Wideband SiN pulse interleaver for optically-enabled analog-to-digital conversion: a device-to-system analysis with cyclic equalization

Abstract

We present the design and experimental characterization of a silicon nitride pulse interleaver based on coupled resonator optical waveguide filters. In order to achieve a targeted free spectral range of 1.44 THz, which is large given the reduced optical confinement of the silicon nitride platform, individual ring resonators are designed with tapered waveguides. Its application to time-interleaved photonically-assisted ADCs is analyzed by combining experimental characterization of the photonic integrated circuit with a comprehensive model of the entire ADC. The impact of fundamental signal distortion and noise sources affecting the converter is investigated and suitable equalization techniques at the digital signal processing level are evaluated. The novel application of a simple but powerful equalization filter in the DSP domain allows for a significant improvement of the digitized signal SNR. An ENOB of 5 over a 75 GHz bandwidth (150 GS/s) and an ENOB of 4.3 over a 100 GHz bandwidth (200 GS/s) are expected to be achievable with compact and off-the-shelf single-section semiconductor mode locked lasers, that can be further improved with lower noise light sources.

Scalable orthogonal delay-division multiplexed OEO artificial neural network trained for TI-ADC equalization

“…9(b), for which we have previously benchmarked linear feed-forward equalization (FFE) with digital electronics. 53 The OE-TI-ADC samples an electrical signal by applying it to a Mach-Zehnder modulator (MZM), to which pulses with different center wavelengths are being fed. 11 Their separation by center wavelength at a following optical processing stage subdivides the modulated pulse train into a number of reduced rate sample streams, that can be analyzed by lower speed electrical ADCs.…”

“…However, time-and frequency-domain signal leakage between the pulse trains results in signal distortion, which was previously addressed with FFE. 53 The MZM must also be driven in the small signal regime to prevent nonlinear distortion of the sampled data, limiting the SNR of the system. 54 To address this limitation, the FFE is replaced by an ANN that allows nonlinear signal equalization and thus driving the MZM with a higher signal strength, reducing the number of optical amplifiers required to maintain the SNR.…”

“…Signal processing in the analog domain, by interposing the ANN between the photoreceivers of the OE-TI-ADC and the digital electronics, allows performing the nonlinear equalization before quantization noise is incurred, so that amplification of quantization noise by the equalizer can be avoided. 53 The overhead associated with implementing the equalization with an OEO-ANN is reduced since integrated photonics are already required for the front-end of the OE-TI-ADC, and the same comb source can be used for both systems. While a higher pulse repetition rate is required for the ANN to avoid aliasing, similarly to what happens in asynchronously clocked time-interleaved electronic architectures, 55 the doubling from 25 GHz to 50 GHz FSR can be straightforwardly obtained by using an interleaver (imbalanced MZI) selecting every second comb line prior to optical amplification.…”

“…The MZM is driven with a high signal strength reaching 80% of the maximum range (corresponding to ±p/4 phase shift in push-pull operation), with the resulting nonlinearity compensated by the ANN. Consequently, two booster optical amplifiers (BOA), that would be otherwise necessary to amplify the complementary outputs of the MZM, 53 have been removed from the network model.…”

“…The stark drop in performance of the unequalized channel is due to increased inter-channel signal leakage at higher signal frequencies. 53 At the same time, the performance is also bounded by the nonlinear distortion introduced by the MZM sampler, which determines the SNR at lower signal frequencies. In the absence of ANN network noise, the single output ANN equalizes the data very well, achieving an ENOB of 5.5.…”