1. Introduction

Driven by a number of popular and bandwidth-hungry broadband services such as internet video, on-line gaming, telemedicine, video surveillance and 3D television, next-generation passive optical networks (NG-PONs) are currently being intensively explored world-wide as a cost-effective, “future-proof” technical strategy to satisfy the exponentially growing end-users’ bandwidth demands [1]. To develop NG-PONs capable of providing desired features, various underlying techniques have been proposed [2–5], of which optical orthogonal frequency division multiplexing (OOFDM) [3–5] has been widely considered as one of the strongest contenders, because of its inherent and unique advantages including, for example, high transmission bit rate, great potential for cost-effective implementation, high spectral efficiency, excellent resistance to various linear channel impairments, adaption to imperfect components and systems, system scalability and flexibility, transmission performance robustness, as well as dynamic provision of hybrid bandwidth allocation in both the frequency and time domains.

To practically implement OOFDM NG-PONs, intensity modulation and direct detection (IMDD) [4–9] is a very promising solution, as it offers further reductions in both the system complexity and the installation and maintenance cost without considerably compromising the system flexibility and performance robustness. Moreover, compared to other intensity modulators such as conventional external intensity modulators, the utilization of directly modulated DFB lasers (DMLs) is preferable due to their many advantages namely low cost, compactness, relatively small driving voltage, low electrical power consumption and high output optical power [10].

In a DML-based IMDD OOFDM PON system utilizing standard single-mode fibres (SMFs), the practically achievable system performance is mainly limited by the following three factors [5,7,9,11]: i) the positive transient frequency chirp associated with the involved DML; ii) the low signal extinction ratio (ER) of the DML modulated OOFDM signal, and iii) subcarrier intermixing upon square-law photon detection in the receiver. To compensate for the DML-induced frequency chirp effect, use can be made of negative dispersion fibres [11,12] and mid-span spectral conversion [13], which, however, require significant changes to installed PON systems. For conventional signal modulation techniques, narrowband wavelength-offset optical filtering [14,15] has been proposed to improve the DML-modulated signal ER via converting unwanted chirp-induced frequency modulation (FM) into useful amplitude modulation (AM). In comparison with negative dispersion fibres and mid-span spectral conversion, the optical filtering approach is more advantageous, as it is capable of reusing the existing PON systems without introducing significant alterations to the installed fiber plants. However, as far as we are aware, detailed investigations of the influence of both narrowband wavelength-offset optical filtering and subcarrier intermixing on the DML-based IMDD OOFDM PON system performance have not been reported.

Here it is worth mentioning that, in comparison with the optical double sideband (DSB) technique, the optical single sideband (SSB) technique enables the transmission system to become more immune to fiber dispersion [9,16–18]. However, it requires a large guard band between the optical carrier and the signal sideband, and a proper suppression of the optical carrier power is also needed for improving the signal ER. This reduces the spectral efficiency. On the contrary, the spectral efficiency remains unchanged when narrowband wavelength-offset optical filtering is utilized, as the use of the guard band is not necessary.

Recently, in DML-based IMDD OOFDM PON systems without incorporating chromatic dispersion compensation and in-line optical amplification, a series of ground-breaking experimental demonstrations have been reported of end-to-end real-time OOFDM transceivers at record-high signal bit rates of up to 11.25Gb/s [5]. More recently, in similar systems, end-to-end real-time automatic OOFDM symbol synchronization has also been achieved experimentally for OOFDM signals encoded using signal modulation formats as high as 128-quadrature amplitude modulation (QAM) [19]. These end-to-end real-time, proof-of-concept experimental demonstrations are significant milestones in driving the OOFDM technique towards practical implementation. In particular, in these experimental demonstrations, great effort has also been given to selecting cost-effective, off-the-shelf electrical and optical components to minimize the transceiver cost and simultaneously maximize the transmission performance. Moreover, a number of functions necessary for a real product such as on-line performance monitoring and live-parameter optimization have also been incorporated into these demonstrations [5,19]. In addition, optical power budgets of approximately 20dB have been proved to be feasible in chromatic dispersion compensation-free real-time OOFDM systems without in-line optical amplification [5,19]. These optical power budgets may, however, not be sufficiently high to satisfy the requirements of NG-PONs.

To further improve the optical power budget of the DML-based IMDD OOFDM PON system, an in-depth understanding of some fundamental issues is of great importance. These issues include: i) what are the key physical mechanisms limiting the maximum achievable optical power budget? and ii) can the identified effects be alleviated effectively by a simple approach without complicating both the real-time OOFDM transceiver designs and the PON system architecture? Addressing these two critical challenges in detail forms the main scope of the present paper.

In this paper, extensive numerical investigations are undertaken, for the first time, of fitting our 11.25Gb/s end-to-end real-time OOFDM experimental results obtained in DML-based IMDD 25km SMF systems [5]. Excellent agreements between theoretical results and experimental measurements are observed in terms of system frequency response, error distribution across subcarriers, subcarrier constellation diagrams and total channel bit error rate (BER) performance. This verifies the validity of the comprehensive theoretical OOFDM system model adopted here. Based on the system model, the impacts of different physical mechanisms are explored on the obtainable optical power budgets. The low ER of the DML intensity modulated OOFDM signal is identified to be the predominant factor limiting the achievable optical power budget. More importantly, the use of a wavelength-offset narrowband optical bandpass filter (OBPF) in the transmitter is proposed to considerably improve the DML-modulated OOFDM signal ER and thus the optical power budget. It is shown that a 0.01nm offset OBPF having a 3dB bandwidth of 0.02nm enables a 7dB increase in optical power budget at a BER of 1 × 10⁻³.

2. Theoretical OOFDM system model

Figure 1 depicts the OOFDM transmission system considered in this paper, which consists of an OOFDM transmitter, an SMF link without involving in-line optical amplification and chromatic dispersion compensation and an OOFDM receiver. The OOFDM transmitter is composed of an electrical OFDM modem, a DML and a variable optical attenuator (VOA). The OOFDM receiver has a square-law photon detector, an electrical low-pass filter (LPF) and an electrical OFDM modem.

 

Fig. 1 Diagram of the transmission system together with the OOFDM transceiver architectures.

Download Full Size | PDF

2.1 OOFDM transceivers

In the transmitter, the generation of an electrical OFDM signal in the OFDM modem is modeled following the procedure presented in [8,9]. The major operations in the modem include data mapping using 64-QAM, inverse fast Fourier transform (IFFT), cyclic prefix insertion, OFDM symbol serialization and digital-to-analog conversion (DAC). Here it is also worth pointing out, in particular, that, to pre-compensate for the system frequency response roll-off effect induced by optical and electrical components involved in the transmission system, variable power loading [5] is applied to the subcarriers prior to the IFFT. Similar to the approach described in [5], each individual 64-QAM-encoded subcarrier power is first normalized using a common scaling factor, and then multiplied with a controllable gain factor to obtain evenly distributed errors across all the subcarriers within a symbol. Having amplified/attenuated the electrical OFDM current to an optimum level, the generated electrical OFDM signal is then combined with an optimum DC current to directly drive the DFB laser. Finally, the output OOFDM signal from the DML is fed into a VOA to fix the launched optical signal power at an optimum level.

After transmitting through the SMF link, the OOFDM signal is first attenuated by a VOA and then detected by a square-law photon detector. An electrical LPF is employed to remove the noise outside the useful OFDM signal band. The received electrical OFDM signal is finally processed in the receiver OFDM modem with an inverse procedure compared to that adopted in the transmitter OFDM modem.

2.2 DML

To simulate the nonlinear properties of the DFB-based DML, here a lumped DFB laser theoretical model developed in [6] is adopted, taking into account a wide range of nonlinear effects namely longitudinal-mode spatial hole-burning, linear and nonlinear carrier recombination as well as nonlinear gain. It is assumed that the influence of the laser linewidth on the link performance is negligible. Such an assumption holds well for the transmission systems considered here [20]. The validity of the DML model has been verified rigorously and used successfully in [6,7,9,11].

2.3 SMF, PIN detectors and LPF

A standard theoretical SSMF model successfully used in [7–9,11] is adopted here, in which the effects of loss, chromatic dispersion and optical power dependence of refractive index are taken into account. The effect of fiber nonlinearity-induced phase noise to intensity noise conversion is also included.

A square-law photon detector is employed in the receiver to detect the optical signals emerging from the transmission systems. Both shot noise and thermal noise are considered, which are simulated following the procedure similar to that presented in [21].

A LPF is considered, which has amplitude and phase responses similar to those employed in the real-time experiments [5]. The LPF introduces the system frequency response roll-off and the frequency dependent phase variation to the received electrical OFDM signal.

2.4 Simulation parameters

In numerical simulations, all parameter values that are made known in the real-time experiments [5] are treated as default constants, which are listed as following:

For simulating the DFB laser operating at 1550nm, the parameters identical to those reported in [9] are considered: a cavity length of 300µm; a cross-section area of the active region of 0.066µm²; a photon lifetime of 3.6ps; a nonlinear gain coefficient of 7.4 × 10⁻²³m³; a linewidth enhancement factor of 3; a transparency carrier density of 1.5 × 10²⁴m⁻³; a carrier lifetime of 10ns; a bimolecular recombination coefficient of 1.0 × 10⁻¹⁶m³/s; an Auger recombination coefficient of 6.5 × 10⁻⁴¹m⁶/s; a linear gain coefficient of 7.5 × 10⁻²⁰m²; an optical width (vertical) of 0.47µm; an optical width (horizontal) of 1.80µm; a confinement factor of 0.07; a group refractive index of 3.7; a phase refractive index of 3.2203 and a 38% coupling efficiency from the laser chip to the SMF.

The SMF parameters are detailed as followings: an effective area of 80µm², a dispersion parameter of 18.0ps/nm/km, a dispersion slope of 0.07ps/nm²/km, a loss of 0.20dB/km and a Kerr coefficient of 2.35 × 10⁻²⁰m²/W.

3. Result comparisons and key limiting factor identification

3.1 Descriptions of experimental setup

Extensive comparisons between theoretical and experimental results are crucial due to the following three reasons: i) rigorously verifying the theoretical OOFDM system model adopted here; ii) establishing a strong base for understanding the underlying physical mechanisms that govern the OOFDM performance, and iii) evaluating newly proposed approaches for further improving the transmission performance. The end-to-end real-time 11.25Gb/s DML-based OOFDM experiment setup is very similar to the diagram shown in Fig. 1. As full descriptions of the FPGA-based OOFDM transceiver architectures and the corresponding experimental setup can be found in [5], an outline of the experimental systems is given below:

In the OOFDM transmitter, pseudo random data is generated as a stream of 84-bit parallel words, which are combined with a fixed 6-bit pilot word used for channel estimation. The combined 90-bit word is mapped onto 15 parallel 64-QAM encoders, each of which has an independent on-line adjustable gain factor for variable power loading [5]. To generate only real-valued samples required for IMDD transmission, Hermitian symmetry is applied to the 15 generated complex values (together with a zero-valued component) and their complex conjugate counterparts. A self-developed 32-pt IFFT logic function is applied to transform the 32 frequency-domain subcarriers to 32 real-valued time-domain samples. Clipping and 8-bit quantisation are then performed, followed by the insertion of an 8 sample cyclic prefix to form a complete 40-sample OFDM symbol. The real-valued symbol is converted from signed to unsigned and subsequently rearranged for transfer over a 32-bit high-speed interface to a 4GS/s 8-bit DAC. The DAC output is adjusted by a variable electrical attenuator to produce an optimized driving signal. After combining the driving signal with an optimized DC bias current, the combined electrical OFDM signal is then employed to directly modulate a 10GHz bandwidth, 1550nm DFB laser. Finally, a variable optical attenuator is utilized to adjust the optical signal power.

At the OOFDM receiver, a 12GHz, −17dBm, linear PIN detector with a MMF pigtail directly detects the transmitted optical signal. The converted and amplified electrical signal level is optimised by a variable electrical attenuator prior to digitisation by a 4GS/s 8-bit ADC. The digitised samples are transferred to the receiver FPGA over an interface similar to that used by the DAC. Manual symbol synchronisation is applied, using the approach reported in [5], to initialize the receiver process. For each received symbol the cyclic prefix is removed before a 32-pt FFT converts the real-valued time-domain symbol into 32 frequency-domain subcarriers from which 15 data-carrying subcarriers are selected. Subsequently, pilot-data detection is performed for channel estimation. Based on the estimated channel transfer function each subcarrier is equalized before data decoding is performed to recover the transmitted data.

The real-time OOFDM transceiver design offers on-line monitoring functionalities, which can monitor the total channel BER, the BER of each individual subcarrier, the system frequency response and the BER distribution across the subcarriers. Such on-line monitoring functions are crucial for optimising not only the subcarrier power loading profile adopted in the transmitter, but also the transceiver operating parameters such as the signal clipping level, the overall RF signal power and the DFB operating conditions. In addition, the SignalTap® II embedded logic analyzer also allows the real-time observation of internal signals in the receiver FPGA such that the aforementioned BERs and system frequency responses can be continuously extracted and viewed. When measuring a BER, the errors are counted over 88,500 symbols (7,965,000 bits in total), and the measured errors are continuously updated and displayed.

3.2 Theoretical and experimental result comparisons

To perform fair comparisons with experimental results presented in [5], apart from the parameters listed in Section 2.4, a system frequency response that is responsible for the entire transmission system from the IFFT input in the transmitter to the FFT output in the receiver is also considered, whose frequency dependent roll-off profile is finely adjusted to provide the best fits with all the key experimental results presented in [5]. In numerical simulations, a total number of 1600 OFDM symbols are employed, which gives rise to 524888 samples after oversampling.

Figure 2 shows the system frequency response comparisons for optical back-to-back (BTB) and 25km SMF transmission. It can be seen in Fig. 2 that excellent agreements between the numerical simulations and experimental measurements are obtained across the whole OOFDM signal spectral region. The observed system frequency response roll-off of approximately 11dB for the optical BTB case is mainly contributed by: inherent sin(x)/x response in the DAC, the DML response and the LPF filtering effects. The SMF further lowers the system frequency response due to the IMDD fibre link frequency response.

 

Fig. 2 Comparisons of system frequency response between numerical simulations and experimental measurements for different link configurations.

Download Full Size | PDF

When equal power loading is applied across all the subcarriers, the presence of the deep system frequency response roll-off shown in Fig. 2 inevitably causes a large variation in received subcarrier power and thus the occurrence of a large number of bit errors on high frequency subcarriers, as seen in Fig. 3 . This leads to a total channel BER as large as 5.0 × 10⁻³ for a received optical power of −6.5dBm. However, by making use of variable power loading with a subcarrier power profile similar to that reported in [5], bit errors are distributed almost evenly across all the subcarriers, as shown in Fig. 3, and the total channel BER is reduced to 8.8 × 10⁻⁴. It is also interesting to note that the simulated bit error distribution and the corresponding average error level are almost identical to those measured in the experiments [5]. For the last four high frequency subcarriers that experience a similar system frequency response roll-off and suffer from the relatively strong DML frequency chirp effect, there exists a significantly large bit error difference between the cases of including and excluding variable power loading. This implies that variable power loading is capable of decreasing the DML frequency chirp-induced signal spectral distortions. This statement is confirmed once again in Fig. 5 and Fig. 8 .

 

Fig. 3 Comparisons of bit error distribution across all the subcarriers between numerical simulations and experimental measurements for equal and variable power loading in various system configurations.

Download Full Size | PDF

 

Fig. 5 Comparisons of BER versus received optical power performance for different system configurations.

Download Full Size | PDF

 

Fig. 8 Impairments of received optical power dependent total channel BER performance for the cases with zero-padding (Z-P) and without Z-P. Different SMF transmission distances are considered for both these cases.

Download Full Size | PDF

Based on the variable power loading profile adopted in simulating Fig. 3, after transmitting through a 25km SMF, the representative subcarrier constellations recorded prior to channel equalization in the receiver are presented in Fig. 4 , where the numerically simulated constellations are shown in Figs. 4(a)-4(c) and the corresponding experimentally measured constellations are given in Figs. 4(d)-4(f). Once again, the simulated results agree very well with the experimental measurements. These constellations also show the residual system frequency roll-off effect for high frequency subcarriers.

 

Fig. 4 Subcarrier constellation comparisons between numerical simulations and experimental measurements: (a)-(c) are simulated results and (d)-(f) are experimental results.

Download Full Size | PDF

Figure 5 shows comparisons of the total channel BER performance as a function of received optical power for optical BTB and 25km SMF transmission subject to variable power loading. Once more, the simulated results are similar to those measured experimentally, in particular, in the vicinity of a BER of 1.0 × 10⁻³. As expected from Fig. 3, after transmitting through the 25km SMF, a negligible optical power penalty is observed in Fig. 5, suggesting that both linear and nonlinear optical-domain distortions are very small.

3.3 Key limiting factor identification

The excellent agreements between the simulated results and experimental measurements presented in Section 3.2 confirm the validity of the theoretical OOFDM system model adopted here. Based on the OOFDM system model and utilizing variable power loading, numerical investigations are extended in this subsection to explore the impacts of various physical mechanisms on the optical power budget to identify key factors limiting the OOFDM performance. Here special attention is focused on three factors, namely OOFDM signal ER, DML frequency chirp and subcarrier intermixing upon square-law detection in the receiver. The simulated results are plotted in Figs. 6 -8.

 

Fig. 6 Impact of OOFDM signal ER on the received optical power dependent total channel BER performance for optical BTB and 25km SMF transmission. The BER performance for ideal intensity modulator (IM)-modulated OOFDM signals is also plotted for comparison.

Download Full Size | PDF

The intensity-modulated optical field emerging from a DML can be expressed as $S(t)=A(t)e^{j\phi (t)}$ with $A(t)$ and $\phi (t)$ being the amplitude and phase of the output optical field. By taking into account the time-domain “noise-like” OOFDM waveform with an approximately Gaussian probability density function, the ER of an OOFDM signal is defined explicitly in [8]. This definition is also utilized in this paper. To alter the signal ER but without considerably affecting both $A(t)$ and $\phi (t)$ , the optical field can be rewritten as

Numerical simulations are undertaken to explore the impact of the OOFDM signal ER on the received optical power dependent BER performance. The simulated results are plotted in Fig. 6 for optical BTB and 25km SMF transmission. For comparison, the BER performance of a DML modulated OOFDM signal (corresponding to C = 0 and a signal ER of 0.2dB) is also plotted, together with the BER performance of an ideal intensity modulator-modulated OOFDM signal (corresponding to a signal ER of 1.65dB). It can be seen in Fig. 6 that the received optical power required for achieving a total channel BER of 1.0 × 10⁻³ decreases significantly with increasing signal ER. For a fixed optical launch power, an ~8dB improvement in optical power budget is feasible when the signal ER is increased from 0.2dB to 1.25dB [the maximum signal ER according to Eq. (1) for the adopted DML operating conditions]. Such significant improvement is due to the fact that the effective optical signal-to-noise ratio (OSNR) of the transmitted OOFDM signal increases with signal ER for a transmission system having a fixed receiver sensitivity [8]. Moreover, Fig. 6 also shows that the optical power budget improvement is transmission link independent, suggesting that the signal ER variation does not change the signal phase for the system considered.

The influence of the DML frequency chirp on the total channel BER performance is also explored theoretically and the simulated results are shown in Fig. 7 for various transmission distances of up to 75km. In obtaining Fig. 7, C = 0 is utilized for all the cases and the cases of excluding the DML frequency chirp is realized by replacing $\phi (t)$ in Eq. (1) with a constant value. It can be seen in Fig. 7 that, for received optical powers of >-8dBm and the cases of including the DML frequency chirp, the received optical power corresponding to a total channel BER of 1.0 × 10⁻³ increases with increasing transmission distance. This is due to the combined effects of chromatic dispersion and DML frequency chirp. Whilst for received optical powers of <-8dBm with the DML frequency chirp still being present, no BER performance differences are observed for different transmission distances. This is because thermal noise associated with the PIN photon detector becomes the dominant factor limiting the system performance. The above analysis is confirmed by the BER performance obtained for the cases of excluding the DML frequency chirp, as shown in Fig. 7. For such cases, no BER differences between different transmission distances are shown over the whole received optical power range. It should also be noted in Fig. 7 that, the DML frequency chirp effect is not as significant as those presented in [11] where equal power loading is applied. This implies that variable power loading cannot only compensate for, to some extent, the system frequency response roll-off effect, as discussed in Fig. 3, but also effectively reduce the DML frequency chirp effect. Experimental verifications of the theoretical predictions are currently being undertaken and results will be reported elsewhere in due course.

 

Fig. 7 Impact of the DML frequency chirp on the received optical power dependent BER performance for different transmission distances. The signal ER of 0.2dB is adopted. FC is frequency chirp.

Download Full Size | PDF

Theoretical explorations of the impairments of subcarrier intermixing on the received optical power dependent BER performance are undertaken and the numerical results are shown in Fig. 8. In obtaining Fig. 8, for the cases where zero-padding is considered, the first 8 subcarriers close to the optical carrier frequency are zero-mapped and no changes are made to all the remaining subcarriers. Given the fact that the subcarrier intermixing effect introduces the strongest spectral distortions to the first subcarrier and becomes weak for higher frequency subcarriers, such a zero-padding approach can, therefore, partly eliminate the subcarrier intermixing effect. Figure 8 reveals that, compared to the cases of excluding zero-padding, zero-padding brings about at least 1dB reduction in received optical power at a total channel BER of 1.0 × 10⁻³ for all the transmission distances considered.

From the above analyses, it is concluded that the low ER of the DML modulated OOFDM signal is the predominant factor limiting the achievable system performance. The subcarrier intermixing effect associated with direct photon detection reduces the optical power budget by at least 1dB. Variable power loading can effectively eliminate the DML-induced frequency chirp effect. Therefore, it is greatly advantageous if a simple and effective approach capable of improving the ER of the DML intensity-modulated OOFDM signal can be identified. This forms the major task of Section 4.

4. Tunable narrowband OBPF-enabled performance improvement

In this section, numerical investigations are undertaken of the feasibility of using a tunable narrowband OBPF to considerably improve the ER of the DML intensity-modulated OOFDM signal and thus the system performance.

The signal ER improvement technique can be easily implemented by inserting a tunable narrowband OBPF between the DML and the VOA in the OOFDM transmitter illustrated in Fig. 1. For a transmission system operating at a specific optical carrier wavelength, the narrowband OBPF should be finely detuned to provide an optimum wavelength offset with respect to the optical carrier wavelength. Such wavelength offset can enhance the conversion from the unwanted FM components into the useful AM components, thus giving rise to a maximized OOFDM signal ER without significantly distorting the OOFDM signal spectrum. For simplicity, in the following numerical simulations, the tunable narrowband OBPF is assumed to have a Gaussian spectral profile and a flat phase response. As an example, Fig. 9 presents the OOFDM signal spectra before and after the OBPF, as well as the Gaussian filtering profile with its peak wavelength being detuned away from the optical carrier wavelength. Generally speaking, for a given wavelength offset, a narrow 3-dB OBPF bandwidth produces a high signal ER.

 

Fig. 9 (a) OOFDM signal spectrum before filtering; (b) OBPF spectral profile with a wavelength offset of 0.01nm (1.25GHz) with respect to the OOFDM carrier wavelength. The 3dB bandwidth is 2.5GHz (0.02nm); (c) filtered OOFDM signal spectrum.

Download Full Size | PDF

The effectiveness of the technique is examined numerically in Fig. 10 , where the total channel BER versus received optical power is plotted for various OBPF bandwidths of 0.02nm (1.44dB ER), 0.03nm (0.8dB ER) and 0.04nm (0.55dB ER). In obtaining Fig. 10, the wavelength offset is set at 0.01nm (1.25GHz) and the transmission distance is taken to be 25km. For comparisons, the BER performance of the DML intensity-modulated OOFDM signal without applying the optical filtering technique is also shown in the same figure. In addition, the filtered OOFDM signal is launched into the SMF link at a fixed optical power of 7dBm.

 

Fig. 10 Total channel BER as a function of received optical power for various bandwidth OBPFs with 0.01nm wavelength detunings.

Download Full Size | PDF

It can be seen in Fig. 10 that a reduction in OBPF bandwidth considerably decreases the received optical power required for achieving a total channel BER of 1.0 × 10⁻³. Compared to the unfiltered case, for a 0.01nm wavelength offset, an OBPF with a 0.02nm bandwidth can increase the signal ER from 0.2dB to 1.44dB, this results in an optical power budget improvement as large as 7dB. It should be pointed out that, when the same OBPFs are used for ideal intensity modulator-modulated OOFDM signals and/or DML intensity-modulated OOFDM signals with $\phi (t)$ being replaced by constant phases, in comparison with the unfiltered cases, no improvements in signal ER are observed and the BER performance becomes worse. This can be understood by considering the fact that these chirp-free signals have flat top spectra, therefore, the use of the OBPF distorts the signal spectra without increasing the signal ER.

Figure 11 explores theoretically the optimum wavelength offsets for various OBPFs with different 3-dB bandwidths of 2.5GHz, 5GHz, 10GHz and 15GHz. Figure 11(a) shows that for a fixed OBPF bandwidth, there exists an optimum wavelength offset, corresponding to which a maximum improvement in optical power budget can be obtained at a total channel BER of 1.0 × 10⁻³. Below the optimum wavelength offset value, the observed increase in optical power budget with increasing wavelength offset is mainly due to the growth of the OOFDM signal ER, as shown in Fig. 11(b); Whilst for wavelength offsets exceeding the optimum value, the BER performance degradation occurs mainly due to the strongly distorted OOFDM spectra. Figure 11(a) also shows that the optimum wavelength offset increases with increasing OBPF bandwidth, and that the maximum achievable optical power budget improvement decreases with increasing OBPF bandwidth. For instance, for a 2.5GHz bandwidth OBPF, a 7dB optical power budget improvement is obtainable, which, however, drops to 4dB when a 15GHz bandwidth OBPF is used. This originates from the fact that a wide bandwidth OBPF gives less steep spectral profile edges, thus leading to small OOFDM signal ERs, as shown in Fig. 11(b).

 

Fig. 11 (a) Optimum wavelength offsets for different 3-dB bandwidth OBPFs. (b) Obtained OOFDM signal ER as a function of frequency offset for different OBPFs. The SMF length is 25km.

Download Full Size | PDF

5. Conclusions

Based on a comprehensive theoretical OOFDM system model rigorously verified by comparing numerical results with 11.25Gb/s end-to-end real-time experimental measurements, detailed explorations have been undertaken, for the first time, of the impacts of various physical factors on the OOFDM system performance over DML-based IMDD SMF systems without in-line optical amplification and chromatic dispersion compensation. It has been shown that the low ER of the DML intensity-modulated OOFDM signal is the predominant factor limiting the maximum achievable optical power budget. The subcarrier intermixing effect associated with direct photon detection in the receiver reduces the optical power budget by at least 1dB, and variable power loading can eliminate the DML-induced frequency chirp effect. Results have also shown that, immediately after the DML in the transmitter, the use of a 0.02nm 3-dB bandwidth OBPF with a 0.01nm wavelength offset can increase the OOFDM signal ER by 1.24dB, thus leading to a 7dB optical power budget improvement at a total channel BER of 1 × 10⁻³.

The proposed tunable narrowband OBPFs have not been made commercially available yet. Our initial experimental results have indicated that use can be made of a few cascaded tunable optical filters to emulate the optical filtering effects required.