Expanding detection bandwidth via a photonic reservoir for ultrafast optical sensing

Basic Information

• Paper ID: 2510.10404
• Title: Expanding detection bandwidth via a photonic reservoir for ultrafast optical sensing
• Authors: Yuito Ito, Tomoaki Niiyama, Tetsuya Asai, Gouhei Tanaka, Atsushi Uchida, Satoshi Sunada
• Classification: physics.optics cs.ET
• Publication Date: October 12, 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2510.10404

Abstract

Detection of ultrafast optical and radiofrequency signals is critical for applications ranging from high-speed communications to advanced sensing. However, conventional detectors face fundamental limitations imposed by their intrinsic bandwidth, constraining accurate wideband signal measurement. This paper demonstrates that a neuromorphic photonic processing approach can overcome this limitation, enabling precise wideband signal detection beyond the detector's bandwidth. The key concept involves spatiotemporal encoding of input waveforms within a photonic reservoir network to reconstruct high-frequency components inaccessible to individual detectors. The research experimentally demonstrates detection of high-speed optical phase signals using an on-chip silicon photonic reservoir, achieving effective bandwidth expansion exceeding 8-fold. This method provides a scalable and integrable platform for high-speed optical and radiofrequency signal processing, opening new opportunities for ultrafast photonics and next-generation communication systems.

Research Background and Motivation

Problem Definition

The core problem addressed in this research is the intrinsic bandwidth limitation of conventional optoelectronic detectors, manifested as:

• Fundamental bandwidth-responsivity trade-off faced by conventional high-speed photodetectors
• Increasing bandwidth typically requires reducing the active area or junction capacitance, which degrades quantum efficiency and signal-to-noise ratio
• Achieving high-speed operation often necessitates increased system complexity and cost

Problem Significance

High-speed optical detection and sensing are critical in modern optical technology, including:

• Growing data rate demands in optical communication systems
• High-speed imaging, millimeter-wave/terahertz generation
• Photonic computing and quantum information processing
• Development of advanced sensing technologies

Limitations of Existing Approaches

1. Photonic-assisted high-speed data acquisition: Primarily targets radiofrequency signal sampling rather than optical signals
2. Conventional photodetector design: Cannot simultaneously achieve ultrafast response and high sensitivity
3. Bandwidth limitation: The condition $B_{sensor} \ll B_{sig}$ severely constrains accurate wideband signal measurement

Research Motivation

Inspired by the reservoir computing paradigm in neuromorphic computing, this work proposes utilizing photonic reservoir networks for spatiotemporal mapping, distributing high-speed temporal information across multiple low-frequency channels, enabling conventional narrowband photodetectors to capture wideband optical signals far exceeding their intrinsic bandwidth.

Core Contributions

1. Proposed a novel wideband signal detection framework based on photonic reservoirs: Achieving bandwidth expansion through spatiotemporal encoding
2. Experimentally verified effective bandwidth expansion exceeding 8-fold: Using on-chip silicon photonic reservoir chips
3. Achieved direct detection of optical phase signals: Without requiring complex interferometric apparatus
4. Provided a scalable integrated platform: Compatible with silicon photonics platforms, supporting low-cost mass production
5. Demonstrated wavelength division multiplexing applications: Further enhancing reconstruction performance

Methodology Details

Task Definition

Input: High-frequency wideband optical signal $u(t)$ with maximum frequency $B_{sig}$Output: Reconstructed wideband signal $y(t)$Constraint: Sensor bandwidth $B_{sensor} \ll B_{sig}$Objective: Achieve $y(t) \approx u(t)$ under the constraint

Model Architecture

Reservoir Network Model

The reservoir network contains $N_r$ nodes, represented as $\mathbf{z}(t) = (z_1, z_2, \cdots, z_{N_r})^T \in \mathbb{C}^{N_r}$, with dynamics:

$\frac{d\mathbf{z}(t)}{dt} = G(\mathbf{z}(t), u(t))$

where $G(\cdot)$ characterizes the reservoir network dynamics.

Sensor Output Model

The output of $N$ sensors is represented as $\mathbf{x}(t) = f(\mathbf{z}(t)) \in \mathbb{R}^N$, where $f(\cdot)$ is the sensor response function. For intensity detection with low-pass filtering:

$f(\mathbf{z}) = \text{LPF}(|\mathbf{z}(t)|^2)$

Linear Reconstruction Model

Employing temporal multiplexing, the output signal is expressed as:

$y(j\Delta t) = \sum_{k=0}^{K-1} \mathbf{w}_k^T \mathbf{x}((j-k)\Delta t)$

where $K$ is the number of temporal multiplexing steps and $\mathbf{w}_k \in \mathbb{R}^N$ is the weight vector for the $k$-th delay step.

Training Optimization

Weight vectors are determined by minimizing mean squared error:

$\{\mathbf{w}_k^*\}_{k=0}^{K-1} = \arg\min_{\{\mathbf{w}_k\}} \left[ \sum_{l=1}^L \sum_{j=1}^{N_T} |y_l(j\Delta t) - u_l(j\Delta t)|^2 + R(\{\mathbf{w}_k\}_{k=0}^{K-1}) \right]$

where $R(\cdot)$ is a regularization term.

Technical Innovations

1. Spatiotemporal encoding strategy: Mapping high-frequency temporal information to spatially distributed narrowband signals
2. Optical phase sensitivity: Exploiting optical interference within the reservoir to convert phase to intensity
3. Nonlinear compensation capability: Inherent reservoir nonlinearity can compensate for nonlinear distortion in photodetectors
4. Wavelength division multiplexing extension: Increasing equivalent channel count through multi-wavelength inputs

Experimental Setup

Hardware Platform

• Photonic reservoir chip: Silicon photonic reservoir based on stadium-shaped microcavities, coupled to 14 single-mode waveguide channels
• Microcavity dimensions: 50 × 200 μm
• Storage length: Approximately 250 ps
• Laser: Alnair Labs TLG-220 (linewidth 100 kHz, power 20 mW)

Experimental Configuration

• Signal generation: Arbitrary waveform generator (Tektronix AWG70002A, 25 GS/s)
• Phase modulation: Lithium niobate phase modulator (16 GHz bandwidth)
• Detector: InGaAs PIN photodetector (Thorlabs RXM25BF)
• Data acquisition: Digital oscilloscope (Tektronix DPO72504DX, 50 GS/s)

Training Parameters

• Number of training signals: $L = 10$
• Signal length: $N_T = 16,384$ ($T = 655.36$ ns)
• Maximum frequency: $B_{sig} = 10$ GHz
• Sampling interval: $\Delta t = 0.02$ ns

Evaluation Metrics

Normalized Mean Squared Error (NMSE): $\text{NMSE} = \frac{\sum_{j=1}^{N_{test}} (y(t_j) - u_{test}(t_j))^2}{\sigma_u^2}$

where $\sigma_u$ is the standard deviation of the target signal.

Experimental Results

Main Results

Bandwidth Expansion Performance

• Numerical verification: 10-fold bandwidth expansion ($B = B_{sig}/B_{sensor} = 10$), NMSE = $3.0 \times 10^{-4}$
• Experimental verification: 10-fold bandwidth expansion (10 GHz signal, 1 GHz filter), NMSE = 0.083
• Practical detector: Detection of 10 GHz signal using 1.2 GHz bandwidth detector, NMSE = 0.25

Sensor Number Dependency

Experiments demonstrate that NMSE monotonically decreases with increasing sensor count $N$, showing particularly good performance when $B \leq 100$. For the extreme case of $B = 1000$, increasing the number of training samples $L$ is necessary to improve performance.

Noise Robustness

In Gaussian noise environments, reconstruction quality degradation can be compensated by increasing sensor count $N$, with NMSE approximately following the relationship $\text{NMSE} \propto N^{-1/2}\exp(-\text{SNR})$.

Diverse Signal Reconstruction

Successfully reconstructed multiple signal types:

• Laser chaos time series (NMSE = 0.06)
• 100 ps pulses (NMSE = 0.63)
• Santa-Fe chaos sequence (NMSE = 0.09)
• Chirp signals (2-8 GHz, NMSE = 0.08)

Wavelength Division Multiplexing Results

Using three wavelengths (1550.0, 1550.1, 1550.2 nm) to achieve $N = 27$ equivalent channels, successfully reconstructing 10 GHz signals under narrowband detector conditions (1.2 GHz), with NMSE < 0.30 when $N > 16$.

Filter Order Impact

Experiments reveal that fourth-order or higher-order filters cause severe attenuation of high-frequency components, resulting in dramatic NMSE increase, indicating the method's sensitivity to extreme bandwidth limitations.

Related Work

Photonic-Assisted Sampling Techniques

• Photonic interleaved sampling: Primarily targets radiofrequency signals rather than optical signals
• Photonic time-stretch technology: Used for high-speed imaging and signal processing
• Photonic compressed sensing: Wideband sparse radiofrequency signal sampling

Reservoir Computing Research

• Photonic reservoir computing: Exploiting inherent dynamics of optical systems for high-dimensional mapping
• Silicon photonic reservoirs: Implementing reservoir computing on chip
• Time series processing: Powerful tool for ultrafast time series processing

High-Speed Photodetectors

• Bandwidth-responsivity trade-off: Fundamental limitation of conventional design
• Traveling-wave photodetectors: Special structural design for performance enhancement
• Microring resonator photodetectors: Enhanced performance in specific spectral bands

Conclusions and Discussion

Main Conclusions

1. Overcoming bandwidth limitations: First demonstration of optical detection bandwidth expansion exceeding 8-fold
2. Optical phase signal detection: Detecting high-speed optical phase dynamics without complex interferometric apparatus
3. Scalability: Reconstruction accuracy improves with increasing detector count
4. Practicality: Compatible with existing silicon photonics platforms, supporting low-cost implementation

Limitations

1. Scattering loss: Current implementation exhibits significant scattering loss exceeding 15 dB
2. Filter order sensitivity: High-order filters severely impact reconstruction performance
3. Training data requirements: Extremely high bandwidth ratios require substantial training samples
4. Output port limitations: Current reservoir output port count is limited

Future Directions

1. Higher frequency applications: Extension to 100 GHz signal detection
2. Integration optimization: Reducing scattering loss and improving integration density
3. Nonlinear compensation: Further exploiting reservoir nonlinearity
4. Multi-wavelength extension: Developing more efficient wavelength division multiplexing schemes

In-Depth Evaluation

Strengths

1. Strong innovation: First application of reservoir computing to optical bandwidth expansion
2. Comprehensive experiments: Complete validation from numerical simulation to actual hardware
3. High practical value: Addresses fundamental limitations of photoelectric detection
4. Advanced technology: Combines advantages of neuromorphic computing and silicon photonics

Weaknesses

1. Limited theoretical analysis: Lacks theoretical analysis of bandwidth expansion limits
2. Insufficient comparison benchmarks: Lacks detailed comparison with other bandwidth expansion techniques
3. Cost-benefit analysis: Insufficient discussion of cost comparison with conventional high-bandwidth detectors
4. Long-term stability: System long-term stability and reliability not evaluated

Impact

1. Academic contribution: Provides new technical pathway for photoelectric detection field
2. Application prospects: Broad application potential in high-speed communications, quantum information, etc.
3. Technology advancement: Promotes further development of reservoir computing in optical domain
4. Commercialization potential: Compatible with existing silicon photonics manufacturing processes

Applicable Scenarios

1. High-speed optical communications: Ultra-high-speed data transmission systems
2. Quantum information processing: High-speed detection of quantum states
3. Wideband optical spectral sensing: Optical spectral measurement requiring high temporal resolution
4. Terahertz technology: Optical detection of terahertz signals
5. Scientific instruments: Research equipment requiring ultrafast detection

References

This paper cites 52 relevant references covering important works in high-speed photoelectric detection, reservoir computing, silicon photonics, and other related fields, providing solid theoretical foundation and technical support for the research.

Summary: This paper proposes an innovative bandwidth expansion method based on photonic reservoirs, achieving important breakthroughs at both theoretical and experimental levels. This work not only addresses fundamental limitations of conventional photodetectors but also opens new directions for neuromorphic photonics development, possessing significant academic value and application prospects.