1. Introduction

High capacity point-to-point wireless links at millimeter wave (mm-Wave) frequencies are expected to be an essential part of mobile backhauling for 5G networks [1]. In fact, speeds beyond 100 Gbit/s will be needed in the backhaul network to support the increasing bandwidth demands [1]. Yet, the generation of these high frequency signals in the electrical domain is challenging [1].

An elegant solution may rely on microwave photonics (MWP) [2–5], where the millimeter wave signals are generated by optical heterodyning in a photodiode [6–8]. MWP enables larger bandwidth, lightweight structures, offer immunity to electromagnetic interference and larger inter-connection distances while simplifying the installation of the system [3]. Recently demonstrated photonic-based wireless links have shown impressive data rates in several frequency bands from 31 GHz up to 245 GHz carrier frequency. At 31 GHz, a 44.3 Gbit/s transmission experiment revealed that very high transmission speeds can be achieved at smaller carrier frequencies when employing pre-equalization to improve the signal quality [9]. At 60 GHz, data rates have been limited to 25 Gbit/s/polarization [10]. In the W-band (75-110 GHz) data rates achieved are 50 Gbit/s/polarization over a 1.2 m wireless link [11], 40 Gbit/s/polarization for a distance of 300 m [12] and even 10 Gbit/s/polarization for 1.7 km transmission [13]. At 245 GHz, data rates beyond 100 Gbit/s have been demonstrated [2]. However: broad adoption of MWP based solutions in next generation backhauling require: 1) A capacity larger than 50 Gbit/s/polarization after forward error correction. 2) Use of frequency bands for which the hardware costs are low (e.g. at 60 GHz, driven by IEEE 802.11ad WiGig standard). 3) Excellent signal quality to enable use of higher order modulation formats (i.e. 32QAM and higher).

In this paper, we demonstrate a data rate of 70 Gbit/s on a single carrier and single polarization at 60 GHz. This high capacity is made possible by an efficient pre-equalization method which is enabled by a one-shot characterization method of the end-to-end system and that ultimately allows the effective use of 14 GBd signals with up to 32QAM. With the consumer oriented IEEE 802.11ad WiGig standard as the mayor driver for 60 GHz technology [14–16], the availability of components for 60 GHz communications could make high-capacity wireless links at 60 GHz affordable and may result in a cost-advantage over more expensive E-band or sub-THz links.

2. Application scenario

Figure 1 depicts a possible scenario for high capacity photonics-based wireless links. We envision a densely populated area with a large number of small cells. In such an environment, the diversity of small cells is needed to deal with the growing capacity requirements and number of users [17]. While each small cell base station could theoretically be deployed with an individual fiber connection to the backhaul network, this may call for expensive and time consuming construction work, and sometimes may not even be possible. Instead, the connection could be established wirelessly from a remote antenna unit (RAU) which is connected to the central office and which provides network access to several small cells, see Fig. 1. Thereby, deploying costly fiber links could be avoided.

 

Fig. 1 – Mobile backhauling employing high capacity wireless links at 60 GHz. (a) In the central office digital signal processing (DSP) is performed on the mobile communications signals which are then transmitted as optical signals over a fiber. (b) In the remote antenna unit (RAU) signals are converted from the optical domain to the electrical domain and converted to the microwave domain. (c) Wireless transmission link and (d) the small cell base station.

Download Full Size | PDF

3. Experimental setup

Following the concept of Fig. 1, our experimental setup is composed of four parts: an optical transmitter, an optical-to-RF converter, a wireless link and an RF receiver mimicking the central office, the remote antenna unit (RAU), the wireless link, and the small cell. The setup is detailed in Fig. 2. The concept has been employed and described in several previous works [2,9–13] as stated in Section 1.

 

Fig. 2 - Experimental setup of the wireless link. (a) In the central office, an IQ modulator driven by an AWG modulates a laser (blue). The data signal is combined with a 60 GHz detuned reference laser (red) and sent to a remote location through 25 km of single-mode fiber. (b) In the remote antenna unit (RAU), a photodiode generates the RF signals which are transmitted over the wireless link (c). (d) In the receiving RAU, the signal is down-converted to an intermediate frequency, and recorded with a digital sampling oscilloscope (DSO).

Download Full Size | PDF

The central office, Fig. 2(a), generates the baseband data with a 65 GS/s arbitrary waveform generator (AWG). The data is fed to separate broadband amplifiers for the in-phase and quadrature components. The amplified signals drive an optical IQ modulator, which encodes the data onto a laser at 1550.12 nm ($f_{SIG}$), see inset Fig. 2(a). Afterwards, the data signal is boosted by two cascaded erbium doped fiber amplifiers (EDFA) with subsequent optical filters. The filters have bandwidths of 0.6 nm. The schematic only shows one EDFA- for the sake of simplicity. Subsequently, the data signal is recombined with a reference laser ($f_{\text{REF}}=f_{\text{Sig}}-60 \text{GHz}$) detuned by 60 GHz. A polarization controller is used to ensure that the data signal and the reference laser are on the same polarization. It should be noted, that the two lasers are not locked to each other. Thus, the receiver requires the standard DSP algorithms of coherent communications to handle the frequency drift and phase noise of the lasers. Next, the signals are sent across 25 km of standard single mode fiber (SMF) to the RAU. Additionally, an optical attenuator (OA) is used to control the launch power to the fiber.

At the RAU, Fig. 2(b), the data and the reference laser are fed to a 70 GHz photodiode. The optical input power to the photodiode is controlled with a variable optical attenuator (VOA). The inset of Fig. 2(b) shows the measured spectra of the signal and reference laser in front of the photodiode with respective power levels of −8 dBm and 0.9 dBm. Through optical heterodyning in the photodiode, a copy of the data signal will be generated at the desired 60 GHz RF carrier. Unwanted signal copies at twice the optical frequency are cut-off by the photodiode while the subsequent V-band low noise amplifier suppresses the baseband copy. The V-band amplifier offers 23 dB gain at 60 GHz.

The wireless link, Fig. 2(c) consists of two high gain antennas by Huber&Suhner, which are set apart by 5 meters. These antennas offer a gain of 38 dBi which is helpful to compensate the added path loss at 60 GHz. Thereby, the link distance of 5 meters is rather limited by the dimensions of our laboratory then by the power budget.

In the small cell base station, Fig. 2(d), the signal is amplified directly after the antenna with a V-band amplifier offering 20 dB gain at 60 GHz. Further, the signal is electrically down-converted to an intermediate frequency of $f_{\text{IF}}=11 \text{GHz}$ using an RF mixer. This is necessary to ensure that the signal lies within the bandwidth of the receiver. Downconversion to baseband can then be realized digitally. Ultimately, the signal is recorded by a digital sampling oscilloscope (DSO) and processed offline.

Figure 3 shows a picture of some parts of the setup in our lab. The main photograph shows the wireless transmission path with the transmitter antenna in the foreground and the receiving antenna in the background. One can instantly see that the RAU only consists of a photodiode, a V-band amplifier and the antenna, compare Fig. 2(b). This setup is very simple since the complexity can be moved to the central office. The inset zooms in on the receiving antenna and reveals a view of the small cell receiver, compare Fig. 2(d). The V-band amplifier and the mixer used for IF downconversion can be identified as well as the digital sampling scope sitting in the background.

 

Fig. 3 – Photographs of the experimental setup. Figure (a) shows the RAU including photodiode (PD), V-band amplifier and 60 GHz antenna. At 60 GHz, the antenna provides 38 dBi gain while the gain of the amplifier is 23 dB. After 5 meter of wireless transmission the signal reaches the small cell that can be identified in the background and is magnified in the inset on the right hand side (b). Here, the 60 GHz signal is collected by an antenna that is identical to the one at the transmitting RAU. After boosting the signal with an amplifier with 20 dB gain at 60 GHz, it is downconverted by an RF mixer to an intermediate frequency (IF) and recorded with a digital sampling oscilloscope (DSO).

Download Full Size | PDF

4. Pre-equalization technique

The overall system performance may be impaired by any component in the signal chain. In our case, the frequency response of the channel is affected by the RF components such as the amplifiers and the antennas. In Fig. 4(a), the received normalized spectrum of a 14 GBd 16QAM signal (roll-off = 0.15) is depicted. The different noise levels to the left and the right of the signal spectrum are due to the low pass characteristics of the RF components. Ideally, the spectrum should be flat but power variations of up to 7 dB can be seen. The post equalizer in the receiver cannot recover the data without pre-equalization, i.e. the decision directed equalizer does not converge [9] because preceding receiver algorithms (e.g. timing and carrier recovery) do not work properly with an overly distorted signal. The demodulation fails under such conditions and data cannot be retrieved, see Fig. 4(a) inset. It should be noted, that the smaller variations in the spectrum are caused by the random character of the data.

 

Fig. 4 – Impact of the one-shot pre-equalization technique and process flow. (a) Received data signal (14 GBd 16QAM) without pre-equalization, strong power variations can be seen in the spectrum (b) Complex channel response (magnitude and phase) measured with our algorithm. (c) Received data signal (14 GBd 16QAM) with pre-equalization. A flat channel spectrum for the received signal is found after applying the complex channel response as a pre-equalization filter. (d) The pre-equalization used in this experiment allows a “single-shot” measurement of the complex response of the system under test. The process consists of the following steps: (1) A frequency comb is generated digitally and sent to an AWG, a random phase is then added to every frequency component to avoid large PAPR. (2) The comb goes through the system, following the very same path that the data will encounter. (3) The comb is acquired by an oscilloscope at the receiver, down-converted, and an FFT is performed. (4) The amplitude peaks of the spectrum are found with a specialized algorithm to recover the frequency components of the comb. (5) The random phase is removed from the results. (6) The complex response of the system is provided. It enables complex pre-equalization of a system under test

Download Full Size | PDF

In literature, pre-equalization or pre-distortion is well known to counteract limitations imposed by the optical transmitter [18–20]. However, pre-equalizing the signal in the transmitter according to the whole system impairments can be used to achieve a flat frequency response at the receiver and make the entire bandwidth usable. To perform pre-equalization, accurate channel response information is needed. A precise channel response can be obtained by offline S-parameter measurement of each RF component individually [9]. However, such an implementation is time consuming. In turn, inline-channel measurements are of great interest as they allow for characterization of a channel without a need for disassembling the link and compensating the measurement equipment’s effects.

For the demonstration in this paper, an inline, one-shot, pre-equalization method was used, which works according to the steps depicted in Fig. 4(d). (1) The AWG generates a frequency comb which is encoded with a random phase pattern to limit the peak-to-average power ratio (PAPR). Note that the random phase pattern is set once, saved and kept for the whole measurement sequence. The frequency spacing as well as the spectral width of the comb can be varied according to the setup specifications. For our measurements, we used a spectral width slightly larger than the desired signal bandwidth and a spacing of 60 MHz for the comb lines, which provided optimal results for the channel pre-equalization. (2) The comb is sent through the system, where it is upconverted and sent through the wireless link while no changes on the setup configuration are made. (3) A sampling oscilloscope acquires the output signal in the receiver and downconversion to the baseband is performed. (4) The peaks of the various comb components are detected in the signal spectrum and used to compute the amplitude response by interpolation. In this work, we used a shape-preserving piecewise cubic interpolation. (5) The linear phase ramp (corresponding to the delay of the system) and the random phase pattern added in the transmitter are removed to extract the phase response. (6) The algorithm output is a complex channel response (amplitude and phase response) that can be used for pre-equalization at the transmitter.

Figure 4(b) shows the channel response of our 60 GHz link before equalization obtained using our one-shot measurement algorithm. The channel measurement was performed over a bandwidth of 20 GHz. The performance of the system is limited by the channel for various reasons, which can be seen from the complex channel response. First, strong amplitude variations (in some cases more than ± 5 dB over 2 GHz) in the magnitude response can be identified. These variations are prone to the RF components, namely the amplifiers and antennas have a big influence on the spectral characteristic. For example, we found that exchanging the V-band amplifier at the transmitter antenna with one of the same models changes the spectral shape significantly. Second, the available bandwidth is limited due to the cut-off of the RF components at around 16 GHz. Third, the phase response shows a very strong dispersion (3000° difference across the full bandwidth).

In Fig. 4(c), the result of wireless transmission after applying this equalization technique is depicted. It can be clearly seen that the spectrum of the 16QAM signal at a symbol rate of 14 GBd (56 Gbit/s) shows less distortions compared to the spectrum without pre-equalization, Fig. 4(a). The filter taps of the decision-directed post-equalizer in the receiver can now converge and the data be recovered, see Fig. 4(c) inset with the constellation diagram. Consequently, our pre-equalization technique counteracts linear distortions in the system to enable usage of standard digital signal processing (DSP) algorithms in the receiver such as timing and carrier recovery or the above mentioned decision-directed post-equalizer. If desired, an additional equalizer to tackle non-linear distortions could be added in the receiver DSP chain. It is worth noting that the pre-equalization technique is not exclusive to applications in V-band and can in principle be used to characterize any system under test.

5. Results

To test the scheme in an experiment, we used root-raised-cosine shaped signals and a 0.15 roll-off with data patterns generated with a DeBruijn-11 sequence. The results presented in this section were obtained with RF components not perfectly suited for this experiment. Yet, thanks to our pre-equalization method very high data rates have been demonstrated. One may however assume that the data rate would be even higher with RF components chosen specifically for the purpose of the demonstration.

First, we analyzed our system performance for various symbol rates and modulation formats. Figure 5 depicts the error vector magnitude (EVM) for BPSK (red), QPSK (yellow), 16QAM (green), 32QAM (light blue), and for a 16QAM reference measurement without the wireless link (back-to-back, violet). Please note that we used the convention where the EVM is normalized to the power of the outermost ideal constellation point [21, 22]. As expected, the signal quality slowly decreases for all modulation formats when increasing the symbol rate, i.e. the average EVM increases. We found that BPSK and QPSK formats were able to perform well up to 15 GBd, 16QAM and 32QAM up to 14 GBd. This is due to the fact that the usable bandwidth is limited to approximately 16 GHz by the RF components. Additionally, BPSK and QPSK have a higher noise tolerance compared to 16QAM and 32QAM which is why they still work up to 15 GBd. The back-to-back case for 16QAM shows that by removing the wireless link from the transmission path, reception of 15 GBd signals is still possible. So, from a technical point of view, using 14 GBd signals is reasonable and doable at 60 GHz. It is yet another question if such bandwidths will be made available by regulation bodies.

 

Fig. 5 EVM for different symbol rates using pre-equalization. Measurements were performed with 4 dBm launch power to the 25 km fiber.

Download Full Size | PDF

As a next step, we investigated the influence on the signal quality of the optical power fed to the photodiode. Figure 6 shows the result for 10 GBd, 12 GBd, and 14 GBd 16QAM signals. A 14 GBd 16QAM measurement without the antennas is included for reference (back-to-back, blue). The signal will initially obtain a better EVM as the optical power increases. Crossing a certain power threshold (between −3.5 dBm and −2 dBm) causes the RF amplifiers at the receiver to go into saturation and distorting the signal. The excess power margin available in this experiment shows that our system will have sufficient power to cover even a larger wireless distance once the limitation of the laboratory room size is removed. We estimate that the wireless link can be extended to 20 meters while keeping the power at the receiver the same by increasing the optical input power by roughly 6 dB.

 

Fig. 6 - Results for 16QAM at symbol rates of 10, 12 and 14 GBd for various optical input powers to the photo diode. Launch power to the fiber was 8.6 dBm.

Download Full Size | PDF

The plot in Fig. 6 further shows how the EVM increases for higher symbol rates whenmeasured near the optimal input power. Note that the measured bit error rate (BER) close to the optimum is below ${10}^{-3}$ for all signals (${10}^5$bits were evaluated preventing statistically relevant claims below). The EVM threshold to obtain a BER of ${10}^{-3}$ for 16QAM is 11% ${10}^{-3}$when additive white Gaussian noise is assumed [21, 22].

Finally, we evaluated the BER of 14 GBd signals for BPSK, QPSK, 16QAM and 32QAM. The received constellation and corresponding BER values can be observed in Fig. 7. For each measurement, at least 10 million bits were analyzed, i.e. BER of ${10}^{-5}$ can be stated at best. Both BPSK and QPSK did not reveal any errors within this recorded sequence (BER<${10}^{-5}$) while 16QAM and 32QAM showed a BER of $7\cdot {10}^{-5}$ and$3.7\cdot {10}^{-3}$, respectively. These raw input BER values are below the threshold for error-free transmission when second-generation hard-decision FEC with 7% overhead is used [23]. In this case, an output BER of below $7\cdot {10}^{-15}$ may be expected if the FEC works correctly and limitations are only due to additive white Gaussian noise. The constellations also show some non-linear distortions which we suspect coming from the RF amplifiers and RF mixer.

 

Fig. 7 – Constellation and measured BER for 14 GBd signals of (a) BPSK, (b) QPSK, (c) 16-QAM, (d) 32-QAM with more than $6\cdot {10}^6$ symbols.

Download Full Size | PDF

The plot in Fig. 7(d) shows a 14 GBd 32QAM transmission experiment. This results in a 70 Gbit/s capacity for a single carrier, single polarization setup at 60 GHz. This is – to the best of our knowledge – the largest capacity so far transmitted in a V-band single polarization experiment.

6. Conclusion

We have successfully shown a 70 Gbit/s photonic wireless link at 60 GHz on a single carrier and single polarization across 25 km of standard single mode fiber and a 5 m wireless transmission link. Usage of 14 GBd signals has been made possible by a pre-equalization technique. The pre-equalization relies on an efficient “single shot” frequency response characterization of the system that might be useful in future systems to repeatedly update the equalizer and thereby adapt to dynamic changes in the system. This work demonstrates that the 60 GHz band may be a good alternative to more expensive E-band or sub-THz links for >100 Gbit/s mobile backhauling.