1. Design and Fabrication

1.1 TFLN Modulator

• Substrate: 360 nm X-cut single-crystalline LiNbO3 (LN) thin film on 2.5 µm SiO2 on quartz (NanoLN).
• Patterning: Waveguides and multi-mode interference (MMI) structures via electron-beam lithography (EBL) and reactive ion etching (fluorine-based).
• Waveguide Specs:
  • Sidewall angle: 72°
  • Slab thickness: 180 nm
  • Rib height: 180 nm
• Edge Couplers: Developed by additional slab etching.
• Cladding: 1.2-µm PECVD SiO2.
• Electrodes:
  • 180-nm NiCr layer for resistor/terminator.
  • 1-µm gold transmission lines and DC electrodes via lift-off.
• Post-Processing: Device end face diced, lapped, polished for enhanced coupling.
• Process: Compatible with foundry standard lithography/etching ensuring reproducibility and scalability.

1.2 Modified UTC-PD

• Epitaxial layers:
  • 180-nm InGaAs absorber with 80-nm undepleted graded doping region for quasi-electric field promoting transport.
  • 100-nm depleted region creating high field at InGaAs/InP heterojunction.
  • 220-nm lightly n-doped InP drift layer for carrier transit.
  • 30-nm heavily doped InP cliff layer to sustain high electric field, enabling velocity overshoot and minimizing carrier pile-up.
• Enhancements:
  • 4-µm benzocyclobutene dielectric layer beneath coplanar waveguides for low parasitic capacitance and robustness.
  • Extended InP drift layer doubles as upper cladding for InGaAsP waveguide.
  • 400-nm InGaAsP waveguide engineered for efficient evanescent coupling.
• Fabrication:
  • Metal–organic chemical vapor deposition of epilayers on 2-inch InP.
  • Ti/Pt/Au p-type contacts, Ge/Au/Ni/Au n-type contacts with thermal annealing.
  • Triple-mesa defined by ICP dry etching and selective wet etching.
  • Benzocyclobutene dielectric passivation and Ti/Au coplanar waveguides deposited.
• Process: Standard III–V semiconductor wafer-scale compatible.

2. Characterization of EO/OE Response

2.1 TFLN Modulator

• Low frequency Vπ (Vπ,LF): Measured as 5.1 V via 100 kHz triangular voltage sweep.
• Frequency-dependent calculation:
  [
  V{\pi, \mathrm{RF}} = V{\pi, \mathrm{LF}} \times 10^{-\frac{\mathrm{EO}S_{21}}{20}}
  ]
• EO bandwidth measurement:
  • Low-frequency (<110 GHz): Vector Network Analyzer (Keysight N5222B + N5292A + N4372E).
  • High frequency (110–220 GHz): Frequency multiplier upconversion, measured by Optical Spectrum Analyzer (OSA).
• Sideband power normalization:
  [
  Pn = \frac{P{\text{carrier}}}{P_{\text{sideband}}}
  ]
• Vπ,RF calculation:
  [
  V_{\pi,\mathrm{RF}} = \frac{\pi}{4} V_p \sqrt{P_n}
  ]
• EO response expressed via:
  [
  \mathrm{EO}S{21} = -20 \log\left(\frac{\pi}{4 V{\pi, \mathrm{LF}}}\right) – 20 \log(V_p) – P_c + P_s
  ]
• Measurement uncertainty dominated by OSA: ±0.4 dBm power translates to ±0.4 dB bandwidth uncertainty.

2.2 Modified UTC-PD

• Setup: Optical heterodyne with two tunable lasers producing beat frequencies:
  [
  f_{\text{beat}} = \frac{c}{\lambda_1} – \frac{c}{\lambda_2}
  ]
• Output RF power under full modulation:
  [
  P_{\text{ideal}} = \frac{1}{2} RL I{\text{ph}}^2
  ]
  where (R_L = 50 \Omega).
• Frequency ranges:
  • DC to 110 GHz: Power meter + 110-GHz GSG probe.

  • 110 GHz: THz power meter + waveguide probes.

• **Losses calibrated/de-embedded for accuracy.

3. Data Transmission Experiments

3.1 Short-Reach IMDD

• Signal generation: Pseudorandom bit sequence by AWG (Keysight M8199B), symbol rates 112–256 Gbaud.
• Amplification: High-bandwidth electronic amplifier (SHF T850 C).
• Modulation: TFLN modulator encodes NRZ, PAM-4.
• Detection:
  • DSO with 120-GHz optical sampling.
  • 70-GHz RTO + 110-GHz PD for real transmission.
• DSP: No bandwidth compensation for eye diagrams; complex-biGRU for signal recovery and BER.

3.2 Wireless Coherent Optical Transparent Relaying

• Optical baseband: Silicon coherent transmitter + 1,550 nm laser.
• IQ modulation: AWG for QPSK, 16-, 32-QAM.
• LO: Tunable ECL offset by 180 GHz generating THz signals by heterodyning in UTC-PD.
• THz Tx/Rx: Horn antennas (26 dBi gain), 20 cm spacing.
• Signal amplification: 145–220 GHz LNA with 24 dB gain.
• EO conversion: TFLN modulator driven by THz RX output.
• Filtering and amplification: Tunable FBG + EDFA.
• Reception: Optical modulation analyzer with LO laser.
• Wireless extension: 4 m link uses 40 dBi lens antennas.

3.3 DSP Processing

• Baseline DSP: Gram–Schmidt normalization, matched filtering, equalization (CMA), carrier recovery, frequency offset estimation, blind phase search, and orthogonalization.
• complex-biGRU: Further equalization using a deep bidirectional GRU neural network.

4. complex-biGRU Algorithm

• Architecture:
  • Five-layer input-processing-output network.
  • Input: Complex symbols separated into I and Q components.
  • Complex-biGRU layer: simultaneous bidirectional processing of I and Q.
  • Fully connected linear layer.
  • Multi-level nonlinear activation function adapted to signal modulation format.
• Activation function: Multi-threshold saturating, enabling better classification of multi-level modulations (PAM-4, PAM-6, 16-QAM, 32-QAM):

  [
  f(x) = \frac{2 \alpha_2}{1 + e^{-\alpha_1 (x – 2\mu)}} – \alpha_2 + 2\mu \alpha_2
  ]

  where (\alpha_1) controls gradient, (\alpha_2) maintains continuity, and (\mu) sets level thresholds depending on modulation.

• Purpose: Mitigate nonlinear distortions causing ‘jail window’ patterns, improving BER and FEC coding performance.

If you need more detailed info on any section, fabrication recipes, analytical equations, or algorithms, feel free to ask!