This section explains the presented mm-wave IFoF setup at the device and system levels and the associated DSP. First, the configuration and characterization of the pair of PAAs employed to realize beam steering on the experimental testbed is discussed. Second, the utilized mm-wave IFoF wireless scheme is presented and explained, highlighting the used device configurations and the 5G entities that are involved in the test bench. Finally, the key aspects of the employed transmitter and receiver DSP are shown together with the used OFDM configuration.

Phased array antenna description and characterization

The PAAs used in the setup are antenna panels supplied by NXP Semiconductors which contain 8-by-8 arrays of dual-polarized circular patch antennas. The antenna elements are separated by half a wavelength at 26 GHz, equal to 5.8mm. The panels contain dual polarized beamforming ICs (MMW9014K³¹) which can drive four channels per polarization. Each channel contains a transmit and a receive chain. Each chain has an 8-bit variable gain amplifier and an 8-bit phase shifter, allowing beamsteering in the intended direction and manipulation of the beam shape. The range in gain that can be achieved is \(\approx\) 30dB, while the range in phase is from 0\(^\circ\) to 360\(^\circ\). The array has two ports, one for the horizontal and one for the vertical polarization, which can be operated independently. For optimal performance, calibration of the array is required. The array calibration entails measuring the gain and phase responses of each channel for each gain and phase setting and creating a map of the actual response for any of the 8-bit weights.

The aforementioned array calibration is performed in an anechoic chamber in a near-field setup, where an open-ended waveguide probe is placed at \(5\lambda\) distance from the panel. For each channel, the probe is moved directly in front of the associated antenna element, and the gain and phase are swept. For each gain and phase setting, and for each of the 64 elements, the \(S_{21}\) (in transmit mode) or \(S_{12}\) (in receive mode) parameters are measured. The map is then obtained by extracting the gain and phase of the measured S-parameter at a single frequency point. In this setup, only the channel under test is turned on, while the remaining channels are disabled. This process is sped up by only measuring 8 gain settings and 16 phase settings out of the \(256^2\) possible combinations, and interpolating the resulting map.

With the resulting maps, some issues with the PAA can be addressed. Firstly, the gain and phase responses of each channel differ. Furthermore, changing the phase on a channel can lead to an unintended change in gain on that channel and vice-versa. This gain-phase coupling can be addressed by using the interpolated map as a lookup table and selecting the setting that matches the intended response most closely. This method is also used to address the phase offsets between the channels. The gain offsets can only be addressed by scaling the most powerful elements to a lower power level, such that all elements radiate the same amount of power. This power scaling decreases the side-lobe level (SLL) at the cost of gain and radiated power. In order to optimize the link budget, the element powers are not scaled, at the cost of increased SLL.

Taking this calibration into account, the phase and gain settings are determined for the transmitter and receiver arrays to steer the beam between \({\pm 35}{^\circ }\) in azimuth in 2.5\(^\circ\) steps and \({\pm 5}{^\circ }\) in elevation in \({5}{^\circ }\) steps. This calibration is done in the standard phased array case and for an additional case where a 20dB Taylor taper is applied to the weights, resulting in a reduction in the SLL³². These cases are compared in terms of error vector magnitude (EVM) and bit error rate (BER) in the outdoor setup. Compared to the standard configuration, the low-SLL case has a gain reduction of about 4dB both for the transmitter and receiver, resulting in a total 8dB link budget reduction. The radiation patterns generated and their corresponding array settings using the presented method are shown in Fig. 2a,b for the standard case, and in Fig. 2d,e for the low-SLL case. The patterns for Fig. 2a,d are normalized to the maximum gain, in this case in the main beam direction. Here an example is shown for the transmitter case, scanning towards -30\(^\circ\) in azimuth and \(0^\circ\) in elevation. The patterns are similar in the receiver case. In Fig. 2c the azimuthal cuts of the transmitter radiation patterns are shown in the standard and low-SLL case when scanning from -30\(^\circ\) to 30\(^\circ\) in azimuth. The different beams in Fig. 2(c) are normalized to the maximum gain at broadside. Observing Fig. 2c, it can be noted that the beamwidth is increased for the low-SLL case by 1.1\(^\circ\) and 1.4\(^\circ\) for transmit and receive respectively, while the SLL is decreased by 5.1dB and 5.0dB for transmit and receive respectively. In Fig. 2f the gain reduction when scanning, compared to the center beams at 0\(^\circ\), is shown, for the transmitter and receiver panels and for standard and low-SLL cases.

Measured radiation patterns in the standard case when scanning to \({-30}{^\circ }\) in azimuth, normalized to the peak amplitude in (a), and the gain settings used to excite the array in (b). In (d) the pattern is shown when an additional Taylor taper is applied, and the respective gain settings are shown in (e). In (c) examples of the beam shapes are shown for the transmitter panel using the standard and Taylor taper case. In (f) the scanning gain is shown for all measured scanning angles, normalized to the center beam of each situation.

The peak SLL averaged across the measured beams for the two under-test configurations, as well as the highest scan loss compared to the center beams at 0\(^\circ\) are given in Table 3. The scan loss, as shown in Fig. 2f, is 0.39dB higher in transmit than in receive for the normal configuration, while with Taylor taper this difference reduces to 0.35dB. The Taylor taper reduces the SLL by 5.0dB to 5.1dB and increases the half power beamwidth (HPBW) by 1.1\(^\circ\) in transmit and 1.4\(^\circ\) in receive direction, respectively. The SLL in the transmitter PAA is almost the same as in the receiver one, and the beamwidths are also equivalent. By using the Friis transmission equation and measuring the cable losses of the anechoic chamber, it is possible to calculate the total integrated gain of the panel for each situation. The integrated gain here is the overall gain of the PAA inclusive of all RF electronics included in the assembly, i.e., array gain, patch gain, splitting losses, beamforming and direction switching, as well as amplification. These results are shown in Table 3 as well. Important to note is that the integrated gain and radiated power are dependent on the array calibration method. The gain difference of the overall PAA assembly in receive and transmit is 2dB. In general, the transmitter and receiver mode arrays perform similarly and their radiation patterns are equivalent.

Experimental setup

Figure 3a depicts the proposed IFoF wireless setup for mm-wave 5G/6G communications. As it can be seen, the schematic of the experimental setup is divided into three different segments which correspond to the different entities of the 5G/6G system³³: CO, RAU, and end-user. The CO function consists of baseband processing, and generating and preparing the data signal for the optical IFoF fronthaul transport. For achieving this, a distributed-feedback (DFB) laser (CoBrite-DX1 tunable laser from ID Photonics) emits an optical carrier at 1550nm with 16dBm of output power. The optical carrier is then used to convert the electrical data signal into the optical domain by using an Avanex Mach-Zehnder modulator (MZM) (P/N: 792 000220). For proper optical data modulation, the MZM is biased in the quadrature point. The electrical data signal, that is introduced into the MZM, is produced with a 6.4 GSa/s arbitrary waveform generator (AWG). The Zynq UltraScale+ RFSoC ZCU111 evaluation kit is used in this experiment as AWG. The intermediate frequency (IF) upconversion of the baseband data signal is digitally performed in the DSP at 2 GHz and is detailed alongside the DSP. After the optical data modulation, the resulting IF optical data signal is sent through a 5 km long standard single-mode fiber (SSMF), emulating the distance between the CO and the RAU.

Experimental IFoF wireless setup for mm-wave 5G/6G communications with a single RAU and end-user: (a) schematic of the setup (downlink only); (b) graphs of the signal spectra at different points of the experimental setup. CO: central office, RAU: remote antenna unit, DSP: digital signal processing, AWG: arbitrary waveform generator, PMF: polarization maintaining fiber, MZM: Mach-Zehnder modulator, SSFM: standard single-mode fiber, PD: photodiode, VSG: vector signal generator, BPF: bandpass filter, OSC: oscilloscope.

At the RAU, the optical signal at the output of the SSMF is detected by a photodiode (PD) from Optilab (P/N: 4323-PD-40-C-DC-ND), generating an electrical bandpass signal at 2 GHz. The resulting electrical signal is upconverted to a center frequency of 27 GHz. For this mm-wave upconversion, a vector signal generator (VSG) and the ADMV1013 evaluation board from Analog Devices are utilized. The ADMV1013 board integrates a local oscillator (LO) quadrupler, RF mixer, and amplification controlled by voltage variable attenuators (VVAs). Hence, the frequency requirements of the VSG are reduced due to the use of the carrier quadrupler. More specifically, the VSG generates a sinusoid of 6.25 GHz. Since the IF mode is the selected configuration on the ADMV1013 board, the upconverted electrical signal is a double-sideband (DSB) with a carrier at 25 GHz (see the spectrum of Fig. 3b1). Moreover, the maximum VVA gain is set in the used ADMV1013 configuration. To condition the output signal for the wireless transmission, a band-pass filter (BPF) is used with a 27 GHz center frequency, \(\approx\) 600 MHz of bandwidth and adequate suppression of unwanted and out-of-band components³⁴. The spectrum of the signal obtained after this filtering process is illustrated in Fig. 3b2. Next, the filtered signal is fed into the transmitter PAA panel, where splitting, amplification, and phase-shifting processes are carried out for transmit beamforming. Finally, the resulting mm-wave signal is transmitted wirelessly at 27 GHz, within the n257 and n258 5G bands.

After wireless transmission, the PAA panel of the end-user catches the signal, and subsequently, phase shifting, amplification, and combining procedures are performed for receive beamforming. The spectrum of the signal at the output of the receiver PAA can be seen in Fig. 3b3. After the end-user PAA, the mm-wave signal is downconverted to a second IF at 1.5 GHz using the ADMV1014 evaluation board from Analog Devices. A carrier quadrupler, RF mixer, and RF amplifiers are integrated on the ADMV1014 board, which is the complementary downconversion model to the ADMV1013 board used in the RAU. In addition, for this downconversion procedure, a second, independent, VSG is required, which produces an LO at 6.375 GHz. In the presented experiment, two N5183B MXG modules are employed for the transmitter and receiver VSGs. Finally, the resulting IF signal is sampled and captured by an oscilloscope with a sampling rate of 10 GSa/s. The Lecroy WavePro 725Zi is utilized as oscilloscope.

It is relevant to point out that the signal bandwidth of the demonstration system is mainly limited by available spectrum and the filter at the transmitter (600 MHz bandwidth) required to stay within emission limits, while the remainder of the system would support substantially larger bandwidths. The PAAs support signals between 24.0 GHz and 27.5 GHz, i.e., up to 3.5 GHz of bandwidth, while the ADMV1014 and ADMV1013 modules support signals with bandwidths up to 5.2 GHz at IF frequencies between 0.8 GHz and6.0 GHz. Finally, the optical IFoF subsystem would support substantially wider bandwidths and higher IF frequencies.

DSP configuration

The same OFDM configuration is employed for all the measurements carried out in this work. This OFDM configuration adheres to the 5G standards and is as follows²: 14 OFDM symbols per slot; 12 subcarriers per resource block (RB); 240 kHz of subcarrier spacing; every OFDM symbol contains 2048 subcarriers of which 416 are null, resulting in a total bandwidth of 391.68 MHz; one dedicated OFDM symbol per slot for channel estimation with all active subcarriers serving as demodulation reference signals (DM-RSs); one phase tracking reference signal (PT-RS) subcarrier every 8 RB for phase noise compensation³⁵; \({0.2976}{\mu }{s}\) of cyclic prefix (CP); and 64-QAM as modulation order on the data subcarriers. With these parameters, the spectral efficiency of the OFDM signal is \(0.86{\log }_{2}(M)\) \(\hbox {bit/s/Hz}\), where M indicates the modulation order. Hence, the final throughput is 2015.5Mbit/s for 64-QAM data modulation and 391.68 MHz of bandwidth.

The DSP block diagram used on the transmitter side is represented on the left side of Fig. 4. The resulting signals of this DSP process are generated by the AWG of the RAU (see Fig. 3a). First, in the DSP transmitter block diagram, the input bits are mapped to 64-QAM symbols. The resulting 64-QAM symbols refer to the data subcarriers. Later, null, PT-RS, and DM-RS subcarriers are inserted respecting the OFDM configuration discussed in the previous paragraph. After this subcarrier insertion, an inverse discrete Fourier transform (IDFT) is performed, moving from the frequency to the time domain. Then, the CP is added to each OFDM symbol. All the aforementioned DSP blocks compose the OFDM transmitter. A preamble is also added at the beginning of the 5G slot frame for fine synchronization on the receiver side. Subsequently, the real and imaginary parts of the OFDM signals are separated and upsampled for a 2 GHz IF upconversion in the digital domain. As a result, an OFDM bandpass signal with an IF of 2 GHz is generated.

DSP block diagrams for transmitter (left) and receiver (right) sides.

On the other hand, the block diagram on the right of Fig. 4 corresponds to the DSP processes performed on the receiver side in order to properly demodulate the captured signal by the oscilloscope. The received signal is filtered with a digital BPF, suppressing any undesired frequency components. Then, an IF demodulation procedure is realized, moving the signal to the baseband. The obtained baseband signal is downsampled. By using the preamble previously inserted on the transmitter side, fine synchronization is performed to find the starting time of the received signal. Subsequently, a rough carrier frequency offset (CFO) compensation is executed to correct for the frequency drift of some devices, such as VSGs and AWG, involved in the experimental setup. At this point, the OFDM receiver block starts by removing the CP. For more accurate CFO compensation, the advanced LI-CPE method³⁶ is used, harnessing the inserted PT-RS symbols. Furthermore, this LI-CPE method allows efficient mitigation of the common phase error (CPE) induced by the phase noise that equally affects all the subcarriers of each OFDM symbol³⁶. After the LI-CPE method, a mean squared error (MSE) channel estimation is carried out by using the DM-RS OFDM symbol contained in every slot³⁷. Thus, the MSE detection is utilized to compensate for the channel on the data subcarriers. Finally, a 64-QAM demodulator is employed to extract the bits from the processed data subcarriers.