High repetition rate frequency combs are predominantly used in optical communications, astronomical spectrographs andmicrowave photonics. Spectral broadening of electro-optic combs based on cascaded intensity and phase modulatorswith highly nonlinear fibers (HNLF) provides broadband combs with tunable repetition-rate and center-frequency.Spectral broadening is achieved using nonlinear effects such as self-phase modulation which requires substantial timedependent intensity at the input. To achieve this, the combs are compressed to a pulse using either fiber-based devices orpulse shapers. However, this has resulted in poor quality spectral broadening. Determining the optimal shaping profile ofthe input electro-optic comb for efficient spectral broadening is not direct due to the complex interplay between multipleparameters such as length, non-linear coefficient and dispersion of the nonlinear media, the initial spectral phases andpower of the comb and modulator biasing conditions. This problem has been addressed here using adaptive pulseshaping. We use cascaded electro-optic modulators to generate a comb with 9 lines (within 20dB) around 1550nm at25GHz repetition-rate. A wave-shaper changes the spectral phase of the comb. Dynamic spectral phase optimization bystochastic perturbations is performed in a closed loop by processing the output spectrum to maximize spectralbandwidth. With an output power of ～210mW, adaptive optimization more than tripled the number of lines to 29 (within20dB) with a smooth spectral envelope while the unoptimized case causes negligible broadening (11 lines). Weanticipate that the demonstrated testbed will enable more advanced methods of machine learning towards optimizationand shaping of frequency combs.
展开▼
