Introduction

Content

Optical Vector Demodulation

Coherent Optical Signal Detection

Abstract

Table Of Contents

Optical Vector Demodulation. 1

Coherent Optical Signal Detection. 1

Abstract 1

Introduction. 2

Coherent Detection For High Capacity Optical Networks 4

Unbalanced Coherent Detector Configuration. 4

Single Balanced Optical Detector Using Optical Hybrids 5

Optical Vector Demodulation Using Optical Hybrids And Balanced Optical Detectors 6

Data Capacity Estimates 8

Physical Implementation Details For The Optical Vector Demodulator 9

Device Matching Strategy. 9

Optical Hybrid Implementation. 10

IC Layout Geometry. 10

Conclusion. 10

Appendix. 10

Introduction

High data capacity fiber optical networks typically combine single laser diode transmitters combined with PIN diode detectors. Relatively simple binary on/off switching is typical; some precautions are made to prevent complete extinction of the transmitting laser diode but the basic configuration is relatively simple.

The configuration has 6 basic signal processing components

Input binary data is converted to a controlled current drive suitable for the laser diode transmitter. This provides a controlled “high = 1” laser diode drive current and an equally controlled “low = 0” laser drive current. A minimum “off” current is maintained to improve switching times from an “off” state to an “on state”. The resulting optical data stream is coupled to a fiber optic cable and presented to a PIN Photo Detector diode that produces a small analog output waveform corresponding to the data stream. A Trans-Impedance Amplifier (TIA) converts this current waveform to an analog voltage waveform whilst providing a low impedance termination required for fast response times. Finally a Threshold Detection Comparator (TDC) decides a binary “1” or “0” state from a potentially indeterminate input waveform.

Whilst the standard approach is simple, it can still achieve data rates up to 10 GB/s with standard optical transceiver modules. However it suffers from several disadvantages

Some efforts have been made to increase data carrying capacity for fiber optic networks using polarization diversity. This approach uses vertical polarization for one channel and horizontal polarization for a secondary channel. This dual channel configuration allows a form of vector signal construction that provides opportunity for QPSK – QAM signal constructions with associated improvements in Bits/Hz data throughput. For example, the use of two independent channels automatically allows the channel capacity to double over that of a single channel.

Although praiseworthy, this approach does not solve SNR issues, nor does it facilitate vector equalization potentially required for long fiber cable networks.

This document describes an alternative approach that might solve some of these residual problems and therefore maximize the data capacity potential of existing fiber cable networks. Although the cost of fiber cable is extremely low, once a network is implemented, adding additional fibers can be cost prohibitive. As we often find, user data capacity demands always exceed initial estimates however well planned. Therefore it is in our best interests to have secondary “back up” plans to better utilize data capacity of optical cable networks already laid down.

Coherent Detection For High Capacity Optical Networks

Unbalanced Coherent Detector Configuration

Coherent detection is used for measurements on weak optical signals. In this approach a Local Oscillator (LO) is added to the incoming optical signal and non-linear transfer characteristics inherent in a Photo Detection Device (e.g. a PIN photo diode) produce a low relatively frequency base-band output signal.

A coherent local oscillator (laser) is used to heterodyne an optical input signal down to a lower, more manageable baseband Intermediate Frequency (IF). This configuration is typically used for weak optical signal detection but this article will show how it can be modified to provide high data capacity transport networks. A coherent, phase locked local oscillator (LO) is assumed.

The frequency conversion different is subtly different from that used in conventional RF signal processing. The input and LO optical waves are added as complex exponentials prior to the photo detector. This component merely follows the envelope variations resulting from the summation;

…(1)

After some trigonometric expansion we find the photo detector output current becomes

 …(2)

We see the emergence of a constant DC term, a frequency difference term n - n^/ and a relative phase term f. We also see that the output current is proportional to the incoming optical signal amplitude and the LO amplitude. This second proportionality shows the potential for detector gain based on the addition of LO energy.

The advantages of this heterodyne approach includes

These features make the heterodyne approach seem attractive for weak signal detection. But how can it be used to solve the higher capacity problem?

Single Balanced Optical Detector Using Optical Hybrids

The solution to capacity enhancement is not based on an offset baseband IF approach (essentially narrow band and low capacity) but instead combines the advantages of heterodyne-based optical amplification with balanced optical detector design. Further, the use of optical hybrids are proposed; in the radio frequency analog these hybrids provide signal splitting and combining functions with varying degrees of amplitude and phase distribution.

Two identical photo detectors are used to produce matched baseband output signals. The difference between each is taken to produce a combined output signal centered at DC = 0 V.

The optical LO is provided to each photo detector with equal amplitude and phase. The Optical signal input is however presented with a 180 degree phase reversal. An optical hybrid is used to produce this (wavelength independent) phase offset.

This configuration provides an electronically amplified DC centered baseband output that can produce positive and negative signal excursions. The effect of LO amplitude variations (amplitude drift, amplitude perturbation noise) is cancelled. Further the advantages of optical amplification is provided due to the energy supplied from the LO source. Finally, the need for a frequency offset is avoided, so that the maximum available photo-detector bandwidth is available.

This single balanced optical detector provides a core component that will be used to construct an optical Vector Demodulation component. It possesses the following favorable characteristics

The previous heterodyne offset method solves the DC offset problem by producing a band-pass baseband signal. However the cost of this solution is severely constrained bandwidth capability. The single balanced optical detector fundamentally solves the DC offset problem through balancing. Further, a secondary “dark diode” is not simply used to provide a DC offset reference; significant is the use of a second photo detector equally employed for Optical to baseband signal conversion.

Optical Vector Demodulation Using Optical Hybrids And Balanced Optical Detectors 

The use of balanced mixers in vector demodulation devices is fundamental; extrapolating the same concept to optical wavelengths appears relatively straightforward. We require an additional optical hybrid component designed to produce dual outputs with a 90 degree phase offset. This can be placed in either the optical or LO paths, but the LO path implementation may provide a better practical realization.

This new configuration represents an optical analog of conventional analog IQ vector demodulators. The independent I + j Q output, centered at DC, contain all the complex signal information contained in the Optical input signal. In contrast to a single detector, the IQ output contains independent spectral information above and below the LO wavelength (frequency). Consequently, exact frequency equalization for long, wavelength dispersive optical cables is possible.

In summary, this optical vector signal demodulation component provides the following key advantages over conventional approaches

Aside:

The LO is assumed to be coherent with the incoming optical signal. As shown, an external, frequency stable reference is assumed to be available for “phase locking” the laser LO source. In principle, this reference can also be propagated along the fiber cable as a “pilot tone” added to the optical signal. Conventional approaches used to impart a narrow band optical carrier on an otherwise unusable part of the fiber cable’s wavelength “windows” are readily available; for example a 10 MHz reference could be used to directly modulate a secondary LED source that occupies some other portion of the optical spectrum and detected at the receiving end, therefore reconstructing the original 10 MHz reference.

Since this reference is continuous, narrow band detection is appropriate. This can use conventional LC band pass filtering followed by Phase Locked Loop (PLL) noise cleanup. The technical difficulties associated with this approach are minimal; also the solution is extremely cost effective and would probably cost less than $10 per system.  

However this overhead, albeit small, is probably not required for high-speed data. As is typical in high capacity QAM systems, symbol package synchronization is used to remove frequency-offset errors. In this strategy, short symbol packages are sent, during each symbol interval, frequency offset errors correspond to relatively small phase offset errors that can be made negligible. Standard carrier phase tracking algorithms operate as background tasks and ensure coherent signal recovery even though LO and incoming RF signals are not coherent. This approach becomes increasingly attractive as bandwidth increases and symbol and package time length decreases.

Data Capacity Estimates

We will assume modern high performance electronic and optical signal processing components are available operating at 10 GB/s. If moderate components are substitute, the same capacity gains will be achieved (by scaling), but the main thrust of this approach is to extract maximum data capacity from existing networks, so the employment of lower performance processing elements would seem to be counter productive to this goal.

The simplest vector constellation is QPSK. This immediately provides 2 Bits/s performance gain. We can therefore predict the improved data capacity to double the previous 10 GB/s limitation to 20 GB/s.

One variation on QPSK is DQPSK. This “differential” constellation offers significant relaxation on frequency and phase stability as each symbol is effectively measured against its predecessor. The penalty for this “robustness” is 3 dB degradation in ultimate SNR; however this deficit may be quite survivable given the high potential already afforded from LO enhanced optical amplification.

It would be desirable to consider higher order constellations however. For example, QAM16 would provide 4 Bits/Hz and predicts a data throughput potential of 80GB/z. Even larger constellations, such as QAM256 provide 8 Bits/Hz and an associated throughput potential of 160 GB/s.

Still higher level constellations may benefit from channel equalization, but as mentioned previously, the vector IQ demodulator allows full optical channel equalization i.e. both sides of the optical spectrum can be treated independently. This is not possible for a single photo detector approach, nor is it available for polarization diversity approaches.

A further serendipitous outcome is that OFDM modulation formats can also be processed. These formats have fundamental “built in” immunity from dispersive channel aberrations based on their multiple sub carrier approach. Each sub carrier could, in principle, carry QAM256 payloads and so the full fiber link capability can be exploited.

Physical Implementation Details For The Optical Vector Demodulator

Device Matching Strategy

A demonstration optical vector demodulator can be fabricated with standard “off the shelf” components. Whilst this would demonstrate “proof in principle”, an Integrated Circuit (IC) implementation would be required for a practical implementation.

The IC implementation is preferred as this ensures maximum possible component matching. The degree of DC offset cancellation, for example, would greatly exceed that possible from discrete components. Further, the long optical path lengths required with relatively large physical processing components would be difficult to maintain with good phase integrity. On an IC scale of dimension, these optical paths would be wavelength comparable and therefore fortuitously easier to manage.

The resulting IC could be mounted in a standard carrier package with separate optical LO and optical signal inputs. Although an internal laser diode could be incorporated, at least in principle, the thermal problems associated with its operation might prove problematic. Further, the silicon fabrication technologies might not be compatible with those required for other signal processing components such as photodiodes and transistors required for high speed electronic processing. For these, the use of Silicon-Germanium (SiGe) is probable, whilst laser diodes might be fabricated on a GaAs process.

Optical Hybrid Implementation

As shown, a number of optical hybrids are needed for photonic distribution. Specifically we need the following

180 Degree Optical Mixer Hybrid

The 180-degree hybrid should ideally introduce its phase inversion as a fundamental property, as opposed to merely adding a propagation time delay – this approach could be overly wavelength dependent and a “broad band” approach would be preferable.

The easiest way to generate is to use reflection. A wave undergoes a 180-degree phase inversion on reflection. The reference throughput path would be equivalent to zero degrees.

This diagram shows the geometric configuration required to realize an equal optical path 180 degree hybrid and a constant phase LO illumination hybrid. The hybrid loss will be 3dB for the optical input (50% “silvered mirror”) but the 0 degree LO hybrid can afford to be far less efficient. Additional LO power can be provided to overcome LO loss but optical input loss cannot be readily compensated.

The 180-degree hybrid will be implemented as a planar structure on the silicon IC substrate. The “overhead lighting” from the “Vertical Light Spreader” will provide equal illumination for each photo diode with equal optical path length and hence equal wavelength independent phase. The key advantage for this single balanced optical mixer is theoretical wavelength independence. The DC balance remains matched despite LO frequency and the phase and amplitude accuracy of the 180 degree input phase shift hybrid is equally wavelength independent.

The electronic baseband output paths will also be of equal length. The Trans Impedance Amplifiers (TIA) will terminate each photo diode in close proximity and an ultra fast difference amplifier will provide the final output (In practice this will be rotated 45 degrees for improved symmetry).

0 and 90 Degree Low Loss Optical Hybrids

The optical input signal is split equally upwards and downwards in this diagram using two 45-degree mirrors presenting a sharp intersecting wedge. The two split waves have equal phase and amplitude and travel towards each balanced mixer. Additional mirrors are placed symmetrically to ensure correct orientation and equal path length.

The LO input also uses this approach but introduces an additional time delay equivalent to 90 degrees phase shift at the center operating wavelength. This distribution can be implemented on a plane just above (or below) the dual diode balanced mixers and then distributed vertically to the PIN diode detectors. Unlike the other hybrids, the phase distribution accuracy of the 90-degree version is directly affected by wavelength. This may not be problematic however, as phase skew errors can be removed at IQ baseband (for example, cross coupling a portion of I into Q and vice versa can introduce an opposite phase skew error).

Summary

A wide band optical vector demodulator appears feasible; although implementation is possible with discrete components, and a prototype “proof in principle” intermediate step would be valuable, a practical usable implementation would best be fabricated on a silicon substrate using conventional die processing methods. The speed advantages afforded from SiGe technologies have been well exploited in recent times and almost every new high frequency components using Bipolar transistors adopts this fabrication. Whilst CMOS is not excluded, certain components, e.g. PIN photodiodes have to use structures similar to bipolar junction devices.

The laser LO source is probably best left as an external, technology specialized component. This may be phase locked to an accurate frequency reference or, in the wide band data cases, might well be adequately left to “free run”.

The use of equivalent matched photodiode pairs provides automatic DC balance and avoids the need for offset heterodyne architectures. Therefore narrow band, high sensitivity applications are enabled as readily as wide band, high data capacity complex modulation formats.

IC Layout Geometry

Conclusion

Appendix

Summary

© Ian Scott 2009