1. Introduction

Existing standards [1,2] for the currently emerging next generation passive optical networks (NG-PON1) provide a maximum transmission capacity of 10Gb/s and employ conventional signal modulation schemes, which are now reaching the limit where further increasing transmission capacity whilst maintaining cost-effectiveness is highly challenging. NG-PON1 technologies are, therefore, considered mid-term solutions for upgrading PON networks. To meet future capacity demands, NG-PON2 technologies are required which provide a long-term solution for PON evolution well beyond 10Gb/s. A critical requirement for any NG-PON2 technology is excellent cost-effectiveness due to the cost sensitive nature of access networks. This imposes heavy restrictions on NG-PON2 optical transceiver designs in terms of: i) limiting individual component costs thus implying that component bandwidths may be constrained; ii) implementing scalable optical line terminal (OLT) solutions to allow incremental capital investment; iii) avoiding unnecessary over-engineering of optical network unit (ONU) capacity in comparison to actual user requirements, and iv) maintaining highly flexible dynamic bandwidth allocation (DBA) functionality whilst employing sub-network-capacity ONUs.

Optical orthogonal frequency division multiplexing (OOFDM) is a strong candidate technology for delivering future high capacity optical networks ranging from in-building networks [3] to long-haul systems [4]. More importantly, OOFDM is seriously being considered as a promising technology for NG-PON2 applications [5–7]. The inherent characteristics of OOFDM in combination with low-cost intensity modulation and direct detection (IMDD) optical transmission also crucially offer a solution well positioned to meet the cost challenges of NG-PON2. Key OOFDM advantages include high spectral efficiency; excellent tolerance to chromatic dispersion and polarisation mode dispersion; adaptive modulation for effective utilisation of channel spectral characteristics, and also its predisposition to fine-granularity DBA. These characteristics are conducive to reducing network costs as in-line optical amplification and dispersion compensating fibers are unnecessary, and the OOFDM characteristics of high spectral efficiency and adaptability cannot only allow the use of low bandwidth components but also compensate for non-ideal and varying component characteristics.

It is technically feasible to employ a single band OOFDM (SB-OOFDM) system in a point-to-point/point-to-multipoint PON architecture such as an OOFDM multiple access (OOFDMA) PON [8]. However, it has several major drawbacks from the perspective of cost-effectiveness. To achieve high capacity transmission with a SB-OOFDM system, component bandwidths must increase in proportion to the total network capacity, thus requiring expensive broadband components. Furthermore, the OLT must support the full SB-OOFDMA network capacity from day one and each individual ONU must also support the whole network capacity to achieve fully flexible DBA, although in practice just a fraction of the available ONU capacity would be utilized for data transport at any one time.

As an example, digital-to-analogue converters (DACs) and analogue-to-digital converters (ADCs) are key components in OOFDM transceivers, their sampling speeds and bit resolutions that increase with SB-OOFDM signal capacity, can severely impact their component costs and consequently have a significant impact on the overall PON installation cost. This is particularly true when high performance DACs/ADCs are utilised in the ONUs, since typically each ONU supports one end-customer and a PON may serve hundreds of end-customers. As a broad indication of the DAC/ADC sampling speeds required to achieve >40Gb/s SB-OOFDM-based PONs in directly modulated DFB-laser based IMDD systems, experimental demonstrations at 52.8Gb/s over 20km SMF, using the non-real-time off-line signal processing approach [6], employed DACs and ADCs operating at 34GS/s and 80GS/s respectively. It is probable that the cost of such high performance devices, even when mass produced, will be prohibitive for their application in cost-sensitive PON scenarios.

However, by employing a multi-band OOFDM (MB-OOFDM) approach, where multiple OFDM signals are frequency division multiplexed (FDM) in the electrical domain, virtually all of the disadvantages associated with the SB-OOFDM approach are eliminated. Firstly, the bandwidth requirements of key components such as DACs/ADCs are dictated only by the sub-band bandwidth and are no longer dependent on total system capacity. This increased flexibility in component choice can be highly beneficial when designing to strict cost constraints. Secondly, by careful selection of sub-band transmission capacity, the ONU performance can be designed to meet the required peak user transmission capacity whilst maintaining sufficient intra-band DBA functionality. This has the important advantage of significantly reducing transceiver complexity in terms of digital signal processing (DSP) logic requirements. The reduction in DSP complexity is particularly pertinent to the fast Fourier transform (FFT) and inverse FFT (IFFT) algorithms. If it is assumed that the total number of subcarriers in the PON system is fixed, the input size of the FFT and IFFT algorithms required to process one sub-band will vary approximately in proportion to 1/n, where n is the number of sub-bands in the MB-OOFDM PON. For an IMDD system employing real-valued time domain signals, for each sub-band the relationship between the number of data-carrying subcarriers N_SC and the IFFT/FFT size N_FFT, is$N_{SC}=\left(N_{FFT}/2\right)-1$. For example, if the total number of subcarriers in a SB-OOFDM PON is 511, a 1024 point IFFT/FFT is required in each ONU. Whilst a corresponding MB-OOFDM PON employing 8 sub-bands has 63 subcarriers per sub-band, thus a 128 point IFFT/FFT is required in each sub-band. For an IFFT/FFT algorithm employing a radix-2, decimation-in-time architecture [9], this reduces the number of complex operations from 15360 for 1024 point to 1344 for 128 point, which corresponds to a considerable 91% saving in the total number of complex operations. Clearly the significant savings in DSP complexity lead to reduced costs. Furthermore the reduced DSP complexity combined with the reduced sampling speeds and logic clock speeds can result in a significant reduction in ONU power consumption. Another important issue to consider when employing ultra-high sampling speed DACs/ADCs in the SB-OOFDM transceivers is the escalation of the data bus capacity between the converters and the DSP logic circuits. These ultra high capacity data buses can be challenging to implement when trying to meet tight cost and power constraints.

In addition, by designing MB-OOFDM optical transceivers that are capable of multiplexing/demultiplexing any OFDM sub-band also brings a unique advantage of improving network operation flexibility and introducing dynamic traffic management at sub-wavelength granularity without requiring extra expensive optical devices such as reconfigurable optical add/drop multiplexers (ROADMs). Apart from that, the same MB-OOFDM transceivers capable of tuning to any OFDM sub-band can also be used in all ONUs to exploit the advantages associated with reduced equipment inventory and simplified system deployment and management. Also the same transceiver electronics employed in the ONU can be utilized in the OLT with only the optical front end specific to the OLT. Moreover, a tunable transceiver provides increased system bandwidth utilisation due to an ONU’s inter-sub-band DBA functionality. Another key advantage of the MB-OOFDM approach employing common electronics for each sub-band is a scalable OLT architecture. By utilising electronic transceiver modules which can be progressively installed in the OLT, the PON network capacity can be incrementally scaled as capacity demand grows with increasing service take up. It is also envisaged that by providing fine adjustment of the sub-band carrier frequency and power level, spectral engineering of the sub-band spectral locations and powers can dynamically optimise system performance via minimising interference from unwanted DBA-dependent spectral products such as upstream optical beat interference (OBI) [10]. Finally, sub-band tunable transceivers can also be exploited by the OLT to achieve power savings through intelligent bandwidth management. During periods of low network usage, traffic can be concentrated into a reduced number of sub-bands, the unused sub-bands allow the associated OLT transceivers to be switched to a low power stand-by state.

To explore the feasibility of OOFDM transmission based on multiple OOFDM sub-bands, recently we have reported a real-time end-to-end dual-band OOFDM transmission system operating at 17.5Gb/s over 25km standard single-mode fiber (SSMF) in an electro-absorption modulated laser (EML)-based IMDD system utilising only 4GS/s DACs/ADCs [11]. In this paper, we report, for the first time, a further increase in the dual-band OOFDM transmission capacity to 19.125GB/s over the EML-based 25km SSMF IMDD system incorporating key components identical to those utilised in [11]. Such improvement is achieved by increasing passband capacity through improvements to the RF circuitry and optimisation of relative RF sub-band power. As in [11] all on-line adaptive transceiver functions are fully utilized to maximize total transmission capacity, additionally here the system is also configured to achieve similar performance in each OFDM sub-band in terms of transmission capacity, bit error rate (BER) performance and receiver sensitivity. In addition, detailed investigations of the impact of major physical factors limiting the system transmission performance are also presented.

2. Real-time dual-band OOFDM experimental system setup

The real-time end-to-end dual-band OOFDM experimental system setup is shown in Fig. 1 with all key system parameters detailed in Table 1 . Independent digital and RF electronics are employed in the transmitters of each OFDM sub-band for the simultaneous generation of two separate OFDM sub-bands. One sub-band with a signal bit rate of 10Gb/s occupies the spectral region from 0 to 2GHz and is therefore a baseband OFDM signal. To generate the second sub-band with a signal bit rate of 9.125Gb/s, a 0-2GHz OFDM baseband signal amplitude modulates a 6.125GHz RF carrier, this generates a double sideband passband OFDM signal occupying the spectral region from 4.125 to 8.125GHz. A single receiver is employed to receive either OFDM sub-band with the inclusion of an RF down-conversion stage for the passband.

 

Fig. 1 Experimental system setup for the 19.125Gb/s real-time dual-band OOFDM system.

Download Full Size | PDF

Table 1. Transceiver and System Parameters

View Table | View all tables in this article

For both OFDM sub-band transceivers, the field programmable gate array (FPGA)-based real-time DSP is identical and is as implemented as in [12]. The transmitter real-time DSP design consists of the following major functions: pseudo-random binary test data generation, pilot-tone insertion, on-line adaptive modulation (bit and power loading) of 15 data-carrying subcarriers with modulation formats selected from 16-quaternatry amplitude modulation (QAM), 32-QAM or 64-QAM, 32-point inverse fast Fourier transform (IFFT) for generation of real-valued OFDM time-domain symbols, on-line adaptive clipping and 8-bit sample quantisation, addition of a 25% cyclic prefix and parallel-to-serial conversion for sample output to a 4GS/s, 8-bit DAC. Conversely, the functions implemented in the corresponding real-time DSP in the receiver are: serial-to-parallel conversion of OFDM signal samples from a 4GS/s, 8-bit ADC, detection of pilot-subcarriers and channel estimation, automatic symbol alignment and cyclic prefix removal, 32-point FFT algorithm for generation of complex-valued frequency domain subcarriers from the received real-valued time domain symbols, channel equalisation, on-line adaptive demodulation of 15 data-carrying subcarriers and BER analysis of total sub-band BER and individual subcarrier BERs. On-line performance monitoring of the receiver measured BERs, system frequency responses and subcarrier constellations is achieved through the FPGA’s embedded logic analyser function. This allows instant analysis of the system transmission performance which, combined with the on-line control of transmitter DSP parameters, RF gain, EML operating conditions and optical launch power, provides rapid optimisation of system performance.

The 10Gb/s baseband RF OFDM transmit signal emerging from the corresponding DAC is first amplified and its power is appropriately adjusted via a variable electrical attenuator. While the 9.125Gb/s passband RF OFDM signal is generated by first amplifying the OFDM signal from another DAC, up-converting with a 6.125GHz RF carrier via a double-balance mixer to generate a double-sideband RF signal, passband filtering to attenuate unwanted out-of-band signals and passing through a second RF gain stage with variable signal power achieved by a variable electrical attenuator. The relative OFDM RF sub-band power levels and the absolute RF dual-band power level can therefore be controlled with a resolution of 1dB. To improve the transmission performance in comparison to [11], each OFDM sub-band is passed through an RF isolator to reduce unwanted signal distortions arising from imperfect RF impedance matching. The two sub-bands with a total signal bit rate of 19.125Gb/s are finally combined in a 6dB resistive RF coupler. The RF amplifiers employed in the passband RF section have improved noise figures (NF) in comparison to those used in [11].

The dual-band OFDM signal is combined with an optimum DC bias voltage in a bias-T to drive the electro-absorption modulator of the EML and generate an intensity modulated (IM) OOFDM signal at 19.125Gb/s. The 1550nm distributed feedback (DFB) laser in the EML is driven with a bias current of 124mA. An erbium doped fiber amplifier (EDFA) at the EML output allows control of optical launch power which, after the optical filter, is optimized to 8.5dBm for launch into the 25km of SSMF.

At the receiver, a variable optical attenuator (VOA) for control of received optical power (ROP), is followed by a 10/90 optical splitter for ROP measurements and a 12.4GHz photodetector with integrated transimpedance amplifier (TIA) for optical to electrical conversion of the dual-band OOFDM signal via direct detection. To receive the baseband (passband) OFDM signal, the down-conversion section shown in Fig. 1 is omitted (included). The received baseband signal or down-converted passband signal passes to a variable RF gain section consisting of a fixed RF gain amplifier and variable electrical attenuators. This is followed by an anti-aliasing low pass filter (LPF) and balun to generate the differential signal required by the 4GS/s, 8-bit ADC. The receiver’s RF gain is manually adjusted according to ROP to maintain an optimum peak-to-peak signal level at the ADC input. In practice this can be achieved by an automatic gain circuit (AGC). The RF down-converter circuit, as shown in Fig. 1, band pass filters the dual-band OFDM signal to remove the baseband OFDM signal and, in comparison to [11], first amplifies the filtered passband OFDM signal before down conversion with a double-balance mixer. The inclusion of a low noise amplifier compensates for the mixer insertion loss and improves, in comparison to [11], the signal-to-noise ratio at the mixer output. The local oscillator (LO) signal for the mixer is derived from the same signal source used for the RF carrier in the passband transmitter. The LO is passed through a variable delay line to correctly align the phase of the LO with that of the received carrier.

In comparison to [11], the design changes implemented in the present paper are summarised as follows: i) employing passband transmit RF amplifiers with reduced NF; ii) incorporation of isolators to improve RF matching at the 6-dB coupler; iii) a low noise pass-band preamplifier inserted before the down-converting mixer; iv) filtering to improve the quality of LO, and v) optimisation of optical launch power. Furthermore, here the relative and absolute RF sub-band power levels are also fully optimised in combination with on-line optimisation of the EML operating conditions, the adaptive loading profiles and clipping levels of each sub-band. The resulting optimised subcarrier bit loading and power loading profiles employed for the system performance analysis are given in Fig. 2(a) and Fig. 2(b) respectively.

 

Fig. 2 Adaptive bit and power loading profiles a) Subcarrier bit loading allocation per OFDM sub-band, b) Relative subcarrier power loading and relative received subcarrier power levels per OFDM sub-band.

Download Full Size | PDF

3. Experimental results

3.1 System frequency response measurements

System frequency response measurements, from the transmitter IFFT input to the receiver FFT output and normalised to the first subcarrier power for each sub-band, are presented in Fig. 3(a) and (b) for various configurations of the baseband and passband OOFDM systems respectively. For the passband transmission an effective system frequency response is measured before (after) RF up(down)-conversion in the transmitter (receiver). This allows the effect of passband transmission-induced relative subcarrier attenuation to be included, but this does not reveal the true system frequency response of the RF and optical channels in the passband region of 4.125-8.125GHz. In Fig. 3, the effect of the DAC and ADC on the system frequency response is measured by connecting the DAC directly to the LPF before the ADC via a 3dB attenuator in an analogue back-to-back configuration. For this case Fig. 3 reveals a maximum roll-off of ~7.5dB for both OOFDM transceivers. This roll-off is known to be mainly caused by the characteristic sin(x)/x response of zero-order hold DACs and the on-chip filtering of the DAC outputs [11,12].

 

Fig. 3 Measured system frequency responses normalised to the first subcarrier power for analogue back-to-back (including and excluding the RF section), optical back-to-back and 25km SSMF. Also shown are the effects on the system frequency response due to the EML + PIN only and due to the SSMF only. a) Baseband system frequency responses, b) Passband system frequency responses.

Download Full Size | PDF

The contribution of the RF electronics to the system frequency responses is explored by configuring an analogue dual-band back-to-back configuration, where the section in Fig. 1, from the bias-T input to the PIN/TIA output is replaced with a wideband 10dB attenuator. Figure 3(a) and 3(b) show that the corresponding baseband and passband systems have large maximum roll-offs of −20.8dB and −23.2dB respectively. It is interesting to note here that, even with these large roll-off values, the aforementioned analogue back-to-back case including the RF sections can exploit adaptive bit and power loading to achieve a real-time dual-band total line rate of 20.375Gb/s with ~10Gb/s in each sub-band and an average BER of 2.45 × 10⁻⁴ (3.4 × 10⁻⁴) in the baseband (passband). The signal line rate is 90% of the maximum possible line rate of 22.5 Gb/s when all subcarriers employ 64-QAM modulation under the transceiver and system parameters listed in Table 1. Figure 3 also shows system frequency responses for each sub-band for the cases of i)an optical back-to-back configuration, where the output of the optical filter in the transmitter is connected directly to the VOA input in the receiver, and ii) the complete transmission system including the 25km SSMF as shown in Fig. 1. For both of these cases there is virtually no change in the maximum system frequency response roll-off. By calculating the difference between the appropriate system frequency responses, the effective change in the normalised system frequency response due to the EML and PIN/TIA only and for the 25kmSSMF only are determined, which are also shown in Fig. 3. Each of these two cases imposes a system frequency response variation of < ± 1dB across the entire frequency response range for both sub-bands. It should also be noted that for the passband measurements these are the effective system frequency responses and do not indicate the real EML, PIN or 25km SSMF responses over the passband region of 4.125-8.125 GHz.

3.2 Dual-band OFDM RF signal characteristics

The dual-band OFDM electrical signal spectra are shown for the transmit signal at the output of the RF coupler in Fig. 4(a) , for the received signal at the output of the PIN/TIA after transmission over the 25km SSMF in Fig. 4(b) and at the output of the PIN/TIA after optical back-to-back transmission in Fig. 4(c). Their measured signal levels are given in Table 2 .

 

Fig. 4 Spectra of dual-band OOFDM RF signals. (a) Transmit signal at output of RF coupler, total RF power −4.25dBm [50Ω]. Received signal at output of PIN after (b) transmission over 25km SSMF, ROP −3.8dBm, total RF power −12.6dBm [50Ω], (c) optical back-to-back transmission, ROP −5.7dBm, total RF power −15.2dBm [50Ω].

Download Full Size | PDF

Table 2. Dual-band OFDM RF Signal Levels

View Table | View all tables in this article

In order to achieve similar performance for each sub-band, it is necessary to employ a passband to baseband RF power ratio of 3.13dB at the transmitter. After transmission through the 25km SSMF, the passband to baseband RF power ratio has fallen to −1.63dB, indicating that the passband OOFDM signal suffers more attenuation than the baseband OOFDM signal. This is consistent with the limited EML modulation bandwidth and IMDD-induced frequency fading effects. Both effects cause an increase in optical signal spectrum roll-off with increasing frequency in the directly detected electrical signal. Evidence of this frequency fading effect is observed if a comparison is made between the dual-band RF spectrum at the PIN/TIA output for the case of i) optical back-to-back transmission as shown in Fig. 4(c), where a symmetrical double sideband spectrum is observed, and ii) entire 25km SSMF transmission system, as shown in Fig. 4(b), where an asymmetrical double sideband spectrum is seen with reduced power in the upper sideband. This effect however does not contribute to the effective system frequency response roll-off of the passband OFDM sub-band, as the mirror-imaged upper and lower RF sidebands result in opposing roll-off when converted to baseband frequencies. Evidence of the above statement is that the measured normalised system frequency responses for each sub-band are very similar, as shown in Fig. 3(a) and 3(b). It should be noted that the absolute system frequency responses for each sub-band would show a higher absolute attenuation in the passband due to the measured decrease in passband to baseband RF power ratio.

Table 2 also indicates the clipping ratios of the baseband and passband OFDM signals at the transmitter and receiver, where clipping ratio is as defined in [13]. For the passband signal the clipping ratio is of the order of 3dB higher than the typical optimum clipping ratio of a baseband OFDM signal of 12 – 14dB [13–15]. This is due to the fact that modulation by the RF carrier reduces the average power by 3dB although its peak power remains unchanged. The clipping ratio of the passband signal measured at the output of the DAC is 13dB which agrees with the expected difference of 3dB between the un-modulated and modulated signals.

3.3 Transmission performance of real-time 19.125Gb/s dual-band OOFDM system

In this Section, extensive experimental measurements are undertaken of the transmission performance of the simultaneously transmitted 19.125Gb/s dual-band OOFDM signals whilst employing the optimised bit and power loading profiles given in Fig. 2, the optimised EML operating conditions specified in Table 1 and the optimised RF signal levels listed in Table 2. It should be highlighted that these optimum system settings cannot only support the best transmission performance of each OFDM sub-band in terms of signal line rate, sub-band BER and ROP, but also attain similar performance in each OFDM sub-band. In the experimental measurements, the baseband and passband transmitter configurations and the EML operating conditions are always fixed whilst the BER performance is measured for each sub-band separately, and only changes to the real-time receiver electrical gain, as previously discussed in Section 2, are made during the measurements. The measured sub-band BER against ROP is shown in Fig. 5 for each sub-band for various system configurations including the optical back-to-back configuration and the 25km SSMF configuration. In addition, Fig. 6 shows the corresponding subcarrier BER distributions across all subcarriers for both the baseband and passband OFDM sub-bands after transmission over the 25km SSMF. It should be pointed out that, in practice, the subcarrier BER variation range shown in Fig. 6 can be further decreased by the appropriate use of the DBA function through intelligent subcarrier BER-aware bit and power allocation.

 

Fig. 5 Baseband and passband BER performance as a function of received optical power for 19.125Gb/s optical back-to-back and 25km SSMF. Passband BER performance over 25km MetroCor^TM is also shown.

Download Full Size | PDF

 

Fig. 6 BER distribution across all subcarriers for baseband and passband OFDM signals after transmission through 25km SSMF.

Download Full Size | PDF

Example received subcarrier constellations before channel equalisation are shown in Fig. 7 and Fig. 8 for the baseband and passband OOFDM signal respectively. The constellations are measured at their minimum sub-band BERs after transmission over the 25km SSMF. The large system frequency response roll-off shown in Fig. 3 is clearly evident from the difference in constellation amplitude from the 1st to 15th subcarriers.

 

Fig. 7 Example received subcarrier constellations before channel equalisation for the baseband OOFDM sub-band.

Download Full Size | PDF

 

Fig. 8 Example received subcarrier constellations before channel equalisation for the passband OOFDM sub-band.

Download Full Size | PDF

As expected from previous discussions and a direct result of adaptive bit and power loading, Fig. 5 clearly shows very similar BER performances between two OFDM sub-bands after 25km SSMF transmission in the EML-based IMDD system. Considering the system parameters listed in Table 1 and ROP levels in Fig. 5, it is easy to find that the achieved 19.125Gb/s over 25km SSMF OOFDM transmission system has an optical power budget of 13.5dB. It is expected that the utilization of an optical amplifier before the PIN or other techniques such as wavelength-offset optical filtering [16] can considerably increase the system power budget. Figure 5 also shows that, in comparison with their optical back-to-back configurations, the optical power penalties for each OFDM sub-band at a forward error correction (FEC) BER limit of 2.3 × 10⁻³ are 0.25dB and 4dB for the baseband and passband respectively, the selected common FEC limit being well within the capability of modern FEC algorithms [17]. The measured significantly larger optical power penalty for the passband OFDM sub-band can be explained by the fact that signal degradation resulting from the EML frequency chirp-enhanced chromatic dispersion (CD) effect increases in proportion to optical signal bandwidth. For the present system, the OFDM baseband signal results in a double sideband OOFDM signal with a spectral range of 4GHz in the optical domain, whereas the passband OFDM signal results in a double sideband OOFDM signal with a large spectral range of ~16GHz. The EML frequency chirp-enhanced CD effect is therefore much more significant for the passband OFDM sub-band.

To experimentally verify the aforementioned physical cause of the passband power penalty, under the condition that the 25km SSMF is simply replaced by a 25km MetroCor^TM fiber and all other transceiver and system parameters are preserved, BER measurements of the passband signal in the dual-band OOFDM system are conducted, and an almost zero optical power penalty at the FEC limit occurs, as shown in Fig. 5. It is well known that the MetroCor^TM fiber with a negative dispersion parameter of −7.6 ps/(nm·km), compared to a positive dispersion parameter of 18ps/(nm·km) for the SSMF, can considerably decrease the positive EML frequency chirp effect, thus giving rise to less severer signal distortions for the MetroCor^TM fiber case. Therefore, it is clear that the EML frequency chirp-enhanced CD effect is a dominant mechanism responsible for the 4dB passband power penalty observed in Fig. 5. The above analysis also suggests that an improved passband transmission performance is feasible if single optical sideband MB-OOFDM transmission is employed in the system.

It can also be seen in Fig. 5 that, for the optical back-to-back case, there exists an approximately 4dB difference in ROP at the FEC limit between the baseband signal and pass band signal. This can be explained by considering the mechanisms described below: as already mentioned previously, the transceiver and system parameter settings are chosen for the entire EML-based 25km SSMF IMDD systems, where the effects of both EML frequency chirp and CD are present. Together with the RF sub-band power optimisation, on-line adaptive subcarrier bit and power loading is then applied to ensure that the maximum signal transmission capacity of each sub-band at the FEC limit is achieved at a minimised ROP. Whilst for the optical back-to-back case where the effects of EML frequency chirp and CD are negligible, the aforementioned performance balance does not hold any more between the sub-band signal bit rate and the corresponding ROP. As a direct result of an intrinsic trade-off between signal line rate and minimum ROP required for a specific BER, adaptive loading thus enables the baseband OFDM signal of 10Gb/s to have a ROP higher than that corresponding to the passband signal of 9.125Gb/s. Our recent experimental work [14] supports this observation where 1Gb/s increase in signal line rate corresponds to a ~3dB increase in ROP.

To explore the impact of inter-sub-band interference on the ROP at the adopted FEC limit after 25km SSMF transmission, the BER performance of each sub-band is measured with the opposing sub-band switched off in the DSP but with all components still connected and powered. Experimental measurements show that the BER performance of the passband OOFDM transmission is almost unaffected by the presence of the baseband OOFDM signal. Whilst the ROP of the baseband OOFDM transmission is improved by approximately 1.2dB when the passband is absent, as shown in Fig. 5. This indicates that, for the present system, the inter-sub-band interference gives rise to an optical power penalty of about 1.2dB. Such a power penalty is in close agreement with that reported in [11]. The physical reason underpinning the above-mentioned behaviours is due to unwanted intermixing frequency products generated upon square-law photon detection in the receiver [18]. According to numerical simulations reported in [18], within the present 8GHz dual-band OOFDM signal spectral region, the unwanted intermixing frequency products occur mainly in two areas: i) a 0-2GHz spectral region with a noise peak located at zero frequency; and ii) a 2-4GHz spectral region with a noise peak located at 3GHz. Clearly, the first spectral region overlaps with the baseband signal, thus resulting in the observed 1.2dB power penalty; whilst the second spectral region can be distinguished from the passband signal spectrum by RF down conversion in the receiver, thus causing the independence of the passband ROP on the baseband signal existence.

Finally, to explore the effect of RF impedance matching on the transmission performance of each individual sub-band, each sub-band is in turn replaced with a 50Ω termination at the RF coupler in the transmitter. Experimental results show that the performance degradation due to impedance mismatching is negligible due to the use of the RF isolators at the input ports of the RF coupler.

4. Conclusions

The fastest ever 19.125Gb/s real-time end-to-end OOFDM transmission, in an EML-based, 25km SSMF, IMDD system has been experimentally demonstrated, for the first time, by simultaneously transmitting two independently generated/detected OOFDM sub-bands. A baseband OFDM sub-band, with an RF spectrum form 0-2GHz supporting 10Gb/s, and a passband OFDM sub-band, with an RF spectrum from 4.125GHz to 8.125GHz supporting 9.125Gb/s, are electrically multiplexed in the frequency domain to intensity modulate a single 1550nm optical carrier via a 10GHz EML.

The performance of the two OOFDM sub-bands is optimised and balanced by exploiting the highly effective on-line performance monitoring and adaptive transceiver features. In particular the on-line adaptive subcarrier bit/power loading and RF sub-band transmit power optimisation allow improvements and equalisation of transmission capacity, BER performance and receiver sensitivity between the sub-bands. As a direct result, the achieved 19.125Gb/s over 25km SSMF OOFDM transmission system has an optical power budget of 13.5dB, and shows almost identical BER performances for both OOFDM sub-bands. In addition, experimental investigations have also indicated that the maximum achievable transmission capacity of the present system is mainly determined by the EML frequency chirp-enhanced CD effect. The passband BER performance is not affected by the two sub-band-induced intermixing effect, which does, however, give a 1.2dB optical power penalty to the baseband signal transmission.

Compared to the corresponding 11.25Gb/s SB-OOFDM system [12,15], the dual-band OOFDM system presented in this paper increases the total system transmission capacity by a factor of 1.7 without increasing either the bandwidths of key components such as DACs and ADCs or the FPGA-based DSP clock speeds. This work thus demonstrates that DAC/ADC bandwidths do not need to increase in proportion to the total OOFDM system capacity. It is also believed that by employing techniques such as I-Q modulation for increased spectral efficiency, further optimization of sub-band bandwidth and single optical sideband transmission, MB-OOFDM may have the potential to provide a cost-effective, low-power and highly-flexible solution for >40Gb/s NG-PON2 systems, without requiring expensive multi-10GS/s broadband converters. In addition, the MB-OOFDM approach can also offer greater flexibility in DAC/ADC bandwidth selection to allow compliance with NG-PON2 cost constraints.