1. Introduction

The characteristics of quantum dot (QD) [1,2] and dash (QDash) [3,4] semiconductor optical amplifiers (SOA) are dictated by the inhomogeneous nature of the gain broadening [5,6] and by the fast (~1 ps) gain recovery times [7,8]. These nano structure based SOAs exhibit very wide gain bandwidths and very complex cross saturation dynamics [9,10]. Cross saturation between channels depends on three parameters [10]: detuning, channel modulation rate and channel placement within the gain spectrum. Proper choice of operating parameters enable multi wavelength amplification with no cross talk [11,12], pattern effect free amplification and regeneration [2,3,13] of 40 Gbit/s. Single channel processing using both cross gain modulation (XGM) [14] and four wave mixing (FWM) [14–16] was also demonstrated in those devices highlighting the fast dynamics and the low α parameter of these nano structure based gain media.

This paper describes another unique capability of inhomogeneously broadened nano structure gain media: the ability to perform simultaneous signal processing operations of different channels. Specifically, we demonstrate simultaneous wavelength conversion (with no cross talk) of two independent 10 Gbit/s channels using FWM. FWM is chosen as the process of choice due to its versatility and independence of modulation format. The two 10 Gbit/s signals are spectrally separated by 30 nm and each is accompanied by a CW pump. One converted (conjugate) 10 Gbit/s signal is detected and its dependence on the second FWM process taking place at a 30 nm detuning is examined. The relatively low gain of the QDash SOA requires large optical powers in order to obtain a properly detectible wavelength converted signal. The high powers cause depletion of high energy barrier and wetting layer states which induce in turn some homogeneity of the gain [10, 16]. Consequently, we observe near each channel cross FWM products induced by the FWM process in the other channel.

The experiment was performed for two types of pump-probe detunings. In the first case, the channels have different detunings and then the two FWM processes are decoupled and no cross talk is observed. When the experiment was repeated such that both channels had the same detuning, the channels affect each other and the detected signal deteriorates.

2. Quantum Dash structure and static characteristics

The gain region of the SOAs comprised six InAs QDash layers separated by InGaAlAs barriers and placed within a GRINCH structure. The AR coated 4 mm long amplifier had a chip gain of 20 dB at a bias of 500 mA with a saturation input power of about +2 dBm. Amplified spontaneous emission (ASE) spectra, measured at different bias levels, are shown in the insert of Fig. 2. The schematic of the two channel FWM experiment is shown in Fig. 1. Each channel consists of a CW pump signal and a 10 Gbit/s modulated probe which is detuned from the pump by 0.5–1 nm.

The wavelength converted (conjugate) signal of channel 1 was filtered at the output and was used for the cross talk examination.

 

Fig. 1. Experimental schematic.

Download Full Size | PDF

A measured output optical spectrum obtained with large input powers of approximately 4 dBm and -6dBm for the two pumps and two probes respectively, is shown in Fig. 2. This rather crowded spectrum (measured with a resolution of 0.05 nm) reveals many of the complexities governing cross saturation dynamics in a QDash SOA. Two sets of spectral lines, separated by approximately 30 nm are seen. Pump and probe (Pmp ^1,2 and Prb ^1,2, the probes being modulated at 10 Gbit/s) are clearly seen together with their corresponding conjugate (converted) signals (Conj ^1,2). Each set contains two additional lines marked in Fig. 2 as X mod ^2→1 and X mod ^1→2. These result from nonlinear processes involving the high energy barrier and wetting layer states which are highly saturated due to the large optical powers within the SOA. For example, gain and index variations at a frequency corresponding to the ~1 nm detuning (125 GHz) stemming from the FWM interaction between Pmp ¹ and Prb ¹, are sensed by high energy states (which are common to all dashes and are coupled to all energy states by carrier capture and escape processes [5,6]). This couples the widely detuned spectral regions. The coupling, while inefficient, induces nevertheless two lines near Pmp ² marked as X mod ^1→2. Similarly, Pmp ² and Prb ² yield X mod ^2→1.

 

Fig. 2. Measured output spectrum, inset shows ASE spectrum.

Download Full Size | PDF

The wide band conversion generating X mod ^1→2 and X mod ^2→1 suggests that the coupling via high energy states homogenizes the gain somewhat. It is therefore important to quantify the FWM capabilities of the QDash SOA at detunings larger than the natural homogeneous spectral width of 8–10 nm [5,6,17]. The experimental setup of Fig. 1 was used with just two CW signals. The pump wavelength was fixed at 1550 nm while the wavelength of the probe was tuned from 1584 to 1535 nm in order to encompass both positive and negative detunings. Using an optical spectrum analyzer we mapped out the FWM efficiency η_FWM defined as η_FWM=P_FWM (L)/P_Probe (0) where P_FWM (L) is the power of the conjugate signal at the output of the amplifier and P_Probe (0) is the power of the probe at the input of the amplifier. The FWM efficiency at a bias current of 390 mA is shown in Fig. 3.

Three distinct regions are observed in the positive detuning curve. The lower efficiency in the negative detuning case limited the detectability of the conjugate signal at large detunings and hence the curve shows only two clear regions. For the detuning region |Δf|<500 GHz, η_FWM rolls off at a rate lower than 20 dB per decade, similar to [14,15]. This indicates that the conjugate field results from multiple nonlinear polarizations arising from the various nonlinearities in the amplifier, such as total carrier density pulsations, coupling between the high energy tails of the density of state functions of different dash types, coupling to higher energy states and spectral hole burning. A non-zero α factor is responsible for the difference in magnitude of the absolute value of FWM conversion efficiency for positive and negative detunings.

A distinct break point appears in both curves at a detuning of 500 GHz. In this detuning region, the summation of all the polarizations adds constructively for the case of positive detuning and destructively for the case of negative detuning. A breakpoint in the FWM efficiency graphs at 500 GHz, suggests that a nonlinear gain mechanism with a characteristic time constant of ~320 fs dominates. Such a short nonlinear gain mechanism was not detected by short pulse pump probe measurements on similar gain media [8] that identified a 2 ps gain recovery time. Pump probe spectroscopy uses moderate saturation levels so as to ensure that the obtained recovery evolution is independent of the perturbation [18]. The present FWM experiment uses very deep saturation conditions which initiate a complex mixture of nonlinearities yielding the apparent very fast response.

 

Fig. 3. Detuning dependence of FWM efficiency in Qdash SOA at a bias of 390 mA. The inset shows the efficiency for positive detuning at a bias of 380 mA.

Download Full Size | PDF

As the detuning is increased beyond 500 GHz, the FWM efficiency levels off somewhat in the positive detuning curve. This is clearer for a lower bias level (380 mA) as seen in the insert of Fig. 3. This flattening does not occur for negative detuning.

For very large detunings, ~3THz, another break point is seen for the positive detuning curve. This detuning is far beyond the homogeneously broadened gain region. FWM requires gain or index modulation due to the coherent addition of the optical fields interacting with the nonlinear medium. Beyond the homogeneous gain region, the pump and probe interact with carrier populations which are weakly coupled through the slow high energy states and therefore η_FWM decreases rapidly at 55 dB per decade. The low efficiency FWM process in this region is mediated by inter-dash relaxation processes [19] which are governed by time constants in the tens of fs regime. No flattening of the conversion efficiency is observed at the widest detuning available (~4THz). Thus we postulate that there is no visible indication of FWM generation due to the intrinsic Kerr nonlinearity of the crystal or two photon absorption, processes that require high signal powers such as those attainable from short optical pulses [20]

3. Dynamic characteristics

The coupling across the entire gain spectrum suggests that multi channel processing may suffer from inter channel cross talk. However, it is crucial to note that Fig. 3 represents the CW case and Fig. 2 describes average spectra. The cross modulation spectral lines in Fig. 2 are mainly generated by nonlinearities of the slow barrier and wetting layer states which do not respond on the time scale of a single 10 Gbit/s bit. Therefore, under most operating conditions, they do not affect the properties of the converted signals as will be shown hereon.

Cross talk effects were examined by observation of the converted 10 Gbit/s in channel 1 and measurements of its bit error rate (BER) performance in the presence of channel 2. The experiment was done for the case where the pump-probe detuning in the two channels was different and then repeated when the detuning was equal.

For the case of different detuning, the eye pattern of the filtered Conj¹ signal was measured with the modulation of Prb² being turned either on or off. The two corresponding eye patterns which are shown in Fig. 4(a) and 4(b), respectively; are identical. Both reveal a larger noise level in the “1-state” due to the detection shot noise contribution. The two corresponding BER curves shown in Fig. 4(c) are indistinguishable. The overall BER performance is limited to about 6·10^-9 and the receiver input powers are rather high due to limitations of the detection system. Nevertheless the key issue of no cross talk of any kind when the two FWM processes take place simultaneously is unequivocally proven.

 

Fig. 4. Measurements of converted channel 1. (a) Channel 2 modulation on. (b) Channel 2 modulation off. (c) BER curves with and without modulation in channel 2.

Download Full Size | PDF

For the case of equal pump probe detunings in the two channels, X mod ^2→1 resides spectrally within the bandwidth of the output optical filter and is detected by the receiver. Beating between the two signals is then possible in the electrical domain. As a result Conj¹ is accompanied by the interfering signal X mod ^2→1 imprinting the data of channel 2 as well as the beating signal. The detected eye pattern under equal detuning conditions is shown in Fig. 5.

 

Fig. 5. Converted channel 1 for equal detuning.

Download Full Size | PDF

A large cross talk is observed causing severe eye closure yielding a non detectable signal and vastly deteriorated BER performance. An additional test in which the pump of channel 2 was turned off eliminated the cross talk completely proving that the deterioration is caused by the equally detuned FWM processes rather than by direct gain modulation due to probe 2.

4. Conclusions

To conclude, we have demonstrated simultaneous signal processing with no cross talk via FWM of two 10 Gbit/s channels separated by 30 nm in a QDash SOA. Large optical powers cause saturation thereby reducing somewhat the overall FWM efficiencies and also deplete high energy and wetting layer states which mediates FWM processes over the entire broad band inhomogeneous gain spectrum. However, this coupling is slow and does not affect the cross-talk-less simultaneous wavelength conversion of the two10 Gbit/s channels except when the pump-probe detunings for the two channels are equal.

To further demonstrate the uniqueness of the inhomogeneously broadened SOA, we repeated the experiment using a homogeneously broadened quantum well SOA. Large cross talk effects were observed in all cases, as expected, mainly due to gain modulation by the 10 Gbit/s probe in channel 2.