Abstract
A significant increase in traffic capacity and routing flexibility for optical free-space networks can be realized by employing WDM and all-optical add-drop multiplexing. Weight, power and cost could possibly be reduced since much of the data not intended for a given node can pass through that node transparently. Moreover, this parallel-wavelength approach allows for hybrid networks and increased variability in service type.

Introduction
There has been a significant resurgence in interest for free-space optical communications due to the potential for much higher bandwidth as compared to RF technologies. The majority of the R&D effort today surrounds the use of point-to-point free-space optical links. However, there might be significant advantages for the free-space community to follow the very successful example set by the optical fiber telecommunications community. From the 1970’s through today, fiber communications progressed as follows:

Employment of Optics
adopted point-to-point optical fiber links: The low loss and high bandwidth of optical fiber allowed much higher communication performance than coaxial cable.

Capacity Enhancement
adopted wavelength-division-multiplexed (WDM): With the invention of the broadband Erbium-doped fiber amplifier, the aggregation of many independent wavelengths on a single fiber was a cost-effective upgrade strategy for high-capacity links.

Flexible Routing
adopted wavelength-selective reconfigurable optical add-drop multiplexing (ROADM): Given that each node has a wavelength as a routing address, only data meant for that node is detected, whereas all the other traffic passes through transparently. Therefore, slower electronic data processing can be used in each node.
This progression took nearly 30 years to reach a critical commercial impact. The same progression may provide a substantial advantage for the free-space optical communications industry, and perhaps the steps could happen at a more rapid pace by leveraging existing technologies.
The potential network performance improvements provided by WDM in an optical satellite network have been investigated by Karafolas [1, 2] and Chan [3]. Multiple wavelengths representing independent lightpaths allow high-capacity data transfer as well as flexible routing (see Fig. 1). Each node can transparently pass, without detection, traffic destined for another node, thereby lowering electronic speed requirements and queuing times. These advantages translate to reduced weight, power and cost, the main concerns for any satellite network.
This paper highlights some of the physical-layer challenges that might arise as this vision of a future WDM network in free space evolves. Technical topics include: transparent multiplexing techniques, physical impairments, and wavelength and format conversion.

Figure 1: Optical Backbone - Accommodates multiple data rates and traffic types. Backbone consists of series of satellites (nodes).

Fig. 2. Hybrid Network Approach – Multiple modulation techniques operating on independent wavelength channels.

Reconfigurable WDM Add/Drop Multiplexers in Free-Space
The utilization of WDM might greatly enhance an optical satellite network. Dynamic routing can be performed by changing the wavelength lightpaths without modifying the physical topology. Moreover, a WDM network of independent parallel wavelengths enables hybrid network connectivity between diverse systems by allowing various modulation formats, data rates, and traffic types (e.g. circuit vs. packet). Figure 2 depicts a simple illustration of diverse data traffic. Each satellite or node can have the capability to modulate/demodulate any subset of these waveforms, while forwarding the remaining waveforms on to other nodes. Such a scenario allows each wavelength channel to be optimized for specific user applications.
As mentioned, one of the key enabling building blocks of today’s fiber-based optical networks is the WDM add/drop multiplexer (ADM), which alleviates the unnecessary detection and processing of channels destined for other nodes. In the configuration shown in Figure 3, the WDM input is demultiplexed into individual wavelength channels. Some of these channels are “dropped” and electrically or optically re-routed (e.g., RF or optical to ground links) or processed onboard. Simultaneously, additional channels are “added” prior to multiplexing with the “through traffic” at the output of the ADM. To ensure network flexibility, it is probably desirable that the ADM be reconfigurable. It should be noted that a free-space ADM system, as opposed to a fiber-based one, might be required to account for the fact that satellite nodes and aggregated users could be constantly moving relative to each other. For example, an aircraft “user” might require a hand-off between two different satellites, thereby requiring a unique wavelength allocation solution.
It is quite possible that the internal routing technology of the satellite would mimic much of what has already been developed by the fiber telecommunications industry. One approach is to leverage existing technologies for the free-space network. However, tighter design specifications may be required to support harsh environmental conditions experienced during launch and orbit, increased power levels to support highly-lossy free-space transmission and novel modulation formats.

Fig. 3. Optical Add/Drop Mux – Key enabler for an efficient optical network. [4]

Physical Impairments
Extension of free-space single-wavelength systems to WDM architectures will probably result in a significant increase in overall transmitted and received power levels. This linear increase in power with the number of channels may have implications of nonlinear effects and physical device limitations within the node itself. Non-negligible power penalties (>1-dB) can arise for high power levels (10 watts per channel) when as few as 4 channels are utilized and propagate over only a few meters of single-mode fiber inside a satellite. Moreover, inter-satellite Doppler shifts might limit the wavelength channel spacing and thus the total number of channels that can be accommodated within the EDFA gain bandwidth [2].
Another physical-impairment issue that might arise in free-space networks involves recovering phase-based modulation formats. Free-space networks could employ either direct detection or coherent detection techniques, depending on the sensitivity limitations and spectral efficiency requirements. Since the vast majority of fiber-based systems are direct detection, free-space coherent systems might encounter unique problems, especially when both the phase and amplitude are used to encode the data. Therefore, it might be necessary to maintain a high-level of phase coherency throughout the system, including passage through the atmosphere, the transmit and receive optics, any wavelength converters and the ROADMs. For example, Fig. 4 shows the system penalty experienced in the presence of a fixed phase offset in the receiver for various coherent waveforms of current interest, including BPSK, QPSK, 8PSK and 16QAM [5]. Extension of coherent signal constellations beyond that of QPSK might require more stringent device specifications, including amplitude distortions and phase tracking.

Fig. 4. Power Penalty due to fixed phase offset for coherent waveforms. Degradation much higher for 8PSK and 16QAM than BPSK and QPSK.

Transparent Conversion and Reconfigurable Routing
In general, it stands to reason that optoelectronic detection and retransmission could be inefficient and that an optically transparent routing interface could be desirable. Therefore, given a reconfigurable and/or hybrid network, there may be a need for optical data format and/or wavelength conversion.

Fig. 5. Critical issues in wavelength conversion (fiber vs. free-space).

Wavelength Conversion
A key motivating factor for all-optical wavelength conversion is the ability to perform dynamic wavelength-based reconfigurable routing and switching. Wavelength conversion enables contention resolution, reduced buffering times, up/down-link and crosslink interface optimization, and multicasting [6]. Specifically, multicasting on different wavelengths can enable data to more assuredly reach its destination through different wavelength-dependent satellite paths.
The critical performance specifications for wavelength converters in free-space systems might be somewhat different from those of fiber-based systems, as shown in Fig. 5. The unique problems in fiber systems tend to be the chirp, polarization dependence and dynamic range. In lossy free-space systems, the DC power consumption, conversion efficiency and block shifting abilities might be of more concern. For nearly all wavelength converters, speed limitations, format dependence and additive noise are potentially important. For any phase-based modulation format, phase preservation through the wavelength converter would be critical [7].

Data Format Conversion
For maximum efficiency and flexibility, the optical backbone might employ different modulation formats for different types of connecting links, thereby raising the possibility of the need for some means of all-optical format conversion (e.g., NRZ to PPM, DPSK to OOK). For example, many existing free-space satellite systems employ PPM (pulse-position-modulation). In order to route an OOK (on-off-keying) signal through a PPM-designed satellite, the signal might first be converted to PPM and then reconverted back to OOK at the destination. All optical format conversion might reduce the need for detection and re-modulation, thus reducing the load for onboard processors.

Fig. 6. Address comparison for a PPM signal using an ultra-fast nonlinear interferometer.

Reconfigurable Routing by Reading Headers
The management and control of lightpaths and the network’s logical topology might be crucial to the success of a reconfigurable network. For a network encompassing vast distances, it may be difficult to isolate lightpath failures or sources of degradation, making communication between nodes even more important. Also, when lightpaths are setup between satellites of different orbital distances (i.e., GEO vs. LEO), the handoff of lightpaths could occur since a satellite at a shorter orbital distance might leave the field-of-view of a given satellite at a higher orbital distance.
If rapid reconfiguration is required in, for example, a packet-switched network, there could be a need for rapidly reading the destination information from the packet’s header. Headers could be subcarriers, separate wavelengths, or simply data bits at the beginning of a packet, as is traditionally done. Since electronic header processing of front-loaded header bits may not be feasible at the extremely high data rates envisioned for future systems, there is interest in all-optical header recognition approaches.
Optics can perform a few operations very rapidly, whereas electronics can do many operations, but slower. Since it is envisioned that orbital optical networks will have much fewer addresses than a typical fiber network, rapid optical processing of only a few key routing bit sequences may be an attractive future choice. Optical header processing implementations could identify, generate and even replace headers in the optical domain without disturbing the payload. As an example, Fig. 6 shows one method by Hamilton, et. al. [8] for performing packet-level, system synchronization and address comparison of a PPM signal. In this technique, a semiconductor ultra-fast nonlinear interferometer (UNI) acting as a Boolean XOR gate is used to achieve four-bit address comparison.

Acknowledgements
We thank DARPA for their generous support. We would also like to thank the following people for insightful discussions: Drs. Tony Acampora, Vincent Chan, Scott Hamilton, Steve Pappert, and Joe Touch.

References
[1] N. Karafolas, S. Baroni, IEEE LEOS Annual Meeting, vol. 1, pp 100-101, 2000.
[2] N. Karafolas, S. Baroni, IEEE/OSA J. of Lightwave Tech., vol. 18, pp 1792-1806, 2000.
[3] V. Chan, IEEE/OSA J. of Lightwave Tech., vol. 21, no. 11,pp 2811-2827, 2003.
[4] Modified from Tingye Li, Private communication.
[5] L.C. Christen, et. al., to be submitted for publication.
[6] Y. Wang, et. al., IEEE/OSA J. of Lightwave Tech., Special Issue on Optical Networks, vol. 23, 2005.
[7] R. Ramaswami and K. Sivarajan, Optical Networks, Morgan Kaufmann Publishers, 1998.
[8] S.A. Hamilton, et. al., IEEE Photonics Tech. Lett., vol. 14, pp 209-211, 2002.