Real-time high-bandwidth mm-wave 5G NR signal transmission with analog radio-over-fiber fronthaul over multi-core fiber

This article presents an experimental demonstration of a high-capacity millimeter-wave 5G NR signal transmission with analog radio-over-fiber (ARoF) fronthaul over multi-core fiber and full real-time processing. The demonstration validates the core of the blueSPACE fronthaul architecture which combines ARoF fronthaul with space division multiplexing in the optical distribution network to alleviate the fronthaul capacity bottleneck and maintain a centralized radio access network with fully centralized signal processing. The introduction of optical beamforming in the blueSPACE architecture brings true multi-beam transmission and enables full spatial control over the RF signal. The proposed ARoF architecture features a transmitter that generates the ARoF signal and an optical signal carrying a reference local oscillator employed for downconversion at the remote unit from a single RF reference at the central office. A space division multiplexing based radio access network with multi-core fibre allows parallel transport of the uplink ARoF signal and reference local oscillator at the same wavelength over separate cores. A complete description of the real-time signal processing and experimental setup is provided and system performance is evaluated. Transmission of an 800 MHz wide extended 5G NR fronthaul signal over a 7-core fibre is shown with full real-time signal processing, achieving 1.4 Gbit/s with a bit error rate <3.8×10-3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$<3.8\times 10^{-3}$$\end{document} and thus below the limit for hard-decision forward error correction with 7% overhead.

the availability of multiple independent fiber channels with very similar transmission characteristics.
Combining SDM transport in the ODN with ARoF fronthaul and optical beamforming is the core technology proposition of the EU H2020 5G-PPP project blueSPACE [14]. A successful demonstration of ARoF fronthaul for high-bandwidth mm-wave 5G NR signals using full real-time processing in an ARoF baseband unit (BBU) and multi-core fiber (MCF) transport was demonstrated and discussed in a paper at EuCNC 2020 [15]. This manuscript expands upon this demonstration, discussing the full blueSPACE architecture and providing a detailed description of the concepts and experimental setup, especially the ARoF BBU and real-time digital signal processing (DSP) which are described in detail for the first time. The demonstrated ARoF fronthaul link achieves real-time transmission of 800 MHz wide extended 5G NR orthogonal frequency division multiplexing (OFDM) signals over 10 km of 7-core MCF and 9 m mm-wave wireless at 26.5 GHz, i.e., in the 3GPP FR2 n258 band [6]. The link further used a remote-fed LO for signal downconversion at the RU, avoiding the need for any RF oscillators at the remote site and maximizing centralization of network resources. Together these prove the viability of ARoF fronthaul and demonstrate the core concepts of the blueSPACE fronthaul architecture.
The remainder of this manuscript is structured as follows: Sect. 2 introduces the blueSPACE fronthaul architecture, while Sect. 3 discusses the demonstration setup, providing a detailed description of the real-time DSP in the ARoF BBU (Sect. 3.1) and introducing the experimental setup (Sect. 3.2). Section 4 discusses the performance of the ARoF transmitter (Sect. 4.1) and provides transmission performance measurements for the overall real-time ARoF fronthaul link (Sect. 4.2). Finally, Sect. 5 summarizes and concludes the manuscript.

Analog fronthaul with optical beamforming for mm-wave 5G
The centralization of resources in RANs has provided substantial reductions in the cost of network ownership, operation and maintenance, by removing baseband (BB) processing from remote sites and centralizing it in a central office (CO) [16]. This centralization of BB processing resources in a BBU pool at the CO allows C-RAN deployments to use statistical multiplexing and sharing of processing resources to avoid overprovisioning. It further simplifies the management and maintenance of the network by reducing complexity placed at remote locations and at the antenna sites. The downside to this centralization of resources is the introduction of a new RAN segment between the BBU at the CO and the RU at the antenna site, referred to as the fronthaul segment [9,16].
In current 4G LTE networks where bandwidths and frequencies are limited, this segment is covered by transmitting digitized in-phase and quadrature (IQ) samples of the target radio waveform using traditional BB optical links, detecting the signal at the RU and performing digital-to-analog conversion there (and vice versa for uplink (UL) direction). Such transmission of digitized samples, typically using CPRI and referred to as digitized radio-over-fiber (DRoF), however requires constant bitrate signals, where the latter scales with signal bandwidth and antenna configuration, is independent of user presence or activity, and is highly sensitive to latency and jitter [17]. In more complex deployments using multi-sector antennas, MIMO transmission or larger aggregated bandwidths, the data rates required for such fronthaul links quickly scales to hundreds of Gbit/s and into the Tbit/s. As a result, they are clearly not a scaleable solution of highbandwidth 5G signals, especially at mm-wave where radio bandwidths will be substantially larger than in the sub-6 GHz spectrum.
Two opposite approaches have emerged to address this fronthaul bottleneck for 5G networks, either remaining with DRoF fronthaul, but moving some functionality/processing steps back to the RU or to some intermediate distributed unit (DU), resulting in different functional splits and giving rise to new protocols such as enhanced common public radio interface (eCPRI) or next generation fronthaul interface (NGFI), or increasing centralization and transitioning to analog fronthaul with ARoF [11,18]. The latter increases centralization by also moving the digital to analog converter (DAC) and analog to digital converter (ADC) from the radio sites to the CO and relies on optical transport of the target RF or intermediate intermediate frequency (IF) waveforms [11,19]. If the RF waveform is transported or optical heterodyning is used for IF to mmwave upconversion, ARoF further allows centralizing all RF oscillators and simplifies the mm-wave generation process [20]. ARoF also minimizes the required optical spectrum in the ODN, as the required bandwidth matches the RF bandwidth-although additional bandwidth is needed if optical heterodyning is used, which in turn may be reduced by interleaving channels, e.g., in ultra-dense wavelength division multiplexing (UD-WDM) passive optical networks (PONs) [21,22].
In addition, ARoF ideally combines with optical beamforming, where analog beamforming is performed in the optical domain, rather than in the RF domain after optical to electrical (O/E) conversion. This brings multiple advantages, including a reduction in spatial, power and cost footprints as well as the possibility to implement full Blass matrices in integrated OBFNs, allowing true multi-beam transmission from a single phased array antenna (PAA) and affording good spatial control over the mm-wave RF signals [23,24].
In blueSPACE, an ARoF fronthaul architecture with optical beamforming is combined with SDM in the ODN to truly solve the capacity bottleneck in the fronthaul segment. The introduction of MCF as fibre medium for the ODN brings a direct increase in capacity, but also allows another degree of flexibility when it comes to resource allocation or network sharing. The blueSPACE ARoF architecture is shown in Fig. 1, with the ARoF BBU pool and IF unit, the ARoF fronthaul segment with OBFNs for both downlink (DL) and UL as well as the radio equipment with amplifiers and antenna array.
When it comes to the use of optical beamforming in ARoF fronthaul as in the blueS-PACE architecture, the placement of the OBFN has significant consequences for the fronthaul transport and may pose specific requirements to the ODN. This is due to the functionality of the OBFN which maps each of its M independent input signals (in DL direction destined for M independent beams) onto each of its N outputs with different differential delays, allowing to independently modify each beam's direction while transmitting all M beams from a single N-element antenna array. As the temporal relation between the N output signals of the OBFN carries the beamforming information, these relations must be precisely preserved from the OBFN to the antenna array element, with the required precision determined by the RF frequency-e.g., in the case of mmwave transmission around 30 GHz, temporal relation between the beamformed signals must be maintained within ≈ 1 ps to avoid significant deformation of the beams [25]. In UL direction, the same is true and temporal alignment must be maintained from the antenna array to the N inputs of the OBFN, which de-maps the received signals and spatially filters them to output the signal from M different received beams on its M outputs. Figure 1a shows the first option, with the OBFNs for DL and UL located with the corresponding ARoF transmitters, i.e., the DL OBFN at the CO and the UL OBFN at the RU. While this configuration allows centralization of the DL OBFN and thus maintains a simpler RU, it does require signals to remain temporally aligned across transport in the ODN. The latter is a major challenge and would be impossible to achieve using separate single-mode fiber (SMF), fibre bundles or single-mode fiber (SMF) ribbon cables and even with the use of MCF, which does provide better temporal alignment, may be problematic [26]. For the DL direction, the same would apply if the OBFN were to be centralized, however with placement at the RU, as shown in Fig. 1, this issue does not arise as the entire critical path-from the antenna array, across the downconversion stage and until the output of the OBFN-is located at the RU. Furthermore, in this scheme the number of channels that are transported across the ODN for DL and UL is asymmetric, with the number of DL channels scaling with the number of antenna elements N and the number of UL channels scaling with the number of beams M . As typically N > M , often substantially so, this poses questions to network occupation and may ultimately limit the size the antenna array can be scaled to, as it may not be possible to equally scale the number of cores in the MCF.
To avoid the necessity to maintain temporal alignment across the ODN, the DL OBFN may be placed at the RU, as shown in Fig. 1, which results in a symmetric number of   Fig. 1 blueSPACE ARoF fronthaul architecture with optical beamforming in both DL and UL: a with DL OBFN at the RU, avoiding signal synchronization issues, but requiring N-core MCF DL, b with DL OBFN placed at the CO and symmetric M-core DL and UL MCF transport. CO: central office, ODN: optical distribution network, RU: remote unit, BBU: baseband unit, IQ: in-phase and quadrature, IF: intermediate frequency, RF: radio frequency, ARoF: analog radio-over-fiber, DL: downlink, UL: uplink, MCF: multi-core fiber, PD: photodiode, TIA: transimpedance amplifier, OBFN: optical beamforming network, PA: power amplifier, LNA: low-noise amplifier channels for DL and UL, both scaling with the number of beams M . There is thus a clear trade-off between complexity at the RU through having the OBFN there and requirements on the ODN. It should further be noted that, if the OBFN is based on optical phase shifting and optical heterodyning is used for mm-wave upconversion, temporal relation may need to be maintained starting at the ARoF transmitter, unless both signal and second tone are transported in the same fiber. Thus, the scheme in Fig. 1b may need to be upgraded, e.g., though the introduction of an additional O/E/O step between MCF and OBFN to guarantee signal alignment while also potentially improving link power budget, but at the cost of increased complexity, cost and power consumption at the RU.

High-capacity ARoF fronthaul demonstration experimental setup
To demonstrate the core concepts of the proposed ARoF fronthaul architecture, an experimental setup is established, including the ARoF BBU with real-time DSP, ARoF signal generation and optical heterodyning for mm-wave generation and the ODN with MCF for signal transmission. While optical beamforming is an integral part of the proposed architecture and improves the mm-wave link power budget through concentration of the radiated energy in a smaller area, as well as greatly enhancing system spectral efficiency (SE) by allowing denser frequency reuse, it is not included in this demonstration, as a simple point-to-point ARoF link must be shown and evaluated first. The following will first detail the architecture, signal processing chain and supported functionality of the ARoF BBU (Sect. 3.1), before discussing the overall fronthaul demonstration setup in detail (Sect. 3.2).

5G NR signal generation and real-time processing
The blueSPACE ARoF BBU core, implemented in a field programmable gate array (FPGA) device [27], provides all the DSP required to generate and handle the 5G NR OFDM signals to be transmitted. The OFDM waveform is represented in BB by two components, the in-phase (I) and quadrature (Q) components, each with half the bandwidth of the desired radio waveform. The BB signals are converted from the digital to the analog domain and vice versa by multi-channel DACs and ADCs [28,29]. The blueS-PACE BBU mainly consists of five modules, i) the digital data interface acting as data source/sink for the DSP chains, ii) the cyclic prefix (CP) OFDM transmitter side DSP datapath, iii) the CP-OFDM receiver side DSP datapath, iv) the analog data interface towards DAC and ADC, and v) the embedded central processing unit (CPU). The interconnection of these modules and their internal structure are shown in Fig. 2.

Digital interface-data source/sink
This bidirectional digital interface acts as a source of data for the transmitting side of the BBU (DL) and as a sink for the receiving side (UL). A 10 Gbit/s ethernet interface is provided for interconnection with the backhaul, while an embedded bit error rate tester (BERT) can be used for debugging and evaluation of the link performance. The BERT employs 32-bit interfaces, both for the transmitter and receiver sides and it is clocked at 245.76 MHz, reaching a peak data rate of 7864.32 Mbit/s in each direction, which is sufficient to support the peak data rate supported by the BBU. The 10 Gbit/s ethernet interface is implemented using the low latency ethernet 10G medium access control (MAC) Intel FPGA IP core [30]. The physical interface for this interface is a small form-factor pluggable (SFP)+ module (10GBASE-R), while the interface with the BBU is a bidirectional 64-bit interface clocked at 156.25 MHz, providing a peak data rate of 10 Gbit/s in each direction. Rate matching between this interface and the BBU digital interface is provided using a pair of dual-clock first-in, first-out (FIFO) memories.

Transmit DSP datapath
The transmitter DSP performs 5G NR OFDM signal generation and supports a number of transmission modes [6,31], as shown in Table 1. This module receives plain digital data, originating from either of the sources in the digital interface, and exports data samples that describe the analog BB signals towards the analog interface which drives the DACs. The incoming bits are grouped into symbols, according to the selected constellation size, using Gray coding. Supported constellations are quadrature phase-shift keying (QPSK) and quadrature amplitude modulation (QAM) (16-QAM, 64-QAM and 256-QAM) [31]. The constellation can be selected through a dedicated register in the register file, which can be updated on the fly through software. From this point on, all data is represented as complex numbers.
The resulting symbols are used to modulate the available OFDM subcarriers. The number of available subcarriers and their position in the spectrum can be programmed on the fly using a dedicated random access memory (RAM), which can be written through software. This module also provides the flexibility to set the number and position of the pilot subcarriers used for channel estimation at the receiver side. Pseudo random bit sequences (PRBSs) based on length-31 Gold sequences are used for pilot subcarriers, providing better cross-correlation properties than maximum length linearfeedback shift register (LFSR) sequences, as defined in [31]. All subcarriers, either carrying data or pilot symbols can be set in blocks of four, as four subcarriers are always processed in parallel during each clock period with a clock frequency of 245.76 MHz, which is equal to 1 4 of the maximum sampling rate of 983.04 MSa/s. The maximum number of modulated subcarriers is 3168 and corresponds to 264 resource blocks (RBs), as defined by 3GPP [6]. Modulated subcarriers (in the frequency domain) are transformed to the time-domain using a 4096-point inverse fast Fourier transform (IFFT) engine. This engine instantiates four 1024-point IFFT engines operating in parallel and combines their outputs using a radix-4 decimation in time (DIT) butterfly unit. The Intel fast Fourier transform (FFT) IP core [32] is used for the 1024-point FFT engines. The framer module introduces the CP and guard time (GT) required to form a complete OFDM symbol. CP and GT are set using dedicated registers in the register file which can be updated on the fly by software. The final transmitter DSP stage includes all processing performed in the time domain and provides a total of 30 dB of digital gain, adjustable in steps of 6 dB, before signals are output towards the analog interface.

Receive DSP datapath
The receiver DSP performs 5G NR OFDM signal demodulation and supports the same transmission modes supported by the transmitter DSP, as shown in Table 1. This module receives data samples of the incoming analog signals, as encoded by the ADCs, from the analog interface and exports the decoded digital data to the digital interface, destined for either the backhaul interface or the built-in BERT for bit error rate (BER) measurements. Rate matching with the external digital interface is again provided using a dual clock FIFO memory.
From the incoming data, the in-phase signal is mapped to the real part of the complex numbers used for further processing, while the quadrature signal is mapped to the imaginary part. The first receiver DSP stage includes all processing performed in the time domain and provides a total of −30 dB of digital gain, adjustable in steps of −6 dB . This stage compensates for the digital gain set at the transmitter side, while the combination of these two stages is used to optimize the signal-to-noise ratio (SNR) of the received signal by adjusting to link power budget conditions. The incoming signal is synchronized using autocorrelation techniques, where the incoming signals are compared with delayed versions of the same signal. This delay can be adjusted and in normal operation is equal to the length of the FFT. After synchronization, the CP is removed from the received data. The carrier frequency offset (CFO) is estimated using a maximum likelihood (ML) technique [33], while the estimated value controls a numerically-controlled oscillator (NCO), which allows the frequency to be set with a resolution of 0.05722 Hz. The CFO is then compensated by simple multiplication of the synchronized data with the output of the NCO. The CFO estimation can be entirely bypassed or combined with a value set at the proper registers in the register file for debugging purposes.
The FFT engine is similar to the IFFT engine in the transmitter DSP, also combining the outputs of four 1024-point FFT engines using a radix-4 DIT butterfly unit to transform an incoming OFDM symbol from the time domain (samples) to the frequency domain (subcarriers). All modulated subcarriers are extracted in the reverse fashion as the subcarrier modulation in the transmitter DSP, while subcarriers not carrying data or pilot information are discarded. Subcarriers carrying payload are directed to the channel equalization unit, while pilot subcarriers are forwarded to the channel estimation unit. The channel estimator compares the received pilot subcarriers with the transmitted pilot subcarriers and feeds a zero-forcing equalizer. The channel estimator, similar to the transmitter side, is equipped with a length-31 Gold sequence generator to generate the pilot symbols on the fly, according to the modulation scheme information that is written to its dedicated RAM. Although the NCO frequency resolution is ≈ 0.05 Hz , this is not small enough for the receiver to acquire and process data for several seconds without re-training the equalizer. The remaining carrier frequency offset is expressed as a tilt in the constellation that slowly rotates, making BER calculation impossible. The residual CFO compensation module continuously estimates the rotation angle and compensates it using vector rotators.
Finally, the symbol estimation/slicer estimates the received symbols by calculating the Euclidean distances between the received symbol and the used constellation and selecting the symbol with the minimum distance. The detected QAM or QPSK symbols are then de-mapped into plain bits using Gray decoding.

Analog interface
This is a digital high bit rate serial interface between the FPGA and the DACs and ADCs. Although still digitally carrying digital signals, this is the FPGA design analog interface as it coordinates the translation to/from the analog domain. The Intel JESD204B FPGA IP [34] has been used for the implementation of this interface. The DAC and ADC interfaces each use a single link JESD204B subclass 1 interface that employs four lanes operating at 9.8304 Gbit/s, reaching a total throughput of 39.3216 Gbit/s including 8b/10b encoding. The employed DAC resolution is 16 bit, while the actual ADC resolution is 12 bit, but 16 bit are sent per sample to ease future upgrades.

Embedded CPU
The CPU embedded with the FPGA allows for control and monitoring of the BBU via a ethernet and universal serial bus (USB) interfaces. The embedded CPU is a dual-core ARM Cortex-A9 hard processor system [35] running at 1.2 GHz that is implemented on the same chip with the Arria 10 FPGA device and has access to 1 GB of DDR4-2133 RAM. Communication between the FPGA and the CPU is realized with multiple advanced eXtensible interfaces (AXIs). The operating system running on the embedded CPU is Angstrom Linux.

Analog radio-over-fiber transmission setup
The ARoF fronthaul demonstration for high-capacity mm-wave 5G signals implements the core components of the blueSPACE ARoF architecture shown in Fig. 1, ARoF signal generation, MCF fronthaul transport and optical heterodyning for mm-wave signal generation-the corresponding reduced architecture is shown in Fig. 3. Beyond the blueS-PACE ARoF BBU unit described in the previous section, this contains the IF unit and ARoF transmitter at the CO, the shared MCF for transport of DL and UL ARoF signals as well as the remote-fed LO, the RU for O/E conversion and RF signal radiation to the UE. The ARoF transmitter includes an RF LO for two-tone generation, while at the RU both the mm-wave RF DL signal and the RF LO used for downconversion of the UL signal are generated through optical heterodyning, avoiding the need for an RF oscillator in the remote location.
The experimental setup employed in this demonstration implements this core ARoF architecture based on commercially available optical and RF components at the CO and RU, while the MCF and corresponding fan-in and -out modules were specifically designed and fabricated for and in the blueSPACE project. The MCF is a 7-core MCF with a central core and six outer cores positioned in a circle around the central core, as seen in the inset of Fig. 4. The MCF is designed to emulate the transmission properties of standard SMF, achieving similar chromatic dispersion ( ≈ 19 ps/nm/km ) and loss ( ≈ 0.25 dB/km ), with an overall insertion loss of ≈ 4.5 dB for 10 km MCF including the fan-in and -out modules from/to separate SMFs.
A detailed schematic of the experimental setup is shown in Fig. 4. The IF unit is the link between the signal processing in the BBU and the ARoF fronthaul link. It upconverts the DL analog BB IQ signals from the DAC in the BBU to an IF f IF in the range of 2.25 GHz to 5.5 GHz based on a tunable locally generated LO, which is also used for downconversion of the received UL IF signal into separate BB IQ signals that can be input to the ADC of the BBU. The IF unit further contains amplifiers to boost the generated DL IF signal as well as to pre-amplify the received UL IF signal before  demodulation; a pair of low-pass filters (LPFs) (not shown in Fig. 4) is used to filter the signal to remove undesired higher frequency components from the generated and received IF signals.
The ARoF transmitter is based on an external cavity laser (ECL) with a linewidth of ≈ 25 kHz and performs two-tone generation with a Mach-Zehnder modulator (MZM) biased at the null point for carrier suppression and driven with an LO at f LO . The generated two-tone signal is amplified to an output power P opt of 18 dBm by an erbium-doped fiber amplifier (EDFA) and split equally into two parts-one is directly transmitted over the MCF to the RU to be used as remote-fed LO, while the other is modulated with the amplified data-carrying IF signal from the IF unit in a second MZM. The latter is biased at the quadrature point to ensure linear transmission of the IF signal. It should be noted, that the selected frequency doubling scheme with suppression of the original carrier eliminates the problem of dispersion induced carrier fading [36], while the use of an IF avoids the issue of signal-signal beat noise [37].
The generated ARoF and LO signals are transmitted to the RU through separate cores of the MCF and through optical heterodyning on a high-speed photodiode (PD) generate the desired modulated DL mm-wave RF signal at f RF = 2f LO + f IF as well as the mm-wave LO at 2f LO used for dowconversion of the received UL mm-wave RF signal. Undesired lower frequency components of the mm-wave RF signal are removed by a high-pass filter (HPF), before amplification by 30 dB using a power amplifier (PA).
Wireless transmission at 25.5 GHz over 9 m distance is performed using a pair of pyramidal horn antennas with a gain of 18.5 dBi gain each, forming a loopback at the RU. The latter both simplifies the experimental setup and realistically emulates a DL transmission link, containing a single optical and mm-wave wireless transmission respectively, as would be experienced by DL transmission to a UE. The received signal at the RU, which in a real scenario would be the received signal from the UE, is amplified by 40 dB using a low-noise amplifier (LNA) and downconverted to IF using the remote-fed LO, which results from the heterodyning of the pure two-tone signal transmitted from the CO with an additional amplification by 30 dB in a medium power amplifier (MPA) to achieve sufficient LO levels to drive the mixer. While, as shown in Fig. 3, in a real scenario the downconverted IF signal would be transported back to the CO over a separate core of the MCF, in the experimental setup it is fed directly to the receiving UL side of the IF unit, effectively using it as UE and maintaining a realistic emulation of a DL transmission.
It should be noted that through the proposed implementation of the ARoF fronthaul architecture shown in Fig. 3 a substantial reduction of complexity of the RU is achieved. First, optical heterodyning for mm-wave generation and the remote-fed LO for downconversion allow centralization of all RF oscillators at the CO, leaving only RF amplifiers and the mixer. Second, optical intermediate frequency-over-fiber (IFoF) transmission of the UL signal will require modulation of an optical signal with the received IF-the required optical carrier can easily be provided from the CO across the MCF, either at a different wavelength sharing a core with the LO or ARoF signals or over a separate core, leaving only the modulator at the RU. Similarly, the proposed setup and architecture avoids the use of any optical amplifiers in the ODN or at the RU, requiring optical amplification only in two-tone generation.

Real-time analog fronthaul demonstration results and discussion
The evaluation of the proposed ARoF setup and its associated transmission performance involves, first, the analysis of the transmission setup, showing associated optical and electrical spectra (Sect. 4.1), and, second evaluation of transmission performance based on BER and error vector magnitude (EVM)/received constellations after real-time signal processing in the ARoF BBU (Sect. 4.2).

Analog radio-over-fiber transmission
The optical two-tone generation and ARoF transmitter setup has a fundamental impact on the performance of the overall link, as it defines the targeted RF frequency and largely determines the levels of phase noise in the final RF signal generated through the optical heterodyning. In this demonstration, the targeted RF frequency is towards the lower end of the 3GPP FR2 n258 band, which ranges from 24.25 GHz to 27.5 GHz, and is set to be f RF = 25.5 GHz , composed of the two-tone signal spacing plus the IF frequency, as: as can be seen in Fig. 5.
First, for the generation of the two-tone signal, it is important to suppress all unnecessary signal components in order to minimize resulting undesired beating products during the optical heterodyning for mm-wave RF signal generation. Careful optimization of bias voltage and driving amplitude allows to achieve ≈ 30 dB of suppression between the desired tones spaced at 2f LO and the original carrier as well as higher order odd and even harmonics respectively, as seen for the two-tone LO output in Fig. 5. Second, modulation with the IF signal should be performed with driving power and bias point optimized  The IF signal, as generated by the IF unit, i.e., before amplification for driving the MZM is shown in Fig. 6a, showing it centered at f IF = 5 GHz as required to achieve the targeted RF frequency according to Eq. 1. It should be noted that the 800 MHz wide OFDM data signal is clearly visible and a carrier-to-signal ratio (CSR) of ≈ 15 dB is observed, while the SNR is ≈ 35 dB. Given the optical ARoF signal shown in Fig. 5, it is clear that the beating on the PD will generate the desired modulated RF signal at f RF = 2f LO + f IF , but will also generate signals at other multiples of f LO and its combinations with f IF . The majority of these are removed through the HPF after the PD, while higher frequency products are mostly outside the bandwidth of the PA used before the radiating antenna. The resulting RF signal after amplification is shown in Fig. 6b, where it should be noted that the tone at 20.5 GHz, i.e., the direct beating of the two-tone signal, is insufficiently suppressed with an approximately equal tone at 30.75 GHz not shown-in a system destined for deployment these could easily be suppressed by replacing the HPF with a band-pass filter (BPF) and limiting PA bandwidth to the required band in order to adhere to out-of-band and spurious emission limits set by relevant standardization [6].
The received IF signal after mm-wave transmission, amplification in the LNA and downconversion using the remote-fed LO is shown in Fig. 6a, showing a similar power level as the transmitter IF signal, but with SNR reduced to ≈ 12 dB . The received IF signals further shows a somewhat reduced CSR which, together with the shape of the noise around the signal, suggests the presence of nonlinearities in the link, likely due to saturation of either the PA or LNA. Finally, the received IF signal appears slightly tilted, with powers of the higher frequency subcarriers somewhat lower than for the lower frequency subcarriers. The latter is likely caused by the LPF used at the input to the IF unit, as it has a 3 dB cut-off frequency of 5.5 GHz and will thus cause some attenuation already at frequencies between 5.0 and 5.5 GHz.

Real-time transmission performance
To evaluate transmission performance, EVM and BER are measured after real-time processing at the BBU for a range of optical powers P opt after two-tone generation. As varying optical power varies both RF and LO power after heterodyning, even small changes in optical power cause substantial changes in the final received IF power: from the square-law detection principle of the PD and as varying P opt varies the power in both tones, a 1 dB change in optical power directly corresponds to a 2 dB change in both RF and LO powers. Furthermore, due to reduced downconversion efficiency in the mixer at the RU for lower LO powers, an additional 1 dB of degradation is estimated for every 2 dB reduction in LO power-causing an overall reduction of 5 dB in received IF power for every 1 dB of optical power reduction.
The BER of the received signal is evaluated continuously by the ARoF BBU and retrieved repeatedly after short time intervals for each measured optical power. The separate measured BER values are shown in Fig. 7a, which also shows their average and corresponding 95% confidence bounds for the BER in any such short time interval, derived from the measured BER statistics. This suggests significant variation of BER over time does occur and that a sufficient margin must correspondingly be considered for any real system implementation.
It also proves successful transmission of a 5G NR OFDM signal using the extended M4bS mode shown in Table 1, showcasing seamless transmission with an RF bandwidth of 800 MHz at 25.5 GHz and thus in the 3GPP FR2 n258 mm-wave band. The system demonstrates transmission at a data-rate of 1.4 Gbit/s after 10 km MCF and 9 m mmwave wireless using fully real-time processing in the ARoF BBU for both signal generation and analysis. Figure 7a further indicates the BER limits of 3.8 × 10 −3 and 1.3 × 10 2 for typical forward error corrections (FECs) with 7% and 25% overhead (OH), respectively. This shows  Fig. 7 Transmission perfromance measurement results for the experimental real-time ARoF fronthaul demonstration: a real-time BER in dependence of optical power after two-tone generation and associated statistics, b observed constellations for the 800 MHz wide 5G NR OFDM signal. BER: bit error rate, OH: overhead, FEC: forward error correction the system easily achieves these limits depending on transmitted power and suggests that with further optimization and the substantially more complex FEC codes routinely used in mobile communications higher order constellations could be achieved-substantially increasing both data rates and SE.
Finally, Fig. 7b shows the observed constellations of the received signal after downconversion to BB, digitization and OFDM processing, with eight subcarrier groups, each corresponding to ≈ 100 MHz of RF spectrum, shown as separate constellations for easier comparison. It should be noticed that although they are shown in separate constellations, all subcarriers are jointly processed during the real-time DSP. The observed constellations match the expectations for the estimated EVM of ≈ 20% . While the latter is above the limit for FR2 base station emission as defined by 3GPP [6] of 17.5% to 19%, it should be noted that the latter is set for transmission from the RU, while that measured here includes wireless transmission and the associated EVM degradation, suggesting the signal radiated from the RU to be within the specified limits.

Conclusions
This article presented the blueSPACE fronthaul architecture with ARoF transport over SDM and optical beamforming and showed a demonstration of a high-bandwidth mm-wave 5G NR signal transmission with ARoF fronthaul over MCF. The blueSPACE architecture introduces the combination of ARoF with SDM in the ODN to solve the fronthaul capacity bottleneck and uses optical beamforming to seamlessly enable multibeam transmission and achieve spatial control over the RF signal.
The demonstrated fronthaul link used the blueSPACE ARoF BBU for real-time signal processing of 800 MHz extended 5G NR OFDM signals, which are upconverted to a tunable IF between 2.25 GHz and 5.5 GHz. Combined with the fixed LO of the ARoF transmitter this allows the demonstrated link to cover the entire 3GPP FR2 n258 band between 24.25 GHz and 27.5 GHz. The real-time fronthaul link achieves data rates of 1.4 Gbit/s with BERs below the limit for a 7% OH FEC after transmission over 10 km 7-core MCF and 9 m mm-wave wireless at 25.5 GHz. The demonstrated ARoF fronthaul link further employs a remote-fed LO for electrical downconversion, thus achieving not only minimum optical bandwidth requirements, but also maximum centralization of network functionality and complexity.
The demonstrated real-time fronthaul link highlights the functionality of the blueS-PACE ARoF BBU and validates the ARoF fronthaul concept at the heart of the blueS-PACE architecture. It further shows a possible use and advantage of SDM using MCF in the RAN. This demonstration proves the feasibility and highlights the potential for timely deployment of such mm-wave ARoF links for high-capacity 5G networks.