Advanced modulation formats for optical access networks

The explosion of Internet-based applications and media delivery over IP in recent years has fuelled the research for faster broadband connections in access networks, or ‘first-mile’ connections, as they are often called. It is widely agreed that fiber-based access networks are the only solution that can efficiently provide the required bandwidth in the long term, however there is a number of competing optical technologies that aim to form the basis for these future, high-performance access networks. In this thesis, the proposal is put forward to use multilevel modulation formats with differential detection and digital signal processing in order to efficiently scale the bit rate up to 40 Gb/s in future optical access networks. With multilevel modulation, the system bit rate can be increased beyond 10 Gb/s, the bit rate specified in the latest industry standards, while keeping the bandwidth of the required photonic and electronic components and the deployed optical distribution network unchanged. By employing differential detection, simpler receivers and digital signal processing are required, compared to the alternative method of detecting multilevel signals (coherent detection). DSP can mitigate impairments and allow for simpler transceivers in the optical part, at the price of increased reliance to digital electronics, which however are subject to Moore’s law. Regarding the implications due to the multiple-access nature of PON networks, the multilevel signals can be treated in the higher layers, responsible for bandwidth allocation, as binary streams with the equivalent bit rate. Alternatively, the order of the modulation can be exploited as an additional degree of freedom, if software-defined transceivers are available, allowing physical layer-aware MAC protocol optimization. Due to the short fiber lengths and power levels involved, no significant degradations to multilevel signals because of nonlinear effects caused by the varying levels of optical power in the fiber (as a result of different user link lengths) are expected. For the upstream channel, good linearity and dynamic range are required from the receiver, but automatic threshold setting can be avoided in DSP-based receivers. Furthermore, multilevel modulation, with its increased spectral efficiency, allows a reduction of packet duration, reducing congestion probability on the upstream link. Four potential candidate multilevel formats are identified and their performance is examined through simulations and experimental work. Two modulation formats, phase pre-integration and coded 16QAM, which enable differential detection of square QAM constellations by means of suitable signal processing, are evaluated though extensive simulations using the VPI software package. Phase pre-integration 16QAM is shown to yield very good receiver sensitivity at 10 Gb/s for both downstream and upstream channels, with a sufficient margin to cover for electrical impairments. Coded 16QAM can achieve 10 and 40 Gb/s error-free transmission for the downstream and 10 Gb/s for the upstream channel. On the other hand, it is more vulnerable to electrical impairments due to the closer symbol spacing of the received constellation diagram. We then proceed to validate the potential of multilevel differential formats experimentally. D8PSK and 16QAM have been chosen for the experimental work, as they are easier to generate at the symbol rates (10 Gsym/s) that are of interest in this work. A versatile experimental set-up has been developed, capable of generating and detecting a variety of differential multilevel formats. The receiver sensitivity of the two aforementioned modulation formats, acting as a downstream channel, is measured at 2.5 and 10 Gsym/s in the context of an access network. A cost-efficient asymmetric OOK signal is used as the upstream channel, with a bit rate equal to the downstream symbol rate. For the final experiment, bidirectional transmission over PON of a 40 Gb/s Star 16QAM downstream channel and a 30 Gb/s D8PSK upstream channel demonstrates the capability of these formats to scale efficiently the bit rate in both directions. The achieved sensitivities indicate that power budgets that allow large splitting ratios and long reach, specifically up to 128 users and 40 km, are possible for both formats, enabling high-speed future access networks. To exploit the possibilities offered by DSP-based detection, algorithms that simplify the receiver and enhance sensitivity are presented and evaluated through simulations and experimental data. By inserting pilot symbols at the start of every transmission, it is shown that there is no need for active alignment of the phase of the incoming signal, by means of optical phase shifters. Gram-Schmidt orthonormalization is used to compensate IQ imbalance and relax phase error requirements on the demodulator. Finally, a multiple symbol phase estimation algorithm that creates a better phase reference is applied to the experimental data collected from the system experiments to enhance the sensitivity of the receiver. Photonic integration is instrumental in enabling transceivers suitable for the cost-sensitive access networks market. Silicon-based devices are especially interesting, as silicon has a favorable cost profile and re-use of infrastructure developed for integrated CMOS electronics is possible. Integrated devices using Silicon-on-Insulator technology, designed and fabricated by external partners, are evaluated with differential formats. Specifically, a 40 GHz demodulator and two complete receivers, at 5 and 10 GHz, incorporating a novel MMI design and zero-biased Ge photodiodes, are extensively tested. Low phase errors and error-free detection of DQPSK is shown with these compact and low-complexity receivers. Results indicate that silicon-based devices can be important building blocks for cost-effective and high-performance multilevel transceivers.