1. Introduction

Optical waveguides with cross sections on the nanometer scale have been shown to exhibit superb nonlinear properties stemming from the high optical intensity and the ability to control propagation parameters e. g. dispersion and group velocity [1, 2]. The most common nanometric waveguides are optical wires, mainly silicon nano wires [1, 3] and photonic crystal (PhC) waveguides. The latter have been fabricated in III–V material [4–6], Silicon [7, 8] and chalcogenide glass [9,10]. Different nonlinear processes have been demonstrated in such waveguides including self phase modulation [6, 11], second and third harmonic generation [7, 12], stimulated Raman scattering [13, 14] and four wave mixing (FWM) [8–10, 15, 16].

Of all those nonlinear processes, the most versatile is FWM and therefore, it lends itself to many signal processing applications. Indeed, FWM was used in silicon nano wires for wavelength conversion [17], including conversion over ultra wide spectral extensions [18], signal regeneration [3] and multi casting [19]. In PhC waveguides, FWM was used to demonstrate wavelength conversion of ps pulses [20] and time demultiplexing of ultra high bit rate data [21].

This paper described dynamical functionalities in a W1 type GaInP PhC waveguide. GaInP is known to be advantageous due to its wide bandgap which prevents nonlinear losses due to two photon absorption at 1550nm [5, 11], enabling the observation of pulse compression and temporal soliton propagation in PhC waveguides [22]. Here we report on two FWM applications; wavelength conversion of 100ps pulses and time domain switching / demultiplexing of a pseudorandom binary sequence (PRBS) signal at 10.56Gbit/s. The PRBS signal is sampled by pulses at 1/16^th and 1/32^nd of its rate to yield demultiplexed channels at 660Mbit/s and 330Mbit/s, respectively. In all cases, the extracted signals at the idler wavelength are low noise high quality pulses which are obtained with moderate pump powers testifying to the high efficiency of the FWM process in the waveguides we used.

2. Device

The device we tested is a 1.5mm long W1 type GaInP PhC waveguide. The fabrication process was detailed in [23]. The 2D PhC structure consists of a triangular lattice of air holes with a periodicity of a = 475nm and a holes radius of r = 0.18a. The GaInP slab is ≃ 170nm thick. The W1 waveguide is created by omitting a single row of air holes in the ΓK direction. The radius of the first row of holes adjacent to the guide was increased to r₁ = 0.21a in order to improve the transmission characteristics of the waveguide. The structure contains mode converters [24] on both facets to improve coupling and eliminate end-facet reflections. The total insertion losses was 9dB of which 8dB accounted for input and output coupling losses while the linear propagation loss across the waveguide for fast modes was only 1dB. The linear transmission of the waveguide and the measured group index versus wavelength obtained by a phase-shift technique are shown in Fig. 1. The figure exhibits the expected behavior except for the dip near 1533nm. That dip is due to scattering into an odd-mode band edge [25], as also found in our earlier works [11].

 

Fig. 1 Left axis: Normalized transmission of the W1 waveguide (TE mode). Right axis: Wavelength dependence of measured group index (markers) and quadratic fit (line).

Download Full Size | PDF

3. Experimental results

The experimental setups for wavelength conversion and time-demultiplexing measurements based on FWM are described in Fig. 2. The pump source is a DFB laser emitting at λ_p which is modulated by a Mach Zendher (MZ) modulator driven from a fast pulse generator (FPG) forming 100ps wide pulses with a variable duty cycle. A tunable laser at λ_s serves as a signal source and operates in CW for the wavelength conversion experiments (Fig. 2(a)). For time demultiplexing measurements, the signal is modulated by a second MZ modulator driven by a PRBS signal generator at a bit rate of 10.56Gb/s which is synchronized to the pump FPG (Fig. 2(b)). The lasers are amplified and filtered before being combined through a 50 : 50 fiber coupler and inserted into the PhC waveguide via a NA = 0.95 microscope objective lens. The polarization of both signals is adjusted to be TE. The transmitted light is collected by a second objective lens and the idler wave, at λ_i, generated through the FWM process, is filtered and measured by a fast photodetector and a sampling or real time oscilloscope.

 

Fig. 2 (a) Experimental setup for (a) switching and (b) time-demultiplexing measurements. PC: polarization controller, mod: Mach Zendher modulator, EDFA: Erbium doped fiber amplifier, OBPF: optical band pass filter.

Download Full Size | PDF

3.1. Optical frequency domain measurements

In order to be able to filter the idler wave at the output of the waveguide while properly removing the residual pump, the detuning between pump and signal was set to 3nm. Such a detuning is close to the limit imposed by dispersion on FWM efficiency in common W1 type waveguides [8, 15]. A careful choice of pump and signal wavelengths enabled efficient operation in the present experiment. The pump and signal wavelengths were set to λ_p = 1537.4nm and λ_s = 1540.4nm, respectively. The group velocity at the pump wavelength was V_g = c/n_g ≃ c/5 (Fig. 1) while the second order dispersion parameter β₂ was almost constant at a value of ≃ −0.42ps²mm⁻¹ in the spectral range over which the experiments took place, 1534.4nm to 1543.4nm. Measurements of FWM conversion efficiency as a function of signal-pump detuning were fitted to a simple analytical model (as in [15]) in order to extract a nonlinear coefficient γ = 400W⁻¹m⁻¹. The γ value obtained here is lower than that found in our previous experiments [15]. The difference stems from the fact that the present experiment used a pump wavelength where the group index is n_g = 5 while in [15], n_g was 7 or higher. A recent theoretical paper by Santagiustina et al. described rigourously the relationship between γ, the field envelope-waveguide overlap and the square of the group index [26]. That calculations are consistent with the γ value we extracted from the experiments.

Figure 3(a) shows output spectra measured for a CW signal and a variety of peak pulse pump powers (obtained by changing the pump pulse duty cycle while keeping the average power constant). The pump peak power values are listed in the legend of Fig. 3(a). The averaged coupled powers in the PhC waveguide were 6.5mW and 6mW, respectively. The FWM conversion efficiency, defined as the ratio between the short wavelength idler output peak power and the signal input powers, takes on the values of −40.7dB, −34dB and −28dB, respectively. Note that the efficiency values take into consideration the duty cycles of the pump in the various experiments. These efficiencies represent an improvement of about one order of magnitude compared to our previously reported results [15]. The improvement is mainly due to a higher input coupling efficiency which naturally enables larger propagating powers.

 

Fig. 3 (a) Output spectra for a pulsed pump and a CW signal for different coupled peak pump powers. (b) Output spectrum for a pulsed pump at a duty cycle 1 : 16 (6.25%) and a coupled peak power of 100mW and a PRBS signal.

Download Full Size | PDF

The output spectrum for the case when the signal is a data sequence is described in Fig. 3(b). The coupled peak pump power was 100mW and the average coupled signal power was 7.5mW. Considering the duty cycles of the pump (6.25%) and the signal (50%), the conversion efficiency is ≃ −35dB which is also very high for a 3nm detuning and the power levels used here.

3.2. Time domain measurements

The two experimental set ups of Fig. 2 were used to demonstrate wavelength conversion of unmodulated 100ps wide pulses and time domain switching / demultiplexing of a 10.56GHz PRBS data stream into 16 and 32 channels.

3.2.1. Wavelength conversion

Using the experimental setup described in Fig. 2(a), we demonstrated wavelength conversion for unmodulated 100ps pulses. The time domain results are illustrated in Fig. 4 and correspond to the case of a pulsed pump at a duty cycle of 1 : 64 having a peak power of 380mW (red spectrum in Fig. 3(a)). The output waveform (Fig. 4(b)) consists of 100ps pulses converted from the input pump (Fig. 4(a)). A single converted pulse is shown in Fig. 4(c); the pulse is undistorted and has a large signal to noise ratio.

 

Fig. 4 Waveforms for 100ps pulses pump at a duty cycle 1 : 64 (peak pump power 380mW) (a) Pump input (b) Converted idler (c) Close up of one pulse of the converted idler.

Download Full Size | PDF

3.2.2. Time domain demultiplexing

We performed a time demultiplexing experiment of a 10.56Gb/s NRZ data stream using the experimental setup depicted in Fig. 2(b). Figure 5(a) shows a portion of the input PRBS signal. This is a data stream which is considered a time multiplexed signal at an aggregate rate of 10.56Gb/s. Figure 5(b) shows the input pump pulses which serve to sample the data. The sampling is at a rate which is 1/16^th of the data rate. The demultiplexed signal, shown in Fig. 5(c) is a PRBS signal at a rate of 660Mb/s. Similarly, Fig. 5(d) shows an extracted demultiplexed idler at 330Mb/s obtained when the sampling rate is 1/32^th of the aggregate data rate. In both cases, the demultiplexed data is clear and noise free due to the efficient FWM process. The demultiplexed pulse trains in Fig. 5 seem to have unequal amplitudes. This is an artifact of the real time oscilloscope which has a limited data acquisition capability. In reality, the pulses have exactly the same amplitude.

 

Fig. 5 (a) Signal input at 10.56Gb/s (b) Sampling pulses (pump) input at a rate of 10.56Gb/s/16 = 660Mb/s. (c) & (d) Demultiplexed idler. Portions of the extracted data pattern for two sampling rates: (c) 10.56Gb/s/16 = 660Mb/s (d) 10.56Gb/s/32 = 330Mb/s.

Download Full Size | PDF

Other measurements on almost identical samples have demonstrated that the FWM conversion is indeed ultra-fast with no memory effects, as a CW signal was converted into 3ps undistorted pulses at 10GHz rate [20]. Therefore, it is expected that the demultiplexing scheme is scalable to bit rates in the hundreds of Gbit/s range.

4. Conclusions

To conclude, we have demonstrated wavelength conversion and time domain demultiplexing using FWM in a GaInP PhC waveguide having a slightly modified hole design and mode converters at the end facets. Efficient and broad band FWM yields clean low noise pulses at the idler wavelength observed directly in the time domain, demonstrating the versatility of FWM and the superb nonlinear properties of the GaInP PhC structures. Also the ability to demultiplex efficiently high bit rate data streams testifies to the potential of PhC integrated optical circuits to be used in signal processing applications.