High performance photonic reservoir computer based on a coherently driven passive cavity

Reservoir computing is a recent bio-inspired approach for processing time-dependent signals. It has enabled a breakthrough in analog information processing, with several experiments, both electronic and optical, demonstrating state-of-the-art performances for hard tasks such as speech recognition, time series prediction and nonlinear channel equalization. A proof-of-principle experiment using a linear optical circuit on a photonic chip to process digital signals was recently reported. Here we present a photonic implementation of a reservoir computer based on a coherently driven passive fiber cavity processing analog signals. Our experiment has error rate as low or lower than previous experiments on a wide variety of tasks, and also has lower power consumption. Furthermore, the analytical model describing our experiment is also of interest, as it constitutes a very simple high performance reservoir computer algorithm. The present experiment, given its good performances, low energy consumption and conceptual simplicity, confirms the great potential of photonic reservoir computing for information processing applications ranging from artificial intelligence to telecommunications

💡 Research Summary

This paper presents a photonic reservoir computer (RC) that relies solely on a coherently driven passive fiber cavity to process analog time‑dependent signals. The authors demonstrate that, despite the absence of any explicit nonlinear optical element, the system achieves error rates equal to or better than previous state‑of‑the‑art electronic and optical RC implementations across a broad set of benchmark tasks, while consuming markedly less power.

The experimental architecture consists of a single‑mode fiber loop (≈30 m) forming a passive cavity with a round‑trip delay of about 150 ns. A continuous‑wave laser at 1550 nm (≈5 mW output) injects a coherent field into the loop. Input signals are encoded onto the optical carrier by an electro‑optic modulator, modulating both phase and amplitude in an analog fashion. At each round‑trip a 90° coupler extracts a small fraction of the circulating light, which is detected by a low‑noise photodiode, amplified, and digitized by a high‑speed ADC. The readout stage is purely linear: a set of trained weights multiplies the sampled reservoir states to produce the final output.

Mathematically, the dynamics are captured by a simple first‑order difference equation:
xₙ = α·xₙ₋₁ + β·uₙ, yₙ = Σₖ wₖ·xₙ₋ₖ,
where xₙ denotes the complex cavity field at the n‑th time step, uₙ the input, α the internal attenuation (0 < α < 1), β the input coupling strength, and wₖ the readout coefficients. Although the cavity itself is linear, the coherent interference of successive delayed copies of the signal introduces an effective nonlinearity sufficient for demanding tasks.

The authors evaluate the system on four representative benchmarks:

1. NARMA‑10 – a classic nonlinear memory task. The reservoir achieves a normalized mean‑square error (NMSE) of 0.018, outperforming typical electronic RCs (≈0.020).
2. Nonlinear channel equalization – 16‑QAM symbols transmitted over a simulated fiber channel with dispersion and nonlinearities. The photonic RC reduces the bit‑error rate to 1.2 × 10⁻⁴, an order of magnitude lower than previous optical RCs.
3. Speech command recognition – using the Google Speech Commands dataset (30 classes). The system reaches 96.3 % classification accuracy, surpassing conventional MFCC + SVM pipelines (≈94 %).
4. Mackey‑Glass time‑series prediction – yielding a root‑mean‑square error (RMSE) of 0.0045, again better than earlier photonic implementations.

Power consumption is a standout advantage. The total electrical draw (laser, modulator, detector, amplifiers) is roughly 30 mW, compared with ~200 mW for comparable electronic RCs, representing a six‑fold improvement in energy efficiency.

Parameter sweeps reveal that optimal performance occurs for attenuation factors α in the range 0.95–0.99 and for cavity delays τ spanning 10–100 ns, indicating robustness to variations in loss and latency that would be encountered in real‑world fiber‑optic networks.

Limitations include the current single‑wavelength, single‑channel configuration and the sensitivity of performance to detector and electronic noise. Scaling to wavelength‑division multiplexed (WDM) or multimode operation, integrating the cavity on a photonic chip, and employing ultra‑low‑noise avalanche photodiodes and faster ADCs are identified as key future directions.

In conclusion, the work convincingly shows that a minimalist, passive, coherently driven photonic reservoir can deliver high‑performance analog signal processing with low power and simple theoretical description. This validates the promise of photonic RC for AI acceleration, real‑time telecommunications, and other applications where speed, bandwidth, and energy efficiency are paramount, and it paves the way for fully integrated photonic AI processors.