Design of a mmWave Sigma-Delta-Based Radio-over-Fiber modulator

(1)

Radio-over-Fiber modulator

Design of a mmWave Sigma-Delta-Based

Academic year 2019-2020 Technology

Master of Science in Electrical Engineering - main subject Communication and Information Master's dissertation submitted in order to obtain the academic degree of

Counsellors: Dr. ir. Haolin Li, Chia-Yi Wu, Prof. dr. ir. Johan Bauwelinck Supervisors: Prof. dr. ir. Guy Torfs, Prof. dr. ir. Sam Lemey

Student number: 01504851

Jakob Declercq

(2)

(3)

Radio-over-Fiber modulator

Design of a mmWave Sigma-Delta-Based

Academic year 2019-2020 Technology

Master of Science in Electrical Engineering - main subject Communication and Information Master's dissertation submitted in order to obtain the academic degree of

Counsellors: Dr. ir. Haolin Li, Chia-Yi Wu, Prof. dr. ir. Johan Bauwelinck Supervisors: Prof. dr. ir. Guy Torfs, Prof. dr. ir. Sam Lemey

Student number: 01504851

Jakob Declercq

(4)

Permission to consult

The author gives permission to make this master dissertation available for consultation and to copy parts of this master dissertation for personal use. In all cases of other use, the copyright terms have to be respected, in particular with regard to the obligation to state explicitly the source when quoting results from this master dissertation.

Jakob Declercq, May 31, 2020

(5)

Acknowledgements

This master thesis has been a genuine learning experience for me, and I would first like to thank my supervisors prof. Guy Torfs and prof. Sam Lemey. Not only did they provide me with this interesting topic to study, they were always available to ask for advice. The same can definitely be said for my counselors dr. ir. Haolin Li, Chi-Yi Wu and prof. Johan Bauwelinck. However, the other members of the IDLAB design group were always ready to help me too. I would also like to thank my promotors, my counsellors, dr. ir. Michiel Verplaetse and dr. ir. Olivier Caytan for attending the biweekly follow-up meetings and assisting me and my fellow student Caro Meysmans, despite the COVID-19 confinement. Apart from this, I would especially like to thank dr. ir. Michiel Verplaetse for always being ready to answer the countless questions I had, and ir. Joris Van Kerrebrouck for his help with the PCB design.

The topics of electronics and communication never cease to interest me, and I am glad to have had the amazing opportunity to study the field of electrical engineering, only deep-ening this interest. However, thanks to the enthusiasm and friendship of my classmates, these past five years in Ghent have been more than just this. I am truly grateful for the good times we had and I am sure there will be more of these moments to come. In the context of this thesis, I would like to express my special thanks towards Borre, Reinier and Caro for both the meaningful and not so meaningful conversations we had in the thesis room.

Last but not least, my profound gratitude goes towards my family and my girlfriend, who have always supported me in pursuing my goals. Without them, none of this would have been possible.

(6)

Design of a mmWave

Sigma-Delta-Based Radio-over-Fiber

modulator

by

Jakob DECLERCQ

Master’s Dissertation submitted to obtain the academic degree of Master of Science in Electrical Engineering

Academic year 2019-2020

Supervisors: prof. dr. ir. Guy TORFS, prof. dr. ir. Sam LEMEY

Counselors: dr. ir. Haolin LI, Chia-Yi WU, prof. dr. ir. Johan BAUWELINCK Faculty of Engineering and Architecture

Ghent University

Department of Information Technology Chairman: Prof. dr. ir. Bart DHOEDT

Summary

All-digital radio-over-fiber provides an efficient way to interconnect the increasing amount of small radio cells in 5G mobile networks. A technique is studied to realize the downlink connection between the central office and a remote radio unit at mmWave frequencies, using sigma delta modulation. The upconversion to a frequency of 28 GHz is accomplished digitally by means of high-speed multiplexers. To reduce the speed requirements of these multiplexers, the in-phase and quadrature signal are combined optically using a parallel EAM structure. Additionally, by optimizing the phase of the optical carrier through the parallel EAM structure, the reach limitation introduced by chromatic dispersion is mitigated.

Keywords

(7)

1

Design of a mmWave Sigma-Delta-Based

Radio-over-Fiber modulator

Jakob Declercq

Supervisors: prof. dr. ir. Guy Torfs, prof. dr. ir. Sam Lemey Councellors: dr. ir. Haolin Li, Chia-Yi Wu, prof. dr. ir. Johan Bauwelinck

Abstract—All-digital radio-over-fiber provides an efficient way to interconnect the increasing amount of small radio cells in 5G mobile networks. A technique is studied to realize the downlink connection between the central office and a remote radio unit at mmWave frequencies, using sigma delta modulation. The upconversion to a frequency of 28 GHz is accomplished digitally by means of high-speed multiplexers. To reduce the speed requirements of these multiplexers, the in-phase and quadrature signal are combined optically using a parallel EAM structure. Additionally, by optimizing the phase of the optical carrier through the parallel EAM structure, the reach limitation introduced by chromatic dispersion is mitigated.

Index Terms—Radio-over-fiber, Sigma Delta modulation, mmWave, Electro-absorption Modulator

I. INTRODUCTION

T

HE fifth generation (5G) of digital cellular networks will enhance mobile connectivity to a new level by enabling higher data rates, massive device connectivity and reduced latency. Higher data rates can be obtained by increasing the bandwidth of the radio signal, but since bandwidth is scarce in the traditionally used sub-6GHz bands, the mostly unoccupied millimeter wave (mmWave) spectrum (above 24 GHz) looks promising. However, these higher frequencies exhibit much higher attenuation, meaning that the cell size will have to be smaller. A vital element in facilitating the connectivity and coordination between all these small cells is an increased level of centralization, yielding a more flexible and cost-effective Centralized Radio Access Network (C-RAN) and reducing the cost of every individual Remote Radio Unit (RRU). Radio-over-fiber (RoF) offers a way to efficiently interconnect the central office (CO) and the RRUs in such a network. Three main types of RoF exist: Analog Radio over Fiber (ARoF), Digitized Radio over Fiber (DRoF) and Sigma Delta over Fiber (SDoF) [1]. ARoF uses analog signals, which are highly sus-ceptible to noise and nonlinearities. By using binary signals, DRoF is more robust against these impairments, but it requires more complex RRUs. SDoF combines the benefits of both technologies by using a digital signal to represent the analog signal in such a way that the latter can be extracted after a simple filtering operation.

Sigma delta modulation of mmWave signals requires a very high sampling rate, necessitating the use of a parallellization technique [2] to enable implementation in digital logic. Mul-tiplexers provide the serialization of these parallel bitstreams into a high-speed binary stream. Previous work [3] showcased a real-time downlink connection covering a frequency range

from 22.75 GHz to 27.5 GHz. This was achieved by imple-menting two low-pass sigma delta modulators (SDMs) on an FPGA which generates four 25 Gb/s bitstreams that are combined in a 4:1 multiplexer (MUX). This MUX performs the digital quadrature upconversion, resulting in a bitrate of four times the carrier frequency. To achieve such a sampling rate (100 GS/s), state-of-the-art MUXs are required, but further frequency bands are hard to reach with this method. Besides, the intensity modulated optical signal will suffer from chro-matic dispersion.

That is why a method [4] is proposed to split the digital quadrature upconversion into two parts. High-speed MUXs perform the digital upconversion of both I and Q signals separately at a rate of twice the carrier frequency, and these two upconverted signals are then combined optically. This principle has been demonstrated experimentally at a carrier frequency of 28 GHz by using parallel electro-absorption modulators (EAMs) to combine both signals optically [4]. An additional benefit of this method is that the influence of chromatic dispersion can be mitigated by tuning the phase of the optical carrier that each EAM branch modulates. In this work, this system is further investigated and a real-time implementation is proposed. Furthermore, a separate upconverter PCB is designed, according to the form factor of a QSFP module such that it can be easily interconnected with an FPGA.

II. SYSTEMDESCRIPTION

The system that is investigated in this work (Fig. 1) operates at a carrier frequency (fc) of 28 GHz and consists of a central office and a remote radio unit which are connected by means of standard single mode fiber (SSMF). The RRU architecture is of low complexity and only requires a photodiode (PD),

Figure 1: System architecture of the mmWave SDoF downlink connection.

(8)

2

Figure 2: Illustration of the output spectrum of a low-pass sigma delta modulator (left) and of the digital upconversion (right), where fs= 2fc.

a power amplifier (PA) and a band-pass filter (BPF), which extracts the RF signal from the binary signal. However, this work focuses on the architecture of the central office. A. Sigma Delta Modulation

Sigma delta modulation can be used to encapsulate an analog signal in a digital bitstream. The digital nature of this stream introduces quantization noise, but owing to the high oversampling and the feedback inside the modulator, it will be shaped outside the frequency band of interest (Fig. 2 left). Figure 3 shows the block diagram of a second-order low-pass SDM with a single-bit quantizer. However, real-time digital implementation of this architecture with sampling rates of up to 56 GS/s is not feasible. Due to the hard non-linearity of the quantizer, typical loop unrolling methods can’t be used. An alternative parallellization technique is proposed in [2], demonstrating SDM operation up to 21 GS/s operating on a Xilinx Virtex Ultrascale VCU108 FGPA. This technique was also adopted in this work to sample the baseband I and Q signals at a rate of 56 GS/s. For each signal (I and Q), the parallel streams generated on the FPGA are multiplexed into two 28 GS/s NRZ streams representing the even and odd bits of the final stream (Fig.1):

[ I(n), Q(n),_{−I(n + 1), −Q(n + 1) ]} n = 0, 2, 4, ... B. Digital Upconversion

These I and Q streams are serialized externally by means of two high-speed multiplexers. The inversion of every odd bit causes the signals to be digitally upconverted to a car-rier frequency (fc) of 28 GHz. The resulting 56 Gb/s NRZ streams thus represent the RF signals I(t) cos(2πfct) and Q(t) cos(2πfct), which could be recovered after band-pass filtering (Fig. 2 right). A tunable delay line causes the Q signal to experience a phase shift of π/2 at the carrier frequency, yielding the quadrature signal: Q(t) sin(2πfct). However, this delay also causes an IQ mismatch, but if the signal bandwidth

+ + z−1 2 z−1 -+

-Figure 3: Block diagram of a second order low-pass SDM [3].

Figure 4: Diagram showing the structure of parallel EAMs, using NRZ driver IC’s.

is small with respect to the carrier frequency, the error is negligibly small. For more broadband signals, the effect can be pre-compensated in the digital baseband.

C. Optical Combination of I and Q signals

An EAM consists of a semiconductor PIN junction and operates based on the Franz-Keldysh effect, where the bandgap of the semiconductor material decreases in the presence of an electrical field, causing an increase in the absorption coefficient at the wavelength of operation [5]. An NRZ driver IC is used to drive the modulator differentially and to reversely bias it while sinking the generated photocurrent [6]. In figure 4 two parallel EAMs are combined with an optical phase shifter, and this structure is used to combine the upconverted I and Q signals and to modulate the optical carrier.

III. INFLUENCE OFCHROMATICDISPERSION

In an optical fiber, the propagation characteristics of elec-tromagnetic waves are frequency dependent. This means that different frequency components of the optical signals travel at different speeds, causing a phase difference between these components when arriving at the other end of the fiber (with λo and ωorespectively the wavelength and pulsation of the optical carrier, L the fiber length and D the dispersion parameter) [7]:

∆φ =_−L λo 2ωo

D(ω_{− ω}o)2 (1)

This effect is called chromatic dispersion and it can cause a severe degradation of the detected signal power at certain fiber lengths. In a RoF system, optical amplifiers (EDFAs) are used to compensate for the fiber attenuation. This restricts the laser wavelength to the C-band, where these EDFAs exhibit gain [8]. However, a SSMF shows a relatively high dispersion parameter (around 16 ps/nm/km) at these wavelengths, render-ing traditional intensity-modulation direct-detection (IM/DD) architectures unusable at certain fiber lengths [3]. This effect can be mitigated by using parallel EAMs (Fig.4), which can be described by the following linearized model (neglecting the modulator chirp): Eout= A 2e jωot_v 1(t) + v2(t)ejθo (2) A DC bias is added to the upconverted I and Q signals (v1(t) = I cos(ωct)+Vband v2(t) = Q sin(ωct)+Vb) and they are each

(9)

3 ∆φ 1 −1 π 2 π 3 π 2 2π cos(∆φ) √₂ 2 cos(∆φ + π/4) √₂ 2 sin(∆φ + π/4)

Figure 5: Curves indicating the performance for the two analyzed cases: 1) θo = 0 (red), 2) θo = −π/2 (blue and green).

applied to an arm of the parallel EAMs, yielding the following optical field: Eout= A 2 I + Qej(θo−π/2) 2 e j(ωo+ωc)t +I− Qe j(θo−π/2) 2 e j(ωo−ωc)t + 2 cos(θo/2)Vbej(ωot+θo/2) ! (3)

The optical field consists of the optical carrier and both the lower and upper sideband, which both experience a phase shift ∆φ due to chromatic dispersion. The photodiode at the RRU detects the optical power |Eout|2, which contains the following signal contribution: cos(θo/2) I cos(ωct) cos(θo/2− ∆φ) + Q sin(ωct) cos(θo/2 + ∆φ) (4)

At fiber lengths where ∆φ ≈ kπ, k ∈ Z, it is optimal to choose θo = 0. In that case (4) reduces to:

cos(∆φ)I cos(ωct) + Q sin(ωct)

(5) However, at fiber lengths where ∆φ ≈ π/2 + kπ, k ∈ Z, the received signal power is heavily reduced. This is very similar to the behavior of optical double sideband modulation [9]. The signal can be recovered by changing θo to a value of −π/2. Then (4) reduces to:

√ 2 2 I cos(ωct) cos(∆φ + π/4) + Q sin(ωct) sin(∆φ + π/4) (6)

Now the I and Q signals are scaled by a different factor. These factors, along with the factor cos(∆φ) of (5) are plotted in figure 5. For other values of ∆φ than those discussed here, optimal results are achieved for different values of θo.

IV. SYSTEMSIMULATIONS

To compare the performance of the proposed system to the reference system, where the full quadrature upconversion is performed electrically [3], simulations are performed using VPIphotonics TransmissionmakerTM_{. A QPSK symbol stream}

0 5 10 15 20 Fiber length [km] 0 5 10 15 20 EVM [%rms] parallel EAMs single EAM

(a) EVM of the received symbols

0 5 10 15 20 Fiber length [km] -50 -45 -40 -35 -30 -25 -20 Received power [dBm] parallel EAMs single EAM (b) Received power

Figure 6: EVM and received power for different fiber lengths, where the input amplitude of the EAMs and the phase shift θo are optimized for a minimal EVM.

of 218.75 MBd is upsampled by a factor 2 and pulse shaped with a SRRC pulse (rolloff 0.2). Next, the I and Q streams are oversampled by a factor of 128 and applied to a second order SDM, followed by a digital upconversion. These 56 Gb/s bitstreams are pulse shaped and the Q stream is delayed to obtain the quadrature signal, after which both signals are applied to the parallel EAMs. The modulated optical carrier (10 dBm at 1550 nm) travels through a single mode fiber of variable length and at the other end, a photodiode (responsivity of 1 A/W) detects the optical power and adds thermal noise (10 pA/√Hz) and shotnoise. A receiver then performs the downconversion and extracts the received symbols. In the case of the reference system, the 56 Gb/s I and Q bitstreams are interleaved and applied to a single EAM at a double rate (after pulse shaping).

For every fiber length, the amplitude of the EAM input signal and the phase shift θo (only for the parallel EAM system) are optimized to minimize the EVM. This yields the results shown in figure 6. The reference system (single EAM) performs better at fiber lengths where the chromatic disper-sion has a limited effect (around 0 km, 10 km and 20 km). However, at fiber lengths of 5 km and 15 km (corresponding to ∆φ = π/2 + kπ, k ∈ Z), the received power (Fig.6b) drops

(10)

4

Figure 7: The architecture of the upconverter system. significantly and hence the EVM (Fig.6a) rises, rendering this system unusable. On the other hand, the proposed system (using parallel EAMs) makes it possible to compensate this reduction in power by changing the value of θo, resulting in more reasonable EVM values for all possible fiber lengths. Still, a gain imbalance between the I and Q signals remains, but this could be solved by tuning the separate amplitudes of the input signals or the splitting ratio of the parallel EAM feeding network.

V. UPCONVERTERPCB DESIGN

In order to enable real-time measurements, a high-speed upconverter PCB is designed. This module interconnects to the FPGA via a QSFP interface and essentially consists of two multiplexers, but some additional components and circuitry are required to get everything working properly. Figure 7 shows the general system that realizes the digital upconversion. A. Component Selection

The main components are the high-speed multiplexer chips [10], denoted as DATU. These chips were developed in-house and are wirebonded onto the PCB. Since their integrated clock and data recovery circuits exhibit too much phasenoise, an external clock must be supplied to perform the multiplexing. In order to synchronize the inputs to these multiplexers, an RF switch (HMC641ALC4) is put in place to feed back the different binary streams to a single high-speed receiver on the FPGA. The DATU chips and the RF switch are controlled by an I2C bridge (SC18IS602B) which communicates with the FPGA over the I2C bus of the QSFP interface. A multi-coax connector (TR70) is used to interconnect the high-speed signals with this PCB.

B. Hardware Design

The components are positioned on the PCB as is indicated in figure 8 and are interconnected using grounded coplanar

Figure 8: Component placement on the top side of the board, along with a topological indication of the most important high speed interconnections.

waveguide technology. The electromagnetic properties of this PCB were simulated with ADS Momentum in order to verify the signal integrity of these high-speed signals.

VI. CONCLUSION

A SDoF downlink connection between the CO and RRU at mmWave frequencies is presented, in which the upconversion is partially performed in the optical domain. The working prin-ciple of this system is demonstrated by means of simulations. Additionally, a strategy to combat the effects of chromatic dispersion is presented, which shows an additional advantage of using this topology, apart from the lower required sampling rate of the external multiplexers. Finally an upconverter PCB was designed, which will enable future experiments and fur-ther research on this topic.

REFERENCES

[1] L. Breyne, G. Torfs, X. Yin, P. Demeester, and J. Bauwelinck, “Compar-ison between analog over-fiber and sigma delta modulated radio-over-fiber,” IEEE Photonics Technology Letters, vol. 29, no. 21, pp. 1808–1811, 2017.

[2] H. Li, L. Breyne, J. Van Kerrebrouck, M. Verplaetse, C. Wu, P. De-meester, and G. Torfs, “A 21-GS/s single-bit second-order delta–sigma modulator for FPGAs,” IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 66, no. 3, pp. 482–486, 2019.

[3] H. Li, M. Verplaetse, J. Verbist, J. Van Kerrebrouck, L. Breyne, C. Wu, L. Bogaert, B. Moeneclaey, X. Yin, J. Bauwelinck, P. Demeester, and G. Torfs, “Real-time 100-GS/s sigma-delta modulator for all-digital radio-over-fiber transmission,” Journal of Lightwave Technology, vol. 38, no. 2, pp. 386–393, Jan 2020.

[4] H. Li, J. V. Kerrebrouck, H. Ramon, L. Bogaert, J. Lambrecht, C.-Y. Wu, L. Breyne, J. Declercq, J. Bauwelinck, X. Yin, P. Ossieur, P. Demeester, and G. Torfs, “Low power all-digital radio-over-fiber transmission for 28-GHz band using parallel electro-absorption modulators,” in Optical Fiber Communication Conference (OFC) 2020. Optical Society of America, 2020, p. M2F.6. [Online]. Available: http://www.osapublishing.org/abstract.cfm?URI=OFC-2020-M2F.6 [5] S. A. Srinivasan, P. Verheyen, R. Loo, I. De Wolf, M. Pantouvaki,

G. Lepage, S. Balakrishnan, W. Vanherle, P. Absil, and J. Van Campen-hout, “50Gb/s C-band GeSi waveguide electro-absorption modulator,” in 2016 Optical Fiber Communications Conference and Exhibition (OFC), 2016, pp. 1–3.

[6] H. Ramon, J. Lambrecht, J. Verbist, M. Vanhoecke, S. A. Srinivasan, P. De Heyn, J. Van Campenhout, P. Ossieur, X. Yin, and J. Bauwelinck, “70 Gb/s low-power DC-Coupled NRZ differential electro-absorption modulator driver in 55 nm SiGe BiCMOS,” Journal of Lightwave Technology, vol. 37, no. 5, pp. 1504–1514, 2019.

(11)

5

[7] G. Meslener, “Chromatic dispersion induced distortion of modulated monochromatic light employing direct detection,” IEEE Journal of Quantum Electronics, vol. 20, no. 10, pp. 1208–1216, 1984.

[8] H. Schmuck, “Comparison of optical millimetre-wave system concepts with regard to chromatic dispersion,” Electronics Letters, vol. 31, no. 21, pp. 1848–1849, 1995.

[9] G. H. Smith, D. Novak, and Z. Ahmed, “Overcoming chromatic-dispersion effects in fiber-wireless systems incorporating external mod-ulators,” IEEE Transactions on Microwave Theory and Techniques, vol. 45, no. 8, pp. 1410–1415, 1997.

[10] J. Verbist, M. Verplaetse, S. A. Srinivasan, J. Van Kerrebrouck, P. De Heyn, P. Absil, T. De Keulenaer, R. Pierco, A. Vyncke, G. Torfs, X. Yin, G. Roelkens, J. Van Campenhout, and J. Bauwelinck, “Real-time 100 Gb/s NRZ and EDB transmission with a GeSi electroabsorption modulator for short-reach optical interconnects,” Journal of Lightwave Technology, vol. 36, no. 1, pp. 90–96, 2018.

(12)

List of Figures

1.1 Infographic of Qualcomm illustrating the use of mmWave frequencies along-side mid band frequencies [2]. . . 2 1.2 Typical implementation of different RoF downlink techniques . . . 3 1.3 System architecture of the mmWave SDoF downlink connection

investi-gated in previous work [6]. . . 5 1.4 System architecture of the mmWave SDoF downlink connection

investi-gated in this thesis. . . 6 2.1 Spectrum of an oversampled digital signal. . . 9 2.2 Block diagrams of a first order low-pass SDM [5]. . . 9 2.3 Spectrum of an oversampled digital signal with noise shaping applied. . . . 10 2.4 Block diagrams of a second order low-pass SDM [6]. . . 11 2.5 Frequency response of the NTF for low-pass SDMs of different order. . . . 11 2.6 Time domain output of an SDM . . . 13 2.7 Power spectra of the output of an SDM . . . 13 2.8 Comparison of the performance of a first and second order SDM for different

values of OSR. . . 14 2.9 Block diagram of a typical analog quadrature upconversion architecture. . 15 2.10 Sampling the quadrature carrier signals at a sampling rate fs = 4fc . . . . 16

2.11 Block diagram of a one-bit digital quadrature upconverter, where fs = 4fc. 16

2.12 Block diagram of an all-digital transmitter, with the sampling rates indicated. 18 xiii

(17)

xiv LIST OF FIGURES

2.13 Simulation of an all-digital transmitter . . . 19

2.14 Diagram showing the upconversion of the I and Q signals using multiplexers. 20 2.15 The power transfer function of an MZM in function of the voltage difference applied to the phase modulators. . . 22

2.16 Diagram showing the structure of the parallel EAMs, using NRZ driver IC’s. 24 3.1 Simulation model of the digital signal generation in VPIphotonics. . . 27

3.2 Simulation of the effect of different pulse shapes on digitally upconverted sigma-delta modulated bitstreams. . . 28

3.3 Converting discrete bitstreams to continuous-time signals in VPIphotonics. 29 3.4 VPI model of the parallel EAMs (figure 2.16). . . 30

3.5 VPI model representing the system described in figure 1.4 . . . 30

3.6 Power spectra of the signals indicated in figure 3.5. . . 32

3.7 Received I and Q signals . . . 32

3.8 EVM and received power in function of the peak-to-peak voltage of the signals applied to the EAMs, for different values of their bandwidth. . . 33

3.9 VPI model representing the reference system described in figure 1.4 . . . . 34

3.10 Power spectra of the signals indicated in figure 3.9, associated with the reference system. . . 35

3.11 EVM and received power of the reference system, in function of the peak-to-peak voltage of the signals applied to the EAMs, for different values of their bandwidth. . . 35

3.12 VPI simulation to show the effect of ODSB modulation on the optical spectrum. . . 39

3.13 VPI simulation to show the effect of OSSB modulation on the optical spec-trum. . . 41

3.14 Performance for the two analyzed cases of θo . . . 44

3.15 EVM and received power for different fiber lengths . . . 45

3.16 Scatter plot of the received symbols when a fiber of 4 km is used, illustrating the gain imbalance. . . 47

(18)

LIST OF FIGURES xv

3.17 EVM for different fiber lengths in the absence of noise . . . 47

4.1 The architecture of the upconverter system. . . 49

4.2 Top view of the VCU108 evaluation board. [23] . . . 50

4.3 Layout of the QSFP+ contact pads [22] . . . 51

4.4 Main building blocks of the DATU chip. . . 51

4.5 Microscope picture of the DATU chip (1.0 × 3.8 mm). . . 52

4.6 Test setup to measure the performance of the integrated CDR of the DATU chip. . . 53

4.7 Constellation diagram of the output of the DATU chip with or without using the internal CDR . . . 54

4.8 DATU multiplexer structure. . . 55

4.9 VNA measurements of the RF switch . . . 56

4.10 Level shifter circuit to transform 3.3 V logic into −5 V logic. . . 57

4.11 Digital communication within the system. . . 57

4.12 Block diagrams of the RF switch and I2C bridge chips . . . 58

5.1 Layer stackup of the PCB. . . 60

5.2 Component placement on the top side of the board, along with a topological indication of the most important high speed interconnections. . . 61

5.3 Using the GRM1555C1H561JA01 capacitor as an AC coupling capacitor. . 62

5.4 GCPW structures used as transmission line in this thesis (not to scale). . . 65

5.5 Via transition between top and bottom layer, simulated using ADS Mo-mentum. . . 67

5.6 Bondwire diagram of the left DATU chip . . . 68

5.7 Transmission line circuit of an equal split Wilkinson power divider. . . 69

5.8 Simulation results of a power divider using ideal quarter-wave (at 56GHz) transmission lines. . . 70

(19)

xvi LIST OF FIGURES 5.9 Layout and performance of a GCPW power splitter designed to operate at

56 GHz. . . 71

5.10 Different ways of representing a differential network. . . 72

5.11 Cutout of the layout containing a differential transmission line with a sig-nificant length difference between both traces. . . 74

5.12 Mixed-mode S-parameter results of the Momentum simulation of the dif-ferential transmission line displayed in figure 5.11. . . 75

5.13 PCB structure to simulate the effect of a transition between differential and single-ended transmission lines. . . 75

5.14 Mixed-mode S-parameter results of the Momentum simulation of the dif-ferential to single-ended transition displayed in figure 5.13. . . 76

5.15 A cutout of the PCB layout in a densely routed area. . . 77

5.16 S-parameter results of transmission lines in a densely routed part of the layout (figure 5.15), obtained using ADS Momentum. . . 78

5.17 Eye diagram of a 28 Gb/s stream (10 ps rise- and falltime) after being sent through the worst performing transmission line of figure 5.15. . . 79

5.18 Panel including Caro’s PCB, the PCB designed in this thesis and some test boards. . . 80

5.19 Photographs of the manufactured boards. . . 81

A.1 Top and bottom view of the ordered PCBs . . . 89

A.2 PCB layout of the top signal layer . . . 90

A.3 PCB layout of the top ground layer . . . 91

A.4 PCB layout of the bottom ground layer . . . 92

(20)

List of Tables

3.1 Parameters related to the signal generation. . . 27

3.2 Parameters related to the pulse shaping. . . 29

3.3 Parameters related to the simulation of the parallel EAM system. . . 31

5.1 Path lengths for each bitstream trace. . . 67

(21)

List of Acronyms

ARoF Analog Radio over Fiber.

AWG Arbitrary Waveform Generator.

BPF Bandpass Filter.

C-RAN Centralized Radio Access Network.

CDR Clock and Data Recovery.

CO Central Office.

CW Continuous Wave.

DAC Digital-to-Analog Converter.

DBB Digital Baseband.

DRoF Digitized Radio over Fiber.

EAM Electro-absorption modulator.

EDFA Erbium Doped Fiber Amplifier.

eMBB Enhanced Mobile Broadband.

EMC Electromagnetic Compliance.

EVM Error Vector Magnitude.

FFE Feedforward Equalizer.

FPGA Field Programmable Gate Array.

(22)

List of Acronyms xix

GCPW Grounded Coplanar Waveguide.

IoT Internet of Things.

MASH Multi-Stage Noise Shaping.

MIMO Multiple-input and Multiple-output.

mMTC Massive Machine Type Communications.

mmWave Millimeter Wave.

MZM Mach-Zehnder Modulator.

NRZ Non-Return-to-Zero.

NTF Noise Transfer Function.

ODSB Optical Double Sideband.

OSR Oversampling Ratio.

OSSB Optical Single Sideband.

PA Power Amplifier.

PCB Printed Circuit Board.

PRBS Pseudo-random Bitstream.

RAN Radio Access Network.

RoF Radio over Fiber.

RRU Remote Radio Unit.

RTO Real-time Oscilloscope.

SDM Sigma Delta Modulator.

SDoF Sigma Delta over Fiber.

SMPA Switched-mode Power Amplifier.

SNR Signal-to-Noise Ratio.

(23)

xx List of Acronyms

SRRC Square-root Raised Cosine.

SSMF Standard Single Mode Fiber.

STF Signal Transfer Function.

TDM Time-Division Multiplexing.

URLLC Ultra-Reliable Low-Latency Communication.

VCO Voltage Controlled Oscillator.

VGA Variable Gain Amplifier.

VNA Vector Network Analyzer.

VSA Vector Signal Analysis.

(24)

1

Introduction

The modern society is evolving to be ever more dependent on its mobile communication infrastructure. Being able to communicate to any person or service on this planet in a matter of milliseconds is increasingly seen as a vital necessity. Without it, our day to day lives would be totally different. However, the current generation of mobile networks (4G) is not capable of providing the services that our ever more mobile and connected society requires. As a result, the fifth generation (5G) of digital cellular networks is meant to enhance mobile connectivity to a new level, bundling the following classes of service into one unified infrastructure [1]:

• Enhanced Mobile Broadband (eMBB) promises higher data rates, required for high definition video streaming and other high-throughput applications.

• Ultra-Reliable Low-Latency Communication (URLLC) enables secure and reliable real-time connectivity, typical applications include self-driving vehicles and remote surgery.

• Massive Machine Type Communications (mMTC) provides connectivity for Internet of Things (IoT) purposes, in which a substantial amount of devices are connected. These devices are characterized by their low data rate requirements, enabling them to be of low complexity and to have a low power consumption.

(25)

2 CHAPTER 1. INTRODUCTION The improvements promised by 5G can only be achieved thanks to various innovative technological advancements of the last years and a lot of these topics are still actively being researched.

1.1 mmWave frequencies

In order to implement the eMBB service class, the 5G network should be able to han-dle high data rates. Alongside other techniques like increasing the spectral efficiency, higher data rates can be obtained by increasing the bandwidth of the radio signal. As additional bandwidth is scarce in the traditionally used low band (below 1 GHz) and mid band (1 to 6 GHz), the mostly unoccupied Millimeter Wave (mmWave) spectrum (above 24 GHz) looks promising (figure 1.1). However, these higher frequencies exhibit much higher attenuation, meaning that the cell size will have to be smaller when working in the mmWave band. If radio cells have to be smaller, more of them will be required in order to serve the same area. These smaller and more numerous cells not only enable the use of mmWave frequencies, by increasing the amount of frequency reuse they also help to fulfill the promises of massive device connectivity and higher data rates.

Figure 1.1: Infographic of Qualcomm illustrating the use of mmWave frequencies alongside mid band frequencies [2].

1.2 Radio over Fiber

A vital element in facilitating the connectivity and coordination between all these small cells is an increased level of centralization, yielding a more flexible and cost-effective Radio Access Network (RAN) and reducing the cost of every individual Remote Radio Unit (RRU). If the interconnections between the Central Office (CO) and the RRUs in such a Centralized Radio Access Network (C-RAN) consist of optical fibers, this is usually

(26)

1.2. RADIO OVER FIBER 3

(a) ARoF.

(b) DRoF.

(c) SDoF.

Figure 1.2: Typical implementation of different RoF downlink techniques. LO: Local Os-cillator; E-O: Electrical-to-optical conversion; O-E: Optical-to-electrical conversion; SER: Serializer; DES: Deserializer; DU: Digital Upconversion

termed Radio over Fiber (RoF). Transporting a radio signal over an optical fiber has the following main benefits [3]:

• The transmission losses over an optical fiber are much lower than those in electrical transmission lines such as coax cables. Especially transmission of high frequency radio signals (e.g. mmWave signals) over electrical cables is very impractical for longer distances.

• Due to the high frequency of the light on which the signals are modulated, fiber-optic communication offers a very large bandwidth. This means that multiple signal streams can be multiplexed onto the same fiber by means of Time-Division Multi-plexing (TDM) or Wavelength-Division MultiMulti-plexing (WDM).

• Fiber-optic communication is practically immune to electromagnetic interference. There exist three main types of Radio over Fiber (RoF): Analog Radio over Fiber (ARoF), Digitized Radio over Fiber (DRoF) and Sigma Delta over Fiber (SDoF). [4]

(27)

4 CHAPTER 1. INTRODUCTION

1.2.1 Analog Radio over Fiber

Conceptually, Analog Radio over Fiber (ARoF) is the most straightforward RoF technique (figure 1.2a). The analog RF signals are generated at a central office and modulated onto a fiber. These optical signals are then distributed to the RRUs. Since no additional signal processing is required, the RRU is quite simple and power efficient. Furthermore, ARoF has great spectral efficiency. On the other hand, analog signals are highly susceptible to noise and nonlinearities, meaning that the signal quality will inevitably deteriorate the further it travels.

1.2.2 Digitized Radio over Fiber

The binary nature of digital signals has a few advantages. Firstly, noise has much less influence on a binary signal compared to an analog signal. Secondly, a binary signal is much more robust against nonlinearities. This is very important in a RoF link since a lot of the components have limited linearity. For example low-cost optical modulators often show a high degree of nonlinearity. In a Digitized Radio over Fiber (DRoF) link, the multi-level baseband signal is first converted to a digital serial bitstream before it is sent over the fiber. At the RRU, this signal must be deserialized, and a high-speed DAC must convert this digital signal back to its analog form. Before amplification, an upconversion step is required.

This whole system is more complex and has a higher power consumption, especially at the RRU. Furthermore, DRoF requires very high data rates [5], resulting in a poor spectral efficiency. To give an example, if a 131.25 MHz quadrature baseband signal (see section 3) is to be transported via DRoF, and assuming that each sample is represented by 16 bits and 10/8 linecoding is required for error correction, a bitrate of 2 · 262.5MS/s · 16b/S · 10 8 = 10.5 Gb/s is required, which is a much higher speed than the 131.25MHz of the analog signal. On top of that, there will be additional latency in the link due to the framing of the digital data and the error correction. A big advantage of DRoF links is that they are very flexible: the same hardware can be easily reused for different communication standards and also other signals (e.g. for signaling, training or calibration purposes) can be sent over this channel.

(28)

1.3. THESIS OBJECTIVE 5

1.2.3 Sigma Delta based Radio over Fiber

Sigma Delta over Fiber (SDoF) tries to combine the advantages of both ARoF and DRoF: it represents the analog signal in a binary way, but without the complexity increase of DRoF. Conceptually, the analog baseband signal is converted into a digital stream by means of a Sigma Delta Modulator (SDM), which is upconverted digitally. This signal is sent over the fiber and recovered at the RRU. A Bandpass Filter (BPF) suffices to extract the analog RF signal from this bitstream, enabling the RRU to be simple and low-cost. For the same 131.25 MHz quadrature baseband signal of the DRoF example above, a significantly higher bitrate of 56 Gb/s is used for the SDoF link of this thesis, but this already includes the upconversion and only the CO must handle this bitrate, in contrast to DRoF.

In practice, the SDoF transmitter can be entirely realized digitally, just like the all-digital transmitter described in section 2.3. This yields a highly flexible RoF transmitter, en-abling more centralization. Since no Digital-to-Analog Converter (DAC) is required and switching amplifiers can be used, the power consumption can be lowered drastically. How-ever, this advantage is partially offset by the fact that the quantization noise constitutes a significant portion of the total power. [4]

1.3 Thesis objective

Figure 1.3: System architecture of the mmWave SDoF downlink connection investigated in previous work [6].

Previous work [6] has shown a novel technique to realize the downlink connection between the CO and a RRU at mmWave frequencies using SDoF. This requires a Sigma Delta Mod-ulator (SDM) with a very high sampling rate. In another paper [7] by the same authors, they show that this is possible by providing enough parallellization. A working system

(29)

6 CHAPTER 1. INTRODUCTION (figure 1.3) was showcased by implementing this SDM on a high-end Field Programmable Gate Array (FPGA) followed by a high-speed multiplexing chip which does the upcon-version to mmWave frequencies. This signal is then modulated onto an optical fiber and an RRU, consisting of a photodiode, a Power Amplifier (PA) and a BPF, subsequently applies the RF signal to an antenna.

Figure 1.4: System architecture of the mmWave SDoF downlink connection investigated in this thesis.

In this thesis, a similar architecture will be investigated, but this time performing a part of the upconversion in the optical domain and thereby relaxing the bitrate requirement of the electrical multiplexing chip and the optical modulator. Figure 1.4 shows the concrete implementation of this concept. Similar to the reference system described above (figure 1.3), the RF signal is generated by the Digital Baseband (DBB) block on the FPGA and the in-phase (I) and quadrature (Q) components are subsequently applied to SDMs (see section 2.1). Two external multiplexers upconvert these signals to a carrier frequency of 28 GHz (see section 2.4) after which the multiplexed outputs are applied (with a phase difference) to a parallel Electro-absorption modulator (EAM) structure, combining the upconverted I and Q signals and modulating the resulting signal onto an optical carrier in one step (see section 2.5.5). This optical carrier is generated by a Continuous Wave (CW) laser source. The modulated optical signal is transported from the CO to the RRU by means of Standard Single Mode Fiber (SSMF), where the RF signal is extracted and put onto an antenna in an identical manner as in the the reference system. Since the FPGA configuration and the RRU are almost entirely similar to the reference system, the main focus of this thesis is the combined upconversion and optical modulation strategy, and the integration of all the parts into a working system. To facilitate further experiments on this topology, the electrical multiplexers are assembled on a separate upconverter Printed Circuit Board (PCB), which is designed as a QSFP module such that it can be easily interconnected with the FPGA.

(30)

1.4. THESIS STRUCTURE 7

1.4 Thesis structure

First, some theoretical concepts will be explored in chapter 2 in order to understand how the proposed system will operate. Subsequently, the validity of this system concept is simulated and the influence of chromatic dispersion is investigated in chapter 3. Next, the hardware architecture for the upconverter is proposed and the required components are selected in chapter 4. With this in mind, the design of this upconverter PCB according to the QSFP form factor is described in chapter 5. Finally, the results of this thesis are discussed and a conclusion is drawn in chapter 6.

1.5 Influence of COVID-19 countermeasures

Due to the COVID-19 pandemic and the associated closure of the university lab, the functionality of the designed PCB could not be tested and no physical system experiments could be performed. Originally, this thesis would have included the adaptation of the FPGA configuration of dr. ir. Haolin Li in order to make these experiments possible. Only the low-speed interfacing between components has been programmed, but this is not reported since it is not relevant in the absence of measurements. As an alternative to these measurements, more elaborate calculations and simulations were performed on the system level (chapter 3) to verify the validity of this system concept.

This preamble was drawn up in mutual consultation between the student and the promo-tors, and was approved by all parties.

(31)

2

Theory

2.1 Sigma Delta Modulation

A digital signal can never fully represent an analog signal due to the fact that each digital sample can only represent a finite set of values, while an analog sample can take on an infinite amount of values. The error that is made by representing an analog signal by a digital code is treated as quantization noise. It can be shown that in a lot of situations, this quantization noise behaves approximately as white noise. Sigma delta modulation is a technique to reduce the effect of this quantization noise, by oversampling the signal and by removing the quantization noise out of the frequency band of interest [8]. This technique is especially powerful when an analog signal must be represented with a low number of bits per sample.

2.1.1 Oversampling

Apart from this quantization, digital signals are also sampled, meaning that the signal is only represented at discrete time instants, unlike analog signals, where the signal value is defined at every possible time. According to the Nyquist-Shannon sampling theorem,

(32)

2.1. SIGMA DELTA MODULATION 9

Figure 2.1: Spectrum of an oversampled digital signal.

a bandlimited analog signal can be represented by a discrete sequence of values without loss of information, provided that the sampling rate fs is higher than twice the highest

frequency f0 of this analog signal. The ratio of this Nyquist frequency fs/2 to the highest

frequency to be represented is called the Oversampling Ratio (OSR): OSR = fs

2f0. As

shown in figure 2.1, the quantization noise is evenly spread out up to the Nyquist fre-quency. If the OSR is large enough, only a small part of the quantization noise falls into the frequency band of interest and low-pass filter can be used to remove all out-of-band noise.

2.1.2 Noise shaping

+

z−1 +

-(a) Actual implementation

+ I(z) z−1 + -+ Q(z) O(z) (b) Linearized model

Figure 2.2: Block diagrams of a first order low-pass SDM [5].

To further reduce the effect of the quantization noise, the introduced quantization error can be fed back to the input in order to compensate it. The simplest example of such a feedback loop is the first order low-pass SDM of figure 2.2a. The effect of the quantizer

(33)

10 CHAPTER 2. THEORY can be approximately modeled as the addition of white quantization noise. This yields the linearized model of figure 2.2b, which simplifies the mathematical analysis. It is easily verified that the following relationship between the output and the input and quantization noise is valid (where I(z), Q(z) and O(z) represent the z-transforms of the input signal, quantization noise and output signal, respectively):

O(z) = I(z) · 1 |{z} ST F +Q(z) · (1 − z−1₎ | {z } N T F (2.1) The output of the SDM consists of two contributions. The Signal Transfer Function (STF) passes the input straight to the output, while the Noise Transfer Function (NTF) applies a first order high-pass filter to the quantization noise. In this way, the white quantization noise is shaped so most of its energy is concentrated in the higher frequencies. This concept is called noise shaping and is illustrated in figure 2.3. Since most of the quantization noise that resided in the signal band is now shifted to higher frequencies, it can be more easily removed by means of low-pass filtering and thereby the analog signal is recovered with a higher Signal-to-Noise Ratio (SNR).

Figure 2.3: Spectrum of an oversampled digital signal with noise shaping applied. The feedback loop of figure 2.4a implements a second order low-pass SDM. Again a linearized model is derived (figure 2.4b) and the output characteristic is described by:

O(z) = I(z) · 1 |{z} ST F +Q(z) · (1 − z−1₎2 | {z } N T F (2.2) Now the Noise Transfer Function (NTF) has a second order high-pass characteristic. In general, an n-th order SDM has an NTF of the form:

(34)

2.1. SIGMA DELTA MODULATION 11 + + z−1 2 z−1 -+

-(a) Actual implementation

+ I(z) + z−1 2 z−1 -+ -+ Q(z) O(z) (b) Linearized model

Figure 2.4: Block diagrams of a second order low-pass SDM [6].

The frequency characteristic of these filters can be obtained by plugging in z = ej2πf /fs into their transfer functions. This characteristic is plotted up to the fourth order in figure 2.5. A first observation is that a higher OSR reduces the quantization noise more because the NTF is very small for low frequencies. Secondly, a higher order SDM further enhances this effect. The plot on the right has a logarithmic frequency axis and it can be observed that for low enough frequencies, the slope is n · 20dB per decade. However, high order SDMs exhibit a very high NTF at high frequencies, which can render the control loop unstable [8]. 0 0.1 0.2 0.3 0.4 0.5 f T s -100 -80 -60 -40 -20 0 20 |NTF| [dB] n = 1 n = 2 n = 3 n = 4 10-4 10-3 10-2 10-1 f T s -100 -80 -60 -40 -20 0 20 |NTF| [dB] n = 1 n = 2 n = 3 n = 4

(35)

12 CHAPTER 2. THEORY

2.1.3 Single bit sigma delta modulation

In the context of this thesis, the main reason to employ sigma delta modulation is because it enables representing multilevel signals in a binary way. Single bit quantization yields a maximal amount of quantization noise, but sigma delta modulation can reduce its impact. Representing a signal using a binary signal can be beneficial since binary signals have a constant envelope which means that nonlinear components can be used without much problems, which can yield much simplified architectures and reduce the cost. To give a typical example: highly efficient Switched-mode Power Amplifiers (SMPAs) can be used instead of linear amplifiers which have much lower power efficiency [9]. In this thesis, it is mostly the optical modulator which has a nonlinear characteristic. On top of this, using a binary signal means that no high-speed DACs are required, the binary output of the digital signal generation platform (e.g. an FPGA) can be used directly.

2.1.4 Simulations

In order to gain more insight into the principle of sigma delta modulation, some simula-tions were carried out in MATLAB. Figure 2.6 shows a typical time domain output for a single-bit SDM, which is binary, in contrast to the input signal. In another simulation, a BPSK signal is upsampled by a factor of 2, pulse shaped (Square-root Raised Cosine (SRRC) pulse with a rolloff factor of 0.2) and applied to the first order low-pass SDM after an oversampling with a factor 16. As can be seen from the power spectrum (figure 2.7a), the output signal consists of the input signal and quantization noise. The effect of noise shaping is clearly visible and by using an appropriate low pass filter, the original analog signal could be extracted with a limited amount of distortion. Figure 2.7b shows the output spectrum when the same signal is applied to a second order SDM, clearly showing the steeper slope of the noise shaping.

Finally, the performance of this first and second order SDM is compared for different values of the OSR. The simulation setup is the same as is used for figure 2.7. The performance is measured in terms of Signal-to-Quantization-Noise Ratio (SQNR), being the ratio of the signal energy to the in-band noise energy. The resulting curves are shown in figure 2.8. The SQNR increases by 9 dB for every doubling of the OSR in the first order case, and by 15 dB in the second order case.

(36)

2.1. SIGMA DELTA MODULATION 13

Figure 2.6: The time domain output (black) of a single-bit first order low-pass SDM when a sinusoidal signal (blue) is applied to its input with an OSR of 16.

-5 0 5 f T 0 -40 -35 -30 -25 -20 -15 -10 -5 0 5 Power spectrum [dB] input SDM output SDM

(a) First order SDM (figure 2.2).

-5 0 5 f T 0 -40 -35 -30 -25 -20 -15 -10 -5 0 5 Power spectrum [dB] input SDM output SDM

(b) Second order SDM (figure 2.4).

Figure 2.7: Power spectra of the output of a single-bit low-pass SDM when a pulse-shaped BPSK signal is applied with an OSR of 16. T0 is the sampling period before oversampling.

(37)

14 CHAPTER 2. THEORY 8 16 32 64 128 256 OSR 20 40 60 80 100 SQNR [dB] 1st order simulation 2nd order simulation

Figure 2.8: Comparison of the performance of a first and second order SDM for different values of OSR.

2.1.5 High-speed SDM

The previous second-order SDM architecture has a simple structure, but in practice more advanced topologies must be adopted. The sampling rate of these SDMs is limited by the speed of the additions. A common way to overcome such speed bottlenecks in real-time digital hardware is to implement several sub-circuits in a parallel manner in such a way that their outputs can be combined to achieve a higher effective rate. However, due to the hard non-linearity of the quantizer, typical loop unrolling methods can’t be used. An approach that does allow for such parallellization is Multi-Stage Noise Shaping (MASH), where multiple low-order SDMs are cascaded to achieve a higher order behavior [10]. These low-order SDMs can be efficiently parallellized. On the other hand, this technique doesn’t yield a single-bit output since it adds up the single-bit outputs of the low-order SDMs. To acquire a binary output signal, a bit reduction step must be introduced after the MASH SDM. This bit reduction process is essentially another SDM, but this time with a reduced input width.

In [7], this technique is proposed to implement a high-speed single-bit second order SDM on a Xilinx Virtex Ultrascale VCU108 FPGA, demonstrating SDM operation up to 21 GS/s. In this paper, a MASH-1-1 SDM is used, which combines two first-order low-pass SDMs to form a two-bit output. A novel bit reduction process was developed in order to convert this 2-bit output into the desired single-bit signal in a parallel manner. Such parallellization techniques have not yet been devised for higher order SDMs. The technique of [7] will also be employed in this thesis since it can provide the necessary high-speed sigma delta modulated signals that are required in a real-time implementation for the targeted mmWave band.

(38)

2.2. QUADRATURE UPCONVERSION 15

I

(t)

cos 2πf

c

t

Q

(t)

sin 2πf

c

t

Figure 2.9: Block diagram of a typical analog quadrature upconversion architecture.

2.2 Quadrature upconversion

2.2.1 Analog quadrature upconversion

Quadrature upconversion is an RF technique in which two baseband signals (I(t) and Q(t)) are combined and upconverted onto a carrier frequency. In the analog domain, this can be accomplished by mixing both streams with quadrature carrier signals, which is depicted in figure 2.9. The output of this upconverter consists of the complex baseband (I(t) + j · Q(t)) signal which is now centered around the carrier frequency instead of around DC. In the receiver, this bandpass signal can then be downconverted by using an analog downconverter, which mixes this signal with both quadrature carrier signals, extracting the I and Q signal again.

2.2.2 Digital quadrature upconversion

As long as the Nyquist-Shannon sampling theorem is fulfilled, it is possible to implement this quadrature upconversion mechanism in the digital domain. By explicitly choosing the sampling frequency fs to be equal to 4 times the carrier frequency fc, a greatly simplified

architecture can be obtained [11]. As can be observed in figure 2.10, both quadrature carrier signals are represented by the sequences [+1, 0, -1, 0] and [0, +1, 0, -1] when sampling at this rate.

Note that the I and Q signals to be upconverted are only generated at a rate of 2fc, while

(39)

t

1 4·f

c

sin 2πf

c

t

cos 2πf

c

t

0 +1

0 -1

+1

0 -1

0

Figure 2.10: Sampling the quadrature carrier signals at a sampling rate fs= 4fc

I

(n)

2f

c

Q

(n)

4f

c

Figure 2.11: Block diagram of a one-bit digital quadrature upconverter, where fs = 4fc.

be upsampled by a factor of 2. If this is done right, the digital output stream is of the following form:

[ I(n), Q(n + 1), −I(n + 2), −Q(n + 3) ] n = 0, 4, 8, ... (2.4) A majorly simplified architecture can be obtained if these streams only consist of 1-bit values. In that case, the sign inversions can be easily implemented with binary inverters and multiplexers can be used to alternately select a sample from the original or inverted I or Q stream. This setup is depicted in figure 2.11.

In practice it is often not possible to do a perfect upsampling of the input signals, because of the sampling rate limitations of a real-time implementation. Another option is to not upsample these signals, which is equivalent to applying zero-order hold interpolation. The digital output stream is then constructed as follows (with the input signals at a rate of 2fc):

(40)

2.2. QUADRATURE UPCONVERSION 17 This is however not a perfect solution since the input Q signal is sampled at the wrong time instant, which introduces an IQ-mismatch equivalent to a slight delay of the baseband Q signal. This delay τ = 1

4fc corresponds to a frequency-dependent phase shift in the frequency domain:

Q(t − τ)−F→ Q(f) · ej2πτ f _(2.6)

Knowing this, the introduced error can be described in the frequency domain:

E(f) = Q(f) − Q(f) · ej2πτ f (2.7)

If the signal Q(f) is assumed to be band-limited (bandwidth B) and its power spectrum is assumed to be flat and normalized in this frequency band: |Q(f)|2 = 1

2B , then the power of this error signal can be approximated as follows:

Pe = B Z −B |E(f)|2_df = B Z −B |Q(f)|2· |1 − ej2πτ f|2df = 1 2B B Z −B 2(1 − cos(2πτf))df = 2 − 2sin(2πτB) 2πτB = 2 1 − sincB(τ) (2.8)

Since the signal power is assumed to be normalised (Ps = 1), the SNR due to the

IQ-mismatch can be determined:

SN R= Ps Pe = 1 2 1 − sincB(τ) (2.9)

In the simulations of chapter 3, the carrier frequency is rather high in comparison to the frequency of the baseband signals: fc= 28 GHz and B = 131.25 MHz (see table 3.1). This

yields an SNR of 47.5 dB, meaning that the effect of the IQ-mismatch will be negligibly small. If the carrier frequency would be lower, or the baseband bandwidth higher, then it might cause problems. However, the effect can be compensated for by means of an extra baseband processing step [5].

(41)

Figure 2.12: Block diagram of an all-digital transmitter, with the sampling rates indicated.

2.3 All-digital transmitter

An all-digital transmitter can be realized by combining the concepts of sigma-delta mod-ulation and digital quadrature upconversion [5]. In the block diagram of figure 2.12 the I and Q symbol streams are first pulse shaped (after being upsampled by a factor Ns).

Then these signals are upsampled by a factor OSR in order to be applied to SDMs. Both single-bit outputs are then upconverted together by means of digital quadrature upcon-version (using the method of equation 2.4). This binary signal can then be applied to a highly efficient SMPA, but before applying this signal to an antenna, a BPF must be applied in order to remove all out-of-band quantization noise.

Up to and including the upconverter, everything is done digitally. This provides a very flexible system that could be altered or tuned on the go. It also avoids the use of inefficient analog components such as high speed DACs and analog mixers. The downside is that the digital system must operate at a very high clock rate (fs = 4fc). This can be alleviated

partly by employing a high degree of parallellization and by making the multiplexers of figure 2.11 external.

This all-digital transmitter has been simulated in MATLAB. QPSK symbols are pulse shaped (SRRC pulse with a rolloff factor of 0.2) and upsampled with an OSR of 32. The real and imaginary part are then applied to second order SDMs and thereafter they are quadrature upconverted (with fs = 4fc). The power spectrum before and after after

(42)

2.4. SEPARATE UPCONVERSION OF THE I AND Q SIGNALS 19 -2 -1.5 -1 -0.5 0 0.5 1 1.5 f T_c -50 -45 -40 -35 -30 -25 -20 -15 -10 -5 0 Power spectrum [dB] I LP SDM output Q LP SDM output upconversion output

Figure 2.13: Power spectra of the input and output of a digital quadrature upconverter in a simulation of an all-digital transmitter (frequency axis referenced to carrier frequency).

2.4 Separate upconversion of the I and Q signals

As described in the introduction, the main objective of this thesis is to construct a SDoF transmitter that works with RF signals around a carrier frequency of 28 GHz. A high-performance FPGA will be used to perform the baseband processing and the low-pass sigma-delta modulation. In order to perform quadrature upconversion, a sampling rate of 4fc = 112 GS/s would be necessary (see section 2.2.2). However, the sigma-delta

processing can happen at a reduced rate thanks to parallellization (see section 2.1.5). Still, the quadrature upconversion must happen at a rate of 4fc, while the line rate of

the high-speed transceivers of the FPGA is limited to 30.5 Gb/s [12]. However, this can be overcome by performing the digital quadrature upconversion externally. As hinted in figure 2.11, an external 4:1 multiplexer might be used for this purpose [6]. In that case, the FPGA has to generate 4 streams at a reduced rate of 28 Gb/s:

(43)

Figure 2.14: Diagram showing the upconversion of the I and Q signals using multiplexers.

As discussed in the introduction, this approach is pushing the limits of what present-day multiplexers are capable of, as they should operate at a sampling rate of 4fc =

112 GS/s. These same limits also apply to the electrical-to-optical and optical-to-electrical converters. As an alternative, a slightly different approach is explored in this thesis (see figure 1.4). Two 2:1 multiplexers (operating at a lower rate of 2fc) combine the positive

and negative I and Q streams separately. This is shown in more detail in figure 2.14. The FPGA still generates the same 4 bitstreams at a rate of 28 Gb/s, and these are upconverted by the multiplexers into bitstreams at a rate of 2fc = 56 Gb/s.

These bitstreams are discrete versions of the following signals:    I(t) cos(2πfct) Q(t) cos(2πfct) (2.10) The diagram of figure 2.14 also features a time delay for the Q signal. In a test setup, this delay would be implemented with a line stretcher. If the delay is set to ∆t = 1

4fc, then this corresponds to a phase shift of π

2, generating the quadrature signal (note that this phase shift is only valid for a narrowband signal around 28 GHz):

   I(t) cos(2πfct) Q(t) sin(2πfct) (2.11)

These signals now have to be combined by an optical IQ modulator, which directly mod-ulates the signal onto an optical carrier as well. The working principle of this technique is explained in the following section.

(44)

2.5. OPTICAL MODULATION TECHNIQUES 21

2.5 Optical Modulation Techniques

2.5.1 Direct versus external modulation

The most straightforward way to send a bitstream over an optical fiber is to modulate the laser current, since the optical output power of a laser depends on it. This tech-nique is called direct modulation. Despite its simplicity, this method is not suited for high bitrate transmissions because of two reasons. Firstly, most lasers have a limited modulation bandwidth, meaning that the optical output power cannot change infinitely fast. Secondly, amplitude modulation of semiconductor lasers is inevitably accompanied by phase modulation. This effect is called frequency chirp, meaning that the frequency of the optical output shifts when the output power changes. This chirp causes the optical output spectrum to broaden significantly, degrading the performance of the optical sys-tem. Due to these two shortcomings, direct modulation is rarely used for bit rates higher than 5 Gb/s.

A second way to modulate a bitstream onto an optical carrier is by using an external modulator. In this case, the laser current is kept constant, resulting in a Continuous Wave (CW) optical output, which is not chirped. This optical carrier is applied to the input of an optical modulator, where an electrical signal modulates the optical carrier. These optical modulators often have a relatively high modulation bandwidth and introduce a lower amount of chirp. Therefore, external modulation will be used in this thesis. There are two main types of these modulators: the Mach-Zehnder Modulator (MZM) and the Electro-absorption modulator (EAM). [13]

2.5.2 Mach-Zehnder Modulator

In a Mach-Zehnder Modulator (MZM) the optical signal is split in two and on each arm a phase modulator applies a phase shift proportional to the applied electrical signal. After combining both optical signals again, these optical fields interfere constructively or destructively, dependent on the applied phase shifts. In this way amplitude modulation is achieved.

The transfer function of the optical field can be described mathematically as follows (with v1(t) and v2(t) the voltages applied to each phase modulator and Vπ the so-called half-wave

(45)

22 CHAPTER 2. THEORY voltage of the phase modulators):

Eout(t) Ein(t) = 1 2 ejv1(t)Vπ π + ejv2(t)Vπ π = cos v1(t) − v2(t) Vπ π 2 ! ejv1+v2Vπ π 2 (2.12)

Note that if the phase shifters are differentially driven (v1(t) = −v2(t)), no phase mod-ulation occurs and the modulator is chirp-free. Since a photodiode detects the optical power, it is instructive to look at the transfer function of the optical power:

Pout(t) Pin(t) = Eout(t) Ein(t) 2 = cos2 v1(t) − v2(t) Vπ π 2 ! = 1₂+ 1₂cos v1(t) − v2(t) Vπ π ! (2.13)

A plot of this transfer function is shown in figure 2.15 and one can observe that it is clearly nonlinear. However, if the signals are biased around v1(t) − v2(t) = 3V₂π (or V₂π), which is called the quadrature point, then the transfer function is approximately linear, provided that the signal swing around this bias is small enough:

Pout(t) Pin(t) ≈ 1 2 + π 2 · v1,ac(t) − v2,ac(t) Vπ (2.14) Note that the bias is often applied to separate low-speed phase modulators, isolated from the high-speed AC signals v1,ac(t) and v2,ac(t).

v

1

(t) − v

2

(t)

Pout(t) Pin(t)

1

0

Vπ 2

V

π

3

Vπ 2

Figure 2.15: The power transfer function of an MZM in function of the voltage difference applied to the phase modulators.

Design of a mmWave Sigma-Delta-Based Radio-over-Fiber modulator

Radio-over-Fiber modulator

Design of a mmWave Sigma-Delta-Based

Jakob Declercq

Radio-over-Fiber modulator

Design of a mmWave Sigma-Delta-Based

Jakob Declercq

Permission to consult

Acknowledgements

Design of a mmWave

Sigma-Delta-Based Radio-over-Fiber

modulator

Summary

Keywords

Design of a mmWave Sigma-Delta-Based

Radio-over-Fiber modulator

T

Contents

List of Figures

List of Tables

List of Acronyms

1

Introduction

1.1

mmWave frequencies

1.2

Radio over Fiber

1.2.1

Analog Radio over Fiber

1.2.2

Digitized Radio over Fiber

1.2.3

Sigma Delta based Radio over Fiber

1.3

Thesis objective

1.4

Thesis structure

1.5

Influence of COVID-19 countermeasures

2

Theory

2.1

Sigma Delta Modulation

2.1.1

Oversampling

2.1.2

Noise shaping

2.1.3

Single bit sigma delta modulation

2.1.4

Simulations

2.1.5

High-speed SDM

I

(t)

cos 2πf

t

Q

(t)

sin 2πf

t

2.2

Quadrature upconversion

2.2.1

Analog quadrature upconversion

2.2.2

Digital quadrature upconversion

t

1

4·f

c

sin 2πf

c

t

cos 2πf

c

t

0

+1

0