GNAT_T9_r3.pdf

Global Network Access Terminal as a Multi-

Modem Solution for Network Convergence

James C. Lyke1, Khanh Pham2, Christian A. Peters Air Force Research Laboratory, Albuquerque,NM 87117, USA

Donald C. Fronterhouse PnP Innovations, Albuquerque, NM 87117, USA

In this paper, we discuss the architecture of a global network access terminal (GNAT) as an information-convergence hub that exists within platforms that participate in such a mobile ad hoc network. The architecture allows for multiple modems to be combined with an intelligent router that brokers/manages the flow of information to and from the platform, whether in the role of data consumer/producer or as an opportunistic relay to improve the performance of an overall distributed network. As a demonstration problem, we discuss a special two-GNAT case, which is to achieve a connection to the “regular” internet, brokered by two GNATs, one located with a user (remote location) and one placed within the internet itself (this is called a bonding server). These GNATs broker an internet connection through multiple modems, each connected to a different service or connection approach. Each of these is considered a soft flow. It is possible to use military and commercial SATCOM as soft flows, but it is also possible to use laser communications links, other

MANETs, and practically any other method so long as both GNATs can exploit the given pathway

(soft flow). The GNATs establish a mutual control plane whose information is superimposed on user (data plane) traffic. The specialized use case of internet access with two GNATs can be generalized to multiple nodes, and this paper discusses a number of possibilities and challenges.

I. Introduction

Based on the current trends of increased computational power (Moore’s Law [1]) and communications, it should not be surprising extrapolation to expect that someday all systems will be connected, pervasively and seamlessly. One author [2] observed that 99% of the information flow across continents is performed through undersea cables; hence is natural to consider the spacecraft networks should play an increasing role in the aspiration for global connectivity.

This realization may in part explain the renewed (Teledesics “plowed this ground” before [3]) and frenetic interest in erecting mass constellations of platforms in the name of Internet distribution worldwide, particularly to the significantly disadvantaged “other 3 billion” (O3B being both an acronym [4] and the name of one of the companies engaged in this pursuit [5]). At the same time, movement is just one of the operations that can be performed on information: there is processing and storage. Moore’s Law gives us plenty of both. Each of the billions of smart phones in existence bear witness to the fact that phones are not merely for communications; they store and re-render information with real-time processing in countless ways to satisfy our needs and whims. They also generate information using arrays of imaging and acoustic sensors. They are in virtually all respects information convergence devices.

Why would we expect less of our leading-edge aerospace systems?

In this paper we consider a net centric view of the modern aerospace system. Through the concept we call the global network access terminal (GNAT), we present the template of an information convergence device. It serves as a nexus for potentially many communications devices, but it also features hooks for internal storage and processing as well as sensors. It can produce, consume, and relay information to peers having the same construction and protocols.

1 Program Manager, AFRL/RVBY, Associate Fellow.

2 Program Manager, AFRL/RVBY, Associate Fellow.

We also believe many things get better with scale, and we discuss how random graphs may be a better model for global network than one based on power laws. Since scale does not happen overnight, we discuss a graduated four-stage approach that we are considering in our own work on creating a strategy and technology basis for “ambient connectivity”.

II. Global Network Access Terminal (GNAT)

If we take apart a modern smart phone, we will find many communication devices, each supporting different physical layers. In addition to the primary radio that provides critical voice and data access through cellular network powers, they normally contain also 802.11 (Wi-Fi) support, global navigation satellite system (GNSS) receivers, Bluetooth, near field communications, as well as optical (infrared) and acoustic transceivers. Smart phones contain a rich variety of processing resources, most likely to include a tightly coupled group of traditional processing cores, an array of graphic processing units (GPUs) and dedicated digital signal processing (DSP) facilities for efficiently managing pipelined algorithms for servicing communications functions. The amounts of storage in even the simplest smart phone is measured in tens of gigabytes, and the most contemporary models (at the time of this writing) approaching the level of a terabyte (eight trillion bits) of nonvolatile flash-based storage, with trifling (but still respectable amounts) of random access memory (RAM) and cache storage. Imaging camera sensors usually adorn the front and back surfaces, each capable of capturing full motion video at impressive resolution.

Figure 1a presents an interpretation of a contemporary smart phone. It is a reasonable schematic of an information convergence device. We can consider an analogous model, slightly re-factored, of an information convergence device for aerospace system (Figure 1b). We refer to this version of an information convergence device as a global network access terminal (GNAT).

Fig. 1 Information convergence devices. (a) Interpretation of smart phone. (b) Global network access terminal (GNAT).

The GNAT can generally be thought of as a hub and spoke architecture, analogous to the so-called “Layer 2” (e.g.

switched Ethernet) network. The GNAT would be like a “smart phone” for an aerospace platform (such as a spacecraft, munition, or user ground terminal). It is a modular construct in that many of the blocks can be combined, omitted, expanded, re-arranged. For this reason, in our work, we have come to refer to any variant of Figure 1b on any platform as a GNAT. A GNAT could be a single radio with a host interface or a bank of transponders and/or intersatellite crosslinks with multiple imaging sensors and banks of processors (as in a flying data center). They are each still a GNAT.

We next review the key elements of the GNAT concept. The ensemble of these components is sometimes referred to as an “enclave”.

A. Physical layer communications devices.

Communication devices transmit information, receive information, or do both (operate as full-duplex transceivers).

They are referred to as “PHYS”, or physical layer devices that they generate information on a usually free-space channel. PHYS can be based on radio frequency or optical technology.

B. Router core switch and GNAT Control Processor.

The GNAT Control Processor (GCP) manages the action of the router core (as well as executing algorithms centrally relevant to GNAT operation). The router core is a switch that handles primitive data forwarding transactions

(or flows) from one port to another, using a database of rules managed by the GCP. This arrangement is very analogous to that used in software defined network (SDN) concepts, such as Open V-Switch (OVS), which employs a similar arrangement of control and data forwarding planes [6].

C. Host interface unit.

For the consumer smartphone, the “host interface unit” is the consumer. For other systems, it is necessary to create an electrical interface that allows the GNAT enclave to be effectively integrated with the hosting platform. This usually requires access to physical power, primary command-and-control interfaces, and sometimes other interfaces that may be idiosyncratic to the hosting platform, to include safety interlocks, auxiliary reset discrete cables, etc. The host interface unit would accommodate these through electronics, including state machines, computers, and software necessary to provide a degree of adaptation necessary for compatibility with the host platform.

D. Optional processing/storage facility.

GNATs can include groups (even banks) of processors and storage devices. For efficiency, as in the case of smart phones, heterogeneous blend of computing types (e.g. general-purpose multicore, GPU, DSP, and field programmable gate array) would be involved in any nontrivial computation facility. Ideally this would support composing computing styles analogous to those in cloud frameworks, especially for the cases where computation might be conceivably distributed across multiple platforms.

E. High-Performance imaging (and / or other) Sensors.

Typical satellite platforms, when they do have high-performance sensors, often downlink sensor data to the ground through dedicated communications devices [7]. In the GNAT approach, this option is still possible, but it is also possible to perform onboard processing and storage functions. Highly processed sensor data can be provided in theory in a form directly consumable to users through one of the communications PHYs.

F. Networking protocols.

We can think of the routing facility of the GNAT, consisting of the GCP and router, as being analogous to a bare metal computer, having latent capability but being ineffective without a “stack” consisting of an operating system and application framework. A GNAT requires network protocols and operating principles, which include resource discovery (i.e., what PHYs and optional devices are plugged in and how to exploit them), peer discovery (other

GNATs) and a set of network concepts that manage the movement of information within and between GNATs.

III. Scaling Considerations: Why we call it a GNAT

We expect most implementations of the GNAT concept involve communications devices for spacecraft that operate under an egalitarian principle, meaning for the most part, that GNATs have similar networking facilities, namely several PHYs and a GCP. The rationale is based on simple observations of network science.

Albert-Laszlo Barabasi’s Linked [8] describes three network graph models: the small-world graph, the power law graph, and the random graph. The “small-world” model is popularly described as the “six degrees of separation” concept, by which any individual on the planet is separated by a social chain of six individuals on average. Power law graphs (which Barabasi refers to as corresponding to a “scale-free” model) refer to network structures having a hierarchy of large nodes connecting to many smaller nodes, these in turn connecting to yet smaller nodes (specifically, the “large-ness” of a node refers to the number of neighbors, also referred to as the degree of the node). Common examples include the structure of airports and biological circulatory systems. Random graphs refer to networks where the size of nodes and structure of the neighborhood of nodes is randomly distributed. We depict notionally the difference between the scale-free and random graph model in Figure 2.

Fig. 2. Notional depiction of graph models. (a) Power-law (scale-free) network, dominated by several large nodes. (b) Random graph model with no single dominant node.

The choice of network model – when choice is possible – has profound impacts on the resiliency of the network.

For example, a power law graph has fragile connectivity. As a thought exercise, consider the susceptibility of air travel in the United States, particularly in bad weather where the closure of major hubs can cripple air traffic. If the airport grid was built using a random graph model, however, the resulting network would be far more robust. Airports would be more egalitarian, for example, Chicago O’Hare would be analogous to the Fayetteville, Arkansas airport. Many airports would have to be closed before an appreciable impact would be experienced in air travel. As Barabasi observed, in random graphs where above a certain (percolation) threshold, nodes are fully connected with high probability), even losing a majority of nodes may not appreciably impact the ability of the network to function.

It is our view that to achieve the condition abandoned connectivity, communication will necessarily aspire to the random graph model. In such an approach, communication nodes seek connection to as many other platforms as possible. Such an approach works better with density, i.e., “the more the merrier”. To succeed, it is necessary to establish that the totality of nodes results in the density that exceeds the percolation threshold, the above this threshold, but once above this threshold the addition of each new platform enriches the overall grid and reinforces resiliency throughout. In the limit argument, every platform takes on the aspect of egalitarian node in a random graph.

Metaphorically, the situation is similar to the experience of jogging on the trail through a swarm of gnats. The experience is miserable in that, no matter how hard one tries to swat the gnats, there are too many and the gesture is futile. This is the condition we strive for in communications networks in general, but especially for the war fighter.

We want to network to be so dense, so resilient, that no adversary will bother to try to break it.

IV. Stages of Progression in GNAT Development Towards Ambient Connectivity

We consider the progression toward the ambient connectivity as being a four stage evolution, depicted in Fig. 3.

Fig. 3 Four-stage evolutionary roadmap to ambient connectivity. (a) Stage 0 – point-to-point connectivity through multiple paths. (b) Stage 1 – mono-domain crosslink network. (c) Stage 2 – intermingled Stage 1 networks. (d) Scaled, dynamic networks on demand – ambient connectivity.

Stage 0 - point-to-point connectivity. The most basic network involves two nodes. In Fig. 3a, we depict the two-node problem as a user connecting to the Internet using a number of pathways. Formally, each path is an edge, denoting a multi-degree (redundant) connection between the two nodes (user and a particular site on the Internet).

Functionally, however we wish to think of the connection is being a single, robust connection. This robust connection is virtualized from potentially many parallel possibilities. Each possibility we refer to as a “soft flow”, and the single virtual connection is formed through a process we refer to as “channel bonding”. In principle, a soft flow may be formed in many ways. It could be a tunnel through the Internet itself using any commodity approach (an 802.11 Wifi connection, a cellular telephone connection, etc). It could be through connections that do not involve the Internet, such as ad hoc networks (including military and commercial satellite communication (SATCOM) networks) and direct beams (radio-frequency or free space optical). The paths are split or aggregated in each direction by a GNAT , located at each node.

Stage 1 - single domain network connectivity. When the Stage 0 network is extended to a larger number of reach-able nodes, and the network traffic is considered to be within the same security domain, we consider the extended topology to be a Stage I network (Fig. 3b). In space, a collection of intersatellite crosslinks and ground stations constitutes an example of a Stage 1 network. While the distinction is subjective, we consider Stage 1 networks to be semipermanent and well-defined from a logical perspective (even though the actual physical network may undergo turbulent changes, as links appear and disappear), as opposed to the more dynamic, temporary short-term on-demand networks.

Stage 2 – multi-domain networks. We consider the Stage 2 network to simply be equivalent to several Stage 1 networks, where the links are comingled, but the networks remain separated through cryptographic methods. In Fig.

3c, for example, the triangle node and square nodes are physically interconnected. However, there are effectively two logically separated networks, so that the triangle nodes cannot effectively reach square nodes. This is analogous to the virtual local area network (VLAN) [9] concept, though in principle each separate network need not behave as the same network type. The concepts of software defined networks (SDN) and network function virtualization (NFV) [10] apply to these networks, and the notion of control planes and management planes become necessary. Various provisioning functions are necessary to establish virtual topologies and manage member relationships, etc. In other respects, the semipermanent nature of the Stage I networks is still present (i.e., the networks are temporally stable, with slowly changing logical topologies relative to the tempo of operation).

Stage 3 – Ambient connectivity. We consider the Stage 3 network (Fig. 3d) to be a sort of limit argument of the

Stage 2 concept. Here, we relax the constraints of well-defined long-term networks, and permit that networks can be fluidly created on demand. Networks involving thousands of users might exist for a few minutes, or could have a small number of nodes for more than a decade. We consider the actual network as that defined by the momentary

(and dynamically) reachable nodes (contact graph) of all modes to be a substrate. The virtual networks are then overlays embedded on the substrate. In such networks, given the high degree of dynamic system that could be possible, special considerations must be given on the notions of overall network provisioning. Concerns of cyber security are also more important in these networks, since the scale factors make it more difficult to understand what represents a well-behaved network from a malicious network.

V. Taking a First Step – Stage 0

In this section, we describe aspects affecting implementation of a particular Stage 0 use case, namely one of

Internet access. This use case involves two endpoints: A and B. Endpoint A is a remote user having no direct access to the Internet. Endpoint B is an Internet-based connection to a website server. Endpoint A is the remote user “A-side” GNAT, and endpoint B is referred to as a bonding server (“B-side” GNAT).

We consider the connection between two points (A and B) in Fig. 4a as a connection mediated by a modem. This is evolved to form the Stage 0 connection. In principle, we can employ multiple modems (Fig. 4b) in parallel that improve the throughput and/or reliability compared to a single modem. In this case, we employ a GNAT at each end of the connection. It manages each modem (m) using algorithms in the GNAT control processor (GCP) (p). The GCP intercepts and synthesizes traffic for the endpoints, splitting and combining information involving the soft flows, which are the specialized data streams sent through each participating modem. The router (r) manages each flow according to rules sets defined by the GCP. The GCP can encrypt the data, partition it, incorporate redundancy, add control plane and quality of service data, resulting in the traffic which is placed in each soft flow. GCPs at each end establish a control plane dialogue to dynamically adjust soft flows based on throughput, latency, and availability.

Fig. 4 Achieving a Stage 0 virtualized wire. (a) Single modem connects A and B. (b) Multi-modem connection.

There are many possibilities for achieving the soft flow, as suggested in Fig. 5. The most common connection type would be an “enterprise” service (Fig. 5a). Representative examples include ordinary 802.11-based WiFi connections (usually hotspots that are connected to other services), cellular telephony networks, and satellite communications (SATCOM) concepts. In most cases, these services have Internet accessible gateways. In the special case where node B is itself on the internet (Fig. 5b), the B-side GNAT does not require a specialized modem for soft flows of this type, but rather can access the SATCOM service through a gateway Internet connection (reflected by the red-dashed line). Many military SATCOM services will not provide general access through their gateway (if such a gateway exists), in which case the B-side GNAT would also require the special modem needed to access the SATCOM service (Fig. 5a).

Fig. 5 Depiction of the diversity of possible soft flows. (a) Black-box enterprise service without Internet gateway. (b) Case where Internet gateway exists. (c) Point-to-point connection. (e) Mobile ad hoc network. (e)

Delay-tolerant network (case where pathway is sporadically available).

The soft flow can be a primitive direct channel, as suggested in (Fig. 5c). Here, the channel can employ a simpler communications protocol, which may not be network based on all but could employ simple framing conventions that do not support the notion of an address space. Another possibility for soft flows would be based on mobile ad hoc networks (MANET, Fig. 5d). In this case the modems participating in the soft flow appear as edge nodes to the

MANET. Finally, a discontinuous delay tolerant network could be used in some cases where the service would permit it (for example, simpler text messages or file transfers). This possibility suggested in Fig.5e.

In implementing the Stage 0 concept, it is necessary to consider a number of factors:

• Disparities in latency. SATCOM services may be low earth orbit based or geostationary, and the use of

MANET and delay-tolerant schemes can have profound impacts on the fundamental efficacy of the

Stage 0 concept for certain applications.

• Flow separation. If Stage 0 itself is used to serve as a hotspot, it will be necessary to groom multiple clients and servers into a single flow. Moreover, other types of service, from text and file transfer, to voice and video may be required. If the soft flows are also different in quality of service, the GCP could conceivably separate flows to more closely align with the mixture of applications.

• Interoperability. There are a set of protocols required between GNATs (GNAT-GNAT), between

GNATs and PHYs (GNAT-PHY), etc. A concern is over vendor neutral schemes, where PHYs and

GNATs can be mixed and are not limited by proprietary interfaces. As such, certain open architecture strategies make sense, but the details for where proprietary and open boundaries remain to be resolved.

• Disparities in bandwidth. Some services have significantly higher bandwidths. Mixing “fat” pipes and

“thin” pipes require potentially sophisticated policies and in some cases it may make sense to not always blend every flow even when possible to simplify implementation.

• Disparities in availability. Some services are not available due to geometric and contractual constraints.

• Provisioning of modems. Most services have some complexities relating to configuring modems to access leased bandwidth pools.

• Relationship to a network operating center. While the formulation of a single stage instance can be performed without the notion of a network operating center (NOC), the situation will likely change with scale factor, especially when each of potentially thousands of Stage 0 instances may involve heterogeneous and unique combinations of soft flows. This is particularly true with leased pools, where the amount of leased bandwidth may need to be rationed across a group of users.

• Security of information on channel. It will be necessarily in general to separate security domains, and protect user information (e.g., keys, content, etc). It is necessary to consider how to deal with terminal compromise when a GNAT installation falls under control or capture by an adversary

• Resiliency to cybersecurity. All manner of cybersecurity attacks must be addressed, which is generally undecideable [11] in the computational complexity sense (i.e., is not possible to guarantee that all attack surfaces are known and protectable) and requires expansion and elaboration on traditional approaches.

• Covertness/traffic obscurity. The ability to harness any or all path connections, while desirable, may have undesired side effects, such as the profile of emissions in communications and the ability to glean insights from the patterns of communications. It is desirable to when possible maintain a low probability of detection/intercept in communications, and it is conceivable that specialty modems or software defined / cognitive radios can be employed to reduce the probability of detection. As such, the

GCP and GNAT operating policies might offer “sliders” for the degree of exposure acceptable in particular operating scenarios.

These factors have become considerations in our on-going research. At the time of this writing, the Air Force

Research Laboratory (AFRL) is working towards a Stage 0 implementation based on using commercial SATCOM services. The development and demonstration effort is referred to as “Topic 9”, a name based on the topic number of a Small Business Innovative Research (SBIR) solicitation [12]. Several of the competing vendors are participating in the present formulation, and the work is being further augmented with other contract and AFRL in-house research activities. It is hoped that a demonstration of these concepts will be completed by mid-year 2020.

As a part of the research, the team is exploring the addition of software-defined radio and reconfigurable antenna concepts. These approaches could reduce the size, weight, and power of fielded equipment by allowing modems and antennas to be re-purposed through software-only changes. In the limit argument, a single sufficiently flexible reconfigurable RF system could implement any existing modem, hence allowing a substantial collapse in the logistics complexity of fielding radio equipment.

VI. Conclusion

Ambient connectivity is the eventual expected end state of all communications technology, whether consumer, industrial, or military. Despite tremendous investments by the military in communications technology (especially

SATCOM), the connectivity of aerospace systems still has significant room for improvement. We believe a self-forming network, in which all platforms participate, is the best way to achieve this vision, as we eventually approximate the robustness of random graph connectivity. Even if this future is obtainable, it may be that nodes want to have the ability to opt in or out based on operational need. We discussed an evolutionary four-stage roadmap to this ambient connectivity concept, highlighting the first part (Stage 0) as mostly being a study in virtualizing point-to-point connectivity through multiple channels (soft flows). Achieving effective formulations will be necessary in these base cases first before scaling the concepts to greater numbers and greater sophistication (i.e., Stages 1-3).

References

[1] G. E. Moore, "Cramming more components onto integrated circuits," Electronics, vol. 38, no.8, pp. 114-117, 19 April 1965.

[2] Starosielski, Nicole. The Undersea Network (Sign, Storage, Transmission), Duke University Press Books, April 1, 2015.

[3] de Selding, Peter B. "Teledesic Plays Its Last Card, Leaves the Game". Space News, July 14, 2003..

[4] “Google and Others Back Satellite Internet Project, Aimed at Connecting 3 Billion People”, Wall Street Journal, September 9, 2008.

[5] “Satellites may connect the entire world to the internet”, The Economist, December 8, 2018.

[6] Open V-Switch web site, https:// https://www.openvswitch.org/ (accessed May 15, 2019).

[7] Singer, J. "USAF To Experiment With Satellite To Improve Ground Communications", Defense News, December 4, 2006.

[8] Barabasi, A. Linked: The New Science of Networks, Perseus Books Group; 1st edition, May 15, 2002.

[9] Tanenbaum, A, Computer Networks, Pearson Education International, New Jersey, 2003.

[10] D. Kreutz, F. M. V. Ramos, P. E. Veríssimo, C. E. Rothenberg, S. Azodolmolky and S. Uhlig, "Software-Defined Networking:

A Comprehensive Survey," in Proceedings of the IEEE, vol. 103, no. 1, pp. 14-76, Jan. 2015.

[11] M. Harrison, W. Ruzzo, and J. Ullman, "Protection in Operating Systems," Communications of the ACM, vol. 19, no. 8, pp.

461-471, August 1976.

[12] “Applications Using New Satellite Communications Constellations”, US Air Force Small Business Innovative Research Topic

AF172-009, (https://www.sbir.gov/sbirsearch/detail/1254397, accessed May 2019) https://www.openvswitch.org/ https://www.sbir.gov/sbirsearch/detail/1254397