1. Introduction

With the exponentially increasing end-users’ demands for broadband services and the availability of enormous transmission capacities in core networks, the existing access networks have become critical bottlenecks for fully utilising the core network bandwidths to provide end-users with desired services [1]. To address such a challenge, great effort has been expended on exploring various techniques for enabling cost-effective, flexible and “future-proof” next generation passive optical networks (NG-PONs) [2]. Of those techniques, optical orthogonal frequency division multiplexing (OOFDM) [3,4] has attracted extensive research and development interests, since it has a number of inherent and unique advantages including, for example, potential for providing cost-effective technical solutions by fully exploiting the rapid advances in modern digital signal processing (DSP) technology, and considerable reduction in optical network complexity owing to its great resistance to dispersion impairments and efficient utilization of channel spectral characteristics. Apart from the above-mentioned advantages, OOFDM is also capable of offering, in both the frequency and time domains, hybrid dynamic allocation of broad bandwidth among various end-users.

To develop NG-PONs with the desired features, intensity modulation and direct detection (IMDD) OOFDM [5] is a very promising solution, as it is capable of offering further reductions in both the network complexity and the installation and maintenance cost without considerably compromising its flexibility and robustness. In addition, compared to other intensity modulators such as conventional external intensity modulators, directly modulated DFB lasers (DMLs) are preferable due to their many advantages, including low cost, compactness, low power consumption, relatively low driving voltage and high output power [6].

The experimental demonstration of real-time OOFDM transceivers is vital for enabling the practical realization of the great potential of OOFDM in NG-PONs. The implementation of highly complex, computationally intense and high-speed signal processing algorithms with sufficient precision and the availability of high-speed data converters with a sufficient number of quantization bits are the major challenges in experimentally implementing real-time OOFDM transceivers. It is, however, noted that real-time OOFDM transmitters [7,8] or receivers [9] have been experimentally demonstrated in external modulator-based transmission systems, where off-line DSP approaches are still adopted in the corresponding transceiver counterpart in the systems.

Our first ground-breaking real-time end-to-end OOFDM transceivers incorporating DMLs were demonstrated experimentally in April 2009 [10], since then the transceiver design has very rapidly evolved in several stages with the achieved net signal bit rates being 1.5Gb/s [10], 3Gb/s [11,12], 5.25Gb/s [13] and 6Gb/s [14,15]. The continuation of the momentum has cumulated in the current cutting-edge OOFDM transceiver design, based on which real-time 64-quadrature amplitude modulation (QAM)-encoded end-to-end OOFDM transmission is successfully demonstrated experimentally, for the first time, at a record-breaking raw signal bit rate of 11.25Gb/s and a large electrical spectral efficiency of 5.625bit/s/Hz in simple DML-based IMDD 25km single-mode fibre (SMF) systems.

Here, it is worth highlighting the key aggregated features incorporated in the present advanced OOFDM transceiver design:

It should also be pointed out, in particular, that the current transceiver design incorporates a DML in a simple IMDD SMF system without the need for in-line optical amplification and dispersion compensation. In addition, the live parameter optimisation ability of the real-time transceivers demonstrated here is an important feature which, in contrast to the partially real-time systems [7–9], allows the rapid exploration of the optimum system operating conditions to facilitate the realisation of a highly optimised transceiver design.

2. Real-time OOFDM transceiver architecture and experimental system setup

Figure 1 shows the detailed architectures of the real-time OOFDM transmitter and receiver implemented in Altera Stratix II GX FPGAs and the 25km SMF system setup, whose key parameters are listed in Table 1 . The transceiver architectures employing real-time DSP for IFFT/FFT algorithms, channel estimation, symbol synchronization, on-line performance monitoring and live parameter optimization, are similar to those reported in [14], except that extensive modifications to the present transceiver design are made in the following three aspects: 1) The subcarrier encoding in the transmitter and decoding in the receiver use 64-QAM modulation; 2) An advanced variable power loading technique in the transmitter is incorporated, which supports, in addition to the live common gain control for all subcarriers, live control of each individual subcarrier amplitude; 3) Analogue noise coupled into the digital-to-analogue converter and analogue-to-digital converter (DAC/ADC) is reduced. As detailed descriptions have already been reported of the real-time transceiver architectures in [14], an outline of the transceiver design and experimental system setup is, therefore, presented below.

 

Fig. 1 Real-time FPGA-based OOFDM transceiver architectures and experimental system setup.

Download Full Size | PDF

Table 1. Transceiver and system parameters

View Table | View all tables in this article

In the transmitter, as each subcarrier supports 6 bits per symbol, the design is adapted to provide an 84 bit wide pseudo random data sequence, of length 88,500 words, and a fixed 6 bit wide pilot data pattern, which is combined into the data sequence, to generate 90 bits for each OFDM symbol containing 15 information-bearing subcarriers in the positive frequency bins. The pilot data sequence is employed for pilot subcarrier assisted channel estimation described explicitly in [11]. Similarly, in the receiver, the design is also modified to accommodate the 90 parallel bits per symbol. The BER analyser continuously counts errors every 88,500 symbols, which corresponds to the total test pattern length of 7,965,000 bits in the present transceiver design. In the experiments, the test pattern is continuously repeated and the corresponding error count value is continuously updated, the FPGA’s embedded logic analyser displays the error count value on a PC via a JTAG connection which is updated roughly once every second. The error count is observed over a long period of time to ensure that an accurate BER is recorded.

The 8 bit digital OFDM samples are generated by the real-time DSP in the transmitter FPGA at a rate of 4GS/s. Four samples are transferred in parallel by a 32 bit wide bus running at 1GHz to an 8 bit, 4GS/s DAC for conversion to an analogue electrical signal. The analogue electrical signal power level is then optimised by a variable electrical attenuator to directly drive, in combination with an adjustable bias current, a 1550nm DFB laser. The output of the DFB laser biased at an optimum current of 36mA is −4.7dBm, which is boosted to 10dBm by a variable gain erbium doped fibre amplifier (EDFA). The optical signal is then band-pass filtered to minimise ASE noise before being injected, at an optical launch power of 7dBm, into the 25km standard SMF (SSMF) or MetroCor SMF system. It should be noted that the use of the EDFA is to vary only the optical launch power for BER performance measurements.

In the receiver, the received optical signal first passes through a variable optical attenuator for the control of the received optical power, and is then coupled into a PIN detector. The electrical output from the PIN is amplified by a fixed 20dB gain RF amplifier plus a variable electrical attenuator to allow the optimisation of the electrical signal power level in preparation for digitisation. The electrical signal is then low-pass filtered and converted to a differential signal before digitisation by an 8 bit, 4GS/s ADC. The digitised samples are then transferred to a receiver FPGA with a bus similar to that adopted in the DAC interface. Finally, the receiver FPGA performs a series of real-time functions as illustrated in Fig. 1, to recover the received data. The most computationally intense function in the receiver, which utilises the majority of the logic resources, is the FFT algorithm. It is also interesting to note that the receiver logic design is currently 82% larger than the transmitter which is mainly due to the channel estimation, symbol alignment and BER measurement functions.

The transmitter and receiver clocks are both generated from a common reference source with the DAC/ADC and FPGAs using a 2GHz and 100MHz clock, respectively. The symbol alignment in the receiver is performed manually following the procedure described in [11–14]. Here it is worth mentioning that, to perform clock information extraction and symbol alignment in the receiver of a real-time OOFDM transmission system, a novel synchronisation technique [16] has been proposed, implemented and experimentally evaluated by our research group. In particular, the proposed synchronisation technique minimizes the sampling clock offset (SCO) effect, which occurs when two independent reference clocks are involved in a single transmission system. The SCO effect may cause the received symbols to be sampled at the non-ideal sampling points, resulting in interference between different subcarriers. Detailed discussions of the operating principle and performance of the synchronisation technique is beyond the scope of the present paper, and will be reported in detail elsewhere in due course.

From the above descriptions, it can be easily derived that, at a symbol rate of 100MHz and a sample rate of 4GS/s, the 64-QAM-encoded OOFDM signal has a record-high raw signal bit rate of 11.25Gb/s and a large electrical spectral efficiency of 5.625bit/s/Hz. As a 25% cyclic prefix (2ns) is utilised here, the net signal bit rate is 9Gb/s. The use of a shorter cyclic prefix (if it can be tolerated) gives a higher net data rate. For example, a 12.5% cyclic prefix (1ns) results in a net signal bit rate of 10.125Gb/s. It should be noted that, for achieving a specific OOFDM transmission capacity, a high electrical spectral efficiency considerably relaxes the requirement on bandwidths of key components such as DACs/ADCs and DMLs.

3. Experimental results

To explore the impairments of different system elements including, digital electrical components, analogue electrical components, optical components and types of fiber on the performance of the developed real-time OOFDM transceivers, in this paper, detailed performance analyses are undertaken for four different system configurations described below with reference to Fig. 1.

3.1 Variable power loading scheme

For the aforementioned various system configurations, the associated system frequency responses are shown in Fig. 2 , which are measured on-line at the subcarrier frequencies as the effective power loss from the input of the IFFT in the transmitter to the output of the FFT in the receiver and subsequently normalised to the first subcarrier power. The case I system frequency response is not plotted in Fig. 2, since it exhibits, as expected, a flat system frequency response. Very similar to that observed in [14,15], case II has a power roll-off of approximately 8dB from the first to last subcarrier, as shown in Fig. 2. This is a direct result of the on-chip filtering in the DAC and its inherent sin(x)/x response. Whilst case III has a power roll-off of approximately11dB over the same signal spectral region. Compared to case II, the extra 3dB power roll-off in case III is due to the positive transient frequency chirp associated with the employed DML [17]. It is also very interesting to note that, in case IV, the system frequency responses of the 25km SSMF link and the 25km MetroCor link occur at different sides of the case III response, this is because chromatic dispersion of SSMF (MetroCor SMF) has an identical (opposite) sign compared to that corresponding to the DML transient frequency chirp, thus giving rise to the enhanced (reduced) chromatic dispersion effect [18].

 

Fig. 2 System frequency responses for various system configurations.

Download Full Size | PDF

The total 12dB power roll-off in case IV within the signal spectral region observed in Fig. 2 means that, if equal subcarrier power is applied in the transmitter, a large variation in the received subcarrier powers in the receiver will occur, thus leading to an unacceptably high total channel BER. However, by using variable power loading in the transmitter, the system frequency response roll-off effect can be pre-compensated. The effectiveness of such a technique is examined in Fig. 3 , where the normalised loaded subcarrier power distribution in the transmitter is presented for the various system configurations, together with the corresponding normalised received subcarrier power distributions in the receiver.

 

Fig. 3 Transmitted and received subcarrier power levels for various system configurations.

Download Full Size | PDF

To gain an in-depth understanding of the physical mechanisms underpinning the resulting loaded/received subcarrier power behaviours shown in Fig. 3, discussions are first made of the implementation of the variable power loading technique in the real-time OOFDM transceivers. The encoded electrical signal amplitude at the output of each of the 15 64-QAM encoders has the same peak value, A, to which two independent multiplication operations, denoted here as ${\stackrel{⌢}{p}}_i$ and ${\stackrel{⌢}{G}}_{com}$, are subsequently applied, here ${\stackrel{⌢}{p}}_i$represents the multiplication by an on-line controlled individual gain factor, $P_i$, of the i-th subcarrier amplitude; and ${\stackrel{⌢}{G}}_{com}$represents the multiplication by an on-line controlled common gain factor, $G_{com}$, of all subcarrier amplitudes. After these two operations, the i-th subcarrier has a peak amplitude of AP_iG_com. Clearly, the loaded subcarrier power profile is determined by P_i . The use of the common gain factor, G_com is to adjust the amplitudes of all the subcarriers simultaneously to ensure that the 32 complex signal values at the input of the IFFT are set at an optimum level. Generally speaking, for achieving the highest calculation precision, the signal level should be as high as possible. However, if the signal level exceeds a specific threshold, internal IFFT parameters can overflow their assigned ranges. To determine an optimum loaded subcarrier power profile for a given system, the initial step is to use an on-line measured system frequency response to estimate the loaded subcarrier powers required for achieving equal subcarrier powers in the receiver. Making use of such an estimated profile, the BER distribution across all the subcarriers is then measured, based on which the loaded subcarrier power profile can be finely optimised on-line to evenly distribute errors across the subcarriers and simultaneously minimise the total channel BER. As an example, a representative optimised BER distribution across all the subcarriers is shown in Fig. 4 .

 

Fig. 4 Error distribution across subcarriers for various system configurations when variable power loading is used. For comparisons, the error distribution obtained under equal power loading is also plotted for case IV with a 25km SSMF.

Download Full Size | PDF

Having described how the variable power loading technique is implemented, attention is then focussed on discussing in detail the loaded and received subcarrier power behaviours observed in Fig. 3. It should be pointed out that, for fair comparisons in Fig. 3, the loaded subcarrier power distributions for case II and III are taken to be similar to that corresponding to case IV, where the loaded subcarrier power profile is optimised for the 25km SSMF. Comparisons between Fig. 3 and Fig. 2 indicate that, for a fixed loaded subcarrier power profile, the difference in the received subcarrier power profiles for the various system configurations considered corresponds to the difference in the corresponding system frequency responses.

In Fig. 3 the loaded (received) subcarrier power profile is seen to have three distinct regions: Region 1 corresponds to the first 4 subcarriers. Over such a region, the loaded subcarrier power level is flat and the received subcarrier power level decays almost linearly. The occurrence of such a power developing trend is due to the effect of direct photon detection-induced unwanted subcarrier intermixing. This effect introduces the strongest spectral distortions to the first subcarrier and relatively weakens for other subcarriers with higher frequencies [19]. This means that the low frequency subcarriers require higher powers to mitigate the effect. Here it is also worth addressing that, for the subcarriers in Region 1, the minimum loaded power level is also determined by the relative quantisation noise, as a reduction in loaded subcarrier power level causes an increase in quantisation noise. The next four subcarriers form Region 2, where the loaded subcarrier power level linearly increases, resulting from the steep decay in the system frequency response. Over such a region, the received subcarrier power level does not vary considerably, suggesting that the subcarrier intermixing effect is negligible. Finally, all the remaining subcarriers are located in Region 3, over which the loaded subcarrier power levels are virtually flat. This is because of two reasons: 1) Region 3 corresponds to a frequency range where the system frequency response roll-off is less steep, and 2) The maximum loaded subcarrier power level is determined by the dynamic power range of the IFFT.

It can also be seen in Fig. 3 that, for the received subcarrier power profiles in all the system configurations, there exist distinguishable subcarrier power peaks centred at the middle subcarriers. This can be explained by considering the effect of imperfect subcarrier orthogonality-induced inter-channel interference (ICI) [20]. Imperfect orthogonality between different subcarriers within a symbol arises due to the quasi-periodic structure of time domain OFDM symbols. The accumulation of the ICI effect brings about the strongest spectral distortions occurring over the middle subcarriers.

From the above analysis, it is clear that, for an optimum loaded subcarrier power profile, there still exists a residual roll-off in the received subcarrier power levels, as shown in Fig. 3. However, such a residual roll-off can be tolerated. This is confirmed in Fig. 4, in which the distribution of errors across the subcarriers is plotted for case II, III and IV with a 25km SSMF. The variable power loading technique can successfully achieve an acceptable total channel BER as shown in Section 3.3, with the residual received subcarrier power roll-off as large as ~6dB (corresponding to a 12dB variation in frequency response) to give a resulting error distribution that varies by just ± 5% from the average level as shown in Fig. 4. In comparison, when equal power loading is employed for case IV with a 25km SSMF, the error distribution increases rapidly for higher subcarrier frequencies, as shown in Fig. 4, and the corresponding total channel BER is increased to an unacceptable level of 8x10⁻³.

3.2 Dependence of optimum clipping ratio on variable power loading profile

The discussions in Section 3.1 indicate that, variable power loading is capable of offering, in an adaptive manner, an optimally loaded subcarrier power profile for a specific system frequency response. This raises a very interesting open question, i.e., whether or not a variation in the loaded subcarrier power profile also alters the signal clipping characteristics. The provision of an answer to the open question is crucial, as the transmission performance of high signal modulation format-encoded OOFDM signals is very sensitive to signal clipping [13,21,22].

The signal samples at the output of the IFFT prior to clipping are signed 12 bit values, which cover the range from −2048 to 2047. The level at which the signal is clipped, C, can thus be set to a value between 0 and 2047. If the unclipped signal is S(t), the clipped signal S_clip(t) is given by

Within the dynamic amplitude range of [-C,C], the clipped signal is then quantised to cover the signed 8 bit range from −128 to + 127. The general definition of clipping ratio ξ in dB has a form of [21]

For fixed G_com, and G_IFFT, as well as the loaded subcarrier power profile optimised for the 25km SSMF link shown in Fig. 3, the measured BER against clipping level C for case II and case IV is presented in Fig. 5 under the condition of variable power loading and also equal power loading for case II. The BER variation with C for the 25km SSMF link with equal power loading is not plotted, as the minimum BER is above an acceptable level. Figure 5 shows that, for case II with equal power loading, the optimum clipping level is 1100, which, however, decreases to 1000 when variable power loading is adopted. These two optimum clipping levels of 1100 and 1000 correspond to signal clipping ratios of 12.2dB and 12.7dB, respectively, measured at the DAC output. This reveals that the optimum clipping ratio is dependent on the loaded subcarrier power profile and there is a significant variation in the clipping level C required to achieve the optimum clipping ratios.

 

Fig. 5 Variation of BER with clipping level for an analogue back-to-back configuration (case II) and a 25km SSMF link (case IV).

Download Full Size | PDF

For case IV with the 25km SSMF link, Fig. 5 shows that the optimum value of C is the same as for case II. However, the BER in case IV is more sensitive to C. This is because the optical signal also experiences additional optical noise and distortions, which make the optical signal less tolerant to both the increasing clipping distortion as C decreases below the optimum value and the increased quantisation noise as C increases above the optimum value. Experimental measurements also show that the optimum C value for case III and case IV with the MetroCor SMF is the same as for case IV with the SSMF link and case II, provided that the same loaded subcarrier power profile and G_com value are utilised.

3.3 Transmission performance of real-time 11.25Gb/s 64-QAM-encoded OOFDM signals

Making use of the optimum loaded subcarrier power profile and the corresponding optimum clipping setting, experimental measurements are undertaken of the transmission performance of 11.25Gb/s real-time 64-QAM-encoded OOFDM signals over DML-based IMDD systems without in-line optical amplification and dispersion compensation. In the experimental measurements, the 25km links of SSMF and MetroCor SMF are employed. For case III and case IV, the measured BER against received optical power is shown in Fig. 6 . In obtaining Fig. 6, the electrical gain at the receiver is adjusted as the received optical power setting varies to maintain the electrical signal amplitude at the ADC input at an optimum level. The minimum achieved BERs and the corresponding received optical powers at a BER of 1x10⁻³ are summarized in Table 2 for the different cases considered here.

 

Fig. 6 BER performance of real-time 11.25Gb/s 64-QAM-encoded OOFDM signal transmission over 25km SSMF, 25km MetroCor SMF and optical back-to-back link configurations.

Download Full Size | PDF

Table 2. 11.25Gb/s real-time OOFDM transceiver performance

View Table | View all tables in this article

For case III, the transmission performance of real-time OOFDM signals is mainly limited by additive white Gaussian noise (AWGN). Over such a channel, by comparing the transmission performance of the present 64-QAM-encoded OOFDM signal with that corresponding to the 16-QAM-encoded signal having the same spectral bandwidth [14], it can be found that, 64-QAM modulation increases the minimum received optical power required for achieving a BER of 1.0 × 10⁻³ by approximately 4dB. Such an optical power increase is well in line with the theoretical prediction [21]. In addition, as shown in Fig. 6, the measured power penalty at a BER of 1x10⁻³ for the 25km SSMF is 0.3dB, whilst the power penalty for the 25km MetroCor SMF reduces to −0.2dB. The observation of negative power penalty is also in excellent agreement with the results obtained for the 3Gb/s real-time 16-QAM-encoded real-time OOFDM transceiver design, where a negative power penalty of −0.6dB was measured for a 25km MetroCor SMF [11]. Furthermore, our numerical simulations have also verified the occurrence of positive and negative power penalties, depending upon the use of SSMFs and MetroCor SMFs, respectively, in the DML-based IMDD transmission systems. The physical origin of the observed power penalty characteristics is mainly attributed to the following two reasons: 1) the fibre chromatic dispersion-induced OOFDM phase shift cannot be preserved perfectly in the electrical domain owing to direct photon detection in the receiver. A MetroCor (SSMF) fibre has a negative (positive) dispersion parameter, which can compensate (enhance) the positive transient frequency chirp effect associated with the DML, thus leading to a reduced (enlarged) total phase shift of the received signal in the electrical domain; 2) the reduced (enlarged) phase shift also decreases (increases) the subcarrier intermixing effect upon direct detection.

To explore the factors limiting the minimum achievable BERs shown in Fig. 6, representative constellations of single subcarriers, recorded prior to performing equalization in the receiver, are presented in Fig. 7 for the various system configurations. Case I gives a zero BER as listed in Table 2, and the corresponding constellation presented in Fig. 7(a) shows very little deviation from the ideal case. Whilst in case II, the minimum BER increases to 6 × 10⁻⁵, and the corresponding constellation plotted in Fig. 7(b) shows an increase in noise and distortion due to the non-ideal sampling, analogue noise and frequency response roll-off of the DAC and ADC. In case III the minimum BER increases approximately by one order of magnitude to 8.0 × 10⁻⁴, and the corresponding constellation of the first subcarrier in Fig. 7(c) shows a significant increase in the noise content. Moreover, for case IV with two types of SMFs being employed, as seen in Fig. 6 and Table 2, the minimum BERs are very similar to that obtained in case III. Figure 7(d)–7(f) show the constellations for the 1st, 8th and 15th subcarriers for the 25km SSMF link, and Fig. 7(g)–7(i) show similar constellations for the 25km MetroCor SMF link. Comparing Fig. 7(d) and Fig. 7(g) with Fig. 7(c) indicates clearly that there is little increase in the noise content. The received constellations for the SMF fibers in Fig. 7(d)–-7(i) also clearly show the residual roll-off in subcarrier amplitude with increasing subcarrier frequency.

 

Fig. 7 Received constellations of a single subcarrier before equalisation (a) Digital back-to-back, total channel BER = 0 (b) Analogue back-to-back, total channel BER = 6.0x10⁻⁵ (c) Optical back-to-back, total channel BER = 8.0x10⁻⁴ (d,e,f) 25km SSMF, total channel BER = 8.5x10⁻⁴,(g,h,i) 25km MetroCor SMF, total channel BER = 8.8x10⁻⁴.

Download Full Size | PDF

All the aforementioned facts indicate that, in addition to the analog electrical component-induced signal distortions, DML-induced signal waveform distortions and subcarrier intermixing upon direct photon detection in the receiver are major factors limiting the minimum achievable total channel BER. To confirm the above statement, numerical simulations are performed. It is shown that, based on numerical parameters identical to those adopted in the experimental system, the simulated minimum BERs agree very well with the experimental results. However, when the DML is replaced by an ideal intensity modulator with the DML-induced positive transient frequency chirp being included, a minimum BER as low as 6.0 × 10⁻⁵ is obtainable. On the other hand, a reduction in the subcarrier intermixing effect by padding zeros at all the subcarriers between subcarrier 1 and subcarrier 7, can lower the minimum total BER to <1.0 × 10⁻⁴ for subcarriers 8 to 15 in the present DML-based IMDD experimental system. The simulation results indicate that further system optimization can still be made to provide a large BER margin for practical system implementation.

4. Effectiveness of variable power loading

From previous discussions, it is clear that variable power loading is essential to allow the successful demonstration of the 11.25Gb/s real-time 64-QAM-encoded OOFDM transceivers. In this section, numerical simulations are undertaken to explore the feasibility of using variable power loading to maximize the OOFDM signal transmission capacity. To conduct such explorations, comparisons are made between three widely used algorithms outlined below:

For fair comparisons between these three algorithms, it is worth highlighting the following three aspects: a) for a given transmission system, the total electrical signal powers generated by all the algorithms are set to be identical, and comparisons of maximum achievable transmission capacity at a BER of 1.0 × 10⁻³ are made; b) In executing algorithms 2) and 3), the signal modulation format taken on each subcarrier varies from differential binary phase shift keying (DBPSK), differential quadrature phase shift keying (DQPSK), 8-QAM to 256-QAM, and c) Any subcarrier suffering a very high transmission loss may be dropped completely if the following condition is met: for algorithm 1 only, errors are too large to achieve the required total channel BER; for algorithms 2) and 3), errors are too large to achieve the required total channel BER even when the lowest modulation format is employed.

A comprehensive theoretical OOFDM system model developed in [20] is adopted, which includes OOFDM transceivers, DMLs, SMFs and square-law photon detectors. All the device and system parameters used in the numerical simulations are identical to those adopted in the present experiments, and all other parameters that are not made known in the experiments are taken from [20].

For IMDD SSMF systems, the simulated signal capacity versus reach performance is shown in Fig. 8 for the three algorithms. It can be seen from Fig. 8 that the three algorithms can support almost identical signal capacities for SSMF links of up to 100km. In particular, at 25km SSMF transmission, variable power loading can achieve 11.25Gb/s compared to 12.25Gb/s supported by combined power and bit loading. Numerical simulations are also performed for MetroCor SMFs and a transmission performance very similar (<1% deviation) to that observed in Fig. 8 is obtained.

 

Fig. 8 Raw signal line rate versus reach for power loading, bit loading and power & bit loading. SSMFs are considered.

Download Full Size | PDF

Variable power loading is very simple and easy to implement in real-time OOFDM transceivers, whist the other two algorithms require extremely sophisticated designs to accommodate variations in both the number of bits per symbol and the selective modulation formats. Therefore, variable power loading is a cost-effective approach for optimising OOFDM transceiver performance to its maximum potential.

5. Conclusions

The fastest ever 11.25Gb/s real-time FPGA-based OOFDM transceivers utilizing 64-QAM encoding/decoding and significantly improved variable power loading on each individual subcarrier have been experimentally demonstrated, for the first time, incorporating advanced functionalities of on-line performance monitoring, live system parameter optimization and automatic channel estimation. The implemented transceivers are constructed entirely from off-the-shelf electrical and optical components. Real-time end-to-end transmission of an 11.25Gb/s 64-QAM-encoded OOFDM signal with a high electrical spectral efficiency of 5.625bit/s/Hz over 25km of standard and MetroCor SMFs has been successfully achieved with respective power penalties of 0.3dB and −0.2dB at a BER of 1.0 × 10⁻³ in a DML-based IMDD system without in-line optical amplification and chromatic dispersion compensation. The impacts of variable power loading as well as electrical and optical components on the transmission performance of the implemented transceivers have been experimentally explored in detail. In addition, numerical simulations have also shown that variable power loading is capable of maximizing the system performance to its fullest potential. By successfully breaking through the 10Gb/s barrier, this work indicates that OOFDM can be justified as a viable and practical physical layer solution for NG-PONs.

Active research activities are currently being undertaken in our research group to implement combined variable power and bit loading in real-time OOFDM transceivers to verify the theoretical predictions. In addition, to further reduce the transceiver cost, investigations are also being conducted of utilizing very cheap optical intensity modulators such as vertical cavity surface emitting lasers (VCSELs) and reflective semiconductor optical amplifiers (RSOAs) in real-time OOFDM transceivers.