. This range of uses suggests that duplex is less a single mechanism than a recurrent structural idea: two coupled directions or layers treated as a single design object.

## 1. Core meanings and taxonomy

In wireless communication, duplexing classically separates or combines uplink and downlink resources. The 5G-oriented discussion of full duplex networking explicitly analyzes **interference scenarios in full duplex**, candidate mitigation approaches, interference-proof frame structures, transceiver structures for channel reciprocity recovery, a **super full duplex base station** where each sector operates in TDD mode, and the extension of **TDD and FDD to full duplex** [1410.5326]. Other works refine this binary picture by introducing **flexible duplex**, **dynamic time-frequency-division duplex**, and **partial-overlap** schemes [1710.00780] [1909.03058] [1509.00976].

Outside wireless networking, the same label appears with domain-specific meanings. In SALM-Duplex, duplex means a speech-to-speech architecture with **continuous user inputs and codec agent outputs with channel fusion** that directly models simultaneous user and agent streams [2505.15670]. In “Duplex quantum communication through a spin chain” [1106.3502], it means two users simultaneously transferring qubit states through one channel. In “Duplex Mikaelian lenses and duplex Maxwell’s fish eye lenses” [2002.03122], it refers to optical instruments obtained by **splicing two half Mikaelian lenses with different periods** and then applying exponential conformal mapping. In Duplex-GS, the term refers to a **dual-hierarchy framework** that combines proxy Gaussian representations with order-independent rendering [2508.03180].

| Domain | Duplex meaning | Representative instances |
|---|---|---|
| Wireless systems | Simultaneous or dynamically partitioned UL/DL operation | full duplex, flexible duplex, D-TFDD, $\alpha$-duplex |
| Speech and graphics | Simultaneous streams or dual hierarchy | SALM-Duplex, Duplex-GS |
| Quantum and optics | Two-way transfer or spliced dual-region construction | spin-chain duplex communication, duplex lenses |

A common misconception is that duplex invariably means **simultaneous transmit and receive on the same carrier**. The surveyed uses are broader. Some are fully simultaneous, such as in-band full duplex and simultaneous speech interaction; others are **virtual**, **adaptive**, or **emulated**, as in rapid on-off-division duplex and CoMPflex [1010.2667] [1507.08574].

## 2. Wireless duplexing: spectrum reuse, partial overlap, and dynamic allocation

The main attraction of wireless duplexing schemes is increased spectrum reuse, but the literature consistently qualifies this promise by interference structure. In small-cell systems with FD base stations and HD user equipment, a **hybrid scheduling policy** assigns FD timeslots when pairing and power levels are favorable and otherwise defaults to HD operation. With practical self-interference cancellation, the reported gains are **94% throughput improvement in the downlink, and 92% in the uplink** in an indoor multi-cell scenario, and **54% in downlink and 61% in uplink** in an outdoor multi-cell scenario, compared to HD [1412.8708]. The same paper also compares energy efficiency, indicating that throughput gain and energy cost must be analyzed jointly.

Flexible duplex generalizes this logic from binary FD/HD selection to dynamic UL/DL resource partitioning. In a multi-cell OFDM system with **minimum resource units (MRUs)**, the paper on dynamic uplink/downlink resource management defines interference-aware algorithms **SAFP** and **RMDI** for max-min resource allocation. These algorithms achieve a **two-fold increase** in worst-case quality of service satisfaction over a fixed UL/DL baseline under low traffic asymmetry, and a **three-fold increase** when traffic is highly asymmetric [1710.00780]. The significance is not merely dynamic partitioning, but dynamic partitioning that is explicitly aware of **inter-mode interference (IMI)**.

The $\alpha$-duplex formulation provides a fine-grained interpretation of duplex as **partial overlap** between uplink and downlink bands. The stochastic-geometry study shows that in-band FD can improve the downlink rate by **up to $97\%$** but can degrade the uplink rate by **up to $94\%$**, demonstrating the vulnerability of uplink to downlink interference [1509.00976]. The proposed $\alpha$-duplex scheme instead allows the overlap to be optimized, and one reported operating point yields a **simultaneous improvement of $28\%$ for the downlink rate and $56\%$ for the uplink rate** [1509.00976]. This directly counters the simplified assumption that maximal overlap is always optimal.

Dynamic time-frequency division duplex pushes this further by combining TDD and FDD. In D-TFDD, a user receives on one frequency band and transmits on another as in FDD, then shares those bands with another user in a D-TDD fashion. The scheme **does not require inter-cell interference knowledge** and only requires CSI of the local BS-U1 and BS-U2 channels. It increases the throughput region, significantly decreases outage probabilities, and **doubles the diversity gain** on both corresponding channels relative to existing duplexing schemes [1909.03058].

## 3. Interference, self-interference cancellation, and transceiver architectures

The main physical obstacle to full duplex is self-interference, but recent literature distinguishes this from the broader problem of **cross-link interference**. In multi-cell communications, a 2020 study argues that, once self-interference is sufficiently suppressed, **cross-link interference**, especially **base station to base station**, becomes the remaining challenge restricting uplink performance. Using low-complexity beamforming, the paper shows that full duplex and dynamic-TDD can substantially increase capacity, and that if cross-link interference can be controlled in full duplex, then it is possible to **almost double the multi cell network capacity as well as user throughput** [2010.11317].

At the transceiver level, several papers study how this suppression is realized. A reduced-complexity in-band full duplex transceiver uses a **Radio Frequency Self-Interference Canceler (RFSIC)** based on the isolation of a circulator together with a **vector modulator** that regenerates the interference signal for destructive combination, plus a **Digital Self-Interference Cancellation (DSIC)** algorithm based on non-linear adaptive filtering. With this simplified analog front-end, the system achieves **90 dB** SI cancellation in the presence of a received signal [1711.07223].

A compact full duplex MIMO radio for D2D underlaid cellular networks combines **dual-polarization with high cross-polarization discrimination (XPD)** for analog cancellation and LTE-based **per-subcarrier digital self-interference cancellation**. In the FPGA-based prototype, the combined analog plus digital SI cancellation reaches **up to 101 dB**; the platform supports **20 MHz** bandwidth, and the reported throughput gain is **up to 3.45x** over half-duplex LTE SISO, with **3.03x at max. Tx power** [1612.06112]. The same work also proposes a **novel timing synchronization method** based on dual-root Zadoff-Chu sequences and NSP index switching.

For flexible duplex OFDM systems, nonlinear cancellation has also moved to the frequency domain. A 2025 paper develops **low-complexity frequency domain nonlinear SIC** for **IBFD**, **SBFD**, and partial-overlap settings, accounting for PA nonlinearity and IQ imbalance. Its method decomposes nonlinear SI into a nonlinear basis, prunes basis functions analytically using an intermodulation-distortion model, uses a specialized pilot sequence, and is validated by both **system-level simulation using 3D ray-tracing** and **proof-of-concept measurement**. The reported residual SI is **close to the noise floor** across duplex configurations while reducing complexity relative to conventional digital SIC [2503.02203].

A second misconception is therefore that duplex performance is limited only by local self-interference cancellation. The cited system-level work shows that even when SI is manageable, **network-level interference geometry** determines whether duplex reuse is beneficial [2010.11317].

## 4. Virtual, emulated, and adaptive duplex operation

Several architectures realize duplex behavior without requiring a monolithic full-duplex radio. **Rapid on-off-division duplex (RODD)** is a canonical example. Each half-duplex radio transmits through a randomly generated **on-off duplex mask** over every frame interval and receives through its own off-slots. Over a single frame, every node can transmit to peers and simultaneously receive from each peer, yielding **virtual full-duplex communication using half-duplex radios** [1010.2667]. The throughput is reported as **significantly larger than that of ALOHA**, especially for simultaneous broadcast to one-hop peers [1010.2667].

**CoMPflex** emulates an in-band full-duplex base station using **two spatially separated and coordinated half duplex base stations**. In the one-dimensional geometric Wyner analysis, the architecture serves one uplink and one downlink mobile station at the same instant and yields gains in **sum-rate and energy efficiency** relative to ordinary FD and a unidirectional baseline [1507.08574]. In a planar stochastic-geometry treatment, CoMPflex is shown to bring BSs closer to the MSs they serve while increasing the distance between a MS and interfering MSs, and thereby improves **communication reliability** relative to FD-BS [1605.02314]. A plausible implication is that duplex emulation can convert a hardware problem into a geometry-and-coordination problem.

Adaptive duplex selection appears in heterogeneous networks and relay systems. In a HetNet, a **received-power-based hybrid-duplex scheme** switches each user between half- and full-duplex mode according to the received power from the serving BS. The thresholds are optimized for sum-rate maximization via a nonlinear integer programming formulation solved approximately by a greedy algorithm, and the reported gain is **up to \~45% sum rate** over a pure half-duplex baseline [1608.06041]. In **X-duplex relaying**, a two-antenna AF relay jointly selects FD or HD mode and its Tx/Rx antenna configuration to minimize SER. The asymptotic analysis shows additional spatial diversity, reduced performance floor at high SNR, and a diversity order that approaches **two** at high SNR [1701.07168].

These works make clear that duplex need not imply a fixed architectural commitment. It can also denote a **switchable operating regime** selected per user, per relay state, or per frame.

## 5. Duplex in speech interaction and graphics

In speech language modeling, duplex denotes direct modeling of **simultaneous human-agent exchange** rather than turn-based alternation. SALM-Duplex introduces a speech-to-speech architecture with **continuous user inputs and codec agent outputs with channel fusion**. User speech is encoded by a pretrained streaming encoder at an **80 ms frame rate**, while agent speech is represented by codec tokens at **12.5 Hz**, aligned in time with the user stream [2505.15670]. The model is described as the **first duplex S2S model without requiring speech pretrain**, because it uses a pretrained streaming CTC encoder and directly fine-tunes the LLM plus encoder on duplex conversational data [2505.15670]. It also reports a codec bitrate of **0.6 kbps**, half the bitrate of the cited prior work, together with **94.5%** barge-in success, **0.69 s** barge-in latency, **0.92 s** first response latency, and GPT-4o scores of **7.8** on ASR-QA, **4.6** on multi-turn roleplay, and **3.5** on UltraChat [2505.15670]. The training and inference code are openly released.

In Gaussian splatting, Duplex-GS uses the term differently. Its contribution is a **dual-hierarchy framework** in which a first layer of **cell proxies** manages groups of Gaussians and a second layer stores the actual 3D Gaussians [2508.03180]. The framework integrates proxy Gaussian representations with **Order-Independent Transparency (OIT)** and a physically inspired weighted-sum rendering formulation. Across real-world datasets, it reports **1.5 to 4 speedup** over existing OIT-based Gaussian splatting methods and a **52.2% to 86.9% reduction of the radix sort overhead** without quality degradation [2508.03180].

This usage suggests an extension of duplex beyond bidirectional signaling to **paired-stream or paired-layer processing**. The commonality is still structural coupling: user and agent streams in SALM-Duplex, proxy and Gaussian hierarchies in Duplex-GS.

## 6. Quantum, optical, sensing, and reciprocity interpretations

In quantum information, duplex describes two-way transmission through a single medium. The spin-chain protocol of Alice and Bob uses an open spin-$1/2$ chain governed by the nearest-neighbor XY Hamiltonian
$$
H= -\frac{J}{2}\sum_{i=1}^{N-1} \left( \sigma_i^x \sigma_{i+1}^x + \sigma_i^y \sigma_{i+1}^y \right) - h \sum_{i=1}^{N} \sigma_i^z .
$$
Alice and Bob encode qubit states at the first and last sites and transmit simultaneously through one channel [1106.3502]. The central reported result is that the **transmission fidelity at each end can usually be enhanced by the presence of a second party**. For **$N=10$**, with $\theta_1/\pi = 0.6$ and $\theta_2/\pi = 0.6$, the maximum fidelity at Bob’s end improves from **$F_{\max}=0.67$** without Bob’s participation to **$F_{\max}=0.91$** with Bob sending simultaneously [1106.3502]. The presence of the second party also **decreases the transfer time**.

In optics, duplex denotes compound gradient-index instruments. The **duplex Mikaelian lens** is constructed by splicing two half Mikaelian lenses of different parameters, with refractive index
$$
n(x)=
\begin{cases}
1/\cosh(px/a), & x<0\\
1/\cosh(qx/a), & x>0 .
\end{cases}
$$
Its period is
$$
L_D=\frac{L}{2p}+\frac{L}{2q}.
$$
The **duplex Maxwell’s fish eye lens** is obtained by exponential conformal transformation of a duplex Mikaelian lens and is an absolute optical instrument only when the ratio of the two periods is rational [2002.03122]. The paper further demonstrates **caustic effects** in geometric optics and the **Talbot effect** in wave optics by numerical calculation [2002.03122].

In integrated sensing and communication, duplex becomes a design question rather than an automatic advantage. The bidirectional ISAC study compares full-duplex and half-duplex operation in both narrowband and wideband systems. It redesigns both modes to account for **sensing echo signals** and develops SCA-based beamforming algorithms that converge to **KKT optimal solutions** [2210.14112]. A central conclusion is that **full-duplex may not always outperform half-duplex**, especially in the **sensing-prior regime** or when the communication channel is **line-of-sight-dominated**; it also concludes that narrowband systems can reuse communication signals for sensing, whereas wideband systems require an **additional dedicated sensing signal** [2210.14112].

A related reciprocity issue appears in FDD massive MIMO. The study of **antenna duplex pattern an-reciprocity** shows that measured contemporary phones exhibit **noticeable duplex pattern divergence**, including gain differences and depolarization, which undermines assumptions of highly correlated average spatial signatures between UL and DL [2011.07492]. The reported implication is that practical handset pattern divergence must be incorporated into realistic DL CSI assessment, and that only **very small duplex separations** of approximately **1% or less fractional offset** allow true UL-DL reciprocity for practical phones [2011.07492].

Across these domains, the recurring lesson is that duplex is not synonymous with unconditional superiority. Whether in wireless links, ISAC, spin chains, or FDD reciprocity, the benefits depend on how the two directions or layers are **coupled, separated, or jointly optimized**.

Source: https://www.emergentmind.com/topics/duplex-aa48752a-b8ff-4f01-9d02-42d65546170f