Over the past few years, traffic patterns in access networks have been propelled to the broadband evolution from voice- and text-based services to video-based interactive and multimedia services due to the continuing remarkable growth in the Internet. By the estimations in [1] and [2], 50% of the revenues of large telephone companies will be based on video services in 2010. In addition to the high-speed, symmetric, and guaranteed bandwidth demands for future video services, the next-generation access networks are driving the needs for the convergence of wired and wireless services to offer end users greater choice, convenience, and variety in an efficient way [3]. This scenario will require the delivery of voice, data, and video services with mobility feature to serve the fixed and mobile users in a unified networking platform.
The most widely deployed access networks based on twisted-pair copper cable are approaching their upper limit of a bandwidth–distance product (10
$\hbox{Mb/s} \cdot \hbox{km}$) [4]. For a distance under 1.5 km, the asymmetric digital-subscriber-loop technology can deliver about 8 Mb/s, whereas the latest very high speed digital-subscriber-loop technology can deliver up to 26 Mb/s for distances under 1 km. Another dominant access medium is the hybrid of fiber and coaxial cables (HFC). The guaranteed bandwidths per subscriber are only 2.8–5.6 Mb/s for the downstream and 0.15–0.3 Mb/s for the upstream due to the bandwidth shared within a cell (500–1000 subscribers) [5]. It is quite obvious that the two technologies cannot meet the bandwidth demands for the future video services and will have limited lifetimes [6]. With the trend to deploy optical fiber deeper and deeper and with the development of highly recognized passive optical network (PON), it is expected that time-division-multiplexing (TDM)-PON and wavelength-division-multiplexing (WDM)-PON will be the most promising candidates for the next-generation access systems. A TDM-PON [7], including an asynchronous-transfer-mode and broadband PON (A/BPON), an Ethernet PON (EPON), and a Gigabit PON (GPON), shares a single transmission channel to be a satisfactory solution for the near-future bandwidth needs. A WDM-PON [8] provides a point-to-point optical connectivity to multiple end users through a single feeder fiber and will be a future-proof access network.
On the other hand, broadband wireless access (BWA) [9] technologies have surged in popularity because they are more convenient, scalable, and flexible for roaming connections. The most widely used technologies are the local multipoint distribution service and the multichannel multipoint distribution service [10]. World Interoperability for Microwave Access is another BWA technology being standardized by IEEE 802.16 [11]. These technologies can provide wireless connection but are severely constrained by the required bandwidth particularly for the video-centric services with high-definition TV (HDTV) quality.
To make full use of the huge bandwidth offered by fiber and the mobile feature presented via a wireless scheme, the integration of wireless and optical networks is a potential solution for increasing the capacity and mobility as well as decreasing the costs in the access network. Thus, the radio-over-fiber (ROF)-based optical–wireless networks came into play and have emerged as an affordable alternative solution in environments such as conference centers, airports, hotels, shopping malls, and ultimately homes and small offices [12]. It has been expected that the millimeter-wave (mm-wave) bands would be utilized to meet the requirement for higher signal bandwidth and to overcome the frequency congestion in the future optical–wireless access networks [13]. In this situation, it is necessary to minimize the cost of the base station (BS) and to shift the system complexity and expensive devices to the central office (CO) because the BS picocell has small coverage due to high atmospheric attenuation in the mm-wave band. Fig. 1 shows the generic architecture of the optical–wireless network and the enabling technologies that we will discuss in this paper. At the CO, the optical mm-wave signals are generated and mixed by using cost-efficient all-optical approaches. Optical networking technologies are leveraged to reach the longer transmission distance over a single-mode fiber (SMF) and to integrate with the WDM-PON between the BS and CO. The BS design goal is to make the full-duplex operation and the dual-service provision possible with a simple and low-cost way, which typically involves the choice of optical upconversion technologies for the overall architecture planning.
This paper is organized as follows. The optical mm-wave generation, upconversion, and transmission in a downlink direction are presented in Section II. The full-duplex system based on wavelength reuse by using a centralized optical source in an uplink direction is described in Section III. Finally, Section IV demonstrates the testbed implementation to simultaneously deliver dual services over the fiber and wireless link.
SECTION II
OPTICAL MM-WAVE UP CONVERSION FOR DOWNSTREAM
Generating mm-wave frequencies using electrical devices is challenging due to the electronic bottleneck. The most promising solution is to use optical means. Over the past few years, many groups have conducted research to develop optical mm-wave generation, upconversion, and transmission techniques. Traditionally, three different methods exist for the generation of mm-wave signals over optical links with intensity modulation: direct intensity modulation, external modulation, and remote heterodyning. Although direct modulation [14], [15] is by far the simplest, due to the limited modulation bandwidth of the laser, this is not suitable for mm-wave bands. The configuration of external modulation is simple, but it has some disadvantages that limit its implementation at mm wave because of high cost for driving signals and low dispersion tolerance. For optical heterodyning technique, two or more optical signals are simultaneously transmitted and are heterodyned in the receiver. However, it requires either a precisely biased electrooptic modulator or a complex laser to reduce the severe phase noise, which greatly adds to the cost and complexity of the system [16].
Recently, several approaches for upconversion of radio signals have been reported. These techniques, based on the nonlinear effects in waveguide device, exhibit low conversion efficiency and need very high input optical power [17]. The scheme, based on cross-gain modulation in semiconductor optical amplifier (SOA) [18], requires a large input power to saturate the gain of the SOA. The scheme, by using cross-phase modulation (XPM) in the SOA Mach–Zehnder (M–Z) interferometer [19], loosens the requirement for the input power; however, the complicated conversion structure and the nonlinear crosstalk among multiple channels are the major hurdles to greatly limit the signal quantity of wireless end users. In this paper, we will demonstrate some new all-optical upconversion schemes in highly nonlinear dispersion-shifted fiber (HNL-DSF), the electroabsorption modulator (EAM), and the external intensity and phase modulator (PM). By using these all-optical upconversion schemes, we will present the integration between optical–wireless systems and WDM PON networks at 8, 16, and all the way to 32 wavelengths.
A. Four-Wave Mixing (fwm) IN HNL-DSF
FWM is one of the important nonlinear effects to generate new waves or parametric amplification, specifically for effectively ultrafast response by using an HNL-DSF. Relying on the third-order electric susceptibility and beating process with the frequency phase-matching condition when the light of three or more different wavelengths is launched into the HNL-DSF, it is possible to realize terahertz all-optical mixing or upconversion, as shown in Fig. 2. Two pumping waves
$\omega_{\rm OCS} - \omega_{\rm RF}$ and
$\omega_{\rm OCS} + \omega_{\rm RF}$ are generated by using optical carrier suppression (OCS) technologies, where
$\omega_{\rm RF}$ is the radio-frequency (RF) clock. The converted signal
$\omega_{\rm con}$ may then be determined by
TeX Source$$\omega_{\rm con} = (\omega_{\rm OCS} - \omega_{\rm RF}) + (\omega_{ \rm OCS} + \omega_{\rm RF}) - \omega_{\rm s} = 2\omega_{\rm OCS} - \omega_{\rm s}\eqno{\hbox{(1)}}$$ where
$\omega_{\rm s}$ is the input signal light. The two pumping waves are set to coincide with the fiber zero-dispersion wavelength to efficiently generate beating grating in the HNL-DSF, which will modulate the input signal
$\omega_{\rm s}$ to produce two sideband waves with the frequency shift
TeX Source$$\omega_{\rm con} \pm \left\vert (\omega_{\rm OCS} - \omega_{\rm RF}) - (\omega_{\rm OCS} + \omega_{\rm RF})\right\vert = \omega_{\rm con}\pm 2\omega_{\rm RF}.\eqno{\hbox{(2)}}$$ The power is determined by the third-order nonlinearity susceptibility
$\chi^{(3)}$ and the fiber parameters [20]. Since FWM is independent of the signal bit rate and coding format, it can be used to realize simultaneous upconversion of multiple WDM signals and can easily be integrated with the passive optical access networks. Because of high Raman gain in HNL-DSF, the conversion efficiency of the FWM process is enhanced [21]. Table I shows the transmission characteristics of the HNL-DSF used in these experiments. The fiber was made by OFS Denmark and has a nonlinear coefficient
$\gamma$ of 10
$\hbox{W}^{-1}\cdot\hbox{km}^{-1}$ with the length of 1 km.
The experimental setup is shown in Fig. 3. Eight WDM signals with a channel spacing of 3.2 nm are generated from eight DFB lasers and modulated by a
$\hbox{LiNbO}_{3}$ M–Z modulator driven by 2.5-Gb/s baseband signals. The modulated WDM signals are then transmitted over a 10-km SMF to decorrelate the signals before the upconversion. The OCS modulation scheme is used to generate a 40-GHz optical local-oscillator (LO) signal as the two FWM pumping signals. It is realized by driving a dual-arm M–Z modulator biased at
$V_{\pi}$ with two complementary 20-GHz RF sinusoidal waveforms. The spectrum and waveform of the 40-GHz LO signal are shown in Fig. 3 as insets (i) and (ii), respectively. The OCS ratio (the ratio of the optical power in the first-order sideband to that of the optical carrier) is larger than 20 dB. Both the optical LO signal and the WDM signals are amplified by erbium-doped fiber amplifiers (EDFAs) before they are launched into the HNL-DSF. The signal power is 4 dBm/channel, whereas the pump power is 14 dBm. To increase the FWM conversion efficiency and to broaden the conversion bandwidth, we use the backward Raman-pump laser diodes (LDs) at 1440 nm with a total power of 750 mW. Two cascaded C-/L-band filters are used to filter out the LO signal and input WDM signals. Insets (iii) and (iv) in Fig. 3 show the optical spectrum and the optical eye diagrams, respectively, at the output of the HNL-DSF. The optical signal-to-noise ratio (OSNR) of all upconverted channels is larger than 20 dB at a noise bandwidth of 0.1 nm. Since the Raman pump is optimized for C-band signal amplification, the upconverted signal with a longer wavelength has a smaller gain. The Raman gains for the longest and shortest wavelengths are 8.5 and 4 dB, respectively. The power of each upconverted channel is larger than
$-$24 dBm; it can well be amplified by an L-band EDFA. A tunable optical filter (TOF) with a 3-dB bandwidth of 0.5 nm is used to extract the desired channel. The extracted channel is then transmitted over a 5-km SMF before being optically to electrically converted by a photodetector (PIN) with a 3-dB bandwidth of 60 GHz. The converted electrical signals are amplified by a narrow-band electrical amplifier (EA) with a bandwidth of 10 GHz centered at 40 GHz. An electrical LO signal at 40 GHz is generated by using a frequency multiplier from 10 to 40 GHz. We use the LO signal and a mixer to downconvert the electrical mm-wave signal.
The power penalty is around 2.5 dB after a 12-km transmission. We then offset the filter center frequency by 30 GHz to realize the vestigial sideband (VSB) filtering, as shown in Fig. 4(a). The downconverted electrical eye diagrams are also shown in Fig. 4(a). The power penalty at a bit error rate (BER) of
$10^{-10}$ is only
$\sim$0.5 dB after the transmission over a 17-km SMF with the VSB filtering. It is also observed that the difference of receiver sensitivity among all the WDM channels at a BER of
$10^{-10}$ is less than 0.5 dB. As a comparison, the BER curve is measured when only the longest channel is present, as shown in Fig. 4(b). The receiver sensitivity at
$10^{-10}$ is degraded by 2 dB in the WDM case as compared to the single-channel situation, which comes from the nonlinear crosstalk among different channels in the FWM process and the reduced OSNR in the WDM case.
The advantage for the FWM-based all-optical upconversion is that FWM is transparent to the signal bit-rate and modulation formats, which is easy to realize upconversion for the different WDM signals. In addition, due to the ultrafast nonlinear response of the fiber, it is possible to realize terahertz waveform all-optical mixing or upconversion. Meanwhile, the HNL-DSF has a higher Raman gain compared with the standard DSF [22], which can be utilized to assist the FWM process.
B. XPM in HNL-DSF
The XPM in the nonlinear optical loop mirror (NOLM) [23] and straight pass [24] in HNL-DSF is also possible to realize all-optical upconversion for more wavelength channels without any interference- and saturation-effect limitation. A conceptual diagram by using an OCS as the control signal is shown in Fig. 5(a) and (b). Regarding the NOLM architecture, the symmetry between the counter-propagation paths of the probe signal is broken due to the XPM-induced phase shift by the control signal. Therefore, the NOLM is changed into the mixer between the probe and control signals. For the straight-line structure, while propagating in the HNL-DSF, the intensity of the control signal modulates the electric field of the probe data signal via an XPM effect. Therefore, the RF sinusoidal clock is imposed onto the probe signal as the sidebands.
In the experiment, we use the identical HNL-DSF depicted in Table I. The results, by using the NOLM architecture, can be found in [25]. Since the NOLM suffers from the stability problems due to the sensitive polarization state in the loop, an XPM-based upconversion in the straight-line HNL-DSF is simpler and more robust. In this experiment, the setup based on the straight-line XPM is shown in Fig. 6. The 16 continuous-wave (CW) channels are multiplexed via an arrayed waveguide grating (AWG) and are simultaneously modulated by an M–Z modulator driven by a 2.5-Gb/s pseudorandom-bit-sequence (PRBS) electrical signal with a word length of
$2^{31} - 1$. The optical spectrum of the WDM signals is shown in Fig. 6 as inset (i). Then, the generated 16
$\times$ 2.5-Gb/s signals are transmitted over a 10-km SMF for signal decorrelation and are then coupled into the HNL-DSF. The optical spectrum of the upconverted signals is also shown in Fig. 6 as inset (ii). The HNL-DSF output is processed via optical-to-electrical (O/E) and downconversions, which are the same modules as the FWM-based setup in Fig. 3.
It is noted that the optical carrier-to-sideband ratio (CSR) has influences on the system performance because the mm-wave signals are generated by the interplay between the optical powers in the carrier and sideband or the two sidebands themselves [26]. CSR refers to the ratio of the optical power in the optical carrier to that of the first-order sideband within a defined resolution bandwidth (here, it is set as 0.01 nm). While maintaining the CSR at 13 dB, the measured receiver sensitivities at a BER of
$10^{-10}$ of all the 16 upconverted channels are around
$-$24 dBm, and the difference among them is less than 0.8 dB, as shown in Fig. 7. Therefore, there is no big crosstalk among multiple channels. The optical and electrical eye diagrams are also shown in Fig. 7 as insets (i) and (ii), respectively. The BER curves at three different control powers while preserving the CSR at 13 dB are shown in Fig. 8. It is observed that 18.7 dBm is the optimal value for high receiver sensitivity, and there is around 0.8-dB power penalty compared with 13.7 dBm. In this experiment, the VSB method is also used to enhance the receiver sensitivity and to simultaneously reduce the occupied bandwidth.
This scheme exhibits very good conversion performance at high data rate and can provide more wavelength channels by being extended into the whole optical-fiber transmission band without any interference- and saturation-effect limitation [17], [18].
C. Cross-Absorption Modulation (XAM) in EAM
The principle of upconversion based on EAM is similar to its wavelength conversion mechanism at high bit rate [27]– [28] [29][30]. The main difference from wavelength conversion is that the modulated data signal will be used to replace the CW source. The experimental setup is shown in Fig. 9.
The EAM (CyOptics: EAM 40) with a 3-dB bandwidth of 32 GHz, a fiber-to-fiber insertion loss of 8 dB, and a polarization sensitivity lower than 1 dB is used to realize the signal upconversion. The optical LO signals are generated by using the OCS modulation scheme with larger than 25 dB of the CS ratio. The electrical signal and waveform are also shown as insets in Fig. 9.
When the dc bias on the EAM is
$-$3 V and the 2.5-Gb/s data signal is
$-$4 dBm, the optical power and the CSR of the mixed optical signal after the EAM as a function of the input power of LO signal are shown in Fig. 10. The measured static transfer curve slope of the EAM at
$-$3V bias is 3.5 dB/V. Due to the absorption effect of the EAM, the data signal is almost absorbed when the LO signal is smaller than 10 dBm. Once the LO signal is larger than 10 dBm, it is mixed with the data signal due to XAM in the EAM. As the power of the LO signal is 12.5 dBm, the CSR is smallest. Fig. 10 also shows that the CSR will be reduced with the LO signal enhancement when the LO signal is larger than 12.5 dBm. The reason is that the LO signal will saturate the EAM, and the CSR could further be reduced as long as we increase the dc bias on the EAM. However, a larger dc bias will reduce the power and OSNR of the upconversion signals. Therefore, the maximum bias in the experiment is
$-$3 V.
The optical spectrum and eye diagrams of the upconverted signal are shown in Fig. 11 when the dc bias on the EAM, the LO power, and the data signal power are
$-$3 V, 12.2 dBm, and
$-$4 dBm, respectively. The CSR of the mixed signals is 19 dB, and the mixed signals will occupy an 80-GHz bandwidth to keep the three frequency tones, which are separated by 40 GHz when the LO power is 12.2 dBm. A high CSR will lead to weak receiver sensitivity [26], whereas a wide bandwidth will reduce the spectral efficiency of the WDM system. In this experiment, the VSB filtering method is used to enhance the receiver sensitivity and to simultaneously reduce the occupied bandwidth. The receiver sensitivity at a BER of
$10^{-9}$ and the CSR of the mixed signals at a different center wavelength of TOF are shown in Fig. 12. As the center wavelength of the TOF is 1560.4 nm, which is the same wavelength as the data signal, the dual-sideband signals can almost be maintained; therefore, the receiver sensitivity is a little higher than other cases. When the center wavelength is tuned longer than 1560.5 nm, the CSR will be reduced because the optical carrier is suppressed, which leads to the improved receiver sensitivity.
This scheme has some advantages such as low power consumption, compact size, polarization insensitivity, easy integration with other devices, and higher speed operation due to the EAM inherent characteristics.
D. External Intensity Modulation
External modulation is another choice to upconvert mm-wave signals for the ROF systems. Essentially, three different schemes exist for the generation and transmission of the mm-wave signals over optical links with intensity modulation: double-sideband (DSB), single-sideband (SSB) [31]– [32] [33] [34] [35][36], and OCS modulation schemes. Fig. 13 shows the optical mm-wave generation based on the DSB, SSB, and OCS modulation schemes, and the corresponding optical spectrum and eye diagrams after mixing with the 40-GHz RF signal are also inserted. Baseband data signal is generated by an M–Z modulator driven by a 2.5-Gb/s PRBS electrical signal with a word length of
$2^{31} - 1$. For the DSB modulation scheme, the M–Z modulator 2 is biased at 0.5
$V_{\pi}$, and the frequency of the driven RF signal is 40 GHz. The generated mm wave will occupy over an 80-GHz bandwidth because it has two sidebands. Since the two sidebands have different velocities in an SMF, the RF power at 40 GHz will disappear after transmitting over a certain length of the SMF. As an example, the eye diagram after a transmission over 2 km is shown in Fig. 13. It is seen that the RF power at 40 GHz is almost faded, which leads to a large power penalty. The measured BER curves in Fig. 14 show that the power penalty is 17 dB at a BER of
$10^{-10}$ after the 2-km transmission. These results indicate that the DSB-modulation-based scheme is not suitable to a large area access network. A dual-arm M–Z modulator is employed to achieve the SSB modulation. The two electrical RF signals to drive the dual-arm M–Z modulator have a phase shift of
$\pi/2$, and the dc bias is at 0.5
$V_{\pi}$. The generated optical mm wave will only occupy a 40-GHz bandwidth, but the optical CSR is generally larger than 15 dB, which means that it is full of dc components at the peak of the center wavelength; hence, it results in low receiver sensitivity [29]. Fig. 14 shows that the receiver sensitivity of back-to-back (B-T-B) mm-wave signal for the SSB modulation is around 10 dB lower than that for the DSB modulation. Although there is no power penalty after a 20-km transmission, it is more than 5 dB after a 40-km transmission due to fiber dispersion and large CSR. When the phase of the two electrical RF signals to drive the dual-arm M–Z modulator is set to
$\pi$ difference and the dc bias is at the minimal intensity-output point or
$V_{\pi}$, the OCS modulation is realized. In this scheme, only a 20-GHz RF signal is needed, and the bandwidth for the M–Z modulator is also only 20 GHz; moreover, the generated optical spectrum just occupies a 40-GHz bandwidth. At a BER of
$10^{-10}$, the receiver sensitivity of the B-T-B mm-wave signal is
$-$39.7 dBm, which is similar to that of the mm-wave-signal generation based on the DSB modulation. There is no power penalty after a transmission over 20 km, and the power penalty is less than 2 dB after a 40-km transmission. The electrical eye diagrams after 10-km and 50-km transmissions are shown as insets (i) and (ii), respectively, in Fig. 14. These results show that mm wave generated by the OCS modulation can be used in large area access networks.
Since the optical mm wave has two peaks after the OCS modulation, it will suffer from dispersion in fiber when transmitting over an SMF. The pulsewidth of the 2.5-Gb/s signal carried by the optical mm wave is approximately 400 ps. The two peaks with a wavelength spacing of 0.32 nm will have a walk-off time of 400 ps caused by fiber dispersion after a transmission over a 74-km SMF with a group-velocity dispersion of 17 ps/nm/km, which means that the eye will fully be closed after this distance. While considering the limited rise and fall times of the optical receiver and EA, the maximum delivery distance will be shorter. Fig. 15 clearly shows the evolution of optical eye diagrams at different transmission distances. The unflat amplitudes of the optical carriers at 40 GHz, as shown in Fig. 15(b), are caused by chromatic dispersion. Wake et al. [37] and Gliese et al. [38] show that fiber dispersion causes the amplitude fluctuation of the carrier, but the RF power at 40 GHz does not disappear when the carrier is a pure dual-mode lightwave. Fig. 15(d) shows that the eye is almost closed after the optical mm wave is transmitted over 60 km.
By using the OCS modulation scheme, the 32
$\times$ 2.5-Gb/s dense-WDM (DWDM) signals after a transmission over 40 km are simultaneously upconverted to integrate with the WDM-PON networks. The experimental setup for DWDM signal transmission, upconversion, and posttransmission is shown in Fig. 16. A 32-DFB-LD laser array is used to realize 32 wavelength signals from 1536.1 to 1560.9 nm with an adjacent 100-GHz spacing. An AWG is used to combine the 32 CW lightwaves before modulation by an M–Z modulator. The optical spectrum of the WDM source is shown in Fig. 16 as inset (i). The generated 32
$\times$ 2.5-Gb/s signals are transmitted over 40 km for simulation of the metro optical network before they are upconverted by using a dual-arm M–Z modulator based on the OCS technique. The optical spectrum of the upconverted signals is also shown in Fig. 16 as inset (ii). The upconverted mm waves are amplified by an EDFA to get a power of 20 dBm before transmission over a variable-length SMF. At the receiver, the desired channel is selected by using the identical O/E and downconversion components as in the forenamed setup. The sensitivities for all channels after a 40-km transmission at a BER of
$10^{-10}$ are shown in Fig. 17. The preamplifier at a short (
$<$1540 nm) or long (
$>$1560.5 nm) wavelength has a smaller gain; therefore, the receiver sensitivity at these wavelengths is a little lower. The power penalty for all channels is roughly 2 dB after a 40-km SMF delivery. It is also observed that the penalty is the same as in the multichannel situation when one channel is kept on with a 10-dBm input power.
The upconversion signals based on the OCS modulation scheme have shown the best performance such as the highest receiver sensitivity, the highest spectral efficiency, and the smallest power penalty over long-distance delivery compared with those on the DSB and SSB modulation schemes.
E. External Phase Modulation
In addition to the intensity modulation, the external phase modulation [39], [40] is also utilized to produce downstream optical mm waves in optical–wireless networks. Fig. 18 shows the principle of using a PM and a subsequent interleaver for mm-wave generation. As a schematic illustration, the case of WDM signals with 100-GHz channel spacing as inset (i) is considered. When the WDM sources are modulated by the PM driven by a 20-GHz sinusoidal RF clock, the signal field of one channel at the output of the PM can be written as a few sidebands
TeX Source$$E_{{\rm output}1}\! = \!A_{\rm s}\! \sum_{n = - \infty}^{+\infty}\!J_{n}(m_{\rm d})\cos \left[(\omega_{\rm s}\! + \!n\omega_{\rm RF})t\! + \!n\pi/2 \right]\eqno{\hbox{(3)}}$$ where
$A_{\rm s}$ is the amplitude of the original optical carrier,
$J_{n}(m_{\rm d})$ is the
$n$th Bessel function of the first kind,
$m_{\rm d} = \pi V_{\rm RF}/V_{\pi}$ is the modulation depth of the PM,
$V_{\rm RF}$ is the driving voltage of the RF signals, and
$n\omega_{\rm RF}$ is the generated sidebands. How many sidebands can be generated depends on the amplitude of the driven RF signal on the PM. Here, we assume that only the first-order sideband is generated through optimization of the modulation depth
$m_{\rm d}$. The peak of the first sideband is 20 GHz away from the original optical carrier, as shown in inset (ii). The interleaver, with one input and two output ports of 25-GHz bandwidth, is used to suppress the optical carrier. When the central wavelengths of the WDM light sources can match up well to the interleaver, the optical carrier of each channel will be suppressed. The output signal of the interleaver is expressed as
TeX Source$$\displaylines{E_{{\rm output}2} = A_{\rm s} \bigg\{\sum_{n = - \infty}^{\infty}J_{n}\left(m_{\rm d}\right)\cos\left[\left(\omega_{\rm c} + n\omega_{{\rm m}}\right)t + n{\pi/2} \right]\hfill\cr\hfill - \alpha J_{0}\left(m_{\rm d}\right)\cos\omega_{\rm c}t \bigg\}\quad\hbox{(4)}}$$ where
$\alpha$ is the attenuation coefficient of the interleave filter at the peak of center frequency. The optical spectrum from output 1 of the interleaver is shown in inset (iii). In this way, the optical mm-wave WDM channels are generated.
The experiment setup for WDM optical mm-wave generation and transmission is shown in Fig. 19. A laser array with eight DFB LDs is used to achieve eight wavelength signals from 1553.7 to 1559.3 nm with 100-GHz spacing. The generated 8
$\times$ 2.5-Gb/s signals are transmitted over a 40-km SMF for decorrelation. The decorrelated WDM signals are modulated by a PM driven by a 20-GHz sinusoidal clock with a peak-to-peak amplitude of 3 V. The optical spectrum after the PM is shown in Fig. 19 as inset (i). The half-wave voltage of the PM is 11 V. Since the driving voltage is much smaller than the half-wave voltage of the PM, the second-order sideband of each channel is 25 dB lower than the first-order sideband; therefore, the second-order sidebands have little effect on the transmission of the optical mm wave in SMFs. An optical interleaver with two output ports and 25-GHz bandwidth is used to suppress the optical carriers and to convert the modulated WDM lightwaves to WDM optical mm waves. After passing through the optical interleaver, the CS ratio of all channels is larger than 15 dB, as shown in inset (iii), and the repetition frequency of the optical mm wave is 40 GHz. The total power of the optical mm-wave signals is 1 dBm. The remained optical carrier from the other port of the interleaver is shown in Fig. 19 as inset (ii). At the receiver, one channel is selected by using the TOF and the identical O/E and downconversion components as in the HNL-DSF setup.
Fig. 20 shows the BER curves and eye diagrams of channel 4 at 1556.1 nm after a transmission over a 40-km SMF. For a BER of
$10^{-10}$, the receiver sensitivity for the B-T-B signal is
$-$36.1 dBm. The penalty after a 20-km transmission is 0.3 dB. At 40-km distance, the power penalty is 2 dB. It is also observed that there is little difference between one single wavelength situation and the simultaneous multichannel case.
Through optimization of the amplitude of the RF signal to drive the PM for suppressing the high-order sidebands of the optical mm wave, the signals can be transmitted over a long distance. This scheme also exhibits better performance on system stability due to the removal of any dc-bias controller.
F. Comparison
Table II summarizes the advantages and disadvantages of several upconversion technologies. It suggests that the upconversion method based on external modulation (intensity and phase modulations) scheme shows the practical advantages in terms of the low cost, the simplicity of system configuration, and the performance over a long-distance transmission. This implies that these schemes will be the desirable candidates for downlink connection in full-duplex optical–wireless networks.
SECTION III
WAVELENGTH REUSE FOR UPSTREAM
It has been expected that mm-wave bands would be utilized to meet the requirement for high bandwidth and to overcome the frequency congestion in the optical–wireless networks. The negative side of mm waves is the need for numerous BSs, which is a consequence of high RF propagation losses in the atmosphere. In this situation, it is necessary to minimize the cost of the BS and to shift the system complexity and expensive devices to the CO. Hence, the overall architecture design, the scheme of RF signal generation, and the transmission for the uplink and downlink play the key roles on the successful deployment in the real networks. Regarding downlink connections, several approaches for mixing or upconverting mm-wave signals have been investigated in Section II. As far as the uplink connection is concerned, some methods have recently been proposed. However, most of them only demonstrated uplink connections over short transmission distances [41], [42]. Full-duplex operation using high RF carrier still raises difficulties that have to be addressed. The network architecture consisting of a single light source at the CO and the reuse of the downlink wavelength at the BS is an attractive solution for low-cost implementation as it requires no additional light source and wavelength management at the BS.
A. Differential Phase-Shift-Keyed (DPSK) Format for Downstream and Remodulated OOK for Upstream
The scheme is shown in Fig. 21. The PM is used to generate DPSK [43], [44] signal, where a bit “1” is coded by a
$\pi$ phase shift of the optical phase, whereas a bit “0” leaves it unaffected. An OCS scheme is employed to simultaneously generate an optical 40-GHz mm wave and upconvert a 2.5-Gb/s baseband signal for the downstream. At the BS, a power splitter (PS) is used to divide the incoming signal into two parts. The first one is demodulated by a Mach–Zehnder delay interferometer (MZDI) and is later detected by a high-speed photodetector. The second one, considered as a CW for the upstream signal, is modulated with the symmetric upstream signal and sent back to the CO where a low-cost receiver with low-frequency response is used for detection.
The experimental setup is shown in Fig. 22. At the CO, a CW is generated by a tunable laser (TL) at 1553.1 nm and is modulated by a PM driven by a 2.5-Gb/s PRBS
$2^{31} - 1$ electrical signal with an amplitude of
$V_{\pi}$. The resulting DSPK signal is upconverted through the OCS modulation scheme. The optical mm-wave signal is then amplified by an EDFA to 13 dBm before it is transmitted over the various lengths of SMF-28. At the BS, the PS is used to divide the incoming signal power for the downlink detection and uplink connection. Because photodetection is phase-insensitive, a phase-to-intensity conversion is needed at the receiver. Therefore, an MZDI is inserted in the downlink path to demodulate the DPSK signal and to retrieve the data. In this experiment, the fiber-based MZDI is made from two couplers that are thermally insulated. The length difference between the two arms, which is needed to perform a 1-b delay, is 7.98 cm. Before detection, the signal is preamplified by a regular EDFA with a 30-dB small-signal gain and is filtered by a 0.5-nm-bandwidth TOF for amplified spontaneous-emission noise reduction. The direct detection is made by a 60-GHz-bandwidth PIN photodiode. After this O/E conversion, we use the identical downconversion setup to retrieve the 2.5-Gb/s baseband signal. For the uplink, the signal is amplified by an EDFA before its remodulation by a symmetric 2.5-Gb/s electrical signal.
In real-network implementations, the diplexer connected with the antenna would act as a circulator to handle the up- and downstream signals at the BS. The baseband upstream signals would be obtained after the downconversion of the end user's information coming from the diplexer in the BS. In this experiment, the same fiber length is used for both up- and downstreams. The uplink signal is detected by a low-frequency response receiver which also filters out the residual part of the mm-wave signals.
Because the OCS modulation creates a spectrum with two peaks separated by 40 GHz, the mm wave will suffer from chromatic-dispersion impairments that are experienced throughout its propagation in the SMF fiber. Fig. 23 shows the optical eye diagrams in the B-T-B configuration and after the 40-km propagation in the SMF. The power fluctuations of the 40-GHz modulation arise from the chromatic dispersion. As far as the 2.5-Gb/s signal is concerned, the difference between the experimental and simulation results is caused by the setup imperfections. In particular, the MZDI, considered as perfect in the simulations, does not produce an exact and stable 1-bit delay, impacting the overall performance.
Fig. 24 shows the BER measured for both up- and downlinks as a function of the received optical power for various propagation distances. For the downlink, the optical power is measured at the output of the MZDI. After the transmissions over the 25- and 40-km SMFs, the power penalty values at the given BER of
$10^{-10}$ are 1.1 and 1.9 dB, respectively. For the uplink, since the signal at the RF carrier [45] is filtered out by the low-speed receiver, it exhibits a better performance than the downlink. After a 40-km downlink transmission, the power penalty for the remodulated optical carrier is less than 0.9 dB over the same uplink transmission distance. The eye diagrams of the B-T-B configuration for both directions are also shown in Fig. 24.
B. OCS for Downstream and Reuse for Upstream
The schematic diagram is shown in Fig. 25. The OCS scheme is employed to simultaneously generate an optical mm wave and upconvert the baseband data signal for the downstream. The original carrier is split prior to an OCS operation and is then coupled with the optical mm-wave signal before they are transmitted to the BS. At the BS, a fiber Bragg grating (FBG) is used to reflect the carrier while passing the optical mm-wave signal to the downlink receiver. The reflected carrier is acted as the CW, remodulated with the symmetric upstream signal, and then transmitted back to CO, where a low-cost receiver with low-frequency response detects the upstream signal [46].
The experimental setup for the full-duplex ROF system is shown in Fig. 26. At the CS, a CW is generated by a TL at 1549.1 nm and split into two parts via a 50:50 optical PS. The first part is modulated via an M–Z modulator driven by the 2.5-Gb/s signals. The optical mm-wave signals are generated by using the OCS scheme. The optical spectrum and eye diagrams after an OCS are measured at points A and B and in insets (i) and (ii) in Fig. 26, respectively. Then, the generated optical mm wave is amplified by an EDFA to get a power of 6 dBm for transmission. The second part is directly sent for amplification by an EDFA to obtain a power of 9 dBm and then combined with the first part via an optical carrier before they are transmitted over an SMF.
At the BS, the FBG is used to take on two roles: One is to reflect the optical carrier to provide a CW light source for uplink connection, whereas the other is to simultaneously pass the two sidebands generated by the OCS, and as a consequence, it increased the CS ratio up to 30 dB due to sharp-notch characteristics. This FBG filter has a 3-dB reflection bandwidth of 0.2 nm and a reflection ratio larger than 50 dB at the reflection-peak wavelength. The eye diagram and the spectrum of the passing and reflected parts are measured at points C, D, and E and in insets (iii), (iv), and (v) in Fig. 26, respectively. Then, we use the identical O/E and downconversion to retrieve the downstream signals. Fig. 27 shows the evolution of the dispersion impact on different distances for two-direction signals. It is clearly seen that the eye still keeps open despite of the transmission over a 40-km SMF, which is long enough for an access-network coverage. However, the uplink might transmit longer distance because most of the component of high frequency is already removed by the FBG filter before it is sent back to the CO. For the uplink, the reflected signal is amplified by an EDFA to compensate the insertion loss of the filter before it is modulated by a symmetric 2.5-Gb/s electrical signal.
The measured BER curves in Fig. 28 show that the power penalty of the downlink signal is negligible after the FBG filter or after a 10-km transmission. For longer transmissions, the power penalty values are 1.4 and 1.9 dB after the 25- and 40-km transmissions, respectively, at a BER of
$10^{-10}$. Because of no large power of the RF component, the uplink exhibited better performance compared with the downlink transmission. After a 40-km downlink transmission, the power penalty for the remodulated optical carrier is roughly 1.2 dB over the same distance uplink transmission due to relatively large chirp transmitter as well as no clock recovery at the receiver side.
C. PM with Subsequent Filter for Downstream and Directly Modulated SOA for Upstream
Fig. 29 shows the schematic diagram of full-duplex optical–wireless system architecture by using the PM and SOA. An optical PM is driven by small RF signal (one-fourth half-wave voltage of the PM) to create first-order sidebands while suppressing the second-order components for increasing dispersion tolerance. An interleaver is employed to separate the optical carrier from the first-order sidebands to generate an optical mm-wave carrier. After modulation by the downlink data, the upconverted optical signal is combined with the original optical carrier and transmitted over the SMF. At the BS, an FBG is used to reflect the optical carrier while passing the optical mm-wave signal to O/E conversion. Then, the boosted electrical mm-wave signal is broadcasted by an antenna through a duplexer acting as a circulator to handle up- and downstream signals at the BS. On the other hand, the upstream signals are downconverted through a mixer and tunable delay (TD) line without the need of the LO signal. The reflected optical carrier is considered as the CW source, directly modulated by baseband upstream signals in the SOA, and sent back to the CS, where a low-cost and low-frequency response receiver is used for detection. In this scheme, the SOA performs the function of both amplification and modulation, which eliminates the requirement of optical amplifier and external modulator at the BS.
Fig. 30 shows the experiment setup for the full-duplex optical–wireless system by using wavelength reuse and directly modulated SOA. At the CO, a CW is generated by a DFB-LD at 1534.4 nm and modulated by an optical PM driven by a 20-GHz RF sinusoidal wave with an amplitude of 1 V (half-wave voltage of the PM is 4 V). The optical spectrum (measured at point a) after modulation is shown in Fig. 30 as inset (a). A 50/25-GHz optical interleaver with 35-dB channel isolation is used to separate the remained optical carrier from the first-order sidebands. The generated optical mm wave is then amplified and modulated by an intensity modulator driven by 2.5-Gb/s PRBS
$2^{31} - 1$ electrical downstream signals. The separated optical carrier is directly sent to combine with the upconverted baseband signals via an optical carrier before its transmission over a 40-km SMF-28 with an 8-dBm input power. The optical spectra of the separated optical carrier, optical mm-wave signal, and combined signals are shown in Fig. 30 as insets (b)–(d), respectively. At the BS, an FBG is used to reflect the optical carrier and to simultaneously transmit the optical mm-wave signals. The FBG filter has a 3-dB reflection bandwidth of 0.2 nm and a reflection ratio larger than 50 dB at the reflection-peak wavelength. The spectra of the reflected carrier and transmitted optical mm-wave signals are shown in Fig. 30 as insets (e) and (f), respectively. The downstream signals are downconverted through a mixer and TD line without requiring any LO signal.
For the uplink, the reflected optical carrier is directly modulated in an SOA driven by a 250-Mb/s (PRBS
$2^{7} - 1$) electrical signal with 3.1-V amplitude and 165-mA bias. The gain is 10 dB in the 34-nm spectral width, and the polarization sensitivity is smaller than 0.5 dB. In this experiment, the same fiber length is used for both up- and downstreams. The uplink signal is detected by a low-frequency response receiver which also filters out the residual part of the high-frequency mm-wave signal due to an imperfect filtering by the FBG.
Fig. 31 shows the optical eye diagrams in the B-T-B configuration and after the 40-km propagation in the SMF. The power fluctuations of the -GHz modulation arise from the chromatic dispersion in the 40 fiber. Fig. 32 shows the BER measured for both the up- and downlinks as a function of the received optical power. For the downlink, the power penalty at the given BER of
$10^{-10}$ is 2.0 dB after a 40-km transmission. For the uplink, the power penalty for the remodulated optical carrier is around 0.5 dB over the same transmission distance. The eye diagrams of the B-T-B configuration for both directions are also shown in Fig. 32.
D. Comparison
Table III summarizes the advantages and disadvantages of the three full-duplex system designs. All schemes use the sole centralized light source for both directions based on the wavelength reuse method. It suggests that the system, by using one colorless SOA as both the amplifier and the modulator for the upstream signal, is a practical solution for the future optical–wireless access networks in terms of the low cost and easy maintenance.
SECTION IV
TESTBED IMPLEMENTATION FOR SIMULTANEOUS DELIVERY OF WIRED AND WIRELESS SERVICES
Currently, the wired and wireless services are separately provided by two independent physical networks. The wired networks based on fiber-to-the-home [47] access technologies provide huge bandwidth to users but are not flexible enough to allow roaming connections. On the other hand, the wireless networks offer mobility to users but do not possess abundant bandwidth to meet the ultimate demand for video services with HDTV quality. Therefore, seamless integration of wired and wireless services for future-proof access networks will lead to convergence of ultimate high bandwidth for both fixed and mobile users in a single low-cost transport platform. This can be accomplished by using our developed hybrid optical–wireless networks, which not only can transmit wireless signals at the BS over fiber but also simultaneously provide the wired and wireless services to the end users due to the cascaded modulation scheme for downstream signals. Optical mm waves can be generated by several all-optical upconversion schemes that we have discussed in Section II. No matter what kind of all-optical up-conversion scheme one chooses, a part of baseband signal still exists in the whole electrical spectrum after the all-optical upconversion [48]. To illustrate this point, we simulate the process of the all-optical upconversion schemes based on FWM in the HNL-DSF and the OCS modulation. As shown in Fig. 33, it shows the electrical spectra after the all-optical upconversion. The LO frequency for the all-optical upconversion scheme based on the FWM in the fiber is 40 GHz, and the baseband signal is 2.5-Gb/s nonreturn-to-zero signal. Fig. 33 shows that there are two components of electrical signals after the all-optical up-conversion: One part occupies the baseband, whereas the other occupies the high-frequency band near 40 GHz. Hence, we propose a novel ROF network architecture to use the baseband signals for broadband optical access at 2.5 Gb/s at low cost and RF frequency for wireless connection.
Fig. 34 shows the architecture for concurrently providing superbroadband wireless and wired services. The content providers or upstream networks send the data to the CO, where the multichannel mm-wave carriers are generated through an external modulator based on our developed all-optical upconverter. This upconversion technique possesses many advantages that allow the data to be transmitted over wired and wireless media in a single platform [48]. First, the generation of the mm-wave carrier and the upconversion of the original data channel are simultaneously performed in the optical domain. Second, as a result of this process, two identical data signals are concurrently generated: one at the electrical baseband and another at the RF-carrier frequency. The upconverted signals are multiplexed before they are transmitted over the fiber to the BS where an AWG in WDM-PON is used to route the signals to customer premise. At the customer premise, the signal is divided by a PS into two. The first part is detected by a high-speed receiver and then electrically amplified using a narrow-band RF amplifier before it is broadcasted by an antenna as a wireless signal. The other part is directly sent to a wall-mounted optical port via a fiber access, and a user can utilize a simple low-cost receiver to detect the baseband data signal by filtering out the high-frequency mm-wave signal. This new hybrid system can allow wired and wireless transmissions of the same content such as HDTV, data, and voice up to 100 times faster than the current networks. The same services will be provided to the customers who will either plug into the wired connection in the wall or access the same information through a wireless system. The customer premises can be conference centers, airports, hotels, shopping malls, and ultimately homes and small offices.
The experimental testbed setup for simultaneously delivering wired and wireless services is shown in Fig. 35. At the BS, the optical mm wave is divided into two parts. The wired part is sent to a low-speed avalanche photodiode (APD) that has a 3-dB bandwidth of 2 GHz. Since the bandwidth of the APD receiver is limited at 2 GHz, the RF signal at high frequency is filtered out. In the case of the wireless part, the converted electrical signal is boosted by an RF EA before it is broadcasted through a double-ridge guide antenna with a gain of 19.2 dBi at a frequency range of 18–40 GHz. After a wireless transmission, the signal is received by another identical mm-wave antenna. The signal is downconverted through a mixer and TD line by using part of the incoming signal as the LO signal. The receiver sensitivities and eye diagrams at different air distances are shown in Fig. 36. The power penalty after a transmission over a 25-km SMF is less than 1.5 dB. The receiver sensitivity quickly degrades when the wireless signals are transmitted beyond a 10-m indoor airspace because the antenna, as an isotropic point source in its radiation direction, spreads its energy over a certain angle of surface area as the wave propagates in the airspace. The path loss is proportional to the reciprocal of distance square. Signal degradation via multiple reflections from the wall is a key factor that limits the maximum transmission distance for 2.5-Gb/s wireless signal in our testbed environment of office building hallway. We then set the data rate to 1.25 Gb/s; the BER curve and eye diagram are measured and shown, respectively, in Fig. 37 after a 25-km fiber transmission. We observed that the receiver sensitivity is increased around 1.5 dB compared to 2.5-Gb/s signals at the 3.2-m air distance.
Using this scheme, the video transmission of 270-Mb/s uncompressed standard-definition TV (SDTV) signal is implemented, as shown in Fig. 38. At the CS, an analog-to-digital converter is used to convert the analog video signal from a DVD player into the 270-Mb/s (SMPTE 259-M SDTV signal) serial digital interface (SDI) signal for driving the M–Z modulator to modulate a CW from a TL. At customer side, the signal is split into two parts for wireless and wired signal deliveries, respectively. For the wired part, after an O/E conversion, a digital-to-analog (D/A) converter is used to convert the SDI signal to analog video signals for display on one TV. For the wireless part, the downconverted signal is converted to analog video signals by a D/A converter and then displayed on another TV. The photograph of this implementation is shown in Fig. 38 as inset (i). The received videos on two TVs are clear, noise-free, and stable.