1. Introduction

Interferometric noise is highly undesired in fiber-optic communication systems, which severely degrades the received signal performance [1]. This kind of noise is generated from beating between the signal and an interferer when the wavelength difference between them is within the electrical bandwidth of the photodetector [2]. The interferer is usually produced by reflections from an unclean fiber facet or non-ideal channel isolations [3]. Due to the square-law detection property of a photodetector, interferometric noise results in severe amplitude fluctuation in the received signal even when the interferer is much weaker than the signal, which in turn dramatically degrades bit error rate (BER) performance [4, 5 ]. Due to the limited channel isolation in optical devices, including wavelength multiplexer/demultiplexer, optical filter, and optical circulators, interferometric noise is inevitable in fiber optics communication systems. Recently, a number of approaches have been demonstrated to mitigate signal degradation resulted from interferometric noise or to avoid the generation of interferometric noise in various types of fiber-optic systems. For example, highly nonlinear bismuth-oxide fiber (Bi-NLF) based dispersion-imbalanced loop mirror (DILM) has been used to remove interferometric noise in optical time division multiplexing (OTDM) system, and over 5 dB signal to noise ratio (SNR) improvement is achieved [6]. With the use of a GeO₂-doped fiber based DILM, both coherent and incoherent interferometric noise in optical code-division multiple access (OCDMA) networks have been suppressed successfully [7, 8 ]. Furthermore, fiber-based Mach-Zehnder interferometer (F-MZI) has been used as an interferometric noise suppressor in dense wavelength-division-multiplexing (DWDM) system [9]. In single fiber bidirectional passive optical networks (PONs), optical or electrical filtering schemes have also been used to suppress interferometric noise [10–12 ]. To prevent the generation of interferometric noise, one can try to avoid beating to occur between the signal and an interferer. One example is to use mode multiplexing in a simple passive single-mode to multimode combiner instead of a standard single-mode fused coupler [13]. Other approaches to mitigate the power penalty of interferometric noise are electrical DC-block filtering, bit pattern misaligning, and optical phase scrambling [4, 14, 15 ].

In this paper, we propose and experimentally demonstrate a bit rate transparent interferometric noise suppression scheme using the nonlinear modulation curve of an electro-absorption modulator (EAM). Bit-rate transparent is a critical property of an interferometric noise mitigation scheme due to the heterogeneous signals supported by modern communication systems. However, it is challenging to achieve due to the phase dependency in DILM [6–8 ] and filtering bandwidth sensitivity of optical/electrical filter [10–12 ], as a result, reconfiguration of the interferometric noise mitigation scheme is needed when there is a change in bit-rate. In our approach, zero-slope region of an EAM nonlinear modulation curve is used to compress the signal amplitude fluctuation resulted from interferometric noise, through electro-optic (EO) conversion at the EAM. The proposed interferometric noise mitigation scheme is bit-rate transparent, while the major device we used is a compact InGaAsP multiple quantum wells (MQW) EAM. Recent researches [16–19 ] have shown that EAM can be easily integrated with other optical or electrical devices through photonic integration circuit (PIC) technology which provides an effective way to further reduce cost and improve stability of whole system. This paper presents an experimental validation of the proposed bit rate transparent interferometric noise mitigation scheme at various bit rates from 2.5 Gb/s to 10 Gb/s. Experimental results show that our scheme can provide more than 8.5 dB improvement in receiver sensitivity for a signal with SNR over 12.5 dB.

2. Principle and experimental setup

Principle of the proposed interferometric noise mitigation scheme using EAM is shown in Fig. 1(a) . When the wavelengths of signal and interferer are the same or are within the electrical bandwidth of a photodetector, interferometric noise is resulted through signal beating between the signal and an interferer at the photodetector, which their relative phase difference is usually rapidly changing over time [13]. This rapidly changing phase difference causes phase-induced amplitude fluctuation in time domain after photodetection, which usually shows up at the high-level of an eye diagram as illustrated in grey color in Fig. 1(a). Since this significant amplitude-varying region occupy a large portion of noise margin in the eye diagram, it is difficult to remove the noise through amplitude thresholding (i.e. both all-optical and optical-to-electrical approaches), resulting in a severely degraded BER performance. With the use of the flat zero-slope region in the nonlinear modulation curve of EAM (depicted in blue solid shade in Fig. 1(a)), large input power fluctuation can be suppressed to relatively small power fluctuation at the output of the EAM. Thus, the reversed bias voltage is set at the linear modulation region and this bias voltage is combined with the degraded signal through a bias-tee. With the set reversed bias voltage, the amplitude fluctuation region at high-level of the eye diagram (gray area) is aligned to the zero-slope region, while the clear portion of the eye diagram is aligned to the linear modulation region (green dotted area in Fig. 1(a)) in the modulation curve of EAM which has steep slope (slope ≫1). With this bias setting, power fluctuation at the high-level is suppressed by the zero-slope region, while the eye-opening is enlarged by the linear modulation region [20]. Thus, through properly biasing the EAM, both the zero-slope region and a portion of linear modulation region are effectively used, a clean and wide-open eye diagram is resulted and the BER performance is significantly improved. The degree of improvement depends on the overall shape of the EAM modulation curve.

 

Fig. 1 (a) Principle and (b) Experimental setup of the proposed interferometric noise suppression scheme. LD: laser diode, PC: polarization controller, MZM: Mach-Zehnder modulator, PPG: pulse pattern generator, OC: optical coupler, VOA: variable optical attenuator, SMF: single-mode fiber, PD: photodetector, EAM: electro-absorption modulator, RF AMP: radio frequency amplifier.

Download Full Size | PDF

Experimental setup of the proposed interferometric noise mitigation scheme is shown in Fig. 1(b). Continuous wave (CW) light is generated from a laser diode (LD 1) and is modulated by a pseudo-random binary sequence (PRBS) signal at the LiNbO3 Mach-Zehnder intensity modulator (MZM). Interferometic noise is generated by passing the modulated signal through a Mach-Zehnder interferometer (MZI) structure and detected by a photodetector. The MZI consists of two optical couplers (OCs) for splitting and combining the signal, where upper branch is the signal and lower branch is the interferer. At the lower branch, a variable optical attenuator (VOA) is used to control the interferer power, which in turn governs the SNR. A span of 1.5 km single-mode fiber (SMF), is used to de-correlate the signal and interferer. To maximize the beating effect for studying the worse case scenario, a polarization controller (PC) is used for adjusting the signal and interferer to have the same polarization state. Interferometric noise is observed after the combined signal is detected by an InP based photodetector (PD 1). To suppress the interferometric noise, a radio frequency amplifier (RF AMP) is used to amplify the signal with interferometric noise at the output of PD 1 to provide enough voltage swing across the EAM modulation curve. The amplified signal is used to drive the EAM for interferometric noise suppression. A bias-tee is used to combine the amplified signal and a DC bias, such that the signal is biased to align with the desired point of the EAM modulation curve. A CW light from LD 2 is used as an input light source for the EAM. Due to the polarization insensitive property of EAM, no polarization controller is needed in our scheme, which is beneficial to future PIC integration. To study the performance of the interferometric noise suppressed signal, an InP based PD 2 is used for detection.

3. Experimental results and analysis

The measured nonlinear modulation curves of EAM at different wavelengths are shown in Fig. 2(a) . The zero-slope region of nonlinear modulation curve is defined as the region which only has less than 1 dB output power fluctuation compare with the zero voltage bias point, despite the amount of input power fluctuation. The upper boundaries of zero-slope region at 1550 nm and 1570 nm are labeled as point A and point B, respectively. When input signal wavelength is at 1530 nm, the nonlinear modulation curve is depicted in pink color and no zero-slope region can be obtained, meaning that 1530 nm is out of the operating region. When the input wavelength moves towards the longer wavelength, the zero-slope region expands accordantly. At 1550 nm (blue triangle curve of Fig. 2(a)), the zero-slope region is enlarged to 0.6 V (from 0 V to −0.6 V, indicated by point A) and the power variation is from −2 dBm to −3 dBm. Through power to voltage conversion (50 Ohms impedance), the power variation corresponds to voltage variation of 54 mV. Therefore, a 600 mV input amplitude fluctuation can be suppressed to 54 mV with the use of the zero-slope region in the EAM modulation curve. If input wavelength of EAM is set to 1570 nm (black square curve in Fig. 2(a)), the measured zero-slope region is increased to 0.92 V (from 0 V to −0.92 V, indicated by point B) and the power variation is from −0.5 dBm to −1.5 dBm which corresponding to voltage variation of 72 mV by using power to voltage calculation. As a result, noise induced amplitude fluctuation of 920 mV can be suppressed to only 72 mV, and noise suppression performance is further enhanced. Although the results show that there is a wavelength dependency in the scheme, mitigation of this wavelength sensitivity can be achieved through temperature control of the EAM [21]. By decreasing the EAM temperature, zero-slope region of EAM at 1550 nm can be enlarged. Thus, signal at various wavelengths can also benefit from the proposed interferometric noise suppression scheme.

 

Fig. 2 (a) Modulation curves of EAM at various wavelengths; (b) Low frequency interferometric noise with and without mitigation with EAM.

Download Full Size | PDF

The measured low frequency beating spectra are shown in Fig. 2(b) for studying the noise mitigation performance in the frequency domain and to identify the optimal bias point for the EAM. In interferometric beating, frequency of the interferometric noise is mainly located at the low frequency region in the order of MHz. To observe noise suppression with our proposed scheme, we measure the frequency spectrum of the signal after beating with and without the EAM, from 0 MHz to 100 MHz. Input signal at 1570 nm is used and the PRBS data source is tuned off to remove any data modulation, such that only beating between two CW light is being studied. Power of the CW interferer is set to result in a SNR of 12.5 dB. Without the EAM for noise suppression, frequency spectrum is measured as shown in cyan in Fig. 2(b). When EAM is used and biased at −0.5 V, the noisy high-level region of the signal is aligned to the zero-slope region. Compared with the frequency spectrum without the EAM, the noise power in the low frequency region is greatly suppressed, as shown by the pink curve in Fig. 2(b), indicating that interferometric noise has been suppressed effectively by the zero-slope region. However, when the bias voltage is increased to between −1.5 V to −2.5 V, the resultant noise floor increases because the bias voltage is too large that eventually biasing the distorted signal to the linear region of the EAM modulation curve, as shown by the blue and red curves in Fig. 2(b). The linear region has a steep slope, which enlarges amplitude fluctuation in the high-level region. When input wavelength is 1570 nm and bias voltage at −3.5 V, no linear modulation can be observed, but a relatively flat region is resulted around −3.5 V with excessive insertion loss of 23.6 dB. Thus, when bias voltage is set at −3.5 V, the excessive 23.6 dB insertion loss attenuate the signal (including noise), resulting in a downshift of the spectrum shown in Fig. 2(b).

To measure the power penalties with and without the proposed EAM based interferometric noise suppression scheme, we use a 10 Gb/s PRBS non-return to zero (NRZ) signal to investigate the noise suppression performance under various extend of interferometric noise. Wavelength of the input NRZ signal launched to the EAM is set to 1570 nm and the bias voltage is set to −2 V (linear modulation region). The peak-to-peak voltage (Vpp) of PRBS signal is amplified to 2.2 V by an RF amplifier, such that the signal can swing across the zero-slope region and the linear modulation region for both high-level noise suppression and eye-opening enlargement. Since the EAM bias voltage (−2 V) and amplified RF driving signal (about 2.2 V) are fixed, the crossing point of the degraded eye diagram is always align at the same point of the nonlinear modulation curve. Thus, the output extinction ratio of EAM is fixed to 9.3 dB and the crossing point of eye diagram is fixed at 50%. After setting the power of LD 1 to 13 dBm and the power of LD 2 to 8.5 dBm, the received power of PD 1 and PD 2 are both fixed at 0 dBm. All these systematic parameters are fixed during power penalty and BER measurements. Measured eye diagrams at different SNR before and after interferometric noise suppression are shown in Fig. 3(a) , while the corresponding power penalties curves are shown in Fig. 3(b). With a SNR at 11.5 dB, interferometric noise is severe such that a high power penalty is obtained with an unacceptable BER and a noisy eye diagram. With the use of EAM based noise mitigation scheme, interferometric noise is greatly suppressed and a widely opened eye diagram is recovered with power penalty of only 0.25 dB, as shown in the red circle curve in Fig. 3(b). By setting SNR to 12.5 dB, interferometric noise is still too high to obtain a good eye diagram and acceptable BER without any noise suppression scheme. With our proposed noise mitigation scheme, power penalty is eliminated, BER reaches 10⁻⁹, and a clear eye diagram is recovered as shown in Fig. 3(a)vi. With SNR of 13.5 and 16 dB, power penalty begin to decrease and error free transmission is achieved for both before and after interferometric noise suppression scheme.

 

Fig. 3 (a) Comparison of eye diagrams at different SNR; (b) Measured power penalties vs. SNR.

Download Full Size | PDF

To verify bit rate transparent operation of our scheme, 2.5 Gb/s, 5 Gb/s and 10 Gb/s 2³¹-1 PRBS signals are used in turn as both the signal and interfere, with SNR at 12.5 dB. Eye diagrams before interferometric noise suppression are shown in Fig. 4(a)i-iii, indicating severe degradation in signal quality. The corresponding BER measurements are shown by the hollow data points in Fig. 4(b) . Without EAM based noise suppression, error floor is observed in all the BER curves at all the data rates we measured, such that error free transmission cannot be achieved. With the use of the proposed noise suppression scheme, the receiver sensitivity of the 2.5 Gb/s, 5 Gb/s and 10 Gb/s signal are at −27.5 dBm, −25.6 dBm and −24.2 dBm, respectively. The resultant eye diagrams are now widely opened as shown in Fig. 4(a)iv-vi. The significant improvements in both eye diagrams and BER measurements for various bit-rates indicate that the proposed scheme is bit-rate transparent and is efficient in suppression interferometric noise.

 

Fig. 4 (a) Eye diagrams at different bit rates before (i)-(iii) and after (iv)-(vi) noise suppression; (b) Corresponding BER measurements for various bit-rate.

Download Full Size | PDF

4. Conclusion

In this paper, we experimentally realized a bit-rate transparent interferometric noise mitigation scheme based on the use of nonlinear modulation curve in EAM. With the use of the zero-slope region of the nonlinear modulation curve, interferometric noise is being suppressed and significant improvement in both eye diagrams and BER measurements are observed. With SNR at 12.5 dB and apply the proposal scheme, experimental results show that the improvement of power penalty at BER of 10⁻⁹ arrives at 8.5 dB, while BER performance are improved significantly at different bit rates with error floor being eliminated from the BER curves. Comparing with traditional OEO regeneration schemes, our system has simple structure and an EAM is used to re-shape the degraded input signal instead of using expensive and complex electronic D-type Flip Flop (DFF) as re-shaping device [16].