Real-Time Experimental Demonstration of Hybrid FSO/Wireless Transmission Based on Coherent Detection and Delta-Sigma Modulation

For the fronthaul link in the fifth-generation (5G) wireless networks, optical fiber is one of the most widely used medium, owing to its large bandwidth and low transmission loss. However, in some scenarios, especially the dense urban areas, the installation of fiber optic cables is costly and difficult to establish. Alternatively, the free space optical (FSO) link can be installed easily and spans complex terrain. In this context, we propose and experimentally demonstrate a hybrid FSO/Wireless transmission system based on bandpass delta-sigma modulator and real-time coherent transponder. Advantage of the proposed technique is twofold. Firstly, coherent detection is applied to improve the system sensitivity compared to conventional direct detection. Secondly, delta-sigma modulation (DSM) is utilized to greatly simplify the wireless transmitters and to enhance the bandwidth flexibility. Over a 50-m FSO and 1-m wireless link, a 64-QAM wireless signal at 1 GHz with 500 MHz bandwidth is successfully transmitted. With 16 wavelength division multiplexing channels, the total throughput of the system is up to 24 Gb/s. Meanwhile, the error vector magnitude (EVM) of all channels is around 2.68% and 6.51%, which are demonstrated after FSO and FSO/Wireless transmission, respectively. The results meet the requirement of the third-generation partnership project release 15 with EVM value of 8% for 64-QAM. Moreover, sensitivity of the real-time coherent receiver is lower than −48 dBm under the 7% forward error correction limit with bit error rate of 3.8e-3.


I. INTRODUCTION
I N RECENT years, the fifth-generation (5G) wireless network has been developed rapidly throughout the world to serve Manuscript  incremental number of end users with emerging high-speed applications, such as virtual reality (VR), augmented reality (AR), 4K/8K ultra-HD video streaming, and internet of things [1], [2]. These applications have put forward higher requirements for capacity, data rate and coverage upon the 5G wireless networks [3]. Therefore, higher density of 5G base stations (BSs) is predictable, which brings some challenges to the fronthaul part of the network [4], [5]. At present, the application of fiber is one of the most suitable solutions as the fronthaul link for satisfying the bandwidth and latency requirements [6]. Nevertheless, in some scenarios such as dense urban areas and cross-river areas, the installation and deployment of fiber are limited by high cost and difficulty of construction [7]. In these cases, communication system based on free space optical (FSO) links has attracted great attention [8], [9], because it transmits signal through free space, which can effectively solve the problem of fiber installation. Moreover, transmitting signal in FSO links instead of fiber may bring many other advantages, such as free bandwidth licenses, lower cost, and immunity to electromagnetic interference [10]. Therefore, integrating FSO links as an alternative medium to fiber in part of the fronthaul link is a possible technical solution [11], [12].
Recently, many hybrid FSO/Wireless structures have been proposed for 5G wireless transmission networks. In [13], [14], [15], [16], a bidirectional fiber-FSO 5G wireless convergent system has been proposed. In [17], [18], [19], [20], the transmission of 64-QAM and 256-QAM signals with different bandwidths have been experimentally demonstrated in a hybrid FSO/Wireless system. In [21], two distinct hybrid architectures for 5G New Radio (NR) Fiber-Wireless (FiWi) systems applying FSO or fiber-based fronthaul have been experimentally evaluated. However, the FSO link is a line-of-sight (LOS) path and will be severely affected in practice by outdoor atmospheric conditions, such as rain, fog and dust [22], leading to severe performance degradation, especially power fading of signal. To diminish the impact of these factors, it is a good way to increase the sensitivity of the FSO system. Therefore, we believe that, compared to traditional intensity modulation/direct detection (IM/DD) FSO systems, the sensitivity improved coherent reception can offer a superior performance in practical FSO systems.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ In parallel, a technique called delta-sigma modulation (DSM) used in wireless fronthaul applications has attracted great attention recently [23], [24]. With the DSM technology, main functions of wireless fronthaul application can be fulfilled in the transmitter, and no additional components are required at the antenna side such as Mixer and digital-to-analog converters (DACs), which can facilitate maintenance and reduce system costs. Moreover, it can realize dynamic configuration only by changing the structure of the DSM modulator rather than changing the associated devices at the base band unite (BBU) side [25], [26]. Thus, applying DSM to the hybrid FSO/Wireless transmission system can significantly simplify the system structure and reduce costs.
Additionally, the DSM is transparent to the modulation format of the analog signal. For instance, a 3.5 Gbps single-carrier 256-QAM signal modulated at 3.5 GHz is employed to verify the feasibility of the real-time delta-sigma radio-over-fiber transmission for a low-cost 5G C-RAN downlink [27]. A carrier aggregation of 32 LTE component carriers (CCs) was realized to support 3GPP release 13 with one-bit and two-bit delta-sigma modulators. It shows that multi-carrier OFDM signals can be applied to DSM-based MFH [24]. In our previous work, we also successfully experimentally demonstrated the digitization and transmission of 65536-QAM baseband OFDM signal with sampling rate of 1.25 GSa/s over 20-km fiber with a variant of DSM [25]. The above work proves that the DSM technique can be used for various signals including single-carrier and multi-carrier signals without any hardware modification.
In this paper, we experimentally demonstrate a real-time hybrid FSO/Wireless transmission system based on coherent detection and bandpass DSM modulator. The original singlecarrier 64-QAM signal with the center frequency of 1 GHz and the bandwidth of 500 MHz is modulated by a bandpass DSM modulator and shaped into a 10 Gb/s 1-bit digital signal. It is then coupled to dual polarization IQ modulator (DP-IQM) and transmitted with 2.5 GBaud/s Polarization-multiplexed Quadrature Phase Shift Keying (PM-QPSK). After transmission over part of a single mode fiber (SMF) and a 50-m FSO link, the digital signal is received by a self-developed coherent transponder and then recovered real-time in a field programmable gate array (FPGA). After FSO transmission, sensitivity of the real-time coherent receiver achieved as low as less than −48 dBm at the bit error rate (BER) threshold of 3.8e-3. As for the wireless transmission, the recovered digital signal is transmitted through the Tx antenna. Over a 1-m wireless link, the 64-QAM signal received by the Rx antenna can be successfully recovered. In addition, wavelength division multiplexing (WDM) technology is also applied to increase the total data capacity. Through 16 WDM channels with a frequency spacing of 0.2 nm, the total throughput for wireless transmission reaches 16 × 1.5 = 24 Gb/s after the hybrid FSO/Wireless links.

A. The Architecture of FSO/Wireless System
A common application scenario of the hybrid FSO/Wireless transmission system is shown in Fig. 1. Generally, the original  signal is generated at the BBU side. It is then modulated and launched into the optical fronthaul link, which consists of fiber and FSO-based links. In this scenario, conventional fiber-based links can be used in normal areas where fiber optic cables are easy to install and route. While FSO link is commonly deployed in environments where the use of fiber optic cables is expensive or hard to build. For instance, in Fig. 1, the FSO links transmit signal outdoor among buildings. Finally, the recovered signal is sent to the antenna and wirelessly transmitted to users in respective regions.
Two typical hybrid FSO/Wireless architectures based on digital and analog signal are demonstrated in Fig. 2(a) and (b), respectively. As shown in Fig. 2(a), digital-based fronthaul link mainly employ common public radio interface (CPRI) with the advantages of high reliability, flexibility, and robustness to nonlinear effects. However, the limited bandwidth, low transmission efficiency, cost and complexity of devices at the remote radio unit (RRU) end restrict the use of CPRI. While as shown in Fig. 2(b), signal is transmitted over the fronthaul link in the analog form, namely analog RoF (A-RoF). The analog signal is generated and upconverted at the BBU. In this way, no additional DAC and up-conversion device are required at the RRU. It simplifies the hardware structure of the RRU and improves the spectral efficiency. However, A-RoF is susceptible to nonlinearity effects, which leads to higher and stricter requirements for the analog electronic equipment and devices, meaning increased cost. Moreover, direct detection with limited sensitivity results in low power budget for conventional systems, which leads to short FSO transmission distance and weak environmental tolerance.
To conquer the drawbacks of the conventional scheme (A) and (B), we propose the DSM-based coherent communication scheme. Compared with the conventional hybrid FSO/Wireless transmission system, the proposed scheme improves the system in two aspects, as shown in Fig. 2(c). Firstly, it utilizes real-time coherent detection instead of direct detection in the FSO system. Secondly, it transmits the DSM encoded digital signal in the fronthaul part. The most important advantage of a coherent detection is its higher receiver sensitivity, which brings in larger power budget and longer transmission distance. A comparison of sensitivity between the two detection methods is shown in Table. I. Moreover, demultiplexing the WDM signal leads to an even larger capacity. Although the coherent receiver causes an increase of cost and power consumption compared with IM/DD, due to the higher requirements for sensitivity and data rate of the WDM FSO communication system, the application of coherent technology is more valuable and attractive to this scenario than IM/DD. Additionally, with the advances in DSP, photonic integration technologies and optical devices, the size, power consumption and cost of the coherent system have been gradually decreased in recent years [29], [30], [31].
On the other hand, the DSM-based fronthaul system mainly combines the advantages of CPRI and A-RoF. Firstly, similar to A-RoF, the main functions such as up-conversion and DSM are centralized in the BBU which helps maintain the simple structure at the RRU side. Only a band-pass filter (BPF) is required to retrieve the original analog waveform before wireless transmission with the antenna. Secondly, the use of DSM technology can convert the analog signal into a 1-bit digital signal, only a high-speed digital port is required to send out the data, as shown in Fig. 2(c). Thirdly, similar to CPRI, the DSM-based transmission will be less affected by nonlinear effects. It means a lower requirement on devices than A-RoF which can reduce the cost of the overall system. Finally, by adjusting the parameters of the DSM structure, it is easy for us to modulate the IF signals with different bandwidths and central frequencies for dynamic configuration. It also lowers the cost and requirements at the BBU side. An overall comparison of the three different schemes is also summarized in Table. II.
Additionally, compared to the A-RoF scheme, the residual quantization noise of DSM leads to lower SNR. However, in this paper, the implemented DSM structure with oversampling ratio (OSR) of 10 can support SNR greater than 31 dB. Therefore, the proposed DSM-based MFH can support most of the wireless services which employs 64-QAM and 256-QAM (with EVM requirement of 8% and 3.5% respectively [32]). It means the proposed scheme can reduce system cost and complexity while  still maintaining good performance. Furthermore, as for the spectral efficiency of the DSM-based scheme, a previous work has shown the comparison with CPRI-based scheme [24]. As for 20 MHz

B. Principle of Bandpass DSM Modulator
DSM is a novel technique for fronthaul transmission. DSM converts analog signal to digital signal, which can be generated by a digital transceiver port. Fig. 3 illustrates the detailed principles of a bandpass DSM modulator. Firstly, the analog IF signal is oversampled to expand its Nyquist zone, which causes the quantization noise to be evenly distributed over a wider frequency range, as shown in Fig. 3(a). Then, the noise shaping technique pushes more quantization noise out of the signal band, leading to less residual noise within the signal band. Meanwhile, the original analog signal will be converted into a 1-bit digital signal with the sampling rate of the DSM, as shown in Fig. 3(b). The DSM encoded digital signal can be transmitted with only digital ports. At the receiver side, a corresponding BPF can be used instead of a DAC to recover the desired signal from digital to analog, as shown in Fig. 3(c). Structure of the 4th-order bandpass DSM modulator used in this article is shown in Fig. 3(d). By adjusting the parameters a1-a4 and g1-g2, we are able to control the passband position and width of the bandpass DSM modulator. As for the wireless fronthaul application, it means that dynamic configuration can be realized in the transmitter using the DSM technology. Frequency response of the noise transfer function (NTF) of the bandpass DSM modulator corresponding to the parameters in Fig. 3(d) is also shown Fig. 3(e). Meanwhile, the signal spectrum after DSM is shown in Fig. 3(f). The center frequency, bandwidth, sampling rate of the bandpass DSM modulator are 1 GHz, 500 MHz and 10 Gb/s. The oversampling ratio (OSR) is 10.

III. EXPERIMENT SETUP
Experimental setup of the proposed hybrid FSO/Wireless structure is shown in Fig. 4. The red and blue links in the diagram represent the transmission link of optical and electrical signals, respectively. Technologies such as WDM, DSM and real-time coherent reception are converged into the hybrid architecture to ensure seamless connections between fiber, FSO and wireless links, enabling the system to achieve higher reception sensitivity and larger communication capacity as much as possible. We intend to transmit a 64-QAM IF signal at 1 GHz with the bandwidth of 500 MHz in the hybrid architecture for largest possible communication capacity under the common SNR constraints of transceiver and DSM.
In the transmitter, a 16-channel massive laser array with 0.2 nm wavelength spacing and ∼75 KHz linewidth is used as the source. The optical power of each channel is set as 13 dBm and the wavelengths are set from 1548.4 to 1551.4 nm. After a 16:1 coupler, the generated lasers are coupled and then injected into a dual polarization IQ modulator (DP-IQM) with the output optical power of −2 dBm. Additionally, a self-developed circuit board integrates DP-IQM and FPGA (Inter Altera Stratix V) on the transmitter side. The original 64-QAM signal for wireless  transmission is shaped by the DSM and converted into a 1-bit digital signal. The modulated digital signal is stored in the random-access memory (RAM) of the FPGA and then coupled to the DP-IQM input through the digital port instead of DACs. The structure of the 4th-order bandpass DSM modulator and its frequency response are shown in Fig. 3(d) and (e). Additionally, the details on how the DSM bits are assigned into PM-QPSK constellations are shown in Fig. 5. The PM-QPSK constellation consists of four parts: Xq, Xi, Yq and Yi. The 1-bit sequence of DSM signal is serial-to-parallel assigned into the four parts of PM-QPSK in order, as shown in Fig. 5.
Afterwards, the modulated WDM optical signal is launched into SMF for transmission and then coupled into the FSO Tx collimator, as shown in Fig. 6(a). The 50-m FSO link is placed in a corridor shown in Fig. 6(b), consisting of two terminals which enables full-duplex communication, implementing the FSO application. In order to complete the capture and alignment process of the two terminals with the highest possible accuracy, we use the charge coupled device (CCD) to acquire the images of beacon light and calculate the pixel coordinates and deflection angles. Then, a proportion-integral-differential (PID) controller is employed for the optical gimbal to change the deflection angle and correct the pointing of the optical antenna. Finally, the platform and fine steering mirror (FSM) are adjusted in this experiment for precise aiming and optimal coupling efficiency. Noteworthy, the CCD (Xenics Bobcat-320-GigE-11856) is located at the Rx platform of the FSO system during the establishment of FSO link, as shown in Fig. 6(b). Due to the limitation of area in the hallway, some active devices like CCD are removed from the platform after the FSO link is established. Additionally, the light beam divergence angle is ∼0.1 mrad and the transmitting beam size is ∼7 mm. After 50-m FSO link, the receiving beam size is about 12 mm. By using an optical antenna with 15mm diameter, the space beam can be well coupled into the fiber. With additional consideration of the pointing error of −60 ∼ +60 μrad, the total FSO link loss is about 10 dB, which means it can support an even longer FSO link in our system.
At the reception, an integrated coherent receiver (ICR), local oscillator laser (LO), analog-to-digital converter (ADC) and FPGA are also integrated in a self-developed circuit board which is shown in Fig. 7(a). The optical beam is captured by the Rx collimator and coupled into the fiber after transmission via the 50-m FSO link. The fiber-coupled optical signal is attenuated by a variable optical attenuator (VOA) and amplified by an erbium doped fiber amplifier (EDFA) before received by the ICR. A VOA is used to emulate the loss to achieve a limit sensitivity test. The EDFA is used to amplify the optical signal and regulate its power in the appropriate reception range of the ICR. Sequentially, the detected signal is sampled by four 6-bit 5 GSa/s ADCs with an analog bandwidth of 6 GHz. The sampled signal is sent to FPGA for real-time DSPs and recovery. The realtime algorithms include ADCs synchronization, clock recovery, constant modulus algorithm (CMA), frequency offset recovery, phase recovery and symbol decision, as shown in Fig. 7(b). Among the algorithms, the Gardner algorithm is utilized for clock recovery. The number of taps for the CMA equalizer is 5. The frequency and phase recovery are based on Viterbi-Viterbi (V-V) algorithm [34]. Noteworthy, the architecture of the main real-time DSPs including CMA, frequency recovery and phase recovery are completed in the Altera DSPbuilder program which is developed by Quartus in Matlab and also shown in Fig. 7(b). Furthermore, the chip planner of the main real-time DSPs in FPGA is shown in Fig. 7(c). Finally, the recovered signal after decision is sent out in the digital form through the transceiver port on the circuit board. In order to analyze the performance of the output signal which can be considered as a communication result after the FSO link, we use a 99:1 power divider to split a part of the signal and send it to the oscilloscope-1 (OSC-1) for data processing and recovery including bandpass filtering, synchronization, down-conversion and symbol decision.
The output signal is transmitted wirelessly through a Tx antenna. As shown in Fig. 6(c), after 1-m wireless link transmission and antenna reception, the signal is acquired and analyzed by an OSC-2 working at 20 GSa/s, which can be considered as a result of the hybrid FSO/Wireless transmission experiment. Note that, both in the data analyses of OSC-1 and OSC-2, we implemented the EVM threshold in accordance with the 3GPP Release 15 recommendations [32], which requires maximum EVM value of 8% for 64-QAM. Additionally, the key system parameters about DSM and experiment scenes are summarized in Table III. IV. RESULTS AND DISCUSSIONS Firstly, the output optical spectrum of the 16-channels WDM signal is shown in Fig. 8(a). The 16 wavelengths from Ch.01 to Ch.16 are set from 1548.4 to 1551.4 nm, spacing with 0.2 nm. Then, the VOA is used to simulate the additional link loss to complete a sensitivity test of the real-time coherent receiver. As shown in Fig. 8(a), three channels (Ch.01-1548.4 nm, Ch.09-1550.0 nm and Ch. 16-1551.4 nm) located at the two ends and middle of the 16-channel WDM signal are chosen to illustrate the BER performance versus received optical power (ROP) of the coherent receiver. The ROP with per channel of the receiver is adjusted from −41 to −48 dBm. With the decrease of ROP, the BER performance of each channel is approaching the 7% overhead hard-decision forward error correction (HD-FEC) threshold (BER = 3.8e-3) and reaching a critical point with ROP of −48 dBm. It can be seen that the sensitivity of the coherent receiver for each channel is better than −48 dBm. Additionally, the recovered constellation diagrams of Ch.09 with ROP at −41 dBm and −48 dBm are shown in Fig. 8(b) and (c), respectively.  Noteworthy, the BER threshold of 3.8e-3 indicated in Fig. 8(a) is considered as a standard [35] for performance evaluation of optical communication receiver in this experiment. If a typical 7% overhead HD-FEC decoder is implemented on the FPGA receiver, the decision errors from QPSK constellation can be well corrected, as long as it is under the corresponding BER threshold. Such function is also commonly implemented by taking well-developed IP cores [36] on the FPGA.
Since no actual FEC decoder is implemented, the impact of error propagation from the received QPSK constellation into the DSM signal and finally the 64-QAM wireless signal cannot be ignored. Therefore, we carry out relevant investigations and analysis about the error propagation relationship between the two signals without FEC decoder, as shown in Fig. 8(d). In Fig. 8(d), with the analyzed data of Ch.09 (1550.0 nm) and ROP from −41 to −48 dBm, the blue curve represents the change of BER of 64-QAM signal and the red one corresponds to the EVM of the constellation. As for the BER performance of 64-QAM, it varies linearly with the BER of QPSK. For EVM results, the decision errors will cause scattered and dislocated points around the 64-QAM constellation in the reconstruction process.
When the BER of QPSK is low, the recovery of 64-QAM constellation is almost unaffected, and the value of EVM remains stable. Further increasing the QPSK-BER (about > 1e-5), the reconstruction process such as filtering and equalization will be impacted, and the EVM of 64-QAM will increase rapidly.
Then, we choose Ch.09 to analyze the performance of the system in different scenarios. The analysis results of constellation diagram and spectrum are shown in Fig. 9(a)-(c). They demonstrate the signal performance for system transmitter, FSO and FSO/Wireless, corresponding to the mark (a)-(c) in Fig. 4, respectively. Fig. 9(a) shows the signal at the transmitter end after delta-sigma modulation. The EVM value of the original 64-QAM is 2.67% (SNR is 31.5 dB). Subsequently, after an FSO fronthaul link, the recovered constellation diagram and spectrum are shown in Fig. 9(b), with EVM of 2.68% and SNR of 31.4 dB. Comparing with the results of Fig. 9(a) and Fig. 9(b), we can see that the signal performance after FSO link is similar to the transmitter side with only a little difference, due to few emerging errors. Finally, via a 1-m wireless link, the signal is received by OSC-2. After analyzing, the constellation diagram and spectrum are shown in Fig. 9(c), with EVM of 6.57% and  SNR of 23.7 dB. All of the results are below the 8% EVM requirements. Furthermore, the transmitted and received QPSK constellation (X-Polarization) and respective spectra are also shown in Fig. 9(d), corresponding to the mark (d) of setup in Fig. 4.
Finally, the EVM value of the recovered 64-QAM signals of all channels are shown in Fig. 10. With a working state at ROP of −41 dBm, the EVM value of 64-QAM signal after FSO link is ∼2.67%, and it is ∼6.51% after FSO/Wireless link. All the test results meet the 8% EVM requirement of 64-QAM after transmission. Additionally, a real-time test result of signal transmission is shown in Fig. 11. After one hour of continuous experiments, data are recorded every three minutes. In both the FSO link and FSO/Wireless link, the 64-QAM signal can be recovered continuously and stably, with EVM of ∼2.67% and ∼6.50% respectively.
Noteworthy, the experiment results are obtained in an indoor environment which is a relatively stable scene. When the system is set outdoors, the impact of beam wandering and scintillation must be considered. Firstly, for the beam wandering, it is a function of the wavelength λ and distance L [37], as follows (1): In (1), the refractive index structure constant C 2 n will change with the climate, and the value range is generally 10 −17 ∼ 10 −13 m −2/3 . If an outdoor FSO link distance of 50 m and C 2 n takes the maximum value in the interval, the offset of the beam is about 0.34 mm around the target position which is small and can be corrected by PID or kalman tracking algorithm. Secondly, for the scintillation effect, it generally occurs when the beam diameter is larger than the turbulence scale. However, the FSO link distance between the buildings is not too long, usually within the range of 50∼500 m. The impact of scintillation effect on receiving power will be very small. Therefore, for the propagation FSO link loss, it is mainly affected by the acceptance end spot size [37], as follows (2): In (2), L S is the space loss, a is the effective receiving area, ω is the acceptance end spot size and ω = z · θ/2. z is transmission length and θ is beam divergence angle. Therefore, considering the outdoor FSO length range of 50∼500 m, the total link loss is mainly affected by the controlling accuracy and can be controlled at 14.53∼22.24 dB. Additionally, we reserve a 2dB margin for power attenuation due to the possible scintillation effect. Regardless of extreme weather conditions like fog, dusty and rainstorm, the FSO link loss under outdoor conditions is about 16.53∼24.24 dB, which can still meet the system requirements. Furthermore, the high sensitivity of coherent receiver (∼ −48 dBm) can also provide sufficient power budget to adapt to the fluctuating outdoor environment than IM/DD (∼ −27 dBm).

V. CONCLUSION
This paper successfully demonstrates a real-time hybrid FSO/Wireless communication system. With the self-developed coherent transponder, the 64-QAM signal with center frequency of 1 GHz and bandwidth of 500 MHz is transmitted in real-time through a 10 Gb/s QPSK FSO link. Noticeably, the sensitivity of the proposed real-time coherent receiver is better than −48 dBm at the BER threshold of 3.8e-3. Compared with the direct detection, it can provide sufficient power budget to cover the loss of spatial optical channel. Furthermore, 16-channel WDM technology is used to increase the total throughput of the transmission system. Finally, over a 50-m FSO link and 1-m wireless link, the total data capacity reaches 16 × 1.5 = 24 Gb/s. Experimental demonstration shows that the combination of coherent detection and DSM would be a promising technique for future optical-wireless scenario.