Back to age-quantum.org AGE-TGI-2S3 + QEC · Photonic quantum research · ES/EN

AGE Correction

AGE Correction Results

Detailed numerical results of AGE re-diagonalisation for temporal modes in waveguide quantum networks

Comparison

Baseline vs AGE Comparison

Baseline Transfer Matrix

f0 f1 f2
f00.8060.0140.001
f10.0160.8380.045
f20.0000.0440.716

AGE-Corrected Transfer Matrix

g0 g1 g2
g00.8280.0020.000
g10.0020.8280.002
g20.0000.0020.826
16.5
× selectivity improvement
Σ
39.1 → 646
⟨diagonal⟩
0.786 → 0.827
Selectivity (dB)
15.9 → 28.1
Off-diag max
0.045 → 0.002
Parameters

Channel Parameters

Values used in the baseline simulation for coherent channel distortion

Parameter Symbol Value
Detuningδω/(2π)1.75 MHz
Coupling rate AκA26.7 MHz
Coupling rate BκB30.7 MHz
Lossploss17%
Shaping amplitudeσshape5%
Shaping correlation timeτcorr5 ns
BandwidthΓ/(2π)24 MHz
Properties

AGE Algorithm Properties

The AGE re-diagonalisation algorithm presents several favorable properties that make it suitable for experimental implementation. The computational cost is minimal: it requires a single 3×3 matrix inversion at calibration time, which is instantaneous on modern hardware. Numerical stability is guaranteed by the good conditioning of the Gram matrix, and Tikhonov regularization is available as a fallback for cases with degraded conditioning.

The maximum amplification |Am,k| = 1.018 is trivial for the AWG, meaning that corrected templates do not require significantly higher pulse powers than the originals. The per-mode template change in L² norm is 13-28%, a range well within the reprogramming capabilities of standard AWGs. The condition number cond(S) = 1.10 confirms that the problem is well-posed and the inversion is stable.

Absorption Matrix A

Well-conditioned
cond(S) = 1.10
Max amplification
|Am,k| = 1.018
Loss limit reached
0.827 / 0.830 = 99.6%
Fundamental limit: AGE correction recovers mode orthogonality but cannot overcome the 17% loss floor. The ratio 0.827/0.830 = 99.6% indicates that AGE reaches nearly the theoretical limit for the transfer matrix diagonal given the channel loss.
1D Sweep

Distortion Sweep Results

The 1D parameter sweeps reveal the differentiated behavior of AGE correction against different distortion sources. In the detuning sweep (0-5 MHz), baseline selectivity Σ degrades drastically from ~1942 to ~4, while ΣAGE stays approximately constant in the ~525-540 range. This stability demonstrates that AGE effectively compensates detuning regardless of its magnitude within the studied range.

In the shaping error sweep (0-20%), baseline selectivity remains nearly flat (~38-39), as shaping errors alone do not significantly break orthogonality. However, ΣAGE increases with shaping (466→1697), a counterintuitive effect explained by the fact that additional distortions provide more information for the AGE algorithm to optimize absorption templates.

Sigma vs Detuning
fig3a · sigma_vs_detuning
Sigma vs Shaping Errors
fig3b · sigma_vs_shaping
2D Robustness

2D Robustness Map

The 2D parameter sweep across 42 operating points shows AGE correction is robust across a wide range of channel conditions. The maximum improvement is 53.1× at (2.0 MHz, 15% shaping), where baseline selectivity has severely collapsed but AGE maintains high selectivity. This result is particularly important because it demonstrates that AGE correction is most valuable in high-distortion regimes where it is needed most.

At zero distortion, AGE can slightly hurt selectivity (improvement < 1×) since modes are already near-orthogonal and the correction introduces minor numerical degradation. At strong distortion (δω = 3 MHz), improvement reaches up to 44×, confirming the scalability of the benefit with channel severity. This smooth transition between regimes confirms the algorithm stability across the entire parameter space.

AGE 2D Robustness Map
fig6 · age_2d_sweep
Limitations & Extensions

Limitations & Extensions

Limitations

  • Does NOT correct loss: loss remains at 17% regardless of AGE correction. Re-diagonalisation only restores mode orthogonality, it cannot recover photons lost in the channel.
  • Phenomenological model: a phenomenological channel model is used that has not been validated against experimental superconducting chip data. Transferability to real hardware requires experimental verification.
  • K=3 modes only: the demonstration is limited to K=3 temporal modes. Scalability to K>3 requires additional analysis of Gram matrix conditioning.
  • Non-Markovian effects: non-Markovian memory effects in the channel are not considered, which could be relevant for waveguide networks with parasitic reflections.

Extensions

  • More modes (K>3): extension to higher-dimensional mode spaces to increase temporal encoding capacity per photon.
  • Cascaded networks: application of AGE to multi-node cascaded networks, where distortions accumulate along the chain.
  • Multiplexed encoding: combination of temporal modes with frequency or polarization encoding for hybrid quantum multiplexing.
  • Optical network adaptation: translation of the AGE framework to fiber optical networks, where distortion sources differ (chromatic dispersion, PMD).
  • Experimental validation: verification on real superconducting chip to confirm the numerical predictions of the phenomenological model.
Reproducibility

Computational Reproducibility

All results presented on this page are fully reproducible using the age_temporal_modes Python package. The package includes complete benchmark scripts, input data, and simulation parameters used. The total runtime is approximately 3 minutes on a standard computer, generating all figures and numerical data presented.

Python package: age_temporal_modes
Open data: data/01…05*.json
DOI repository: in preparation (Zenodo)

Output data (JSON) and simulation parameters are published on this page to enable independent verification; the clone commands were withdrawn until a public DOI repository is available (see the errata log, entry 7).

TM-04 → TM-06 ladder on age-quantum.org

age_temporal_modes

Language
Python 3
Dependencies
NumPy, SciPy, Matplotlib
Runtime
~3 minutes
Output format
JSON + PNG
Licencia
MIT

Reference Data

Download the complete parameter sweep and AGE correction data for independent verification.