A unified resource-pool architecture for high-dimensional direct-detection optical communication

Abstract: Increasing optical communication capacity without proportionally increasing receiver complexity remains a key challenge for direct-detection links. Conventional systems typically assign wavelength, polarization and intensity to fixed, separately recovered functions, so that alphabet expansion is accompanied by additional demultiplexing, polarization handling, receiver branches and electronic processing. Here we introduce a unified resource-pool architecture for high-dimensional direct-detection optical communication, in which wavelength, polarization and intensity are jointly organized as a composite optical symbol space and recovered through optical-domain joint projection rather than dimension-by-dimension separation. The receiver is implemented with an integrated disordered photonic processor that transforms each composite optical state into a reproducible multi-output electrical fingerprint for single-shot direct recovery. In a dual-wavelength transmission experiment, the system resolves 4096 composite symbols, corresponding to 12 bits per symbol slot, with a bit error rate of 4.25e-4 after 10 km standard-fiber transmission. Additional experiments demonstrate dense polarization alphabets, wavelength-indexed state-space expansion and high-launch-power operation over hollow-core fiber. These results establish disorder-enabled joint projection in an integrated photonic processor as a route to hardware-efficient high-dimensional direct-detection communication beyond conventional dimension-partitioned receiver architecture.

• The paper demonstrates that integrating wavelength, polarization, and intensity into a composite symbol space increases mutual information per symbol.
• It employs a disordered photonic processor with a neural network for efficient joint symbol recovery and significant reduction in receiver complexity.
• Experimental results reveal a 50% reduction in physical resource scale and an 83% decrease in hardware complexity compared to conventional IMDD systems.

A Unified Resource-Pool Architecture for High-Dimensional Direct-Detection Optical Communication

Introduction and Motivation

Conventional optical communication systems, particularly intensity-modulation direct-detection (IMDD) links, are fundamentally dimension-partitioned: wavelength, polarization, and intensity are demultiplexed and processed by dedicated hardware subsystems. This paradigm results in hardware complexity scaling at least linearly with the number of exploited physical degrees of freedom, hampering further increases in spectral efficiency and communication density. Alphabet expansion, essential for increased throughput, invariably results in additional complexity, optical filtering, and electronic processing. Even recent high-dimensional photodetectors targeting advanced state discrimination have primarily addressed sensing and pattern recognition, rarely extending their architectures to true communication frameworks evaluated by bit error rate (BER) over practical fiber links.

This paper presents a unified resource-pool architecture for high-dimensional direct-detection optical communication, in which wavelength, polarization, and intensity are not assigned to statically separated roles. Instead, these optical variables form a composite symbol space, directly projected and recovered using an integrated disordered photonic processor. This paradigm shift enhances mutual information per detected symbol and numerically decouples alphabet cardinality from receiver complexity.

Unified Resource-Pool Framework and Theoretical Underpinnings

The described architecture organizes wavelength, polarization, and intensity into a direct-product symbol space, enabling multiplicative rather than additive alphabet scaling. The resulting symbol $S=(\Lambda, P, I)$ resides in a composite state-space whose entropy increases as $H(S) = \log_2(|\Lambda| \cdot |P| \cdot |I|)$. Critically, receiver processing no longer requires isolation of physical channels; instead, distinguishing the resulting multidimensional states after a fixed optical projection is sufficient.

The architecture’s physical-layer receiver is an integrated silicon-on-insulator photonic chip embedding a disordered scattering region, coupled to a multi-port detector array. This passive structure implements an optical-domain joint projection: incoming optical symbols are mapped to reproducible, high-dimensional electrical ‘fingerprints’, which are uniquely associated with each transmitted composite symbol. Digitally, a lightweight neural network operating on these fingerprints accomplishes symbol recovery.

Critically, information loading (IL)—the mutual information per symbol slot—scales multiplicatively in the resource-pool architecture, while physical resource scale (PRS) and hardware complexity (HC) remain largely fixed. Numerical analysis demonstrates that for a twelve-bit 4096-symbol alphabet, the resource-pool framework achieves a 50% reduction in PRS and an 83% reduction in HC compared to a conventional polarization-division-multiplexed PAM4 baseline, with even greater savings as the symbol space expands.

Photonic Processor Implementation and Characterization

The integrated disordered photonic processor serves as the key enabling hardware. By leveraging disorder-induced multi-path interference in a micron-scale footprint, the chip generates distinct, high-dimensional electrical responses for each composite optical input. Experimental characterization reveals:

• Polarization-state discrimination with 1.2° precision on the Poincaré sphere,
• Spectral decorrelation bandwidth of 0.4 nm, enabling efficient wavelength discrimination,
• Optoelectronic bandwidth of 22.6 GHz, favorable for deployment in high-speed applications,
• Robust four-level PAM (PAM4) recovery, confirming dynamic performance.

These results confirm the ability to resolve a high-cardinality composite state space not through orthogonal physical demultiplexing and separation, but through learned digital classification of complex response patterns.

Experimental Demonstration and Transmission Results

A prototype system encodes information jointly across dual-wavelengths (1550, 1551 nm), four intensity levels, and sixteen polarization states per wavelength, yielding a 4096-state composite alphabet (12 bits/symbol slot). After propagation over standard single-mode fiber (SMF), the system demonstrates reliable recovery (99.49% accuracy, BER $\sim 4.25 \times 10^{-4}$ after 10 km), with all physical degrees of freedom individually classified at $\geq$99% accuracy.

The system maintains an acceptable BER well below the 7% hard-decision FEC threshold at received powers down to −9 dBm and for transmission distances up to 30 km. The architecture’s resilience is further validated in high-power regimes: in conventional SMF, degradation occurs for powers $>$12 dBm, dominated by Brillouin scattering and impairments in the intensity dimension, whereas hollow-core fiber (HCF) enables stable operation to 17 dBm, thus confirming the importance of nonlinear-tolerant transmission media for large-alphabet state recovery.

Alphabet Scalability: Polarization and Wavelength Expansion

The scalability of the joint-projection principle is empirically validated by extending:

• Polarization Densification: A single-wavelength, 128-state (32 polarization $\times$ 4 intensity) alphabet is robustly recognized with $\geq$99% symbol accuracy, demonstrating that polarization can serve as a dense symbolic coordinate rather than merely a binary channel.
• Wavelength Indexing: Sequential increase from 2 to 16 wavelength states (spanning 1542–1557 nm), while maintaining polarization and intensity encoding, yields up to 1024-symbol alphabets (10 bits/symbol). Even at the highest cardinality, joint classification accuracy remains above 99.1%, confirming sufficient spectral decorrelation within the photonic processor.

These findings substantiate the generality of the architecture for leveraging available degrees of freedom, limited only by the decorrelation properties of the integrated disorder transfer function.

Practical and Theoretical Implications

The resource-pool communication architecture shifts the scaling law for optical receivers: rather than requiring physically proportionate hardware to resolve each exploited DOF, a fixed hardware configuration can resolve exponentially growing alphabets by exploiting joint photonic projection and compact digital post-processing. This is achieved without violating fundamental channel capacity limits. The projection strategy allows for adaptive co-design of modulation alphabets, photonic hardware, and signal processing.

Dimension-resolved analysis under fiber nonlinearities reveals that the architecture exposes which symbol dimensions are most vulnerable—a potential aid for link adaptation and physical layer security. The approach is generalizable beyond communication, enabling new strategies in multidimensional optical-state sensing, photonic computation, and interface design for multimodal measurement systems.

Limitations and Future Directions

This work does not extend the theoretical capacity of the optical channel itself but reorganizes information encoding for practical, hardware-efficient recovery of high-dimensional classical alphabets. Real-world deployment will require high-baud, RF-optimized integrated packaging, low-insertion-loss design, and real-time DSP-optimized classification. Environmental robustness and calibration stability remain practical challenges.

The demonstrated architecture provides a systematic framework for future studies into mutual information maximization over composite optical state spaces, intelligent projection design, and modulation format co-optimization with adaptive machine learning-based recovery.

Conclusion

This paper establishes a unified resource-pool architecture for high-dimensional direct-detection optical communication, combining wavelength, polarization, and intensity into a joint symbol space directly projected by an integrated disordered photonic processor. The architecture achieves substantial reductions in hardware complexity and resource scaling for large symbol alphabets, validated by experimental recovery of 4096-symbol (12-bit) composite alphabets over realistic transmission conditions. Robustness to nonlinear impairments and scalability along polarization and wavelength dimensions are demonstrated, addressing core limitations of dimension-partitioned architectures. This approach opens new avenues for pragmatic, hardware-efficient, high-dimensional optical communication and generalized multidimensional photonic state classification (2606.20004).