W-band photonic-wireless link with a Schottky diode envelope detector and bend insensitive fiber

: The performance and potential of a W-band radio-over-ﬁber link is analyzed, including a characterization of the wireless channel. The presented setup focuses on minimizing complexity in the radio frequency domain, using a passive radio frequency transmitter and a Schottky diode based envelope detector. Performance is experimentally validated with carriers at 75–87GHz over wireless distances of 30–70m. Finally the necessity for and impact of bend insensitive ﬁber for on-site installation are discussed and experimentally investigated.


Introduction
The immense growth in the usage of mobile and wireless data -driven by the use of bandwidth intensive applications on mobile consumer devices -has lead to an increasing demand for highspeed multi-gigabit wireless connections.In order to accommodate this need, new strategies must be employed both in direct wireless access and in wireless point-to-point links leading up to the access point, such as in mobile front-and backhaul, wireless building-to-building links or wireless bridges over obstacles [1][2][3][4].A migration away from the already congested conventional wireless bands and towards higher frequency bands -such as the millimeterwave (mmW) region -is generally considered a necessity to support the required bandwidths and mmW radio-over-fiber (RoF) links have been identified as a key candidate, combining the advantages of optical and wireless communications [4][5][6][7].
Hybrid photonic-wireless links in the IEEE V-(50-75 GHz) [3,8,9] and W-bands (75-110 GHz) [1,10,11] and beyond [12] have been demonstrated with a variety of setups, achieving transmission distances ranging from a few to multiple hundred meters.While carriers at 60 GHz have been in focus for Gbit/s indoor wireless [3,9] this region is affected by a large peak in atmospheric absorption [6], severely limiting it to short-range transmissions.The W-band however does not share this limitation and may thus be a candidate for indoor and buildingto-building transmission scenarios as well as mobile front-and backhaul [1][2][3] as shown in Fig. 1.
To ensure the feasibility of mmW wireless links in all the aforementioned scenarios however, three issues require additional attention and are addressed in this article: • characterization of the wireless channel: most efforts for channel characterization and modelling have regarded frequencies in and below V-band or only estimated path loss  [13,14].A characterization of the W-band wireless channel has been performed [15] and is presented in section 2.
• complexity of the setup: Gbit/s class W-band wireless transmissions reaching long distances have been performed with high complexity in both the optical and radio frequency (RF) domains; setups with reduced complexity employing envelope detection for (RF) downconversion were suggested [10,16], but only recently [11,17,18] reached distances beyond 10 m wireless transmission.A setup with a passive RF transmitter and a Schottky diode based envelope detector to minimize complexity in the RF domain has been developed [11] and is presented in section 3. Section 4 presents experimental results, successfully transmitting 2.5 Gbit/s of data over wireless distances of 30 to 70 m with carriers at 75 to 87 GHz as shown in section 4.1.
• fiber installation: the requirements for fiber to be installed directly to the transmitting antenna have not been regarded; previous investigations have focussed on the transmission fiber and associated impairments [19,20] or on capacity enhancements through fiber spatial multiplexing and multiple-input multiple-output (MIMO) RF transmission [21].The necessity for bend insensitive fiber (BIF) for on-site installation is obvious and similar to the case of fiber-to-the-x [22,23] and the impact of including it in the transmission line is analyzed and discussed in section 4.2.
Finally section 5 summarizes and concludes the article.

Wireless Channel Modelling
The characterization of the phenomena affecting communication signals during propagation in wireless systems is fundamental to the definition and development of the digital signal processing required for their compensation or mitigation.The manifestations of fading channels are commonly classified in two groups: large-scale and small-scale fading [24], with the former mainly due to the distance between the transmitting and receiving antennas, the given environment and the resulting attenuation, while the latter is described by a number of characteristics including the time and frequency responses of the channel.Large-scale fading is characterized statistically by calculating the mean of the observed attenuation in dependence on transmission distance -path loss -and its variations around this mean -shadowing.While a variety of models for path loss estimation is found in the literature [25], two of the most commonly employed have been selected for comparison.The free-space loss model (FSL, referred to as Friis model) defines path loss as in (1), where d is the distance between transmitting and receiving antenna, f the carrier frequency and c the speed of light in The generic empirical path loss model (GEPL) in (2) expresses total path loss as the observed loss L 0 at a reference distance d 0 and a term dependent on both distance d and the path loss exponent n, with the latter adapting the model to the studied environment.
Shadowing is easily calculated from experimental path loss data as a random variable -the shadowing parameter (σ S ) -with log-normal distribution additive to path loss [25].
The small-scale fading behavior of a channel includes its frequency selectivity as well as possible multipath phenomena and is characterized by the respective associated parameters, including coherence bandwidth (B c ), root mean squared delay (τ rms ), Doppler shift ( f d ) and coherence period (T c ) [24][25][26].

Channel Measurement Setup and Raw Results
For the characterization of a short distance indoor W-band wireless link a setup as depicted in Fig. 2 is used, in which a vector signal generator (VSG) in combination with a frequency sextupler generates the signal to be transmitted, while at the receiver the signal is down-converted to intermediate frequency (IF) using a harmonic mixer driven by a 4-6 GHz local oscillator (LO).The resulting received spectra and power levels are recorded on an electrical spectrum analyzer (ESA).Wireless transmission is based on a pair of pyramidal waveguide horn antennas with a mid-band gain of 25 dBi each.
In order to ensure reliability and comparability of the measurement results, the involved VSG and ESA were adjusted to give agreeing and consistent power readings and the frequency response of the mixer was accounted for according to manufacturer measurements; it should be noted however that not all uncertainty may be calibrated out of the setup and a certain effect of the equipments' response over frequency may affect the measurement results.For the antennas a flat frequency response is assumed across the entire sounded band.
A single-tone method as described in [26] is used to sweep the complete W-band spectrum with a 1 MHz resolution, recording received power levels at distances between 0.5 m and 4 m.The recorded data allows for both large and small scale analysis, but already in their raw form as shown in Fig. 3 gives a clear indication of the distance dependency of path loss; the reduced performance at the upper frequency edge of the analyzed band is caused by frequency limitations of the employed equipment.

Large-Scale Fading Analysis
The experimentally obtained W-band path loss values are compared to the predictions of the Friis model in Figs.4(a) and 4(b) for distance and frequency dependency respectively.In both cases good agreement between prediction and experimental data is found although a loss over distance slightly higher than predicted by the Friis model is observed in Fig. 4(a), while in Fig. 4  The path loss exponent (n) and shadowing parameter (σ S ) were calculated from the experimental data as discussed in [25], finding the GEPL to approximate the experimental data well for path loss exponents of n ≈ 2.2 and n ≈ 2.5 at 75 GHz and 110 GHz respectively, confirming the expectations for a short directive link in an indoor environment [25].Due to the short distances and moderate directivity of the link, the observed shadowing is very small at 0.05 dB and 0.12 dB for 75 GHz and 110 GHz respectively.
Overall it is clear that for short distance links under the present conditions the dependency of path loss on distance is by far the dominant effect, with an increase in distance from 0.5 m to 4 m resulting in an additional 13.1 dB to 16.2 dB of path loss depending on frequency.This frequency dependency of path loss seen in Fig. 4(b) implies an increase in path loss of 3.8 dB at 0.5 m, 3.3 dB at 2 m and 1.9 dB at 4 m for an increase of frequency from 75 GHz to 110 GHz.While small in comparison to the impact of varying distance and easily compensated in narrow band systems, such frequency dependent variations in path loss may be detrimental to signal integrity for systems with bandwidths as large as expected for future communication systems in W-band.Large-scale fading analysis of the channel is thus insufficient to describe the encountered channel effects in the required detail and small-scale fading analysis becomes a necessity.

Small-Scale Fading Analysis
For signals with bandwidths of multiple GHz -as expected in future systems and when using photonic upconversion -the relevant channels are likely to be characterized by a multipath rich profile, resulting in significantly lower power levels for each multipath component.Analysis of the experimentally obtained S 21 parameters using the inverse fast Fourier transformation (IFFT) and the Hermitian method [26]   It is clear that with increasing transmission distance the delays grow -as the difference in distance traveled by the multipath components increases -and hence coherence bandwidth reduces.For a modulated signal of 6 GHz bandwidth this would indicate that while at 0.5 m the encountered channel may be quasi flat, at 4 m severe frequency fading will cause a significant degradation of signal quality.The assumed threshold of −60 dBm may appear low when considering observed path loss and typical receiver sensitivities and with a higher threshold all significant components might belong to the first tap of the PDP; when considering links with lower directivity or the introduction of point-to-multipoint transmission however, the limitations posed by multipath propagation will be significant and may result in coherence bandwidths considerably smaller than those found.

Discussion
The results obtained from the channel characterization suggest that the well known models for path loss estimation -i.e.Friis and the GEPL models -retain their validity in predicting largescale phenomena and received power levels.Multipath propagation effects have been observed to potentially pose limits on the usable signal bandwidths, depending on link directivity, the considered threshold level for the significance of multipath contributions and on whether transmission takes place in a point-to-point or point-to-multipoint scenario.The specific channel analyzed may be considered line-of-sight and -even though only covering short distancesthe channel characteristics found may give an indication towards conditions for longer distance transmissions.This is especially the case if an increase in transmission distance is accompanied by an increase in antenna directivity, such as in the following sections where wireless distances of 30-70 m are covered by employing a pair of highly directive parabolic antennas.
While on one hand for short distance transmissions or systems with highly directive antennas the channel may be considered as line-of-sight only, for systems with low directivity on the other hand it may be necessary to compensate for the signal degradation resulting from path loss and shadowing -potentially through employing a degree of diversity, gaining additional uncorrelated estimates of the signal -and to mitigate the observed frequency selectivity, either through pre-processing or by applying equalization.

Low RF Complexity Fiber-Wireless Transmission System Setup
The radio-over-fiber (RoF) transmission system employed for the analysis of transmission performance and the study of mitigating fiber bending impacts through the use of bend insensitive fiber (BIF) is shown in Figs. 6 and 7.The setup consists of three distinct parts which in a deployment case would be geographically separate: (I) optical signal generation and modulation linked by optical fiber transmission to (II) optical to radio-frequency (RF) conversion linked by W-band RF transmission to (III) the receiver.Signal generation -i.e. the left hand side of Fig. 6 -consists of an external cavity laser (ECL) at λ = 1550 nm, followed by a Mach-Zehnder modulator (MZM) biased at its minimum transmission point and driven with a sinusoidal at f RF /2 to generate two spectral lines spaced at f RF .In order to allow performance analysis at different RF carrier frequencies the frequency of the driving signal is varied, generating line spacings of 75-87 GHz.The signal is amplified by 22 dB using an erbium doped fiber amplifier (EDFA) with a noise figure (NF) 5.8 dB and an arrayed waveguide grating (AWG) separates the two spectral lines, allowing to modulate one with data.A 2 15 − 1 bit long pseudo-random bit sequence (PRBS15) non-return-to-zero (NRZ) signal at speeds between 1 Gbit/s and 2.5 Gbit/s -generated with a pulse pattern generator (PPG) -directly drives the second MZM, biased in the linear part of its transmission curve for optimum extinction ratio.Two variable optical attenuators (VOAs) ensure equal power of the two arms and allow control over the output optical power.The experimental setup for signal A second EDFA with a 4.3 dB NF amplifies the signal by 20 dB before it is transmitted through 10 km of ITU-T G.652 standard single-mode fiber (SMF) with one of three spool samples of 5 km Sterlite BIF added to the transmission line for the investigation of performance when subjected to macrobending.After fiber transmission the optical signal is converted to an RF signal by beating the two spectral lines on a photodiode (PD) with a nominal 3 dB bandwidth of 100 GHz and a responsivity of 0.5 A/W.The generated RF signal is transmitted without further amplification through a pair of parabolic antennas -shown in Figs.7(c) and 7(d) -with a gain of 48 dBi each.
At the receiver site the signal is amplified by a low noise amplifier (LNA) providing 40 dB gain before down-conversion takes place in a Schottky diode based envelope detector (ED) with a 3 dB bandwidth of 3 GHz, yielding a baseband signal as shown in Fig. 8(b) for an exemplary case.Employing envelope detection rather than down-conversion with an electrical mixer alleviates the need for a local oscillator at the receiver and thus -together with the a passive wireless transmitter and only a single LNA at the receiver -minimizes complexity of the RF chain at both ends.Finally the received baseband signal is recorded using a digital storage oscilloscope (DSO) and bit-error rate (BER) values are determined through offline processing, consisting of simple thresholding and error counting over four recorded sequences with a length >2.5 Mbit each.
For the analysis of transmission performance the power incident on the PD is varied between 0-8 dBm through the use of the second VOA while wireless distance and carrier frequency are varied between 30-70 m and 75-87 GHz respectively.During the analysis of the effects of fiber bending and their mitigation through the use of BIF, carrier frequency and wireless distance are fixed to 84 GHz and 70 m respectively.Macrobending is induced by winding the different fiber samples for 1-10 turns around mandrels with radii between 5-15 mm -shown in Fig. 7(e) and performance is analyzed by monitoring the resulting incident power on the PD with a 1 % tap and by recording BER values.

W-Band Transmission Performance Analysis
The performance of the reduced RF complexity W-band photonic-wireless link is analyzed for wireless links of 30-70 m and utilizing carrier frequencies between 75-87 GHz, while the optical transmission link consists of 10 km SMF in all cases.Figure 9 shows BER as a function  of power incident on the PD and compares performance dependency on wireless distance and carrier frequency.Figure 9(a) allows analysis of frequency dependency at a fixed distance and data rate, finding it to be low while following the trend predicted by the Friis model; with the additional free space loss amounting to only 1.3 dB when comparing RF carriers at 75 GHz and 87 GHz performance variations of the involved RF equipment clearly dominate when regarding frequency dependency.Figure 9(b) details evolution of transmission performance over distance with a fixed carrier frequency of 84 GHz, while Fig. 10 shows corresponding transmitter and receiver eye diagrams for a fixed optical power of 5 dBm; Fig. 8(b) gives the received electrical spectrum after 70 m wireless transmission.It should be noted that in all of Fig. 9 a plotted BER of 10 −7 indicates successful and error free transmission of four sequences with a length of >2.5 Mbit each.Assuming a BER of 10 −6 as reference minimum, optical powers on the PD of 0.7 dBm and 4.4 dBm are found for distances of 40 m and 70 m respectively, translating into an increase in available RF power adequate to overcome the additional free space loss of 4.9 dB, while also allowing for a slightly less optimal alignment of the antennas at the larger distances.
Overall the system is found to be power limited, as is also seen in Fig. 9(c) with the surfaces -corresponding to optical powers on the PD -neatly nested and allowing different distancefrequency combinations depending on selected BER limit.With no error observed at 5 dBm and an additional 3 dB in optical power available it is expected that transmission distances beyond the mark of 100 m would be within reach of the system.The first step in analyzing how BIF can help mitigate the impact of fiber bending is a comparison of macrobending loss of the different fibers under test with the specifications of ITU-T G.657 as shown in Fig. 11.As the number of turns in an installation may easily reach the equivalent of ten turns, the observed bending loss of SMF suggests minimum bending radii of 15 mm or above, depending on system power budget.With losses well below those required by ITU-T G.657.A1 and as much as an order of magnitude lower at radii of 10-15 mm than those of SMF, sample A allows significantly tighter radii.Samples B and C compare favorably to the requirements of G.657.A2/B2 and G.657.B3 respectively with sample C showing a five times lower loss at 7.5 mm radius, suggesting macrobending loss at this radius to be near negligible and radii down to 5 mm to be allowable while maintaining lower loss per turn values than SMF at 15 mm bending radius.
Further analysis of transmission performance is dependent on the system under test, but may give a good indication of the benefit obtained through the used of BIF.At a distance of 70 m and with a carrier of 84 GHz, setting P opt = 6 dBm as reference power and assuming a required maximum BER of 10 −6 suggests a margin for allowable loss through fiber bending of about 2 dB.Figures 12(a) and 12(c) show the dependency of system BER performance on the optical power P opt incident on the PD at 1 Gbit/s and 2.5 Gbit/s respectively.Figures 12(b) and 12(d) visualize the corresponding numbers of turns at different radii that were found to be allowable with the different fiber samples while maintaining system performance with a BER < 10 −6 , while setting a BER limit of 10 −3 does not significantly alter the picture with only a single data point being changed for each of the data rates, as indicated by the and in Figs.12(b) and 12(d) respectively.Both clearly show the benefit with respect to bending tolerance that may be obtained with BIF and correspond well to the loss margin of 2 dB.It is clear that at radii of 10 mm G.657.A1 type fiber may sufficiently reduce the encountered macrobending losses, while for even smaller radii A2/B2 and B3 fibers may be required.While in the present case no difference was seen between samples B and C, an extrapolation of their encountered loss per turn values from Fig. 11 up to the loss margin suggests as many as 11 and 14 turns to be allowable at 5 mm, while at 7.5 mm up to 38 and 126 turns respectively are expected be possible without degrading system performance.
While the margin of 2 dB for allowable bending loss is specific to the assessed system and link, the indication of requiring BIF for radii of 10 mm and below is a general one as is the scale of the benefit to be obtained by the different fiber types.This serves to highlight the tradeoff to be found between system parameters such as the available power budget and logistical challenges such as the required tolerance to bending and ease of installation on one side and economical considerations such as the additional cost of BIF on the other side.

Conclusions
The wireless channel in W-band was characterized and a radio-over-fiber transmission link with reduced complexity in the RF domain demonstrated, achieving successful transmission of a 2.5 Gbit/s signal.The wireless channel is found to be affected by multipath propagation and resulting impairments, however in the case of short distances or high link directivity it may be regarded as purely line-of-sight.The complexity of the RoF link setup was reduced by employing a passive RF transmitter and a Schottky diode based envelope detector for signal downconversion, minimizing the number of required RF amplifiers and alleviating the need for RF mixers and a local oscillator at the receiver.Successful transmission is demonstrated for RF carriers at 75-87 GHz -i.e. the lower half of the W-band -and over wireless distances of 30-70 m, with distances of 100 m and beyond predicted to be achievable.The system is thus shown to be applicable in both indoor and outdoor scenarios, including building-to-building connections and mobile front-and backhaul.
Furthermore the performance impact of including bend insensitive fiber for on-site installation was experimentally analyzed, employing three samples of Sterlite bend insensitive fiber.A clear benefit in terms of tolerance to macrobending is found, with bend radii as small as 5 mm and more than ten turns allowable without significant degradation of system BER performance.
Through the reduction in complexity and the inclusion of bend insensitive fiber, the applicability of W-band hybrid photonic-wireless links as a key enabler for future wireless and mobile networks is highlighted and with the obtained transmission results their performance is experimentally confirmed.

Fig. 1 .
Fig. 1.Network scenarios for the hybrid fiber-wireless link.(a)-(d) outdoor medium distance links: (a) building-to-building communication, (b) recovery and protection of fiber links, (c) mobile front-/backhaul, (d) spanning obstacles and providing broadband access to rural ares; (e) short-range indoor wireless distribution

Fig. 2 .
Fig. 2. Schematic of the experimental setup for W-band channel characterization and photograph of the pair of pyramidal waveguide horns.LO: local oscillator, ESA: electrical spectrum analyzer

FrequencyFig. 3 .FrequencyFig. 4 .
Fig. 3. Received power over frequency at distances of 0.5-4 m; close up: showing the aggregated effects of path loss, shadowing and small-scale fading

Fig. 8 .
Fig. 8. Signal spectra: (a) Optical signal at the output of the 3 dB coupler before fiber transmission and photonic upconversion, for the generation of a signal with an 80 GHz carrier; (b) Received electrical spectrum after envelope detection of the 2.5 Gbit/s NRZ signal on an 84 GHz carrier and 70 m wireless transmission with P opt = 5 dBm