Noise-suppressed Mutually Injected Fabry-perot Laser Diodes for 10-gb/s Broadcast Signal Transmission in Wdm Passive Optical Networks References and Links Broadcast Signal Transmission for Wdm-pon with Ase Injection Seeding to a Reflective Modulator, " in Noise Suppression for Fiber Radio Transmissi

We propose and demonstrate 10-Gb/s broadcast signal transmission in a wavelength-division-multiplexing passive optical network (WDM-PON) by employing a modular-type mutually injected Fabry-Perot laser diodes (MI F-P LDs) as a cost-effective multi-wavelength light source (MWS). We introduce a simple interferometric noise suppression technique with proper electrical filtering to improve transmission performance. The noise suppression doubles the number of supported subscribers with a single MI F-P LDs. Fiber to the home using a PON infrastructure, " J. Lightwave Technol. High-speed point-to-point and multiple broadcast services delivered over a WDM passive optical network, " IEEE Photon. demonstration of a cost-effective broadcast overlay for a commercial WDM-PON, " in Proceedings of the " 2.5-Gb/s broadcast signal transmission in a WDM-PON by using a mutually injected Fabry-Pérot laser diodes, " in Cost-effective colorless WDM-PON delivering up/down-stream data and broadcast services on a single wavelength using mutually injected Fabry-Perot laser diodes, " Opt. Reflective SOA-based bidirectional WDM-PON sharing optical source for up/downlink data and broadcasting transmission, " IEEE Photon. A multicast WDM-PON architecture using DPSK/NRZ orthogonal modulation, " IEEE Photon. A broadcast-capable WDM-PON based on polarization-sensitive weak-resonant-cavity Fabry–Perot laser diodes, " J. Decision threshold control method for optical receiver of WDM-PON, " J. multiple wavelength, room temperature, Raman fiber ring laser with external 19 channel, 10 GHz pulse generation in a single electro-absorption


Introduction
A wavelength-division-multiplexing passive optical network (WDM-PON) has been attracting a considerable attention for an access network to handle traffic explosions with growth of video-centric services such as high definition (HD) and ultra HD, since it can provide unlimited bandwidth, high security, and protocol transparency with a virtual point-topoint connection [1].However, it is rather inefficient to provide broadcast services which are essentially based on a point-to-multipoint service.In recent years, a number of methods has been proposed to deliver broadcast data over the WDM-PON by employing various multiplexing techniques in wavelength [2][3][4][5][6][7], radio frequency [8][9][10], and polarization [11] domains.Among them, the wavelength band multiplexing can be preferable to the costsensitive WDM-PON due to easy separation of the broadcast signal from the existing downstream signal and unnecessary modification of a remote node (RN), thanks to a cyclic property of an arrayed-waveguide grating (AWG).In this scheme, a low-cost, simple, and compact multi-wavelength light source (MWS) is essential as an economical alternative to single mode laser arrays.In the early stage, a light emitting diode (LED) was adopted for a cost-effective implementation [2].However, there are challenges related to limited launch power and high intensity noise originated from amplified spontaneous emission (ASE)-ASE beating after spectrum slicing by the AWG.Recently, a high-power superluminescent diode (SLED) was used to transmit the 1.25-Gb/s non-return-to-zero (NRZ) broadcast signal by help of a forward error correction (FEC) [3].Taking advantage of the unpolarized ASE from an Erbium-doped fiber amplifier (EDFA), the relative intensity noise (RIN) can be improved by 3 dB relative to the polarized SLED [4].A similar amount of the RIN can be also suppressed with an interferometric structure [5].Even though error-free transmission can be accomplished without the FEC, data rate is limited as ever.As the alternative MWS, a mutually injected Fabry-Pérot laser diodes (MI F-P LDs) has been proposed [6][7][8].Most recently, 2.5-Gb/s broadcast signal transmission without the FEC has been demonstrated using the low-noise characteristics of the MI F-P LDs [7].Here, a Manchester modulation format was used to locate signal spectrum within the low-noise region between two noise peaks (i.e., dc and the fundamental noise of the external cavity resonance).However, transmission of the 10-Gb/s broadcast signal was questionable, since both noise peaks were within receiver bandwidth.
In this paper, we propose and demonstrate 10-Gb/s NRZ broadcast signal transmission in the WDM-PON based on the MI F-P LDs with a simple interferometric noise suppression technique and an electrical high-pass filter (HPF).The noise suppressor (NS) located at a central office (CO) can be shared by users in a WDM-PON.To evaluate a noise suppression performance, we simulate and measure RIN spectrum as a function of effective bandwidth of the NS.Thanks to the effective noise reduction, it is able to provide 10-Gb/s broadcast services to more than doubled subscribers with better receiver sensitivities.The origin of power penalty according to channel is also investigated by measuring the RIN and dynamic extinction ratio (ER).

Proposed broadcast capable WDM-PON based on a noise-suppressed MI F-P LDs
The schematic architecture of the proposed broadcast capable WDM-PON based on the noisesuppressed MI F-P LDs is shown in Fig. 1.The broadcast signal transmitter consists of the MI F-P LDs as the MWS and the NS based on the interferometric structure (dash box) at the CO.The linearly polarized light output from the MI F-P LDs is split into two equal-intensity beams through a 50/50 optical coupler.A polarization controller (PC1 and PC2) is inserted to align the state of polarization (SOP) of the polarized beam to each principal axes of a polarization beam combiner (PBC), respectively.The optical beam of the lower branch is sent to a fiber optic delay line (ODL) to make a time delay difference (Δt d ) between two branches.The optical beams orthogonally polarization-multiplexed by the PBC are modulated with the 10-Gb/s broadcast signal by using a polarization-insensitive external modulator (EM), since two orthogonally polarization-multiplexed beams should be simultaneously modulated for an operation of the noise suppression.The MI F-P LDs spectrum for the broadcast signal (λ B1:n ) is combined with WDM-PON wavelength bands for the up/downstream signal (λ U1:n /λ D1:n ) using a WDM filter and separated by integer multiples of the cyclic AWG's free spectral range (FSR) from the WDM-PON wavelength bands.After propagating through a feeder fiber (FF), the multi-wavelength light is then spectrally sliced by the AWG at the RN and each slice is sent to each optical network unit (ONU) through a distribution fiber (DF).The broadcast signal is then received at each optical receiver (Rx B1:n ).
Since the optical signals into the receiver are orthogonally polarized, the average photocurrent (i.e., signal) at a photodiode (PD) in the Rx B is a simple summation of individual photocurrent generated by two equal-intensity optical signals.In the case when there is no time delay between two arms of the interferometer, the photocurrent fluctuations (i.e., noises) are also summation of the x-and y-polarized optical fluctuation.Thus, the signal-to-noise ratio (SNR) cannot be improved.On the other hand, if there is time delay between two arms of the NS, the photocurrent fluctuations interfere with each other constructively (in-phase) and destructively (out-of-phase) creating a series of low-noise windows.The time delay Δt d determines center frequencies of low-noise windows corresponding to n/2Δt d , where n is an odd number.The structure of the MI F-P LDs is based on optical sub-assembly as schematically shown in Fig. 2(a).It consists of two anti-reflection coated F-P LDs (front and rear facet reflectivity: 0.1% and 30%, respectively), aspheric lens, and 50/50 tap filter.The cavity length of each F-P LD was 600 µm.Thus, mode spacing was 0.6 nm (75 GHz).The external cavity length (the addition of L1 and L2 denoted in Fig. 2(a)) between two F-P LDs was about 19.3 mm (optical path length: 24.25 mm) which corresponds to the FSR of 6.1 GHz.All of these components were packaged into a single compact module [7]. Figure 2(b) shows the measured overall emission spectrum of the MI F-P LDs by an optical spectrum analyzer (OSA) with a resolution bandwidth of 0.06 nm.The bias current of each F-P LD was set at 2.3 × I th , where I th is the lasing threshold current.It can be seen that more than 18 broadcast channels were generated with 75 GHz mode spacing determined by the F-P LD. Figure 2(c) shows the measured RIN spectra for the total and single mode by using an avalanche PD (APD) with a 3 dB bandwidth of 5 GHz and an electrical spectrum analyzer (ESA) with a resolution bandwidth of 300 kHz.In the case of the single mode (red curve), a tunable optical band pass filter (TOBPF) with a 3 dB bandwidth of 0.64 nm (80 GHz) was used to select one mode.The MI F-P LDs has superior noise characteristics between noise peaks (RIN <−130 dB/Hz).Unfortunately, there exist 1/ƒ noise whose 3 dB (10 dB) bandwidth is 22 MHz (62 MHz) and periodic noise peaks with intervals of 6.1 GHz determined by a length of the external cavity.This rapid increase should be attributed to the mode partition noise (MPN).The average RIN (obtained by dividing the integrated RIN by a frequency span from 9 kHz to 7 GHz) of the total mode and single mode are −134 and −112 dB/Hz, respectively.It is clear that the high-speed baseband data transmission over 10-Gb/s is limited by not only the 1/ƒ noise but also the 1st periodic noise peak included in the bandwidth of the optical receiver.
As described in the beginning of this section, to improve the noise performance of the MI F-P LDs, we needed to introduce the NS based on the interferometric structure due to its effectiveness of the noise suppression at certain frequencies.Firstly, the noise suppression effect was investigated through a RIN simulation.Figure 2(d) shows the RIN spectra as a function of the time delay difference of two arms in the NS.Besides, the inset shows the extended views in low frequency range to observe the 1/ƒ noise suppression.It may be noted that the time delay difference determines an effective bandwidth of low-noise window.Surprisingly, the 1st periodic noise peak is completely suppressed for about 35 dB while maintaining the 1/ƒ noise, when the time delay difference was 0.246 ns (solid line).Furthermore, both the 1/ƒ noise and the 1st periodic noise are simultaneously suppressed at the time delay difference of 3.030 ns (dotted line).However, the average RIN indicates relatively small improvement of 3.4 dB, since the NS is unable to suppress the 1/ƒ noise for more than 3 dB in the low frequency region, since the constructive interference always occurs at the dc photocurrent.Thus, for the considerable average RIN improvement, we need to utilize an electrical HPF at the optical receiver.
In order to evaluate the noise suppression performance by the NS and the HPF, we conducted a RIN simulation again as a function of the time delay difference with a time step of 1 picosecond and it was then repeated according to a selected 3 dB low-cutoff frequency of the 1st order HPF.As shown in Fig. 3(a) and its inset, we can see a periodic feature (solid line) at two times 0.082 ns interval corresponding to the inverse of the 1st periodic noise frequency.As the time delay difference increases from 0 (without the NS) to 15 ns, the overall trend of the decreasing average RIN was observed, since several low-noise windows with rather small effective bandwidth suppress not only the periodic noise but also the still dominant 1/ƒ noise located above the low-cutoff frequency of the HPF.As we increase the cutoff frequency to 43.0, 83.8, and 144.7 MHz, the minimum average RIN dramatically decreases at the short time delay difference satisfying the RIN requirement of −120 dB/Hz for 10-Gb/s signal transmission [12] as shown in Figs.3(b)-3(d).Moreover, the maximum noise suppression increases as indicated in insets, since the 1st periodic noise peak dominates as we increase the cutoff frequency.The maximum suppression was 10.1 dB at the cutoff frequency of 144.7 MHz.Therefore, when the NS and the HPF are utilized to suppress the 1st periodic noise and the 1/ƒ noise, individually, the best RIN performance is achievable.
On the other hand, in the case of the spectrally sliced ASE from the EDFA considered for comparison (dash line), noise suppression ratio was fixed by 3 dB regardless of both the time delay difference and the selected low-cutoff frequency, because the ASE-ASE beat noise of the EDFA has a flat spectral shape over the entire frequency much less than the optical spectral width.Unfortunately, the HPF which has the higher low-cutoff frequency compared with the 1st signal spectrum brings about unavoidable signal distortion [13].Thus, we need trade-off between the noise suppression and the signal distortion.It may be noted that even though the suppressed RIN by the combined use of the NS and HPF gives the average RIN less than -120 dB/Hz, it would be necessary to utilize the FEC.sequence (PRBS) pattern length of 2 7 -1 from a pulse pattern generator.Here, we selected RS (255, 239) code and RS (1901, 1855) with Extended Hamming Product Code (512, 502) × (510, 500) as the 1st and 2nd generation FEC, respectively [14].The modulated optical signal propagated through a dispersion compensating fiber module (DCM) to compensate the chromatic dispersion and a 20 km standard single-mode fiber (SSMF).After spectrum slicing by the TOBPF emulating the 0.8 nm (100 GHz) channel spacing AWG with a 3 dB bandwidth of 0.64 nm (80 GHz) at the RN, it was then received at the APD whose 3 dB bandwidth was about 5 GHz.The 1st order HPFs which had various low-cutoff frequencies were simply constructed using a single capacitor.Figure 5(a) displays the measured RIN spectra w/o and w/ the NS to show performance of the noise suppression.Firstly, the RIN spectrum of the EDFA with a polarizer and the NS was measured (green curve) to set the time delay difference of 0.246 ns yielded the 2nd low noise window at 6.1 GHz for the best RIN performance as understood through the previous RIN simulation.As we expected, the 1st periodic noise peak was significantly suppressed for 31.1 dB with the NS (red curve) compared to that without the NS (blue curve).In addition, the 1/ƒ noise suppression was confirmed by the average RIN measurement as a function of the low-cutoff frequency of the used HPF as shown in Fig. 5(b).Results showed that measured average RIN values w/o and w/ NS were well matched with simulation results.It may be noted that again, the effect of HPF with the NS is remarkable compared to that without the NS reducing RIN below −120 dB/Hz.For the investigation of the transmission performance, we measured bit-error ratio (BER) curves and eye diagrams as a function of the low-cutoff frequency w/o and w/ the NS as shown in Figs.6(a) and 6(b), respectively.These measurements were conducted in back-toback (B-to-B) configuration (without the 20 km SSMF and DCF in Fig. 4).As the low-cutoff frequency increases from 10.6 to 83.8 MHz, the receiver sensitivity at the 2nd generation FEC threshold (FEC th ) of 4.6 × 10 −3 is enhanced due to the more suppressed 1/ƒ noise.As displayed in insets, we can observe the clearer eye diagram corresponding to each BER curve at −20 dBm received power.By comparing performance w/o and w/ the NS, the significant BER improvement can be achieved at the low-cutoff frequency higher than 10.6 MHz.In the case of the 83.8 MHz cutoff frequency (▲), the NS can improve more than two orders of magnitude of BER at −20 dBm received power.Furthermore, it enables the 1st generation FEC whose threshold is 1.8 × 10 −4 to be utilized.Obviously, we can also observe the clearly opened eye diagrams compared to that without the NS.However, the further increase of the cutoff frequency to 144.7 MHz () rather degrades the BER performance by the more distorted NRZ signal components appearing at 86.6 MHz interval.Thus, there exists the optimized low-cutoff frequency around 83.8 MHz.Obviously, the NS improves BER performances of all 18 channels satisfying the 2nd generation FEC th .Moreover, it is able to utilize the 1st generation FEC for more than 8 channels.Based on these BER curves, the receiver sensitivities at the FEC th were measured and simulated as shown in Fig. 8(a).In the simulation, a 5th order Bessel filter which has the most similar frequency response with the APD was designed as an electrical low-pass filter (LPF).In order to maximize the number of useable channels, we selected the 2nd generation FEC.From the results, if we allow 3 dB changes in the receiver sensitivity, the number of useable channels can be remarkably expanded from 8 w/o the NS to 18 w/ the NS ().
In addition, to figure out the origin of the power penalty according to the channel, dynamic ER and the average RIN were also measured.As shown in Fig. 8(a), ER ( × ) was proportional to the spectral shape of the MI F-P LDs because of modulation characteristics in the used EAM [15].Unfortunately, all measured ERs were less than 6 dB, since the emission spectrum of the MI F-P LDs was not-matched to designed wavelength of the EAM (~1550 nm).Thus, it is clear that if we utilize a suitable EAM, the higher ER can be achieved (We observed 10.0 dB ER at 1550 nm with a different light source).As seen in Fig. 8(b), the measured average RIN (from 9 kHz to 7 GHz), in common with ER, also showed a spectral shape dependency.To confirm whether the measured RIN originated from only the 1/ƒ noise induced by the MPN () or the addition of the MPN and the ASE beat noise () from the used OA, we calculated the maximum beat noise from a measured optical SNR (OSNR).The RIN degradation by the ASE beat noise was insignificantly less than 0.5 dB resulting from the OSNR ( × ) larger than 20 dB.Thus, dominant factors of the power penalty are both the MPN and ER.For the receiver bandwidth optimization, we conducted the simulation of the receiver sensitivity at the 2nd generation FEC th with the NS as a function of the low-and high-cutoff frequencies for the HPF and the LPF, respectively.As illustrated in Fig. 9(a), the optimum low-cutoff frequency range for the HPF was from 66.3 MHz to 93.6 MHz including the experimentally optimized cutoff frequency of 83.8 MHz.However, for the LPF, we observed that the high-cutoff frequency of 5.5 GHz () relative to the used APD bandwidth of 5 GHz () was optimized value.This can be interpreted as trade-off between filtering out the 2nd periodic noise peak at 12.2 GHz and an intersymbol interference (ISI) effect.It may be noted that the NS cannot suppress the 2nd periodic noise due to the occurrence of the constructive interference when the low-noise window is set at the 1st periodic noise frequency.Even though the improvement in the receiver sensitivity (~0.2 dB) was relatively small, we could observe the more opened eye diagram as shown in the inset.Thus, the results imply that the utilization of the higher-speed receiver can enhance the transmission performance.
To investigate the feasibility of higher-speed transmission, the bit-rate was doubled to 22-Gb/s, while keeping the MI F-P LDs which has the periodic noise peaks with intervals of 6.1 GHz.Although the NS suppressed the fundamental noise peak at 6.1 GHz, the performance degradation by the noise peak at 2nd harmonic was expected, since a more wideband receiver was needed for this case as shown in Fig. 9(b).The inset shows the degraded eye diagram () compared with 10-Gb/s eyes.However, it is possible to transmit the 20-Gb/s broadcast signal with proper HPF and LPF.The optimum high-cutoff frequency was less than a half of the bit-rate to suppress the 2nd periodic noise peak.For the low-cutoff frequency of the HPF, we expected the increase of the optimum cutoff frequency by two times, since the first spectrum of the signal (86.6 MHz) was doubled.However, it was less than two times that for the 10-Gb/s case.This can be explained by the narrow 1/f noise bandwidth of 22 MHz as shown in Fig. 2(d).Based on this simulation result, we expect much higher bit-rate transmission using a MI F-P LDs with higher the fundamental noise peak.In other words, the MI F-P LDs with more compact package is needed for higher bit-rate broadcast signal transmission.

Fig. 1 .
Fig. 1. 10-Gb/s broadcast capable WDM-PON architecture based on the MI F-P LDs with a noise suppressor employing the interferometric structure.

Fig. 2 .
Fig. 2. (a) Schematic, (b) measured optical spectrum, (c) measured RIN spectra of the MI F-P LDs, and (d) simulated RIN spectra with the NS as a function of the time delay difference and their extended views with a 500 MHz span (inset).

Fig. 3 .
Fig. 3. Simulation results of the average RIN from 9 kHz to 7 GHz as a function of time delay difference according to the 3 dB low-cutoff frequency of (a) 10.6 MHz, (b) 43.0 MHz, (c) 83.8 MHz, (d) 144.7 MHz and their extended views with 0.3 ns span (inset).

Fig. 4 .
Fig. 4. Experimental setup for transmission using the noise suppressed MI F-P LDs.

Fig. 5 .
Fig. 5. (a) Measured RIN spectra of the MI F-P LDs w/o and w/ the NS and (b) average RIN from 9 kHz to 7 GHz according to the 3-dB low-cutoff frequency of the constructed HPF.

Fig. 6 .
Fig. 6.Measured BER curves (a) w/o the NS, (b) w/ the NS as a function of low-cutoff frequency and their corresponding eye diagrams at −20 dBm received power (inset).As seen in Figs.7(a) and 7(b), we also measured the BER curves according to channel numbered in Fig.2(b) with the 20 km SSMF and the DCM.The optimized HPF with the 83.8 MHz low-cutoff frequency was used.It should be noted that the power penalty induced by dispersion after 20 km transmission was negligible compared to the back-to-back configuration by help of the DCM as shown in Fig.7(b).Obviously, the NS improves BER performances of all 18 channels satisfying the 2nd generation FEC th .Moreover, it is able to utilize the 1st generation FEC for more than 8 channels.Based on these BER curves, the receiver sensitivities at the FEC th were measured and simulated as shown in Fig.8(a).In the simulation, a 5th order Bessel filter which has the most similar frequency response with the APD was designed as an electrical low-pass filter (LPF).In order to maximize the number of useable channels, we selected the 2nd generation FEC.From the results, if we allow 3 dB changes in the receiver sensitivity, the number of useable channels can be remarkably expanded from 8 w/o the NS to 18 w/ the NS ().In addition, to figure out the origin of the power penalty according to the channel, dynamic ER and the average RIN were also measured.As shown in Fig.8(a), ER ( × ) was proportional to the spectral shape of the MI F-P LDs because of modulation characteristics in the used EAM[15].Unfortunately, all measured ERs were less than 6 dB, since the emission spectrum of the MI F-P LDs was not-matched to designed wavelength of the EAM (~1550 nm).Thus, it is clear that if we utilize a suitable EAM, the higher ER can be achieved (We observed 10.0 dB ER at 1550 nm with a different light source).As seen in Fig.8(b), the measured average RIN (from 9 kHz to 7 GHz), in common with ER, also showed a spectral shape dependency.To confirm whether the measured RIN originated from only the 1/ƒ noise induced by the MPN () or the addition of the MPN and the ASE beat noise () from the used OA, we calculated the maximum beat noise from a measured optical SNR (OSNR).The