1. Introduction

The proliferation of mobile data traffic in the fifth-generation (5G) era has led to significant challenges to the radio access network (RAN) [1]. As the key RAN architecture towards 5G and beyond, the centralized, collaborative, cloud and clean RAN (C-RAN) consolidates the baseband processing functions into a baseband unit (BBU) pool in a single equipment room, which greatly simplifies the antenna site [2]. The BBU pool is connected with multiple remote radio units (RRUs) via mobile fronthaul networks (MFNs). Given the merits of optical fiber transmission systems, such as broad bandwidth and low latency, the radio-over-fiber (RoF) becomes a promising technology for fronthaul networks [3].

Analog radio-over-fiber (ARoF) links transport the mobile signals in an analog format over optical links, providing outstanding benefits such as high spectrum efficiency, simple infrastructure and low cost. It also simplifies the RRU through the avoidance of high-speed digital-to-analog converters (DACs), which is crucial for the deployment of small cell antenna sites. However, the analog signals are vulnerable to nonlinear distortions, resulting in stringent requirements for electronic and optical devices. Digital radio-over-fiber (DRoF) is currently deployed in fronthaul links based on common public radio interface (CPRI), where the continuous waveforms are digitized into the binary bit sequence. DRoF transmission is highly immune to nonlinearities, unfortunately, it unavoidably suffers from inefficient bandwidth, limited scalability and high RRU complexity. Combining the advantages of both ARoF and CPRI-based DRoF, the DRoF based on the delta-sigma modulation (DSM) draws considerable attention [4–6]. Unlike the CPRI with 15-bit quantization, the DSM technique exploits a one- or two-bit quantizer, corresponding to the non-return-to-zero (NRZ) or four-level pulse amplitude modulation (PAM4) signal, which introduces severe in-band quantization noise. Therefore, both oversampling and noise shaping techniques are utilized to place most of the quantization noise out of the signal band. At the receiver, only a passive analog filter is required to remove the out-of-band quantization noise and recover the original signal, rather than relying on conventional high-resolution DACs. The high-efficiency switch-mode power amplifier can be used before the filter to amplify the binary digital signal. Furthermore, many users are able to share the cost of the delta-sigma modulator in the fronthaul downlink, maintaining the simplicity and power-efficiency of the RRUs [7].

In recent studies, the DSM-based digital mobile fronthaul links via optical intensity modulation and direct detection (IM-DD) systems have been widely investigated [8,9]. Despite its simplicity, the IM-DD solution has poor receiver sensitivity and low spectral efficiency. These issues can be addressed by the coherent detection, but it is not low-cost enough to satisfy MFNs applications. Compared with traditional coherent systems, self-homodyne coherent systems require simpler implementations, where the modulated signal is transmitted together with a continuous wave tone originating from the same laser [10–12]. Additionally, the injection-locked laser (ILL) based coherent reception further simplifies the digital signal processing (DSP), obviating the need for carrier frequency estimation and phase noise compensation [13–15]. To this end, an electro-absorption modulated laser (EML) serving as a coherent receiver provides an attractive solution, which has been commonly used as the transmitter in optical access networks and data center interconnect owing to its low cost and small form size. The electro-absorption modulator (EAM) acts as a high-speed photodetector when operating at strong absorption [16], and the injection-locked distributed feedback (DFB) laser simultaneously yields the required local oscillator (LO) [17]. An EML-based coherent receiver co-integrated at the die-level with a transimpedance amplifier (TIA) is able to provide higher power budget [18]. Moreover, the EML can serve as receiver and transmitter at the same time to enable cost-effective transceiver integration, which has been investigated for analog radio-over-fiber, analog radio-over-air and access networks by implementing frequency division duplexing (FDD) [19–21].

In this paper, we propose a novel coherent digital mobile fronthaul scheme employing a band-pass delta-sigma modulator in BBU and an EML-based self-homodyne coherent receiver in RRU. We experimentally demonstrate the feasibility of the injection-locked operation via residual carrier of the DSM signal and further propose two modified schemes for the DSM signal to increase the locking range with a small sensitivity penalty. According to the results, the EML-based coherent receiver provides better receiver sensitivity than a PIN photodiode. In addition, by applying an equalizer at the receiver, the DSM-based DRoF system exhibits an excellent performance with a 5-dB loss budget improvement over the ARoF system. The proposed scheme provides a low-cost potential solution for coherent digital MFNs.

2. Principle

The principle of the EML-based coherent reception for the DSM signal is shown in Fig. 1. The optical signal injected into the EML via a three-port optical circulator (OC) contains both the carrier tone ${\lambda _{inj}}$ and the data signal, as shown in Fig. 1(a). The residual optical carrier from the tunable semiconductor laser (TSL) serves as the master light. A polarization controller (PC) is utilized to control the polarization of the injected optical carrier, which should be aligned with the polarization of the slave DFB laser in the EML. Alternatively, the polarization diversity reception through a tandem-EML can be adopted in realistic deployment [22]. The EAM detector along with a integrated DFB laser enables the EML to serve as a coherent receiver, as shown in Fig. 1(b). It shall be stressed that the integrated optical isolator in the EML is removed to allow sufficient injected optical power. The optical signal is partially injected into the DFB laser, which is locked to the residual carrier component in the middle of the injected signal. Thus, the DFB emission ${\lambda _{LO}}$ is pulled towards the externally injected optical carrier ${\lambda _{inj}}$, eliminating the frequency offset between the optical signal and LO. The LO provided by the injection-locked DFB laser is coupled with another part of the optical signal and then coherently detected by the EAM photodiode, which is reversely biased through a bias tee. The detected binary DSM signal is converted to the radio signals through a band-pass filter (BPF), as shown in Fig. 1(c), which can be subsequently applied to a power amplifier and an antenna. The output of the EML is sent to a photodetector and then captured by a electrical spectrum analyzer, through which the locked and unlocked states can be observed. Figure 1(d) provides the spectra in the free-running and injection-locked states, respectively. The beat frequency between the master and slave lasers appears at low frequencies in the free-running state, while this beat frequency vanishes when the DFB laser tracks the injected optical frequency. For convenience, we can vary the incident optical wavelength ${\lambda _{inj}}$ to allocate it close to the DFB laser wavelength ${\lambda _{LO}}$ in the experiment. In the practical implementation, the locked state can be obtained by adjusting the EML emission through temperature or DFB bias current control.

 

Fig. 1. The principle of the EML-based coherent reception for the DSM signal in virtue of injection locking technique.

Download Full Size | PDF

The locking range is defined as the maximum range of the frequency detuning between the master and slave lasers for stable injection-locked operation, which is proportional to the square root of the injection ratio [23]. The injection ratio is the ratio of the injected optical power to the slave laser output power. Note that the optical power injected into the DFB laser is strongly related to the actual absorption at EAM section, and the output power of the DFB laser depends on DFB bias current. Therefore, the appropriate EAM bias voltage ${V_{EAM}}$ and DFB bias current ${I_{DFB}}$ enable effective trade-off between EVM performance and locking range. Besides, the increase in the residual carrier component indeed allows a larger locking range, which requires reduced quantization noise around the carrier. For this purpose, two schemes implemented by offline processing are considered, i.e., quantization noise reduction and high-pass filtering, as shown in Fig. 1(a). Multi-bit quantization is a common approach to reduce quantization noise, but leads to large data rate. Alternatively, a part of the quantization noise can be subtracted from the DSM signal. The quantization noise is first extracted and later attenuated through multiplied by a factor ${\alpha (0<\alpha <1)}$. Finally, the reduced quantization noise is added to the desired signal to obtain the modified DSM signal. Figure 2(a) presents the power spectral density (PSD) of the DSM signals without, ${\alpha =0}$, and with quantization noise reduction for ${\alpha =0.1}$ and ${\alpha =0.2}$. The time-domain waveforms of the aforementioned signals are shown in Fig. 2(c). As can be seen, the fluctuations have occurred in the two-level DSM signal, which will become more severe as the value of ${\alpha }$ increases. Another method is to directly filter out the partial quantization noise at low frequencies, as illustrated in Fig. 2(b). The 1st-order high-pass Butterworth digital filters with the cutoff frequency ${F_0}$ of 0.35 and 0.7 GHz are adopted. Compared with the first scheme, the waveform distortion caused by this method is more serious, as shown in Fig. 2(d). Although the time-domain waveform is no longer two-level, the moderate resolution DAC is sufficient to generate the delta-sigma modulated signal with quantization noise suppression since the waveform fluctuations only appear within a certain range.

 

Fig. 2. (a)-(b) The electrical spectra and (c)-(d) the waveforms of the DSM signals before and after quantization noise reduction or high-pass filtering.

Download Full Size | PDF

3. Experimental setup

Figure 3 illustrates the experimental setup for the proposed mobile fronthaul downlink. The delta-sigma modulation and recovery processes of the orthogonal frequency multiplexed (OFDM) signal are shown in the Tx and Rx off-line DSP blocks, respectively. In BBU, a 400-MHz wide OFDM signal with 256 subcarriers is generated in MATLAB, among which 206 subcarriers are loaded with data. The modulation format is set to 16-point quadrature amplitude modulation (16-QAM) on each subcarrier. After over-sampled to 14 GSa/s, the OFDM signal is up-converted to a RF carrier frequency of 3.5 GHz, which is a quarter of the sampling rate. Subsequently, the OFDM radio signal is fed into a fourth-order one-bit band-pass delta-sigma modulator (BP-DSM) for noise shaping and quantization. The electrical spectra of the baseband OFDM signal and the DSM signal are shown in Figs. 3(a) and (b), respectively. Figure 3(c) shows the zeros and poles of the noise transfer function (NTF), which can be expressed by Eq. (1). The number of zero-poles is four, which is equal to the order of the delta-sigma modulator. The magnitude response of the BP-DSM NTF is shown in Fig. 3(d). The root mean-square (RMS) gain of the discrete-time NTF in the signal band [3.3 GHz, 3.7 GHz] is −34 dB. The generated DSM signal is uploaded into an arbitrary waveform generator (AWG, Keysight M8195A) with an amplitude of 400 mVpp. The AWG output is amplified by a 14-GHz linear electrical amplifier (EA) with 12 dB gain before driving a 12.5-Gb/s Mach–Zehnder modulator (MZM) biased at its linear modulation region. A tunable semiconductor laser (Santec TSL-710) with 100 kHz linewidth emits at around 1575.2 nm, which is close to the emission wavelength of the EML at the receiver.

 

Fig. 3. Experimental setup for the proposed coherent digital mobile fronthaul system. The electrical spectra of (a) the baseband OFDM signal and (b) the DSM signal. The (c) zero-poles and (d) magnitude response of the noise transfer function. (e)-(g) The device schematic, locking range and emission spectrum of the EML receiver.

Download Full Size | PDF

A 25-km standard single mode fiber (SSMF) carries the optical signal to the RRU, where a 10-GHz commercial isolator-free EML is used as the coherent receiver. As shown in Fig. 3(e), the EML is implemented in a transmitter optical sub-assembly (TOSA) package. The locking range reaches 130 MHz for an injected power of −30 dBm when an unmodulated optical signal is injected into the EML receiver with an EAM bias of −1.2 V, as provided in Fig. 3(f). The received optical signal is injected into the EML through an OC by adjusting the received optical power (ROP) and polarization with a variable optical attenuator (VOA) and a manual PC. After injection-locked to part of the injected signal, the DFB laser serving as the LO is coupled with another part of that and homodyne-detected by the EAM. Notably, the temperature and DFB current with tuning coefficients of around 2.6 GHz/K and 0.6 GHz/mA enable both coarse and fine tuning of the EML emission wavelength to realize the locked state. Then, the detected signal is amplified by a 6-GHz EA and digitized at 80-GSa/s by a 33-GHz real-time digital signal analyzer (DSA, Keysight DSAZ594A). The output from port three of the OC is for injection-locked state observation using an EAM and an electrical spectrum analyzer (ESA). At the Rx-side DSP, re-sampling and synchronization are firstly performed, and then a 35-tap feed forward equalizer (FFE) with a 5-tap decision feedback equalizer (DFE) is applied for NRZ signal recovery. Subsequently, a band-pass Butterworth filter covering 3.3 to 3.7 GHz eliminates the out-of-band quantization noise and retrieves the desired OFDM signal from the binary DSM signal. After down-converted to baseband, the OFDM signal is down-sampled and demodulated. Finally, the EVM is calculated to evaluate the performance.

4. Results and discussion

4.1 Parameters of the EML-based receiver

The EAM bias voltage is a crucial parameter for an EML-based receiver. The EAM detector exhibits different absorption characteristics when applying different bias voltages. As it can be expected from the quantum confined Stark effect (QCSE), the EAM absorption edge will be red-shifted when applying a large reverse bias, which leads to large absorption. Figure 4(a) shows the RF power of 1-GHz sine wave detected by an EAM versus the EAM bias voltage at various wavelengths, where the DFB current is set to 0 mA. The RF power measured by an ESA reflects the opto-electronic absorption, i.e., the responsivity of EAM photodiode, and the value measured at ${{V_{EAM}} = -2.2}$ V and ${\lambda = 1575}$ nm is indicated as a reference. When injecting an optical signal at short wavelength, there is no significant magnitude variance with increased reverse EAM bias. Nevertheless, the emission wavelength of the EML used in our experiments is around 1575 nm, where the absorption is severely affected by reverse EAM bias. Figure 4(b) indicates the receiver sensitivity of the DSM signal when using an EML-based coherent receiver. It is obvious that the highest sensitivity is obtained when the EAM is biased at −1.2 V at the required EVM limit of 12.5% for 16-QAM. The reverse bias voltage should be sufficiently large to assure large absorption, which will eventually reach saturation. As the reverse bias increases, the noise floor of the EML receiver will slightly increase, while the increase in the detected signal magnitude is negligible when the reverse bias voltage exceeds 1.2 V, consequently resulting in sensitivity degradation. As provided in Fig. 4(c), a higher reverse EAM bias leads to lower optical power injected into the DFB section and hence a narrower locking range. In the following measurements, the EAM bias voltage is chosen at −1.2 V. It is worth noting that if an EML with a shorter operating wavelength is available, the receiver sensitivity will be further improved due to stronger EAM absorption, but at the cost of a narrower locking range.

 

Fig. 4. (a) The absorption characteristics of the EAM detector. (b) EVM and (c) locking range versus ROP when the EAM detector is biased at different voltages.

Download Full Size | PDF

The bias current of the DFB laser has a noticeable impact on EML-based coherent detection as well. Figure 5(a) depicts that the threshold current of the DFB laser is approximately 10 mA, above which the output optical power increases linearly with the increased current. The DFB bias current is not only related to the emission wavelength, but also to the EVM performance and locking range, as provided in Figs. 5(b) and (c). The larger the bias current, the larger the LO power in coherent detection, resulting in the higher receiver sensitivity. Therefore, the best sensitivity is achieved when the DFB laser is biased at 90 mA. The emission wavelength of DFB laser is 1575.2 nm at the applied bias current of 90 mA and temperature of 25$^{\circ }$C. However, with the increase of bias current, the injection ratio gradually drops, resulting in a slightly reduced locking range. The bias current of the DFB laser is set to 90 mA in the following experiments.

 

Fig. 5. (a) The emission characteristics of the DFB laser. (b) EVM and (c) locking range versus ROP when the DFB laser is biased at different currents.

Download Full Size | PDF

4.2 Modified delta-sigma modulated signal

Apart from being affected by EAM bias voltage and DFB bias current, the locking range is also dependent on the residual carrier component of the injected signal. As mentioned in the principle, a part of the quantization noise can be subtracted from the DSM signal to increase the locking range by applying a factor ${\alpha }$. Figure 6(a) shows the EVM versus ROP for the DSM signal with and without quantization noise reduction. The EVM performance degradation caused by quantization noise reduction is negligible when ${\alpha }$ is set to 0.1, while around 1-dB sensitivity penalty is observed for ${\alpha = 0.2}$ with more severe signal distortion. In comparison with the original DSM signal, an effective improvement in locking range is obtained when employing quantization noise reduction, as depicted in Fig. 6(b). The DSM signals with ${\alpha = 0.1}$ and ${\alpha = 0.2}$ widen the locking range by 330 MHz and 670 MHz at ROP of −7 dBm, respectively.

 

Fig. 6. (a) EVM and (b) locking range versus ROP with and without quantization noise reduction.

Download Full Size | PDF

In addition, the DSM signals can also be high-pass filtered with 0.35-GHz and 0.7-GHz cutoff frequencies to allow a wider locking range with a slight sensitivity degradation, as shown in Figs. 7(a) and (b). Although the high-pass filtering brings more severe waveform distortion compared with the above scheme, it still exhibits a tolerable sensitivity penalty with the help of equalization. Besides, the larger residual carrier component can be obtained by high-pass filtering, thereby greatly increasing the locking range. By utilizing a high-pass filter with 0.35-GHz cutoff frequency, the locking range is increased by 900 MHz at the ROP of −7 dBm, which cannot further increase significantly with increased cutoff frequency due to the substantially similar residual carrier component. Considering the trade-off between receiver sensitivity and locking range, the aforementioned two modified schemes can be adopted in the injection-locked operation.

 

Fig. 7. (a) EVM and (b) locking range versus ROP with and without high-pass filter.

Download Full Size | PDF

4.3 Evaluation on receiver sensitivity

Figure 8(a) illustrates the EVM versus ROP for received 16-QAM and 64-QAM DSM signals under different transmission links. Compared with the optical back-to-back (OB2B) case, no noticeable sensitivity penalty is observed after 25-km SSMF transmission. The power fading effect induced by the fiber chromatic dispersion can be negligible at a carrier frequency of 3.5 GHz. In consideration of the required EVM limits of $12.5\%$ for 16-QAM and $8\%$ for 64-QAM, the receiver sensitivity is −16.5 dBm and −15.1 dBm for 16-QAM and 64-QAM cases over 25-km fiber, respectively. Figure 8(b) presents the EVMs for all recovered 16-QAM OFDM data sub-carriers at the ROP of −7 dBm, which has good consistency with the magnitude response of DSM NTF in Fig. 3(d). As the ROP reduces to −13 dBm, the transmission performance is more susceptible to noise, resulting in a worse EVM performance and a more dispersed distribution of the EVM values, as shown in Fig. 8(c). The constellation diagrams of the fifth sub-carrier ($\bullet$) in Figs. 8(b) and (c) are shown in Figs. 8(d) and (e), respectively. Moreover, the corresponding constellation diagrams for the 64-QAM DSM signals are shown in Figs. 8(f) and (g).

 

Fig. 8. (a) EVM versus ROP at different cases. (b)-(c) EVMs of recovered 16-QAM OFDM data sub-carriers at −7 and −13 dBm ROP. (d)-(g) The 16-QAM and 64-QAM constellations of the fifth sub-carrier at −7 and −13 dBm ROP.

Download Full Size | PDF

The sensitivity comparison for different receivers is presented in Fig. 9(a). The eye diagrams of the DSM signals detected by different receivers at ROP of −5dBm are shown in the insets of Fig. 9(a), respectively. Due to the in-band quantization noise introduced by delta-sigma modulation, the EVM stabilizes at a fixed floor at high ROP. When using a PIN or EAM receiver, the sudden increase in EVM values at low ROP is caused by low signal-to-noise ratio, where the two-level signal cannot be effectively recovered by equalization. Compared with direct detection, the EVM performance degrades relatively slowly at low ROP when using EML coherent receiver owing to the existence of the DFB-based LO. As the DFB laser is left unbiased, the EAM direct-detection receiver provides a sensitivity of −11.8 dBm at the EVM threshold of $12.5\%$. In comparison, a sensitivity improvement of 5 dB is obtained by utilizing an EML receiver, evidencing the sensitivity gain obtained by coherent detection. The 18-GHz PIN shows higher responsivity compared with the EAM photodiode. Besides, the reception performance of the PIN with lower noise floor is slightly better than that of the EML receiver at high ROP. Even so, the EML receiver still outperforms the PIN photodiode with 1.5-dB sensitivity superiority. As we can see, the achievable sensitivity is not comparable to other works using EML-based coherent receivers [18,19], but it can be further improved if an EML is co-integrated with a TIA circuit or operates at short wavelengths. Figure 9(b) shows the electrical spectra of the DSM signals received by the EML and EAM receivers at the ROP of −13 dBm, respectively. The frequency response of the system composed of the 12.5-Gb/s MZM and different receivers is illustrated in Fig. 9(c), from which we can find that the high-frequency response is significantly improved by virtue of light injection when using the EML-based coherent receiver, resulting in a 3-dB end-to-end system bandwidth of 14 GHz and a quite open eye diagram.

 

Fig. 9. (a) EVM versus ROP when using different receivers. (b) The electrical spectra of the DSM signals received by the EML and EAM receivers at the ROP of −13 dBm. (c) Measured frequency response of the system with different receivers at OB2B case.

Download Full Size | PDF

In addition, the performance comparison between OFDM and DSM signals is also conducted. Figure 10(a) shows the EVM and locking range versus MZM bias voltage at a fixed ROP of −9 dBm when the 400-MHz bandwidth OFDM signal is transmitted at a 3.5 GHz carrier. The half-wave switching voltage $V_\pi$ of the MZM is roughly 1.8 V with the quadrature and null point of 1.4 V and 2.3 V, respectively. The insets present the time-domain waveforms of the received OFDM signals at different bias voltages. When the MZM is biased at 1.4 V, the relative power of the signal sideband is low, resulting in a poor EVM performance. As the bias voltage increases, the EVM performance gets better with the increased signal power, but at the cost of a narrower locking range due to lower carrier power. In the following comparative experiment, the OFDM signal is transmitted with bias voltages from 1.6 V to 1.9 V stepped by 0.1 V, where the optical carrier is gradually suppressed but not totally suppressed. For the DSM signal, a high-pass filtering scheme with 0.35-GHz cutoff frequency is adopted and the MZM bias is set to 1.4 V to obtain the locking range of 1.48 GHz at an injected power of −9 dBm, which is comparable to that of the OFDM signal. Figure 10(b) shows the EVM versus loss budget for the filtered DSM and OFDM signals, varying ROP from −9 to −17 dBm. As the MZM is biased away from quadrature point, the sensitivity improves at the expense of the reduced launch power when transmitting the OFDM signal. Compared to the quadrature bias case, there exist 2.2, 4.06, 6.01, 9.8 dB power penalty with the bias voltages from 1.6 V to 1.9 V, respectively. As a result, a maximum loss budget of 20 dB is achieved in the DSM-based digital transmission system with a 5-dB improvement over the OFDM analog system, enabling a 1:32 split fronthaul network over a 25 km reach. Furthermore, the achievable EVM reaches $1.34\%$ at ROP of −9 dBm when DSM is employed, which obviously prevails over the ARoF to support higher order modulation formats.

 

Fig. 10. (a) EVM and locking range versus MZM bias voltage when transmitting the OFDM signal. (b) EVM versus loss budget for the filtered DSM and OFDM signals.

Download Full Size | PDF

5. Conclusion

In this paper, we experimentally demonstrate a cost-effective mobile fronthaul scheme based on a band-pass delta-sigma modulator and an EML-based homodyne coherent receiver. We validate the feasibility of injection-locked operation through the center carrier tone of the DSM signal. To increase the locking range, we also propose and justify two modified schemes for the DSM signal without introducing significant sensitivity penalty. Moreover, we optimize the EAM bias voltage and DFB bias current of the EML-based receiver to obtain superior sensitivity. The experimental results show that, 1) the EML-based coherent receiver achieves better receiver sensitivity than a PIN photodiode; 2) compared to the ARoF system, the DSM-based digital transmission system provides a 5-dB improvement in loss budget by means of equalization techniques. These results indicate that the proposed coherent DRoF scheme not only improves receiver sensitivity, but also enhances the resistance to transmission impairments. Note that the receiver sensitivity can be further improved if an EAM with higher responsivity or an EML co-integrated with a TIA circuit is available. To further take advantage of the EML serving as a transceiver, full-duplex transmission with DSM-based downlink and analog uplink is to be investigated in future work.